Merge branch 'master' into fix-temp-table-drop

.github/workflows/backport_branches.yml

@@ -359,15 +359,11 @@ jobs:
     steps:
       - name: Set envs
         run: |
-          DEPENDENCIES=$(cat << 'EOF' | jq '. | length'
-          ${{ toJSON(needs) }}
-          EOF
-          )
-          echo "DEPENDENCIES=$DEPENDENCIES" >> "$GITHUB_ENV"
           cat >> "$GITHUB_ENV" << 'EOF'
           CHECK_NAME=ClickHouse build check (actions)
           REPORTS_PATH=${{runner.temp}}/reports_dir
           TEMP_PATH=${{runner.temp}}/report_check
+          NEEDS_DATA_PATH=${{runner.temp}}/needs.json
           EOF
       - name: Download json reports
         uses: actions/download-artifact@v2
@@ -382,8 +378,11 @@ jobs:
         run: |
           sudo rm -fr "$TEMP_PATH"
           mkdir -p "$TEMP_PATH"
+          cat > "$NEEDS_DATA_PATH" << 'EOF'
+          ${{ toJSON(needs) }}
+          EOF
           cd "$GITHUB_WORKSPACE/tests/ci"
-          python3 build_report_check.py "$CHECK_NAME" "$DEPENDENCIES"
+          python3 build_report_check.py "$CHECK_NAME"
       - name: Cleanup
         if: always()
         run: |

.github/workflows/jepsen.yml

@@ -7,11 +7,8 @@ concurrency:
 on: # yamllint disable-line rule:truthy
   schedule:
     - cron: '0 */6 * * *'
-  workflow_run:
-    workflows: ["PullRequestCI"]
-    types:
-      - completed
   workflow_dispatch:
+  workflow_call:
 jobs:
   KeeperJepsenRelease:
     runs-on: [self-hosted, style-checker]

.github/workflows/master.yml

@@ -970,16 +970,12 @@ jobs:
     steps:
       - name: Set envs
         run: |
-          DEPENDENCIES=$(cat << 'EOF' | jq '. | length'
-          ${{ toJSON(needs) }}
-          EOF
-          )
-          echo "DEPENDENCIES=$DEPENDENCIES" >> "$GITHUB_ENV"
           cat >> "$GITHUB_ENV" << 'EOF'
           CHECK_NAME=ClickHouse build check (actions)
           REPORTS_PATH=${{runner.temp}}/reports_dir
           REPORTS_PATH=${{runner.temp}}/reports_dir
           TEMP_PATH=${{runner.temp}}/report_check
+          NEEDS_DATA_PATH=${{runner.temp}}/needs.json
           EOF
       - name: Download json reports
         uses: actions/download-artifact@v2
@@ -994,8 +990,11 @@ jobs:
         run: |
           sudo rm -fr "$TEMP_PATH"
           mkdir -p "$TEMP_PATH"
+          cat > "$NEEDS_DATA_PATH" << 'EOF'
+          ${{ toJSON(needs) }}
+          EOF
           cd "$GITHUB_WORKSPACE/tests/ci"
-          python3 build_report_check.py "$CHECK_NAME" "$DEPENDENCIES"
+          python3 build_report_check.py "$CHECK_NAME"
       - name: Cleanup
         if: always()
         run: |
@@ -1018,15 +1017,11 @@ jobs:
     steps:
       - name: Set envs
         run: |
-          DEPENDENCIES=$(cat << 'EOF' | jq '. | length'
-          ${{ toJSON(needs) }}
-          EOF
-          )
-          echo "DEPENDENCIES=$DEPENDENCIES" >> "$GITHUB_ENV"
           cat >> "$GITHUB_ENV" << 'EOF'
           TEMP_PATH=${{runner.temp}}/report_check
           REPORTS_PATH=${{runner.temp}}/reports_dir
           CHECK_NAME=ClickHouse special build check (actions)
+          NEEDS_DATA_PATH=${{runner.temp}}/needs.json
           EOF
       - name: Download json reports
         uses: actions/download-artifact@v2
@@ -1041,8 +1036,11 @@ jobs:
         run: |
           sudo rm -fr "$TEMP_PATH"
           mkdir -p "$TEMP_PATH"
+          cat > "$NEEDS_DATA_PATH" << 'EOF'
+          ${{ toJSON(needs) }}
+          EOF
           cd "$GITHUB_WORKSPACE/tests/ci"
-          python3 build_report_check.py "$CHECK_NAME" "$DEPENDENCIES"
+          python3 build_report_check.py "$CHECK_NAME"
       - name: Cleanup
         if: always()
         run: |

.github/workflows/pull_request.yml

@@ -1025,15 +1025,11 @@ jobs:
     steps:
       - name: Set envs
         run: |
-          DEPENDENCIES=$(cat << 'EOF' | jq '. | length'
-          ${{ toJSON(needs) }}
-          EOF
-          )
-          echo "DEPENDENCIES=$DEPENDENCIES" >> "$GITHUB_ENV"
           cat >> "$GITHUB_ENV" << 'EOF'
           CHECK_NAME=ClickHouse build check (actions)
           REPORTS_PATH=${{runner.temp}}/reports_dir
           TEMP_PATH=${{runner.temp}}/report_check
+          NEEDS_DATA_PATH=${{runner.temp}}/needs.json
           EOF
       - name: Download json reports
         uses: actions/download-artifact@v2
@@ -1048,8 +1044,11 @@ jobs:
         run: |
           sudo rm -fr "$TEMP_PATH"
           mkdir -p "$TEMP_PATH"
+          cat > "$NEEDS_DATA_PATH" << 'EOF'
+          ${{ toJSON(needs) }}
+          EOF
           cd "$GITHUB_WORKSPACE/tests/ci"
-          python3 build_report_check.py "$CHECK_NAME" "$DEPENDENCIES"
+          python3 build_report_check.py "$CHECK_NAME"
       - name: Cleanup
         if: always()
         run: |
@@ -1073,15 +1072,11 @@ jobs:
     steps:
       - name: Set envs
         run: |
-          DEPENDENCIES=$(cat << 'EOF' | jq '. | length'
-          ${{ toJSON(needs) }}
-          EOF
-          )
-          echo "DEPENDENCIES=$DEPENDENCIES" >> "$GITHUB_ENV"
           cat >> "$GITHUB_ENV" << 'EOF'
           TEMP_PATH=${{runner.temp}}/report_check
           REPORTS_PATH=${{runner.temp}}/reports_dir
           CHECK_NAME=ClickHouse special build check (actions)
+          NEEDS_DATA_PATH=${{runner.temp}}/needs.json
           EOF
       - name: Download json reports
         uses: actions/download-artifact@v2
@@ -1096,8 +1091,11 @@ jobs:
         run: |
           sudo rm -fr "$TEMP_PATH"
           mkdir -p "$TEMP_PATH"
+          cat > "$NEEDS_DATA_PATH" << 'EOF'
+          ${{ toJSON(needs) }}
+          EOF
           cd "$GITHUB_WORKSPACE/tests/ci"
-          python3 build_report_check.py "$CHECK_NAME" "$DEPENDENCIES"
+          python3 build_report_check.py "$CHECK_NAME"
       - name: Cleanup
         if: always()
         run: |
@@ -3272,6 +3270,13 @@ jobs:
           # shellcheck disable=SC2046
           docker rm -f $(docker ps -a -q) ||:
           sudo rm -fr "$TEMP_PATH"
+#############################################################################################
+###################################### JEPSEN TESTS #########################################
+#############################################################################################
+  Jepsen:
+    needs: [BuilderBinRelease]
+    uses: ./.github/workflows/jepsen.yml
+
   FinishCheck:
     needs:
       - StyleCheck
@@ -3336,6 +3341,7 @@ jobs:
       - SplitBuildSmokeTest
       - CompatibilityCheck
       - IntegrationTestsFlakyCheck
+      - Jepsen
     runs-on: [self-hosted, style-checker]
     steps:
       - name: Clear repository

.github/workflows/release_branches.yml

@@ -442,16 +442,12 @@ jobs:
     steps:
       - name: Set envs
         run: |
-          DEPENDENCIES=$(cat << 'EOF' | jq '. | length'
-          ${{ toJSON(needs) }}
-          EOF
-          )
-          echo "DEPENDENCIES=$DEPENDENCIES" >> "$GITHUB_ENV"
           cat >> "$GITHUB_ENV" << 'EOF'
           CHECK_NAME=ClickHouse build check (actions)
           REPORTS_PATH=${{runner.temp}}/reports_dir
           REPORTS_PATH=${{runner.temp}}/reports_dir
           TEMP_PATH=${{runner.temp}}/report_check
+          NEEDS_DATA_PATH=${{runner.temp}}/needs.json
           EOF
       - name: Download json reports
         uses: actions/download-artifact@v2
@@ -466,8 +462,11 @@ jobs:
         run: |
           sudo rm -fr "$TEMP_PATH"
           mkdir -p "$TEMP_PATH"
+          cat > "$NEEDS_DATA_PATH" << 'EOF'
+          ${{ toJSON(needs) }}
+          EOF
           cd "$GITHUB_WORKSPACE/tests/ci"
-          python3 build_report_check.py "$CHECK_NAME" "$DEPENDENCIES"
+          python3 build_report_check.py "$CHECK_NAME"
       - name: Cleanup
         if: always()
         run: |

contrib/amqpcpp-cmake/CMakeLists.txt

@@ -5,6 +5,11 @@ if (NOT ENABLE_AMQPCPP)
     return()
 endif()
 
+if (NOT TARGET ch_contrib::uv)
+    message(STATUS "Not using AMQP-CPP because libuv is disabled")
+    return()
+endif()
+
 set (LIBRARY_DIR "${ClickHouse_SOURCE_DIR}/contrib/AMQP-CPP")
 
 set (SRCS

contrib/cassandra-cmake/CMakeLists.txt

@@ -5,6 +5,11 @@ if (NOT ENABLE_CASSANDRA)
     return()
 endif()
 
+if (NOT TARGET ch_contrib::uv)
+    message(STATUS "Not using cassandra because libuv is disabled")
+    return()
+endif()
+
 if (APPLE)
     set(CMAKE_MACOSX_RPATH ON)
 endif()

docker/test/fuzzer/run-fuzzer.sh

@@ -355,22 +355,8 @@ fi
 cat > report.html <<EOF ||:
 <!DOCTYPE html>
 <html lang="en">
-<link rel="preload" as="font" href="https://yastatic.net/adv-www/_/sUYVCPUAQE7ExrvMS7FoISoO83s.woff2" type="font/woff2" crossorigin="anonymous"/>
   <style>
-@font-face {
-    font-family:'Yandex Sans Display Web';
-    src:url(https://yastatic.net/adv-www/_/H63jN0veW07XQUIA2317lr9UIm8.eot);
-    src:url(https://yastatic.net/adv-www/_/H63jN0veW07XQUIA2317lr9UIm8.eot?#iefix) format('embedded-opentype'),
-            url(https://yastatic.net/adv-www/_/sUYVCPUAQE7ExrvMS7FoISoO83s.woff2) format('woff2'),
-            url(https://yastatic.net/adv-www/_/v2Sve_obH3rKm6rKrtSQpf-eB7U.woff) format('woff'),
-            url(https://yastatic.net/adv-www/_/PzD8hWLMunow5i3RfJ6WQJAL7aI.ttf) format('truetype'),
-            url(https://yastatic.net/adv-www/_/lF_KG5g4tpQNlYIgA0e77fBSZ5s.svg#YandexSansDisplayWeb-Regular) format('svg');
-    font-weight:400;
-    font-style:normal;
-    font-stretch:normal
-}
-
-body { font-family: "Yandex Sans Display Web", Arial, sans-serif; background: #EEE; }
+body { font-family: "DejaVu Sans", "Noto Sans", Arial, sans-serif; background: #EEE; }
 h1 { margin-left: 10px; }
 th, td { border: 0; padding: 5px 10px 5px 10px; text-align: left; vertical-align: top; line-height: 1.5; background-color: #FFF;
 td { white-space: pre; font-family: Monospace, Courier New; }
@@ -378,7 +364,6 @@ border: 0; box-shadow: 0 0 0 1px rgba(0, 0, 0, 0.05), 0 8px 25px -5px rgba(0, 0,
 a { color: #06F; text-decoration: none; }
 a:hover, a:active { color: #F40; text-decoration: underline; }
 table { border: 0; }
-.main { margin-left: 10%; }
 p.links a { padding: 5px; margin: 3px; background: #FFF; line-height: 2; white-space: nowrap; box-shadow: 0 0 0 1px rgba(0, 0, 0, 0.05), 0 8px 25px -5px rgba(0, 0, 0, 0.1); }
 th { cursor: pointer; }
 

docs/en/development/contrib.md

@@ -92,7 +92,7 @@ The list of third-party libraries can be obtained by the following query:
 SELECT library_name, license_type, license_path FROM system.licenses ORDER BY library_name COLLATE 'en';
 ```
 
-[Example](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUIGxpYnJhcnlfbmFtZSwgbGljZW5zZV90eXBlLCBsaWNlbnNlX3BhdGggRlJPTSBzeXN0ZW0ubGljZW5zZXMgT1JERVIgQlkgbGlicmFyeV9uYW1lIENPTExBVEUgJ2VuJw==)
+[Example](https://play.clickhouse.com/play?user=play#U0VMRUNUIGxpYnJhcnlfbmFtZSwgbGljZW5zZV90eXBlLCBsaWNlbnNlX3BhdGggRlJPTSBzeXN0ZW0ubGljZW5zZXMgT1JERVIgQlkgbGlicmFyeV9uYW1lIENPTExBVEUgJ2VuJw==)
 
 ## Adding new third-party libraries and maintaining patches in third-party libraries {#adding-third-party-libraries}
 

docs/en/getting-started/example-datasets/brown-benchmark.md

@@ -411,6 +411,6 @@ ORDER BY yr,
          mo;
 ```
 
-The data is also available for interactive queries in the [Playground](https://gh-api.clickhouse.com/play?user=play), [example](https://gh-api.clickhouse.com/play?user=play#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).
+The data is also available for interactive queries in the [Playground](https://play.clickhouse.com/play?user=play), [example](https://play.clickhouse.com/play?user=play#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).
 
 [Original article](https://clickhouse.com/docs/en/getting_started/example_datasets/brown-benchmark/) <!--hide-->

docs/en/getting-started/example-datasets/cell-towers.md

@@ -126,6 +126,6 @@ SELECT count() FROM cell_towers WHERE pointInPolygon((lon, lat), (SELECT * FROM
 1 rows in set. Elapsed: 0.067 sec. Processed 43.28 million rows, 692.42 MB (645.83 million rows/s., 10.33 GB/s.)
 ```
 
-The data is also available for interactive queries in the [Playground](https://gh-api.clickhouse.com/play?user=play), [example](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUIG1jYywgY291bnQoKSBGUk9NIGNlbGxfdG93ZXJzIEdST1VQIEJZIG1jYyBPUkRFUiBCWSBjb3VudCgpIERFU0M=).
+The data is also available for interactive queries in the [Playground](https://play.clickhouse.com/play?user=play), [example](https://play.clickhouse.com/play?user=play#U0VMRUNUIG1jYywgY291bnQoKSBGUk9NIGNlbGxfdG93ZXJzIEdST1VQIEJZIG1jYyBPUkRFUiBCWSBjb3VudCgpIERFU0M=).
 
-Although you cannot create temporary tables there.
+Although you cannot create temporary tables there.

docs/en/getting-started/example-datasets/menus.md

@@ -351,4 +351,4 @@ At least they have caviar with vodka. Very nice.
 
 ## Online Playground {#playground}
 
-The data is uploaded to ClickHouse Playground, [example](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUCiAgICByb3VuZCh0b1VJbnQzMk9yWmVybyhleHRyYWN0KG1lbnVfZGF0ZSwgJ15cXGR7NH0nKSksIC0xKSBBUyBkLAogICAgY291bnQoKSwKICAgIHJvdW5kKGF2ZyhwcmljZSksIDIpLAogICAgYmFyKGF2ZyhwcmljZSksIDAsIDUwLCAxMDApLAogICAgYW55KGRpc2hfbmFtZSkKRlJPTSBtZW51X2l0ZW1fZGVub3JtCldIRVJFIChtZW51X2N1cnJlbmN5IElOICgnRG9sbGFycycsICcnKSkgQU5EIChkID4gMCkgQU5EIChkIDwgMjAyMikgQU5EIChkaXNoX25hbWUgSUxJS0UgJyVjYXZpYXIlJykKR1JPVVAgQlkgZApPUkRFUiBCWSBkIEFTQw==).
+The data is uploaded to ClickHouse Playground, [example](https://play.clickhouse.com/play?user=play#U0VMRUNUCiAgICByb3VuZCh0b1VJbnQzMk9yWmVybyhleHRyYWN0KG1lbnVfZGF0ZSwgJ15cXGR7NH0nKSksIC0xKSBBUyBkLAogICAgY291bnQoKSwKICAgIHJvdW5kKGF2ZyhwcmljZSksIDIpLAogICAgYmFyKGF2ZyhwcmljZSksIDAsIDUwLCAxMDApLAogICAgYW55KGRpc2hfbmFtZSkKRlJPTSBtZW51X2l0ZW1fZGVub3JtCldIRVJFIChtZW51X2N1cnJlbmN5IElOICgnRG9sbGFycycsICcnKSkgQU5EIChkID4gMCkgQU5EIChkIDwgMjAyMikgQU5EIChkaXNoX25hbWUgSUxJS0UgJyVjYXZpYXIlJykKR1JPVVAgQlkgZApPUkRFUiBCWSBkIEFTQw==).

docs/en/getting-started/example-datasets/ontime.md

@@ -5,20 +5,9 @@ description: Dataset containing the on-time performance of airline flights
 
 # OnTime 
 
-This dataset can be obtained in two ways:
+This dataset contains data from Bureau of Transportation Statistics.
 
--   import from raw data
--   download of prepared partitions
-
-## Import from Raw Data {#import-from-raw-data}
-
-Downloading data:
-
-``` bash
-wget --no-check-certificate --continue https://transtats.bts.gov/PREZIP/On_Time_Reporting_Carrier_On_Time_Performance_1987_present_{1987..2021}_{1..12}.zip
-```
-
-Creating a table:
+## Creating a table
 
 ``` sql
 CREATE TABLE `ontime`
@@ -29,140 +18,138 @@ CREATE TABLE `ontime`
     `DayofMonth`                      UInt8,
     `DayOfWeek`                       UInt8,
     `FlightDate`                      Date,
-    `Reporting_Airline`               String,
+    `Reporting_Airline`               LowCardinality(String),
     `DOT_ID_Reporting_Airline`        Int32,
-    `IATA_CODE_Reporting_Airline`     String,
-    `Tail_Number`                     String,
-    `Flight_Number_Reporting_Airline` String,
+    `IATA_CODE_Reporting_Airline`     LowCardinality(String),
+    `Tail_Number`                     LowCardinality(String),
+    `Flight_Number_Reporting_Airline` LowCardinality(String),
     `OriginAirportID`                 Int32,
     `OriginAirportSeqID`              Int32,
     `OriginCityMarketID`              Int32,
     `Origin`                          FixedString(5),
-    `OriginCityName`                  String,
+    `OriginCityName`                  LowCardinality(String),
     `OriginState`                     FixedString(2),
-    `OriginStateFips`                 String,
-    `OriginStateName`                 String,
+    `OriginStateFips`                 FixedString(2),
+    `OriginStateName`                 LowCardinality(String),
     `OriginWac`                       Int32,
     `DestAirportID`                   Int32,
     `DestAirportSeqID`                Int32,
     `DestCityMarketID`                Int32,
     `Dest`                            FixedString(5),
-    `DestCityName`                    String,
+    `DestCityName`                    LowCardinality(String),
     `DestState`                       FixedString(2),
-    `DestStateFips`                   String,
-    `DestStateName`                   String,
+    `DestStateFips`                   FixedString(2),
+    `DestStateName`                   LowCardinality(String),
     `DestWac`                         Int32,
     `CRSDepTime`                      Int32,
     `DepTime`                         Int32,
     `DepDelay`                        Int32,
     `DepDelayMinutes`                 Int32,
     `DepDel15`                        Int32,
-    `DepartureDelayGroups`            String,
-    `DepTimeBlk`                      String,
+    `DepartureDelayGroups`            LowCardinality(String),
+    `DepTimeBlk`                      LowCardinality(String),
     `TaxiOut`                         Int32,
-    `WheelsOff`                       Int32,
-    `WheelsOn`                        Int32,
+    `WheelsOff`                       LowCardinality(String),
+    `WheelsOn`                        LowCardinality(String),
     `TaxiIn`                          Int32,
     `CRSArrTime`                      Int32,
     `ArrTime`                         Int32,
     `ArrDelay`                        Int32,
     `ArrDelayMinutes`                 Int32,
     `ArrDel15`                        Int32,
-    `ArrivalDelayGroups`              Int32,
-    `ArrTimeBlk`                      String,
-    `Cancelled`                       UInt8,
+    `ArrivalDelayGroups`              LowCardinality(String),
+    `ArrTimeBlk`                      LowCardinality(String),
+    `Cancelled`                       Int8,
     `CancellationCode`                FixedString(1),
-    `Diverted`                        UInt8,
+    `Diverted`                        Int8,
     `CRSElapsedTime`                  Int32,
     `ActualElapsedTime`               Int32,
-    `AirTime`                         Nullable(Int32),
+    `AirTime`                         Int32,
     `Flights`                         Int32,
     `Distance`                        Int32,
-    `DistanceGroup`                   UInt8,
+    `DistanceGroup`                   Int8,
     `CarrierDelay`                    Int32,
     `WeatherDelay`                    Int32,
     `NASDelay`                        Int32,
     `SecurityDelay`                   Int32,
     `LateAircraftDelay`               Int32,
-    `FirstDepTime`                    String,
-    `TotalAddGTime`                   String,
-    `LongestAddGTime`                 String,
-    `DivAirportLandings`              String,
-    `DivReachedDest`                  String,
-    `DivActualElapsedTime`            String,
-    `DivArrDelay`                     String,
-    `DivDistance`                     String,
-    `Div1Airport`                     String,
+    `FirstDepTime`                    Int16,
+    `TotalAddGTime`                   Int16,
+    `LongestAddGTime`                 Int16,
+    `DivAirportLandings`              Int8,
+    `DivReachedDest`                  Int8,
+    `DivActualElapsedTime`            Int16,
+    `DivArrDelay`                     Int16,
+    `DivDistance`                     Int16,
+    `Div1Airport`                     LowCardinality(String),
     `Div1AirportID`                   Int32,
     `Div1AirportSeqID`                Int32,
-    `Div1WheelsOn`                    String,
-    `Div1TotalGTime`                  String,
-    `Div1LongestGTime`                String,
-    `Div1WheelsOff`                   String,
-    `Div1TailNum`                     String,
-    `Div2Airport`                     String,
+    `Div1WheelsOn`                    Int16,
+    `Div1TotalGTime`                  Int16,
+    `Div1LongestGTime`                Int16,
+    `Div1WheelsOff`                   Int16,
+    `Div1TailNum`                     LowCardinality(String),
+    `Div2Airport`                     LowCardinality(String),
     `Div2AirportID`                   Int32,
     `Div2AirportSeqID`                Int32,
-    `Div2WheelsOn`                    String,
-    `Div2TotalGTime`                  String,
-    `Div2LongestGTime`                String,
-    `Div2WheelsOff`                   String,
-    `Div2TailNum`                     String,
-    `Div3Airport`                     String,
+    `Div2WheelsOn`                    Int16,
+    `Div2TotalGTime`                  Int16,
+    `Div2LongestGTime`                Int16,
+    `Div2WheelsOff`                   Int16,
+    `Div2TailNum`                     LowCardinality(String),
+    `Div3Airport`                     LowCardinality(String),
     `Div3AirportID`                   Int32,
     `Div3AirportSeqID`                Int32,
-    `Div3WheelsOn`                    String,
-    `Div3TotalGTime`                  String,
-    `Div3LongestGTime`                String,
-    `Div3WheelsOff`                   String,
-    `Div3TailNum`                     String,
-    `Div4Airport`                     String,
+    `Div3WheelsOn`                    Int16,
+    `Div3TotalGTime`                  Int16,
+    `Div3LongestGTime`                Int16,
+    `Div3WheelsOff`                   Int16,
+    `Div3TailNum`                     LowCardinality(String),
+    `Div4Airport`                     LowCardinality(String),
     `Div4AirportID`                   Int32,
     `Div4AirportSeqID`                Int32,
-    `Div4WheelsOn`                    String,
-    `Div4TotalGTime`                  String,
-    `Div4LongestGTime`                String,
-    `Div4WheelsOff`                   String,
-    `Div4TailNum`                     String,
-    `Div5Airport`                     String,
+    `Div4WheelsOn`                    Int16,
+    `Div4TotalGTime`                  Int16,
+    `Div4LongestGTime`                Int16,
+    `Div4WheelsOff`                   Int16,
+    `Div4TailNum`                     LowCardinality(String),
+    `Div5Airport`                     LowCardinality(String),
     `Div5AirportID`                   Int32,
     `Div5AirportSeqID`                Int32,
-    `Div5WheelsOn`                    String,
-    `Div5TotalGTime`                  String,
-    `Div5LongestGTime`                String,
-    `Div5WheelsOff`                   String,
-    `Div5TailNum`                     String
+    `Div5WheelsOn`                    Int16,
+    `Div5TotalGTime`                  Int16,
+    `Div5LongestGTime`                Int16,
+    `Div5WheelsOff`                   Int16,
+    `Div5TailNum`                     LowCardinality(String)
 ) ENGINE = MergeTree
-      PARTITION BY Year
-      ORDER BY (IATA_CODE_Reporting_Airline, FlightDate)
-      SETTINGS index_granularity = 8192;
+  ORDER BY (Year, Quarter, Month, DayofMonth, FlightDate, IATA_CODE_Reporting_Airline);
+```
+
+## Import from Raw Data {#import-from-raw-data}
+
+Downloading data:
+
+``` bash
+wget --no-check-certificate --continue https://transtats.bts.gov/PREZIP/On_Time_Reporting_Carrier_On_Time_Performance_1987_present_{1987..2022}_{1..12}.zip
 ```
 
 Loading data with multiple threads:
 
 ``` bash
-ls -1 *.zip | xargs -I{} -P $(nproc) bash -c "echo {}; unzip -cq {} '*.csv' | sed 's/\.00//g' | clickhouse-client --input_format_with_names_use_header=0 --query='INSERT INTO ontime FORMAT CSVWithNames'"
+ls -1 *.zip | xargs -I{} -P $(nproc) bash -c "echo {}; unzip -cq {} '*.csv' | sed 's/\.00//g' | clickhouse-client --input_format_csv_empty_as_default 1 --query='INSERT INTO ontime FORMAT CSVWithNames'"
 ```
 
 (if you will have memory shortage or other issues on your server, remove the `-P $(nproc)` part)
 
-## Download of Prepared Partitions {#download-of-prepared-partitions}
+## Import from a saved copy
 
-``` bash
-$ curl -O https://datasets.clickhouse.com/ontime/partitions/ontime.tar
-$ tar xvf ontime.tar -C /var/lib/clickhouse # path to ClickHouse data directory
-$ # check permissions of unpacked data, fix if required
-$ sudo service clickhouse-server restart
-$ clickhouse-client --query "select count(*) from datasets.ontime"
+Alternatively, you can import data from a saved copy by the following query:
+
+```
+INSERT INTO ontime SELECT * FROM s3('https://clickhouse-public-datasets.s3.amazonaws.com/ontime/csv_by_year/*.csv.gz', CSVWithNames) SETTINGS max_insert_threads = 40;
 ```
 
-:::note    
-If you will run the queries described below, you have to use the full table name, `datasets.ontime`.
-:::
-
-
-!!! info "Info"
-    If you are using the prepared partitions or the Online Playground replace any occurrence of `IATA_CODE_Reporting_Airline` or `IATA_CODE_Reporting_Airline AS Carrier` in the following queries with `Carrier` (see `describe ontime`).
+The snapshot was created on 2022-05-29.
 
 ## Queries {#queries}
 
@@ -398,7 +385,7 @@ ORDER BY c DESC
 LIMIT 10;
 ```
 
-You can also play with the data in Playground, [example](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUIERheU9mV2VlaywgY291bnQoKikgQVMgYwpGUk9NIG9udGltZQpXSEVSRSBZZWFyPj0yMDAwIEFORCBZZWFyPD0yMDA4CkdST1VQIEJZIERheU9mV2VlawpPUkRFUiBCWSBjIERFU0M7Cg==).
+You can also play with the data in Playground, [example](https://play.clickhouse.com/play?user=play#U0VMRUNUIERheU9mV2VlaywgY291bnQoKikgQVMgYwpGUk9NIG9udGltZQpXSEVSRSBZZWFyPj0yMDAwIEFORCBZZWFyPD0yMDA4CkdST1VQIEJZIERheU9mV2VlawpPUkRFUiBCWSBjIERFU0M7Cg==).
 
 This performance test was created by Vadim Tkachenko. See:
 

docs/en/getting-started/example-datasets/opensky.md

@@ -417,4 +417,4 @@ Result:
 
 ### Online Playground {#playground}
 
-You can test other queries to this data set using the interactive resource [Online Playground](https://gh-api.clickhouse.com/play?user=play). For example, [like this](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUCiAgICBvcmlnaW4sCiAgICBjb3VudCgpLAogICAgcm91bmQoYXZnKGdlb0Rpc3RhbmNlKGxvbmdpdHVkZV8xLCBsYXRpdHVkZV8xLCBsb25naXR1ZGVfMiwgbGF0aXR1ZGVfMikpKSBBUyBkaXN0YW5jZSwKICAgIGJhcihkaXN0YW5jZSwgMCwgMTAwMDAwMDAsIDEwMCkgQVMgYmFyCkZST00gb3BlbnNreQpXSEVSRSBvcmlnaW4gIT0gJycKR1JPVVAgQlkgb3JpZ2luCk9SREVSIEJZIGNvdW50KCkgREVTQwpMSU1JVCAxMDA=). However, please note that you cannot create temporary tables here.
+You can test other queries to this data set using the interactive resource [Online Playground](https://play.clickhouse.com/play?user=play). For example, [like this](https://play.clickhouse.com/play?user=play#U0VMRUNUCiAgICBvcmlnaW4sCiAgICBjb3VudCgpLAogICAgcm91bmQoYXZnKGdlb0Rpc3RhbmNlKGxvbmdpdHVkZV8xLCBsYXRpdHVkZV8xLCBsb25naXR1ZGVfMiwgbGF0aXR1ZGVfMikpKSBBUyBkaXN0YW5jZSwKICAgIGJhcihkaXN0YW5jZSwgMCwgMTAwMDAwMDAsIDEwMCkgQVMgYmFyCkZST00gb3BlbnNreQpXSEVSRSBvcmlnaW4gIT0gJycKR1JPVVAgQlkgb3JpZ2luCk9SREVSIEJZIGNvdW50KCkgREVTQwpMSU1JVCAxMDA=). However, please note that you cannot create temporary tables here.

docs/en/getting-started/example-datasets/recipes.md

@@ -334,6 +334,6 @@ Result:
 
 ### Online Playground
 
-The dataset is also available in the [Online Playground](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUCiAgICBhcnJheUpvaW4oTkVSKSBBUyBrLAogICAgY291bnQoKSBBUyBjCkZST00gcmVjaXBlcwpHUk9VUCBCWSBrCk9SREVSIEJZIGMgREVTQwpMSU1JVCA1MA==).
+The dataset is also available in the [Online Playground](https://play.clickhouse.com/play?user=play#U0VMRUNUCiAgICBhcnJheUpvaW4oTkVSKSBBUyBrLAogICAgY291bnQoKSBBUyBjCkZST00gcmVjaXBlcwpHUk9VUCBCWSBrCk9SREVSIEJZIGMgREVTQwpMSU1JVCA1MA==).
 
 [Original article](https://clickhouse.com/docs/en/getting-started/example-datasets/recipes/) <!--hide-->

docs/en/getting-started/example-datasets/star-schema.md

@@ -26,7 +26,6 @@ $ ./dbgen -s 1000 -T c
 $ ./dbgen -s 1000 -T l
 $ ./dbgen -s 1000 -T p
 $ ./dbgen -s 1000 -T s
-$ ./dbgen -s 1000 -T d
 ```
 
 Creating tables in ClickHouse:
@@ -109,10 +108,8 @@ Converting “star schema” to denormalized “flat schema”:
 SET max_memory_usage = 20000000000;
 
 CREATE TABLE lineorder_flat
-ENGINE = MergeTree
-PARTITION BY toYear(LO_ORDERDATE)
-ORDER BY (LO_ORDERDATE, LO_ORDERKEY) AS
-SELECT
+ENGINE = MergeTree ORDER BY (LO_ORDERDATE, LO_ORDERKEY)
+AS SELECT
     l.LO_ORDERKEY AS LO_ORDERKEY,
     l.LO_LINENUMBER AS LO_LINENUMBER,
     l.LO_CUSTKEY AS LO_CUSTKEY,

docs/en/getting-started/example-datasets/uk-price-paid.md

@@ -646,4 +646,4 @@ no projection: 100 rows in set. Elapsed: 0.069 sec. Processed 26.32 million rows
 
 ### Test It in Playground {#playground}
 
-The dataset is also available in the [Online Playground](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUIHRvd24sIGRpc3RyaWN0LCBjb3VudCgpIEFTIGMsIHJvdW5kKGF2ZyhwcmljZSkpIEFTIHByaWNlLCBiYXIocHJpY2UsIDAsIDUwMDAwMDAsIDEwMCkgRlJPTSB1a19wcmljZV9wYWlkIFdIRVJFIGRhdGUgPj0gJzIwMjAtMDEtMDEnIEdST1VQIEJZIHRvd24sIGRpc3RyaWN0IEhBVklORyBjID49IDEwMCBPUkRFUiBCWSBwcmljZSBERVNDIExJTUlUIDEwMA==).
+The dataset is also available in the [Online Playground](https://play.clickhouse.com/play?user=play#U0VMRUNUIHRvd24sIGRpc3RyaWN0LCBjb3VudCgpIEFTIGMsIHJvdW5kKGF2ZyhwcmljZSkpIEFTIHByaWNlLCBiYXIocHJpY2UsIDAsIDUwMDAwMDAsIDEwMCkgRlJPTSB1a19wcmljZV9wYWlkIFdIRVJFIGRhdGUgPj0gJzIwMjAtMDEtMDEnIEdST1VQIEJZIHRvd24sIGRpc3RyaWN0IEhBVklORyBjID49IDEwMCBPUkRFUiBCWSBwcmljZSBERVNDIExJTUlUIDEwMA==).

docs/en/interfaces/third-party/gui.md

@@ -147,6 +147,16 @@ Features:
 
 [Zeppelin-Interpreter-for-ClickHouse](https://github.com/SiderZhang/Zeppelin-Interpreter-for-ClickHouse) is a [Zeppelin](https://zeppelin.apache.org) interpreter for ClickHouse. Compared with JDBC interpreter, it can provide better timeout control for long running queries.
 
+### ClickCat {#clickcat}
+
+[ClickCat](https://github.com/open-botech/ClickCat) is a firendly user interface that lets you search, explore and visualize your ClickHouse Data.
+
+Features:
+
+-   An online SQL editor which can run your SQL code without any installing.
+-   You can observe all processes and mutations. For those unfinished processes, you can kill them in ui.
+-   The Metrics contains Cluster Analysis,Data Analysis,Query Analysis.
+
 ## Commercial {#commercial}
 
 ### DataGrip {#datagrip}

docs/en/sql-reference/dictionaries/external-dictionaries/external-dicts-dict-layout.md

@@ -21,7 +21,7 @@ ClickHouse generates an exception for errors with dictionaries. Examples of erro
 -   The dictionary being accessed could not be loaded.
 -   Error querying a `cached` dictionary.
 
-You can view the list of external dictionaries and their statuses in the `system.dictionaries` table.
+You can view the list of external dictionaries and their statuses in the [system.dictionaries](../../../operations/system-tables/dictionaries.md) table.
 
 The configuration looks like this:
 
@@ -48,6 +48,35 @@ LAYOUT(LAYOUT_TYPE(param value)) -- layout settings
 ...
 ```
 
+Dictionaries without word `complex-key*` in a layout have a key with [UInt64](../../../sql-reference/data-types/int-uint.md) type, `complex-key*` dictionaries have a composite key (complex, with arbitrary types).
+
+[UInt64](../../../sql-reference/data-types/int-uint.md) keys in XML dictionaries are defined with `<id>` tag.
+
+Configuration example (column key_column has UInt64 type):
+```xml
+...
+<structure>
+    <id>
+        <name>key_column</name>
+    </id>
+...
+```
+
+Composite `complex` keys XML dictionaries are defined `<key>` tag.
+
+Configuration example of a composite key (key has one element with [String](../../../sql-reference/data-types/string.md) type):
+```xml
+...
+<structure>
+    <key>
+        <attribute>
+            <name>country_code</name>
+            <type>String</type>
+        </attribute>
+    </key>
+...
+```
+
 ## Ways to Store Dictionaries in Memory {#ways-to-store-dictionaries-in-memory}
 
 -   [flat](#flat)
@@ -98,6 +127,8 @@ LAYOUT(FLAT(INITIAL_ARRAY_SIZE 50000 MAX_ARRAY_SIZE 5000000))
 
 The dictionary is completely stored in memory in the form of a hash table. The dictionary can contain any number of elements with any identifiers In practice, the number of keys can reach tens of millions of items.
 
+The dictionary key has the [UInt64](../../../sql-reference/data-types/int-uint.md) type.
+
 If `preallocate` is `true` (default is `false`) the hash table will be preallocated (this will make the dictionary load faster). But note that you should use it only if:
 
 - The source support an approximate number of elements (for now it is supported only by the `ClickHouse` source).
@@ -125,6 +156,8 @@ LAYOUT(HASHED(PREALLOCATE 0))
 
 Similar to `hashed`, but uses less memory in favor more CPU usage.
 
+The dictionary key has the [UInt64](../../../sql-reference/data-types/int-uint.md) type.
+
 It will be also preallocated so as `hashed` (with `preallocate` set to `true`), and note that it is even more significant for `sparse_hashed`.
 
 Configuration example:
@@ -181,6 +214,8 @@ LAYOUT(COMPLEX_KEY_SPARSE_HASHED())
 
 The dictionary is completely stored in memory. Each attribute is stored in an array. The key attribute is stored in the form of a hashed table where value is an index in the attributes array. The dictionary can contain any number of elements with any identifiers. In practice, the number of keys can reach tens of millions of items.
 
+The dictionary key has the [UInt64](../../../sql-reference/data-types/int-uint.md) type.
+
 All types of sources are supported. When updating, data (from a file or from a table) is read in its entirety.
 
 Configuration example:
@@ -220,6 +255,7 @@ LAYOUT(COMPLEX_KEY_HASHED_ARRAY())
 
 The dictionary is stored in memory in the form of a hash table with an ordered array of ranges and their corresponding values.
 
+The dictionary key has the [UInt64](../../../sql-reference/data-types/int-uint.md) type.
 This storage method works the same way as hashed and allows using date/time (arbitrary numeric type) ranges in addition to the key.
 
 Example: The table contains discounts for each advertiser in the format:
@@ -360,6 +396,8 @@ RANGE(MIN StartDate MAX EndDate);
 
 The dictionary is stored in a cache that has a fixed number of cells. These cells contain frequently used elements.
 
+The dictionary key has the [UInt64](../../../sql-reference/data-types/int-uint.md) type.
+
 When searching for a dictionary, the cache is searched first. For each block of data, all keys that are not found in the cache or are outdated are requested from the source using `SELECT attrs... FROM db.table WHERE id IN (k1, k2, ...)`. The received data is then written to the cache.
 
 If keys are not found in dictionary, then update cache task is created and added into update queue. Update queue properties can be controlled with settings `max_update_queue_size`, `update_queue_push_timeout_milliseconds`, `query_wait_timeout_milliseconds`, `max_threads_for_updates`.
@@ -420,6 +458,8 @@ This type of storage is for use with composite [keys](../../../sql-reference/dic
 
 Similar to `cache`, but stores data on SSD and index in RAM. All cache dictionary settings related to update queue can also be applied to SSD cache dictionaries.
 
+The dictionary key has the [UInt64](../../../sql-reference/data-types/int-uint.md) type.
+
 ``` xml
 <layout>
     <ssd_cache>
@@ -452,7 +492,7 @@ This type of storage is for use with composite [keys](../../../sql-reference/dic
 
 The dictionary is not stored in memory and directly goes to the source during the processing of a request.
 
-The dictionary key has the `UInt64` type.
+The dictionary key has the [UInt64](../../../sql-reference/data-types/int-uint.md) type.
 
 All types of [sources](../../../sql-reference/dictionaries/external-dictionaries/external-dicts-dict-sources.md), except local files, are supported.
 

docs/ru/development/contrib.md

@@ -92,7 +92,7 @@ sidebar_label: "Используемые сторонние библиотеки
 SELECT library_name, license_type, license_path FROM system.licenses ORDER BY library_name COLLATE 'en';
 ```
 
-[Пример](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUIGxpYnJhcnlfbmFtZSwgbGljZW5zZV90eXBlLCBsaWNlbnNlX3BhdGggRlJPTSBzeXN0ZW0ubGljZW5zZXMgT1JERVIgQlkgbGlicmFyeV9uYW1lIENPTExBVEUgJ2VuJw==)
+[Пример](https://play.clickhouse.com/play?user=play#U0VMRUNUIGxpYnJhcnlfbmFtZSwgbGljZW5zZV90eXBlLCBsaWNlbnNlX3BhdGggRlJPTSBzeXN0ZW0ubGljZW5zZXMgT1JERVIgQlkgbGlicmFyeV9uYW1lIENPTExBVEUgJ2VuJw==)
 
 ## Рекомендации по добавлению сторонних библиотек и поддержанию в них пользовательских изменений {#adding-third-party-libraries}
 

docs/ru/getting-started/example-datasets/brown-benchmark.md

@@ -411,5 +411,4 @@ ORDER BY yr,
          mo;
 ```
 
-Данные также доступны для работы с интерактивными запросами через [Playground](https://gh-api.clickhouse.com/play?user=play), [пример](https://gh-api.clickhouse.com/play?user=play#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).
-
+Данные также доступны для работы с интерактивными запросами через [Playground](https://play.clickhouse.com/play?user=play), [пример](https://play.clickhouse.com/play?user=play#U0VMRUNUIG1hY2hpbmVfbmFtZSwKICAgICAgIE1JTihjcHUpIEFTIGNwdV9taW4sCiAgICAgICBNQVgoY3B1KSBBUyBjcHVfbWF4LAogICAgICAgQVZHKGNwdSkgQVMgY3B1X2F2ZywKICAgICAgIE1JTihuZXRfaW4pIEFTIG5ldF9pbl9taW4sCiAgICAgICBNQVgobmV0X2luKSBBUyBuZXRfaW5fbWF4LAogICAgICAgQVZHKG5ldF9pbikgQVMgbmV0X2luX2F2ZywKICAgICAgIE1JTihuZXRfb3V0KSBBUyBuZXRfb3V0X21pbiwKICAgICAgIE1BWChuZXRfb3V0KSBBUyBuZXRfb3V0X21heCwKICAgICAgIEFWRyhuZXRfb3V0KSBBUyBuZXRfb3V0X2F2ZwpGUk9NICgKICBTRUxFQ1QgbWFjaGluZV9uYW1lLAogICAgICAgICBDT0FMRVNDRShjcHVfdXNlciwgMC4wKSBBUyBjcHUsCiAgICAgICAgIENPQUxFU0NFKGJ5dGVzX2luLCAwLjApIEFTIG5ldF9pbiwKICAgICAgICAgQ09BTEVTQ0UoYnl0ZXNfb3V0LCAwLjApIEFTIG5ldF9vdXQKICBGUk9NIG1nYmVuY2gubG9nczEKICBXSEVSRSBtYWNoaW5lX25hbWUgSU4gKCdhbmFuc2knLCdhcmFnb2cnLCd1cmQnKQogICAgQU5EIGxvZ190aW1lID49IFRJTUVTVEFNUCAnMjAxNy0wMS0xMSAwMDowMDowMCcKKSBBUyByCkdST1VQIEJZIG1hY2hpbmVfbmFtZQ==).

docs/ru/getting-started/example-datasets/cell-towers.md

@@ -125,4 +125,4 @@ SELECT count() FROM cell_towers WHERE pointInPolygon((lon, lat), (SELECT * FROM
 1 rows in set. Elapsed: 0.067 sec. Processed 43.28 million rows, 692.42 MB (645.83 million rows/s., 10.33 GB/s.)
 ```
 
-Вы можете протестировать другие запросы с помощью интерактивного ресурса [Playground](https://gh-api.clickhouse.com/play?user=play). Например, [вот так](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUIG1jYywgY291bnQoKSBGUk9NIGNlbGxfdG93ZXJzIEdST1VQIEJZIG1jYyBPUkRFUiBCWSBjb3VudCgpIERFU0M=). Однако, обратите внимание, что здесь нельзя создавать временные таблицы.
+Вы можете протестировать другие запросы с помощью интерактивного ресурса [Playground](https://play.clickhouse.com/play?user=play). Например, [вот так](https://play.clickhouse.com/play?user=play#U0VMRUNUIG1jYywgY291bnQoKSBGUk9NIGNlbGxfdG93ZXJzIEdST1VQIEJZIG1jYyBPUkRFUiBCWSBjb3VudCgpIERFU0M=). Однако, обратите внимание, что здесь нельзя создавать временные таблицы.

docs/ru/getting-started/example-datasets/recipes.md

@@ -337,6 +337,6 @@ WHERE title = 'Chocolate-Strawberry-Orange Wedding Cake';
 
 ### Online Playground
 
-Этот набор данных доступен в [Online Playground](https://gh-api.clickhouse.com/play?user=play#U0VMRUNUCiAgICBhcnJheUpvaW4oTkVSKSBBUyBrLAogICAgY291bnQoKSBBUyBjCkZST00gcmVjaXBlcwpHUk9VUCBCWSBrCk9SREVSIEJZIGMgREVTQwpMSU1JVCA1MA==).
+Этот набор данных доступен в [Online Playground](https://play.clickhouse.com/play?user=play#U0VMRUNUCiAgICBhcnJheUpvaW4oTkVSKSBBUyBrLAogICAgY291bnQoKSBBUyBjCkZST00gcmVjaXBlcwpHUk9VUCBCWSBrCk9SREVSIEJZIGMgREVTQwpMSU1JVCA1MA==).
 
 [Оригинальная статья](https://clickhouse.com/docs/ru/getting-started/example-datasets/recipes/) <!--hide-->

docs/ru/sql-reference/dictionaries/external-dictionaries/external-dicts-dict-layout.md

@@ -21,7 +21,7 @@ sidebar_label: "Хранение словарей в памяти"
 -   При обращении к словарю, который не удалось загрузить.
 -   При ошибке запроса к `cached`-словарю.
 
-Список внешних словарей и их статус можно посмотреть в таблице `system.dictionaries`.
+Список внешних словарей и их статус можно посмотреть в таблице [system.dictionaries](../../../operations/system-tables/dictionaries.md).
 
 Общий вид конфигурации:
 
@@ -48,6 +48,36 @@ LAYOUT(LAYOUT_TYPE(param value)) -- layout settings
 ...
 ```
 
+Ключ словарей не имеющих слово `complex-key*` в названии имеет тип [UInt64](../../../sql-reference/data-types/int-uint.md), `complex-key*` словари позволяют произвольный тип ключа (составной, и из разных типов).
+
+[UInt64](../../../sql-reference/data-types/int-uint.md) ключи в XML словарях задаются тегом `<id>`.
+
+Пример конфигурации (поле key_column имеет тип UInt64):
+```xml
+...
+<structure>
+    <id>
+        <name>key_column</name>
+    </id>
+...
+```
+
+Cоставные `complex` ключи в XML словарях задаются тегом `<key>`.
+
+Пример конфигурации составного ключа (ключ состоит из одного элемента с типом [String](../../../sql-reference/data-types/string.md)):
+```xml
+...
+<structure>
+    <key>
+        <attribute>
+            <name>country_code</name>
+            <type>String</type>
+        </attribute>
+    </key>
+...
+```
+
+
 ## Способы размещения словарей в памяти {#ways-to-store-dictionaries-in-memory}
 
 -   [flat](#flat)
@@ -98,6 +128,8 @@ LAYOUT(FLAT(INITIAL_ARRAY_SIZE 50000 MAX_ARRAY_SIZE 5000000))
 
 Словарь полностью хранится в оперативной памяти в виде хэш-таблиц. Словарь может содержать произвольное количество элементов с произвольными идентификаторами. На практике количество ключей может достигать десятков миллионов элементов.
 
+Ключ словаря имеет тип [UInt64](../../../sql-reference/data-types/int-uint.md).
+
 Если `preallocate` имеет значение `true` (по умолчанию `false`), хеш-таблица будет предварительно определена (это ускорит загрузку словаря). Используйте этот метод только в случае, если:
 
 - Источник поддерживает произвольное количество элементов (пока поддерживается только источником `ClickHouse`).
@@ -125,6 +157,8 @@ LAYOUT(HASHED(PREALLOCATE 0))
 
 Аналогичен `hashed`, но при этом занимает меньше места в памяти и генерирует более высокую загрузку CPU.
 
+Ключ словаря имеет тип [UInt64](../../../sql-reference/data-types/int-uint.md).
+
 Для этого типа размещения также можно задать `preallocate` в значении `true`. В данном случае это более важно, чем для типа `hashed`.
 
 Пример конфигурации:
@@ -181,6 +215,8 @@ LAYOUT(COMPLEX_KEY_SPARSE_HASHED())
 
 Словарь полностью хранится в оперативной памяти. Каждый атрибут хранится в массиве. Ключевой атрибут хранится в виде хеш-таблицы, где его значение является индексом в массиве атрибутов. Словарь может содержать произвольное количество элементов с произвольными идентификаторами. На практике количество ключей может достигать десятков миллионов элементов.
 
+Ключ словаря имеет тип [UInt64](../../../sql-reference/data-types/int-uint.md).
+
 Поддерживаются все виды источников. При обновлении данные (из файла, из таблицы) считываются целиком.
 
 Пример конфигурации:
@@ -220,6 +256,7 @@ LAYOUT(COMPLEX_KEY_HASHED_ARRAY())
 
 Словарь хранится в оперативной памяти в виде хэш-таблицы с упорядоченным массивом диапазонов и соответствующих им значений.
 
+Ключ словаря имеет тип [UInt64](../../../sql-reference/data-types/int-uint.md).
 Этот способ размещения работает также как и hashed и позволяет дополнительно к ключу использовать дипазоны по дате/времени (произвольному числовому типу).
 
 Пример: таблица содержит скидки для каждого рекламодателя в виде:
@@ -355,6 +392,8 @@ RANGE(MIN StartDate MAX EndDate);
 
 Словарь хранится в кэше, состоящем из фиксированного количества ячеек. Ячейки содержат часто используемые элементы.
 
+Ключ словаря имеет тип [UInt64](../../../sql-reference/data-types/int-uint.md).
+
 При поиске в словаре сначала просматривается кэш. На каждый блок данных, все не найденные в кэше или устаревшие ключи запрашиваются у источника с помощью `SELECT attrs... FROM db.table WHERE id IN (k1, k2, ...)`. Затем, полученные данные записываются в кэш.
 
 Если ключи не были найдены в словаре, то для обновления кэша создается задание и добавляется в очередь обновлений. Параметры очереди обновлений можно устанавливать настройками `max_update_queue_size`, `update_queue_push_timeout_milliseconds`, `query_wait_timeout_milliseconds`, `max_threads_for_updates`
@@ -414,6 +453,8 @@ LAYOUT(CACHE(SIZE_IN_CELLS 1000000000))
 
 Похож на `cache`, но хранит данные на SSD, а индекс в оперативной памяти. Все параметры, относящиеся к очереди обновлений, могут также быть применены к SSD-кэш словарям.
 
+Ключ словаря имеет тип [UInt64](../../../sql-reference/data-types/int-uint.md).
+
 ``` xml
 <layout>
     <ssd_cache>
@@ -446,7 +487,7 @@ LAYOUT(SSD_CACHE(BLOCK_SIZE 4096 FILE_SIZE 16777216 READ_BUFFER_SIZE 1048576
 
 Словарь не хранит данные локально и взаимодействует с источником непосредственно в момент запроса.
 
-Ключ словаря имеет тип `UInt64`.
+Ключ словаря имеет тип [UInt64](../../../sql-reference/data-types/int-uint.md).
 
 Поддерживаются все виды [источников](external-dicts-dict-sources.md), кроме локальных файлов.
 

src/AggregateFunctions/AggregateFunctionAvg.h

@@ -224,8 +224,16 @@ public:
         ++this->data(place).denominator;
     }
 
-    void
-    addBatchSinglePlace(
+    void addManyDefaults(
+        AggregateDataPtr __restrict place,
+        const IColumn ** /*columns*/,
+        size_t length,
+        Arena * /*arena*/) const override
+    {
+        this->data(place).denominator += length;
+    }
+
+    void addBatchSinglePlace(
         size_t row_begin,
         size_t row_end,
         AggregateDataPtr place,

src/AggregateFunctions/AggregateFunctionCount.h

@@ -53,6 +53,15 @@ public:
         ++data(place).count;
     }
 
+    void addManyDefaults(
+        AggregateDataPtr __restrict place,
+        const IColumn ** /*columns*/,
+        size_t length,
+        Arena * /*arena*/) const override
+    {
+        data(place).count += length;
+    }
+
     void addBatchSinglePlace(
         size_t row_begin,
         size_t row_end,

src/AggregateFunctions/AggregateFunctionMinMaxAny.h

@@ -880,8 +880,9 @@ struct AggregateFunctionMinData : Data
 {
     using Self = AggregateFunctionMinData;
 
-    bool changeIfBetter(const IColumn & column, size_t row_num, Arena * arena) { return this->changeIfLess(column, row_num, arena); }
-    bool changeIfBetter(const Self & to, Arena * arena)                        { return this->changeIfLess(to, arena); }
+    bool changeIfBetter(const IColumn & column, size_t row_num, Arena * arena)     { return this->changeIfLess(column, row_num, arena); }
+    bool changeIfBetter(const Self & to, Arena * arena)                            { return this->changeIfLess(to, arena); }
+    void addManyDefaults(const IColumn & column, size_t /*length*/, Arena * arena) { this->changeIfLess(column, 0, arena); }
 
     static const char * name() { return "min"; }
 
@@ -907,8 +908,9 @@ struct AggregateFunctionMaxData : Data
 {
     using Self = AggregateFunctionMaxData;
 
-    bool changeIfBetter(const IColumn & column, size_t row_num, Arena * arena) { return this->changeIfGreater(column, row_num, arena); }
-    bool changeIfBetter(const Self & to, Arena * arena)                        { return this->changeIfGreater(to, arena); }
+    bool changeIfBetter(const IColumn & column, size_t row_num, Arena * arena)     { return this->changeIfGreater(column, row_num, arena); }
+    bool changeIfBetter(const Self & to, Arena * arena)                            { return this->changeIfGreater(to, arena); }
+    void addManyDefaults(const IColumn & column, size_t /*length*/, Arena * arena) { this->changeIfGreater(column, 0, arena); }
 
     static const char * name() { return "max"; }
 
@@ -935,8 +937,9 @@ struct AggregateFunctionAnyData : Data
     using Self = AggregateFunctionAnyData;
     static constexpr bool is_any = true;
 
-    bool changeIfBetter(const IColumn & column, size_t row_num, Arena * arena) { return this->changeFirstTime(column, row_num, arena); }
-    bool changeIfBetter(const Self & to, Arena * arena)                        { return this->changeFirstTime(to, arena); }
+    bool changeIfBetter(const IColumn & column, size_t row_num, Arena * arena)     { return this->changeFirstTime(column, row_num, arena); }
+    bool changeIfBetter(const Self & to, Arena * arena)                            { return this->changeFirstTime(to, arena); }
+    void addManyDefaults(const IColumn & column, size_t /*length*/, Arena * arena) { this->changeFirstTime(column, 0, arena); }
 
     static const char * name() { return "any"; }
 
@@ -962,8 +965,9 @@ struct AggregateFunctionAnyLastData : Data
 {
     using Self = AggregateFunctionAnyLastData;
 
-    bool changeIfBetter(const IColumn & column, size_t row_num, Arena * arena) { return this->changeEveryTime(column, row_num, arena); }
-    bool changeIfBetter(const Self & to, Arena * arena)                        { return this->changeEveryTime(to, arena); }
+    bool changeIfBetter(const IColumn & column, size_t row_num, Arena * arena)     { return this->changeEveryTime(column, row_num, arena); }
+    bool changeIfBetter(const Self & to, Arena * arena)                            { return this->changeEveryTime(to, arena); }
+    void addManyDefaults(const IColumn & column, size_t /*length*/, Arena * arena) { this->changeEveryTime(column, 0, arena); }
 
     static const char * name() { return "anyLast"; }
 
@@ -1024,6 +1028,8 @@ struct AggregateFunctionSingleValueOrNullData : Data
         return false;
     }
 
+    void addManyDefaults(const IColumn & column, size_t /*length*/, Arena * arena) { this->changeIfBetter(column, 0, arena); }
+
     void insertResultInto(IColumn & to) const
     {
         if (is_null || first_value)
@@ -1098,6 +1104,12 @@ struct AggregateFunctionAnyHeavyData : Data
         return false;
     }
 
+    void addManyDefaults(const IColumn & column, size_t length, Arena * arena)
+    {
+        for (size_t i = 0; i < length; ++i)
+            changeIfBetter(column, 0, arena);
+    }
+
     void write(WriteBuffer & buf, const ISerialization & serialization) const
     {
         Data::write(buf, serialization);
@@ -1158,6 +1170,15 @@ public:
         this->data(place).changeIfBetter(*columns[0], row_num, arena);
     }
 
+    void addManyDefaults(
+        AggregateDataPtr __restrict place,
+        const IColumn ** columns,
+        size_t length,
+        Arena * arena) const override
+    {
+        this->data(place).addManyDefaults(*columns[0], length, arena);
+    }
+
     void addBatchSinglePlace(
         size_t row_begin,
         size_t row_end,

src/AggregateFunctions/AggregateFunctionSum.h

@@ -489,6 +489,33 @@ public:
         }
     }
 
+    void addManyDefaults(
+        AggregateDataPtr __restrict /*place*/,
+        const IColumn ** /*columns*/,
+        size_t /*length*/,
+        Arena * /*arena*/) const override
+    {
+    }
+
+    void addBatchSparse(
+        size_t row_begin,
+        size_t row_end,
+        AggregateDataPtr * places,
+        size_t place_offset,
+        const IColumn ** columns,
+        Arena * arena) const override
+    {
+        const auto & column_sparse = assert_cast<const ColumnSparse &>(*columns[0]);
+        const auto * values = &column_sparse.getValuesColumn();
+        const auto & offsets = column_sparse.getOffsetsData();
+
+        size_t from = std::lower_bound(offsets.begin(), offsets.end(), row_begin) - offsets.begin();
+        size_t to = std::lower_bound(offsets.begin(), offsets.end(), row_end) - offsets.begin();
+
+        for (size_t i = from; i < to; ++i)
+            add(places[offsets[i]] + place_offset, &values, i + 1, arena);
+    }
+
     void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, Arena *) const override
     {
         this->data(place).merge(this->data(rhs));

src/AggregateFunctions/AggregateFunctionUniq.h

@@ -237,6 +237,15 @@ public:
         detail::OneAdder<T, Data>::add(this->data(place), *columns[0], row_num);
     }
 
+    void addManyDefaults(
+        AggregateDataPtr __restrict place,
+        const IColumn ** columns,
+        size_t /*length*/,
+        Arena * /*arena*/) const override
+    {
+        detail::OneAdder<T, Data>::add(this->data(place), *columns[0], 0);
+    }
+
     void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, Arena *) const override
     {
         this->data(place).set.merge(this->data(rhs).set);

src/AggregateFunctions/IAggregateFunction.h

@@ -123,6 +123,10 @@ public:
      */
     virtual void add(AggregateDataPtr __restrict place, const IColumn ** columns, size_t row_num, Arena * arena) const = 0;
 
+    /// Adds several default values of arguments into aggregation data on which place points to.
+    /// Default values must be a the 0-th positions in columns.
+    virtual void addManyDefaults(AggregateDataPtr __restrict place, const IColumn ** columns, size_t length, Arena * arena) const = 0;
+
     /// Merges state (on which place points to) with other state of current aggregation function.
     virtual void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, Arena * arena) const = 0;
 
@@ -377,6 +381,16 @@ public:
 
     AddFunc getAddressOfAddFunction() const override { return &addFree; }
 
+    void addManyDefaults(
+        AggregateDataPtr __restrict place,
+        const IColumn ** columns,
+        size_t length,
+        Arena * arena) const override
+    {
+        for (size_t i = 0; i < length; ++i)
+            static_cast<const Derived *>(this)->add(place, columns, 0, arena);
+    }
+
     void addBatch( /// NOLINT
         size_t row_begin,
         size_t row_end,
@@ -413,13 +427,9 @@ public:
     {
         const auto & column_sparse = assert_cast<const ColumnSparse &>(*columns[0]);
         const auto * values = &column_sparse.getValuesColumn();
-        auto offset_it = column_sparse.begin();
+        auto offset_it = column_sparse.getIterator(row_begin);
 
-        /// FIXME: make it more optimal
-        for (size_t i = 0; i < row_begin; ++i, ++offset_it)
-            ;
-
-        for (size_t i = 0; i < row_end; ++i, ++offset_it)
+        for (size_t i = row_begin; i < row_end; ++i, ++offset_it)
             static_cast<const Derived *>(this)->add(places[offset_it.getCurrentRow()] + place_offset,
                                                     &values, offset_it.getValueIndex(), arena);
     }
@@ -468,17 +478,16 @@ public:
         const IColumn ** columns,
         Arena * arena) const override
     {
-        /// TODO: add values and defaults separately if order of adding isn't important.
         const auto & column_sparse = assert_cast<const ColumnSparse &>(*columns[0]);
         const auto * values = &column_sparse.getValuesColumn();
-        auto offset_it = column_sparse.begin();
+        const auto & offsets = column_sparse.getOffsetsData();
 
-        /// FIXME: make it more optimal
-        for (size_t i = 0; i < row_begin; ++i, ++offset_it)
-            ;
+        auto from = std::lower_bound(offsets.begin(), offsets.end(), row_begin) - offsets.begin() + 1;
+        auto to = std::lower_bound(offsets.begin(), offsets.end(), row_end) - offsets.begin() + 1;
 
-        for (size_t i = 0; i < row_end; ++i, ++offset_it)
-            static_cast<const Derived *>(this)->add(place, &values, offset_it.getValueIndex(), arena);
+        size_t num_defaults = (row_end - row_begin) - (to - from);
+        static_cast<const Derived *>(this)->addBatchSinglePlace(from, to, place, &values, arena, -1);
+        static_cast<const Derived *>(this)->addManyDefaults(place, &values, num_defaults, arena);
     }
 
     void addBatchSinglePlaceNotNull( /// NOLINT

src/CMakeLists.txt

@@ -87,6 +87,7 @@ add_headers_and_sources(clickhouse_common_io IO/S3)
 list (REMOVE_ITEM clickhouse_common_io_sources Common/malloc.cpp Common/new_delete.cpp)
 
 add_headers_and_sources(dbms Disks/IO)
+add_headers_and_sources(dbms Disks/ObjectStorages)
 if (TARGET ch_contrib::sqlite)
     add_headers_and_sources(dbms Databases/SQLite)
 endif()
@@ -113,16 +114,16 @@ endif()
 
 if (TARGET ch_contrib::aws_s3)
     add_headers_and_sources(dbms Common/S3)
-    add_headers_and_sources(dbms Disks/S3)
+    add_headers_and_sources(dbms Disks/ObjectStorages/S3)
 endif()
 
 if (TARGET ch_contrib::azure_sdk)
-    add_headers_and_sources(dbms Disks/AzureBlobStorage)
+    add_headers_and_sources(dbms Disks/ObjectStorages/AzureBlobStorage)
 endif()
 
 if (TARGET ch_contrib::hdfs)
     add_headers_and_sources(dbms Storages/HDFS)
-    add_headers_and_sources(dbms Disks/HDFS)
+    add_headers_and_sources(dbms Disks/ObjectStorages/HDFS)
 endif()
 
 add_headers_and_sources(dbms Storages/Cache)

src/Client/ClientBase.cpp

@@ -285,11 +285,11 @@ void ClientBase::setupSignalHandler()
     sigemptyset(&new_act.sa_mask);
 #else
     if (sigemptyset(&new_act.sa_mask))
-        throw Exception(ErrorCodes::CANNOT_SET_SIGNAL_HANDLER, "Cannot set signal handler.");
+        throwFromErrno("Cannot set signal handler.", ErrorCodes::CANNOT_SET_SIGNAL_HANDLER);
 #endif
 
     if (sigaction(SIGINT, &new_act, nullptr))
-        throw Exception(ErrorCodes::CANNOT_SET_SIGNAL_HANDLER, "Cannot set signal handler.");
+        throwFromErrno("Cannot set signal handler.", ErrorCodes::CANNOT_SET_SIGNAL_HANDLER);
 }
 
 
@@ -492,7 +492,8 @@ try
         String pager = config().getString("pager", "");
         if (!pager.empty())
         {
-            signal(SIGPIPE, SIG_IGN);
+            if (SIG_ERR == signal(SIGPIPE, SIG_IGN))
+                throwFromErrno("Cannot set signal handler.", ErrorCodes::CANNOT_SET_SIGNAL_HANDLER);
 
             ShellCommand::Config config(pager);
             config.pipe_stdin_only = true;

src/Columns/ColumnSparse.cpp

@@ -772,6 +772,14 @@ size_t ColumnSparse::getValueIndex(size_t n) const
     return it - offsets_data.begin() + 1;
 }
 
+ColumnSparse::Iterator ColumnSparse::getIterator(size_t n) const
+{
+    const auto & offsets_data = getOffsetsData();
+    const auto * it = std::lower_bound(offsets_data.begin(), offsets_data.end(), n);
+    size_t current_offset = it - offsets_data.begin();
+    return Iterator(offsets_data, _size, current_offset, n);
+}
+
 ColumnPtr recursiveRemoveSparse(const ColumnPtr & column)
 {
     if (!column)

src/Columns/ColumnSparse.h

@@ -215,6 +215,7 @@ public:
 
     Iterator begin() const { return Iterator(getOffsetsData(), _size, 0, 0); }
     Iterator end() const { return Iterator(getOffsetsData(), _size, getOffsetsData().size(), _size); }
+    Iterator getIterator(size_t n) const;
 
 private:
     using Inserter = std::function<void(IColumn &)>;

src/Common/ErrorCodes.cpp

@@ -628,6 +628,7 @@
     M(657, UNSUPPORTED_MEILISEARCH_TYPE) \
     M(658, MEILISEARCH_MISSING_SOME_COLUMNS) \
     M(659, UNKNOWN_STATUS_OF_TRANSACTION) \
+    M(660, HDFS_ERROR) \
     \
     M(999, KEEPER_EXCEPTION) \
     M(1000, POCO_EXCEPTION) \

src/Common/FileCache.cpp

@@ -72,6 +72,8 @@ void IFileCache::assertInitialized() const
 
 LRUFileCache::LRUFileCache(const String & cache_base_path_, const FileCacheSettings & cache_settings_)
     : IFileCache(cache_base_path_, cache_settings_)
+    , max_stash_element_size(cache_settings_.max_elements)
+    , enable_cache_hits_threshold(cache_settings_.enable_cache_hits_threshold)
     , log(&Poco::Logger::get("LRUFileCache"))
 {
 }
@@ -404,9 +406,42 @@ LRUFileCache::FileSegmentCell * LRUFileCache::addCell(
             "Cache already exists for key: `{}`, offset: {}, size: {}.\nCurrent cache structure: {}",
             keyToStr(key), offset, size, dumpStructureUnlocked(key, cache_lock));
 
-    auto file_segment = std::make_shared<FileSegment>(offset, size, key, this, state);
-    FileSegmentCell cell(std::move(file_segment), this, cache_lock);
+    auto skip_or_download = [&]() -> FileSegmentPtr
+    {
+        if (state == FileSegment::State::EMPTY && enable_cache_hits_threshold)
+        {
+            auto record = records.find({key, offset});
 
+            if (record == records.end())
+            {
+                auto queue_iter = stash_queue.add(key, offset, 0, cache_lock);
+                records.insert({{key, offset}, queue_iter});
+
+                if (stash_queue.getElementsNum(cache_lock) > max_stash_element_size)
+                {
+                    auto remove_queue_iter = stash_queue.begin();
+                    records.erase({remove_queue_iter->key, remove_queue_iter->offset});
+                    stash_queue.remove(remove_queue_iter, cache_lock);
+                }
+
+                /// For segments that do not reach the download threshold, we do not download them, but directly read them
+                return std::make_shared<FileSegment>(offset, size, key, this, FileSegment::State::SKIP_CACHE);
+            }
+            else
+            {
+                auto queue_iter = record->second;
+                queue_iter->hits++;
+                stash_queue.moveToEnd(queue_iter, cache_lock);
+
+                state = queue_iter->hits >= enable_cache_hits_threshold ? FileSegment::State::EMPTY : FileSegment::State::SKIP_CACHE;
+                return std::make_shared<FileSegment>(offset, size, key, this, state);
+            }
+        }
+        else
+            return std::make_shared<FileSegment>(offset, size, key, this, state);
+    };
+
+    FileSegmentCell cell(skip_or_download(), this, cache_lock);
     auto & offsets = files[key];
 
     if (offsets.empty())
@@ -471,7 +506,7 @@ bool LRUFileCache::tryReserve(
     std::vector<FileSegmentCell *> to_evict;
     std::vector<FileSegmentCell *> trash;
 
-    for (const auto & [entry_key, entry_offset, entry_size] : queue)
+    for (const auto & [entry_key, entry_offset, entry_size, _] : queue)
     {
         if (!is_overflow())
             break;
@@ -603,10 +638,6 @@ void LRUFileCache::remove(const Key & key)
 
     if (fs::exists(key_path))
         fs::remove(key_path);
-
-#ifndef NDEBUG
-    assertCacheCorrectness(cache_lock);
-#endif
 }
 
 void LRUFileCache::remove()
@@ -619,7 +650,7 @@ void LRUFileCache::remove()
     std::vector<FileSegment *> to_remove;
     for (auto it = queue.begin(); it != queue.end();)
     {
-        const auto & [key, offset, size] = *it++;
+        const auto & [key, offset, size, _] = *it++;
         auto * cell = getCell(key, offset, cache_lock);
         if (!cell)
             throw Exception(
@@ -637,6 +668,10 @@ void LRUFileCache::remove()
             }
         }
     }
+
+    /// Remove all access information.
+    records.clear();
+    stash_queue.removeAll(cache_lock);
 }
 
 void LRUFileCache::remove(
@@ -882,6 +917,7 @@ LRUFileCache::FileSegmentCell::FileSegmentCell(
             queue_iterator = cache->queue.add(file_segment->key(), file_segment->offset(), file_segment->range().size(), cache_lock);
             break;
         }
+        case FileSegment::State::SKIP_CACHE:
         case FileSegment::State::EMPTY:
         case FileSegment::State::DOWNLOADING:
         {
@@ -898,7 +934,7 @@ LRUFileCache::LRUQueue::Iterator LRUFileCache::LRUQueue::add(
     const IFileCache::Key & key, size_t offset, size_t size, std::lock_guard<std::mutex> & /* cache_lock */)
 {
 #ifndef NDEBUG
-    for (const auto & [entry_key, entry_offset, _] : queue)
+    for (const auto & [entry_key, entry_offset, entry_size, entry_hits] : queue)
     {
         if (entry_key == key && entry_offset == offset)
             throw Exception(
@@ -918,6 +954,12 @@ void LRUFileCache::LRUQueue::remove(Iterator queue_it, std::lock_guard<std::mute
     queue.erase(queue_it);
 }
 
+void LRUFileCache::LRUQueue::removeAll(std::lock_guard<std::mutex> & /* cache_lock */)
+{
+    queue.clear();
+    cache_size = 0;
+}
+
 void LRUFileCache::LRUQueue::moveToEnd(Iterator queue_it, std::lock_guard<std::mutex> & /* cache_lock */)
 {
     queue.splice(queue.end(), queue, queue_it);
@@ -934,7 +976,7 @@ bool LRUFileCache::LRUQueue::contains(
 {
     /// This method is used for assertions in debug mode.
     /// So we do not care about complexity here.
-    for (const auto & [entry_key, entry_offset, size] : queue)
+    for (const auto & [entry_key, entry_offset, size, _] : queue)
     {
         if (key == entry_key && offset == entry_offset)
             return true;
@@ -947,7 +989,7 @@ void LRUFileCache::LRUQueue::assertCorrectness(LRUFileCache * cache, std::lock_g
     [[maybe_unused]] size_t total_size = 0;
     for (auto it = queue.begin(); it != queue.end();)
     {
-        auto & [key, offset, size] = *it++;
+        auto & [key, offset, size, _] = *it++;
 
         auto * cell = cache->getCell(key, offset, cache_lock);
         if (!cell)
@@ -969,7 +1011,7 @@ void LRUFileCache::LRUQueue::assertCorrectness(LRUFileCache * cache, std::lock_g
 String LRUFileCache::LRUQueue::toString(std::lock_guard<std::mutex> & /* cache_lock */) const
 {
     String result;
-    for (const auto & [key, offset, size] : queue)
+    for (const auto & [key, offset, size, _] : queue)
     {
         if (!result.empty())
             result += ", ";

src/Common/FileCache.h

@@ -7,6 +7,7 @@
 #include <mutex>
 #include <unordered_map>
 #include <unordered_set>
+#include <boost/functional/hash.hpp>
 #include <boost/noncopyable.hpp>
 #include <map>
 
@@ -165,6 +166,7 @@ private:
             Key key;
             size_t offset;
             size_t size;
+            size_t hits = 0;
 
             FileKeyAndOffset(const Key & key_, size_t offset_, size_t size_) : key(key_), offset(offset_), size(size_) {}
         };
@@ -194,6 +196,8 @@ private:
 
         Iterator end() { return queue.end(); }
 
+        void removeAll(std::lock_guard<std::mutex> & cache_lock);
+
     private:
         std::list<FileKeyAndOffset> queue;
         size_t cache_size = 0;
@@ -223,8 +227,26 @@ private:
     using FileSegmentsByOffset = std::map<size_t, FileSegmentCell>;
     using CachedFiles = std::unordered_map<Key, FileSegmentsByOffset>;
 
+    using AccessKeyAndOffset = std::pair<Key, size_t>;
+
+    struct KeyAndOffsetHash
+    {
+        std::size_t operator()(const AccessKeyAndOffset & key) const
+        {
+            return std::hash<UInt128>()(key.first) ^ std::hash<UInt64>()(key.second);
+        }
+    };
+
+    using AccessRecord = std::unordered_map<AccessKeyAndOffset, LRUQueue::Iterator, KeyAndOffsetHash>;
+
     CachedFiles files;
     LRUQueue queue;
+
+    LRUQueue stash_queue;
+    AccessRecord records;
+    size_t max_stash_element_size;
+    size_t enable_cache_hits_threshold;
+
     Poco::Logger * log;
 
     FileSegments getImpl(

src/Common/FileCacheSettings.cpp

@@ -11,6 +11,7 @@ void FileCacheSettings::loadFromConfig(const Poco::Util::AbstractConfiguration &
     max_elements = config.getUInt64(config_prefix + ".data_cache_max_elements", REMOTE_FS_OBJECTS_CACHE_DEFAULT_MAX_ELEMENTS);
     max_file_segment_size = config.getUInt64(config_prefix + ".max_file_segment_size", REMOTE_FS_OBJECTS_CACHE_DEFAULT_MAX_FILE_SEGMENT_SIZE);
     cache_on_write_operations = config.getUInt64(config_prefix + ".cache_on_write_operations", false);
+    enable_cache_hits_threshold = config.getUInt64(config_prefix + ".enable_cache_hits_threshold", REMOTE_FS_OBJECTS_CACHE_ENABLE_HITS_THRESHOLD);
 }
 
 }

src/Common/FileCacheSettings.h

@@ -14,6 +14,8 @@ struct FileCacheSettings
     size_t max_file_segment_size = REMOTE_FS_OBJECTS_CACHE_DEFAULT_MAX_FILE_SEGMENT_SIZE;
     bool cache_on_write_operations = false;
 
+    size_t enable_cache_hits_threshold = REMOTE_FS_OBJECTS_CACHE_ENABLE_HITS_THRESHOLD;
+
     void loadFromConfig(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix);
 };
 

src/Common/FileCache_fwd.h

@@ -7,6 +7,7 @@ namespace DB
 static constexpr int REMOTE_FS_OBJECTS_CACHE_DEFAULT_MAX_CACHE_SIZE = 1024 * 1024 * 1024;
 static constexpr int REMOTE_FS_OBJECTS_CACHE_DEFAULT_MAX_FILE_SEGMENT_SIZE = 100 * 1024 * 1024;
 static constexpr int REMOTE_FS_OBJECTS_CACHE_DEFAULT_MAX_ELEMENTS = 1024 * 1024;
+static constexpr int REMOTE_FS_OBJECTS_CACHE_ENABLE_HITS_THRESHOLD = 0;
 
 class IFileCache;
 using FileCachePtr = std::shared_ptr<IFileCache>;

src/Common/FileSegment.cpp

@@ -59,6 +59,10 @@ FileSegment::FileSegment(
             downloader_id = getCallerId();
             break;
         }
+        case (State::SKIP_CACHE):
+        {
+            break;
+        }
         default:
         {
             throw Exception(ErrorCodes::REMOTE_FS_OBJECT_CACHE_ERROR, "Can create cell with either EMPTY, DOWNLOADED, DOWNLOADING state");
@@ -525,6 +529,14 @@ void FileSegment::complete(std::lock_guard<std::mutex> & cache_lock)
 
 void FileSegment::completeUnlocked(std::lock_guard<std::mutex> & cache_lock, std::lock_guard<std::mutex> & segment_lock)
 {
+    bool is_last_holder = cache->isLastFileSegmentHolder(key(), offset(), cache_lock, segment_lock);
+
+    if (is_last_holder && download_state == State::SKIP_CACHE)
+    {
+        cache->remove(key(), offset(), cache_lock, segment_lock);
+        return;
+    }
+
     if (download_state == State::SKIP_CACHE || is_detached)
         return;
 
@@ -542,8 +554,7 @@ void FileSegment::completeUnlocked(std::lock_guard<std::mutex> & cache_lock, std
         /// Segment state can be changed from DOWNLOADING or EMPTY only if the caller is the
         /// downloader or the only owner of the segment.
 
-        bool can_update_segment_state = isDownloaderImpl(segment_lock)
-            || cache->isLastFileSegmentHolder(key(), offset(), cache_lock, segment_lock);
+        bool can_update_segment_state = isDownloaderImpl(segment_lock) || is_last_holder;
 
         if (can_update_segment_state)
             download_state = State::PARTIALLY_DOWNLOADED;

src/Common/ProfileEvents.cpp

@@ -144,6 +144,13 @@
     M(MergeTreeDataWriterBlocks, "Number of blocks INSERTed to MergeTree tables. Each block forms a data part of level zero.") \
     M(MergeTreeDataWriterBlocksAlreadySorted, "Number of blocks INSERTed to MergeTree tables that appeared to be already sorted.") \
     \
+    M(InsertedWideParts, "Number of parts inserted in Wide format.") \
+    M(InsertedCompactParts, "Number of parts inserted in Compact format.") \
+    M(InsertedInMemoryParts, "Number of parts inserted in InMemory format.") \
+    M(MergedIntoWideParts, "Number of parts merged into Wide format.") \
+    M(MergedIntoCompactParts, "Number of parts merged into Compact format.") \
+    M(MergedIntoInMemoryParts, "Number of parts in merged into InMemory format.") \
+    \
     M(MergeTreeDataProjectionWriterRows, "Number of rows INSERTed to MergeTree tables projection.") \
     M(MergeTreeDataProjectionWriterUncompressedBytes, "Uncompressed bytes (for columns as they stored in memory) INSERTed to MergeTree tables projection.") \
     M(MergeTreeDataProjectionWriterCompressedBytes, "Bytes written to filesystem for data INSERTed to MergeTree tables projection.") \

src/Common/ThreadFuzzer.cpp

@@ -1,3 +1,5 @@
+// NOLINTBEGIN(readability-inconsistent-declaration-parameter-name)
+
 #include <csignal>
 #include <sys/time.h>
 #if defined(OS_LINUX)
@@ -317,3 +319,5 @@ FOR_EACH_WRAPPED_FUNCTION(MAKE_WRAPPER)
 #    undef MAKE_WRAPPER
 #endif
 }
+
+// NOLINTEND(readability-inconsistent-declaration-parameter-name)

src/Common/filesystemHelpers.cpp

@@ -1,10 +1,8 @@
 #include "filesystemHelpers.h"
 
-#include <sys/stat.h>
 #if defined(__linux__)
 #    include <cstdio>
 #    include <mntent.h>
-#    include <sys/stat.h>
 #    include <sys/sysmacros.h>
 #endif
 #include <cerrno>
@@ -13,6 +11,7 @@
 #include <filesystem>
 #include <fcntl.h>
 #include <unistd.h>
+#include <sys/stat.h>
 #include <sys/types.h>
 #include <utime.h>
 #include <IO/ReadBufferFromFile.h>

src/Coordination/KeeperStorage.cpp

@@ -1166,12 +1166,12 @@ KeeperStorage::ResponsesForSessions KeeperStorage::processRequest(const Coordina
                     ? list_watches
                     : watches;
 
-                watches_type[zk_request->getPath()].emplace_back(session_id);
+                watches_type[zk_request->getPath()].emplace(session_id);
                 sessions_and_watchers[session_id].emplace(zk_request->getPath());
             }
             else if (response->error == Coordination::Error::ZNONODE && zk_request->getOpNum() == Coordination::OpNum::Exists)
             {
-                watches[zk_request->getPath()].emplace_back(session_id);
+                watches[zk_request->getPath()].emplace(session_id);
                 sessions_and_watchers[session_id].emplace(zk_request->getPath());
             }
         }
@@ -1206,13 +1206,7 @@ void KeeperStorage::clearDeadWatches(int64_t session_id)
             if (watch != watches.end())
             {
                 auto & watches_for_path = watch->second;
-                for (auto w_it = watches_for_path.begin(); w_it != watches_for_path.end();)
-                {
-                    if (*w_it == session_id)
-                        w_it = watches_for_path.erase(w_it);
-                    else
-                        ++w_it;
-                }
+                watches_for_path.erase(session_id);
                 if (watches_for_path.empty())
                     watches.erase(watch);
             }
@@ -1222,13 +1216,7 @@ void KeeperStorage::clearDeadWatches(int64_t session_id)
             if (list_watch != list_watches.end())
             {
                 auto & list_watches_for_path = list_watch->second;
-                for (auto w_it = list_watches_for_path.begin(); w_it != list_watches_for_path.end();)
-                {
-                    if (*w_it == session_id)
-                        w_it = list_watches_for_path.erase(w_it);
-                    else
-                        ++w_it;
-                }
+                list_watches_for_path.erase(session_id);
                 if (list_watches_for_path.empty())
                     list_watches.erase(list_watch);
             }
@@ -1250,7 +1238,7 @@ void KeeperStorage::dumpWatches(WriteBufferFromOwnString & buf) const
 
 void KeeperStorage::dumpWatchesByPath(WriteBufferFromOwnString & buf) const
 {
-    auto write_int_vec = [&buf](const std::vector<int64_t> & session_ids)
+    auto write_int_container = [&buf](const auto & session_ids)
     {
         for (int64_t session_id : session_ids)
         {
@@ -1261,13 +1249,13 @@ void KeeperStorage::dumpWatchesByPath(WriteBufferFromOwnString & buf) const
     for (const auto & [watch_path, sessions] : watches)
     {
         buf << watch_path << "\n";
-        write_int_vec(sessions);
+        write_int_container(sessions);
     }
 
     for (const auto & [watch_path, sessions] : list_watches)
     {
         buf << watch_path << "\n";
-        write_int_vec(sessions);
+        write_int_container(sessions);
     }
 }
 

src/Coordination/KeeperStorage.h

@@ -96,7 +96,7 @@ public:
     using Container = SnapshotableHashTable<Node>;
     using Ephemerals = std::unordered_map<int64_t, std::unordered_set<std::string>>;
     using SessionAndWatcher = std::unordered_map<int64_t, std::unordered_set<std::string>>;
-    using SessionIDs = std::vector<int64_t>;
+    using SessionIDs = std::unordered_set<int64_t>;
 
     /// Just vector of SHA1 from user:password
     using AuthIDs = std::vector<AuthID>;

src/Core/ColumnNumbers.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <unordered_set>
 #include <vector>
 
 
@@ -7,6 +8,8 @@ namespace DB
 {
 
 using ColumnNumbers = std::vector<size_t>;
+using ColumnNumbersSet = std::unordered_set<size_t>;
 using ColumnNumbersList = std::vector<ColumnNumbers>;
+using ColumnNumbersSetList = std::vector<ColumnNumbersSet>;
 
 }

src/Core/Names.h

@@ -16,6 +16,7 @@ using NameOrderedSet = std::set<std::string>;
 using NameToNameMap = std::unordered_map<std::string, std::string>;
 using NameToNameSetMap = std::unordered_map<std::string, NameSet>;
 using NameToNameVector = std::vector<std::pair<std::string, std::string>>;
+using NameToIndexMap = std::unordered_map<std::string, size_t>;
 
 using NameWithAlias = std::pair<std::string, std::string>;
 using NamesWithAliases = std::vector<NameWithAlias>;

src/Core/NamesAndTypes.cpp

@@ -1,3 +1,4 @@
+#include <cstddef>
 #include <Core/NamesAndTypes.h>
 
 #include <base/sort.h>
@@ -214,4 +215,17 @@ std::optional<NameAndTypePair> NamesAndTypesList::tryGetByName(const std::string
     }
     return {};
 }
+
+size_t NamesAndTypesList::getPosByName(const std::string &name) const noexcept
+{
+    size_t pos = 0;
+    for (const NameAndTypePair & column : *this)
+    {
+        if (column.name == name)
+            break;
+        ++pos;
+    }
+    return pos;
+}
+
 }

src/Core/NamesAndTypes.h

@@ -105,8 +105,11 @@ public:
     /// Check that column contains in list
     bool contains(const String & name) const;
 
-    /// Try to get column by name, return empty optional if column not found
+    /// Try to get column by name, returns empty optional if column not found
     std::optional<NameAndTypePair> tryGetByName(const std::string & name) const;
+
+    /// Try to get column position by name, returns number of columns if column isn't found
+    size_t getPosByName(const std::string & name) const noexcept;
 };
 
 using NamesAndTypesLists = std::vector<NamesAndTypesList>;

src/Core/Settings.h

@@ -86,6 +86,7 @@ static constexpr UInt64 operator""_GiB(unsigned long long value)
     M(UInt64, s3_max_connections, 1024, "The maximum number of connections per server.", 0) \
     M(Bool, s3_truncate_on_insert, false, "Enables or disables truncate before insert in s3 engine tables.", 0) \
     M(Bool, s3_create_new_file_on_insert, false, "Enables or disables creating a new file on each insert in s3 engine tables", 0) \
+    M(Bool, enable_s3_requests_logging, false, "Enable very explicit logging of S3 requests. Makes sense for debug only.", 0) \
     M(UInt64, hdfs_replication, 0, "The actual number of replications can be specified when the hdfs file is created.", 0) \
     M(Bool, hdfs_truncate_on_insert, false, "Enables or disables truncate before insert in s3 engine tables", 0) \
     M(Bool, hdfs_create_new_file_on_insert, false, "Enables or disables creating a new file on each insert in hdfs engine tables", 0) \
@@ -566,7 +567,7 @@ static constexpr UInt64 operator""_GiB(unsigned long long value)
     \
     M(UInt64, remote_fs_read_max_backoff_ms, 10000, "Max wait time when trying to read data for remote disk", 0) \
     M(UInt64, remote_fs_read_backoff_max_tries, 5, "Max attempts to read with backoff", 0) \
-    M(Bool, enable_filesystem_cache, true, "Use cache for remote filesystem. This setting does not turn on/off cache for disks (must me done via disk config), but allows to bypass cache for some queries if intended", 0) \
+    M(Bool, enable_filesystem_cache, true, "Use cache for remote filesystem. This setting does not turn on/off cache for disks (must be done via disk config), but allows to bypass cache for some queries if intended", 0) \
     M(UInt64, filesystem_cache_max_wait_sec, 5, "Allow to wait at most this number of seconds for download of current remote_fs_buffer_size bytes, and skip cache if exceeded", 0) \
     M(Bool, enable_filesystem_cache_on_write_operations, false, "Write into cache on write operations. To actually work this setting requires be added to disk config too", 0) \
     M(Bool, enable_filesystem_cache_log, false, "Allows to record the filesystem caching log for each query", 0) \

src/Core/SettingsFields.cpp

@@ -9,6 +9,8 @@
 #include <IO/WriteHelpers.h>
 #include <boost/algorithm/string/predicate.hpp>
 
+#include <cmath>
+
 
 namespace DB
 {
@@ -16,6 +18,7 @@ namespace ErrorCodes
 {
     extern const int SIZE_OF_FIXED_STRING_DOESNT_MATCH;
     extern const int CANNOT_PARSE_BOOL;
+    extern const int CANNOT_PARSE_NUMBER;
 }
 
 
@@ -176,27 +179,75 @@ UInt64 SettingFieldMaxThreads::getAuto()
     return getNumberOfPhysicalCPUCores();
 }
 
+namespace
+{
+    Poco::Timespan::TimeDiff float64AsSecondsToTimespan(Float64 d)
+    {
+        if (d != 0.0 && !std::isnormal(d))
+            throw Exception(
+                ErrorCodes::CANNOT_PARSE_NUMBER, "A setting's value in seconds must be a normal floating point number or zero. Got {}", d);
+        return static_cast<Poco::Timespan::TimeDiff>(d * 1000000);
+    }
 
-template <SettingFieldTimespanUnit unit_>
-SettingFieldTimespan<unit_>::SettingFieldTimespan(const Field & f) : SettingFieldTimespan(fieldToNumber<UInt64>(f))
+}
+
+template <>
+SettingFieldSeconds::SettingFieldTimespan(const Field & f) : SettingFieldTimespan(float64AsSecondsToTimespan(fieldToNumber<Float64>(f)))
 {
 }
 
-template <SettingFieldTimespanUnit unit_>
-SettingFieldTimespan<unit_> & SettingFieldTimespan<unit_>::operator=(const Field & f)
+template <>
+SettingFieldMilliseconds::SettingFieldTimespan(const Field & f) : SettingFieldTimespan(fieldToNumber<UInt64>(f))
+{
+}
+
+template <>
+SettingFieldTimespan<SettingFieldTimespanUnit::Second> & SettingFieldSeconds::operator=(const Field & f)
+{
+    *this = Poco::Timespan{float64AsSecondsToTimespan(fieldToNumber<Float64>(f))};
+    return *this;
+}
+
+template <>
+SettingFieldTimespan<SettingFieldTimespanUnit::Millisecond> & SettingFieldMilliseconds::operator=(const Field & f)
 {
     *this = fieldToNumber<UInt64>(f);
     return *this;
 }
 
-template <SettingFieldTimespanUnit unit_>
-String SettingFieldTimespan<unit_>::toString() const
+template <>
+String SettingFieldSeconds::toString() const
+{
+    return ::DB::toString(static_cast<Float64>(value.totalMicroseconds()) / microseconds_per_unit);
+}
+
+template <>
+String SettingFieldMilliseconds::toString() const
 {
     return ::DB::toString(operator UInt64());
 }
 
-template <SettingFieldTimespanUnit unit_>
-void SettingFieldTimespan<unit_>::parseFromString(const String & str)
+template <>
+SettingFieldSeconds::operator Field() const
+{
+    return static_cast<Float64>(value.totalMicroseconds()) / microseconds_per_unit;
+}
+
+template <>
+SettingFieldMilliseconds::operator Field() const
+{
+    return operator UInt64();
+}
+
+template <>
+void SettingFieldSeconds::parseFromString(const String & str)
+{
+    Float64 n = parse<Float64>(str.data(), str.size());
+    *this = Poco::Timespan{static_cast<Poco::Timespan::TimeDiff>(n * microseconds_per_unit)};
+}
+
+template <>
+void SettingFieldMilliseconds::parseFromString(const String & str)
 {
     *this = stringToNumber<UInt64>(str);
 }
@@ -204,6 +255,13 @@ void SettingFieldTimespan<unit_>::parseFromString(const String & str)
 template <SettingFieldTimespanUnit unit_>
 void SettingFieldTimespan<unit_>::writeBinary(WriteBuffer & out) const
 {
+    /// Note that this returns an UInt64 (for both seconds and milliseconds units) for compatibility reasons as the value
+    /// for seconds used to be a integer (now a Float64)
+    /// This method is only used to communicate with clients or servers older than DBMS_MIN_REVISION_WITH_SETTINGS_SERIALIZED_AS_STRINGS
+    /// in which the value was passed as binary (as a UInt64)
+    /// Later versions pass the setting values as String (using toString() and parseFromString()) and there passing "1.2" will
+    /// lead to `1` on releases with integer seconds or `1.2` on more recent releases
+    /// See https://github.com/ClickHouse/ClickHouse/issues/36940 for more details
     auto num_units = operator UInt64();
     writeVarUInt(num_units, out);
 }

src/Core/SettingsFields.h

@@ -124,7 +124,7 @@ struct SettingFieldTimespan
     operator std::chrono::duration<Rep, Period>() const { return std::chrono::duration_cast<std::chrono::duration<Rep, Period>>(std::chrono::microseconds(value.totalMicroseconds())); } /// NOLINT
 
     explicit operator UInt64() const { return value.totalMicroseconds() / microseconds_per_unit; }
-    explicit operator Field() const { return operator UInt64(); }
+    explicit operator Field() const;
 
     Poco::Timespan::TimeDiff totalMicroseconds() const { return value.totalMicroseconds(); }
     Poco::Timespan::TimeDiff totalMilliseconds() const { return value.totalMilliseconds(); }

src/Daemon/BaseDaemon.cpp

@@ -68,6 +68,14 @@
 
 namespace fs = std::filesystem;
 
+namespace DB
+{
+    namespace ErrorCodes
+    {
+        extern const int CANNOT_SET_SIGNAL_HANDLER;
+    }
+}
+
 DB::PipeFDs signal_pipe;
 
 
@@ -76,7 +84,8 @@ DB::PipeFDs signal_pipe;
   */
 static void call_default_signal_handler(int sig)
 {
-    signal(sig, SIG_DFL);
+    if (SIG_ERR == signal(sig, SIG_DFL))
+        DB::throwFromErrno("Cannot set signal handler.", DB::ErrorCodes::CANNOT_SET_SIGNAL_HANDLER);
     raise(sig);
 }
 
@@ -498,9 +507,8 @@ BaseDaemon::~BaseDaemon()
     signal_listener_thread.join();
     /// Reset signals to SIG_DFL to avoid trying to write to the signal_pipe that will be closed after.
     for (int sig : handled_signals)
-    {
-        signal(sig, SIG_DFL);
-    }
+        if (SIG_ERR == signal(sig, SIG_DFL))
+            DB::throwFromErrno("Cannot set signal handler.", DB::ErrorCodes::CANNOT_SET_SIGNAL_HANDLER);
     signal_pipe.close();
 }
 

src/Databases/MySQL/DatabaseMySQL.cpp

@@ -89,7 +89,7 @@ bool DatabaseMySQL::empty() const
         return true;
 
     for (const auto & [table_name, storage_info] : local_tables_cache)
-        if (!remove_or_detach_tables.count(table_name))
+        if (!remove_or_detach_tables.contains(table_name))
             return false;
 
     return true;
@@ -103,7 +103,7 @@ DatabaseTablesIteratorPtr DatabaseMySQL::getTablesIterator(ContextPtr local_cont
     fetchTablesIntoLocalCache(local_context);
 
     for (const auto & [table_name, modify_time_and_storage] : local_tables_cache)
-        if (!remove_or_detach_tables.count(table_name) && (!filter_by_table_name || filter_by_table_name(table_name)))
+        if (!remove_or_detach_tables.contains(table_name) && (!filter_by_table_name || filter_by_table_name(table_name)))
             tables[table_name] = modify_time_and_storage.second;
 
     return std::make_unique<DatabaseTablesSnapshotIterator>(tables, database_name);
@@ -120,7 +120,7 @@ StoragePtr DatabaseMySQL::tryGetTable(const String & mysql_table_name, ContextPt
 
     fetchTablesIntoLocalCache(local_context);
 
-    if (!remove_or_detach_tables.count(mysql_table_name) && local_tables_cache.find(mysql_table_name) != local_tables_cache.end())
+    if (!remove_or_detach_tables.contains(mysql_table_name) && local_tables_cache.find(mysql_table_name) != local_tables_cache.end())
         return local_tables_cache[mysql_table_name].second;
 
     return StoragePtr{};
@@ -349,11 +349,11 @@ void DatabaseMySQL::attachTable(ContextPtr /* context_ */, const String & table_
 {
     std::lock_guard<std::mutex> lock{mutex};
 
-    if (!local_tables_cache.count(table_name))
+    if (!local_tables_cache.contains(table_name))
         throw Exception("Cannot attach table " + backQuoteIfNeed(database_name) + "." + backQuoteIfNeed(table_name) +
             " because it does not exist.", ErrorCodes::UNKNOWN_TABLE);
 
-    if (!remove_or_detach_tables.count(table_name))
+    if (!remove_or_detach_tables.contains(table_name))
         throw Exception("Cannot attach table " + backQuoteIfNeed(database_name) + "." + backQuoteIfNeed(table_name) +
             " because it already exists.", ErrorCodes::TABLE_ALREADY_EXISTS);
 
@@ -372,11 +372,11 @@ StoragePtr DatabaseMySQL::detachTable(ContextPtr /* context */, const String & t
 {
     std::lock_guard<std::mutex> lock{mutex};
 
-    if (remove_or_detach_tables.count(table_name))
+    if (remove_or_detach_tables.contains(table_name))
         throw Exception("Table " + backQuoteIfNeed(database_name) + "." + backQuoteIfNeed(table_name) + " is dropped",
             ErrorCodes::TABLE_IS_DROPPED);
 
-    if (!local_tables_cache.count(table_name))
+    if (!local_tables_cache.contains(table_name))
         throw Exception("Table " + backQuoteIfNeed(database_name) + "." + backQuoteIfNeed(table_name) + " doesn't exist.",
             ErrorCodes::UNKNOWN_TABLE);
 
@@ -412,7 +412,7 @@ void DatabaseMySQL::detachTablePermanently(ContextPtr, const String & table_name
 
     fs::path remove_flag = fs::path(getMetadataPath()) / (escapeForFileName(table_name) + suffix);
 
-    if (remove_or_detach_tables.count(table_name))
+    if (remove_or_detach_tables.contains(table_name))
         throw Exception(ErrorCodes::TABLE_IS_DROPPED, "Table {}.{} is dropped", backQuoteIfNeed(database_name), backQuoteIfNeed(table_name));
 
     if (fs::exists(remove_flag))

src/Dictionaries/FlatDictionary.cpp

@@ -643,7 +643,7 @@ void registerDictionaryFlat(DictionaryFactory & factory)
 
         const auto dict_id = StorageID::fromDictionaryConfig(config, config_prefix);
 
-        return std::make_unique<FlatDictionary>(dict_id, dict_struct, std::move(source_ptr), std::move(configuration));
+        return std::make_unique<FlatDictionary>(dict_id, dict_struct, std::move(source_ptr), configuration);
     };
 
     factory.registerLayout("flat", create_layout, false);

src/Dictionaries/MySQLDictionarySource.cpp

@@ -16,6 +16,13 @@
 #include <Storages/ExternalDataSourceConfiguration.h>
 #include <Storages/MySQL/MySQLHelpers.h>
 #include <Storages/MySQL/MySQLSettings.h>
+#include <Columns/ColumnString.h>
+#include <DataTypes/DataTypeString.h>
+#include <IO/WriteBufferFromString.h>
+#include <IO/WriteHelpers.h>
+#include <Common/LocalDateTime.h>
+#include <Common/logger_useful.h>
+#include "readInvalidateQuery.h"
 
 
 namespace DB
@@ -118,15 +125,6 @@ void registerDictionarySourceMysql(DictionarySourceFactory & factory)
 
 
 #if USE_MYSQL
-#    include <Columns/ColumnString.h>
-#    include <DataTypes/DataTypeString.h>
-#    include <IO/WriteBufferFromString.h>
-#    include <IO/WriteHelpers.h>
-#    include <Common/LocalDateTime.h>
-#    include <Common/logger_useful.h>
-#    include "readInvalidateQuery.h"
-#    include <mysqlxx/Exception.h>
-#    include <Core/Settings.h>
 
 namespace DB
 {

src/Dictionaries/RangeHashedDictionary.cpp

@@ -104,7 +104,7 @@ ColumnPtr RangeHashedDictionary<dictionary_key_type>::getColumn(
 
     /// Cast range column to storage type
     Columns modified_key_columns = key_columns;
-    auto range_storage_column = key_columns.back();
+    const ColumnPtr & range_storage_column = key_columns.back();
     ColumnWithTypeAndName column_to_cast = {range_storage_column->convertToFullColumnIfConst(), key_types.back(), ""};
     modified_key_columns.back() = castColumnAccurate(column_to_cast, dict_struct.range_min->type);
 
@@ -314,7 +314,7 @@ ColumnUInt8::Ptr RangeHashedDictionary<dictionary_key_type>::hasKeys(const Colum
     }
 
     /// Cast range column to storage type
-    auto range_storage_column = key_columns.back();
+    const ColumnPtr & range_storage_column = key_columns.back();
     ColumnWithTypeAndName column_to_cast = {range_storage_column->convertToFullColumnIfConst(), key_types.back(), ""};
     auto range_column_updated = castColumnAccurate(column_to_cast, dict_struct.range_min->type);
     auto key_columns_copy = key_columns;
@@ -513,7 +513,7 @@ void RangeHashedDictionary<dictionary_key_type>::getItemsImpl(
 
     size_t keys_found = 0;
 
-    auto range_column = key_columns.back();
+    const ColumnPtr & range_column = key_columns.back();
     auto key_columns_copy = key_columns;
     key_columns_copy.pop_back();
 
@@ -984,7 +984,7 @@ Pipe RangeHashedDictionary<dictionary_key_type>::read(const Names & column_names
         Columns result;
         result.reserve(attribute_names_size);
 
-        auto key_column = key_columns.back();
+        const ColumnPtr & key_column = key_columns.back();
 
         const auto * key_to_index_column = typeid_cast<const ColumnUInt64 *>(key_column.get());
         if (!key_to_index_column)

src/Disks/AzureBlobStorage/AzureBlobStorageAuth.h

@@ -1,18 +0,0 @@
-#pragma once
-
-#include <Common/config.h>
-
-#if USE_AZURE_BLOB_STORAGE
-
-#include <Disks/IDiskRemote.h>
-#include <azure/storage/blobs.hpp>
-
-namespace DB
-{
-
-std::shared_ptr<Azure::Storage::Blobs::BlobContainerClient> getAzureBlobContainerClient(
-    const Poco::Util::AbstractConfiguration & config, const String & config_prefix);
-
-}
-
-#endif

src/Disks/AzureBlobStorage/DiskAzureBlobStorage.cpp

@@ -1,168 +0,0 @@
-#include <Disks/AzureBlobStorage/DiskAzureBlobStorage.h>
-
-#if USE_AZURE_BLOB_STORAGE
-
-#include <Disks/RemoteDisksCommon.h>
-#include <Disks/IO/ReadBufferFromRemoteFSGather.h>
-#include <Disks/IO/AsynchronousReadIndirectBufferFromRemoteFS.h>
-#include <Disks/IO/ReadIndirectBufferFromRemoteFS.h>
-#include <Disks/IO/WriteIndirectBufferFromRemoteFS.h>
-#include <Common/getRandomASCIIString.h>
-
-
-namespace DB
-{
-
-namespace ErrorCodes
-{
-    extern const int AZURE_BLOB_STORAGE_ERROR;
-}
-
-
-DiskAzureBlobStorageSettings::DiskAzureBlobStorageSettings(
-    UInt64 max_single_part_upload_size_,
-    UInt64 min_bytes_for_seek_,
-    int max_single_read_retries_,
-    int max_single_download_retries_,
-    int thread_pool_size_) :
-    max_single_part_upload_size(max_single_part_upload_size_),
-    min_bytes_for_seek(min_bytes_for_seek_),
-    max_single_read_retries(max_single_read_retries_),
-    max_single_download_retries(max_single_download_retries_),
-    thread_pool_size(thread_pool_size_) {}
-
-
-DiskAzureBlobStorage::DiskAzureBlobStorage(
-    const String & name_,
-    DiskPtr metadata_disk_,
-    std::shared_ptr<Azure::Storage::Blobs::BlobContainerClient> blob_container_client_,
-    SettingsPtr settings_,
-    GetDiskSettings settings_getter_) :
-    IDiskRemote(name_, "", metadata_disk_, nullptr, "DiskAzureBlobStorage", settings_->thread_pool_size),
-    blob_container_client(blob_container_client_),
-    current_settings(std::move(settings_)),
-    settings_getter(settings_getter_) {}
-
-
-std::unique_ptr<ReadBufferFromFileBase> DiskAzureBlobStorage::readFile(
-    const String & path,
-    const ReadSettings & read_settings,
-    std::optional<size_t>,
-    std::optional<size_t>) const
-{
-    auto settings = current_settings.get();
-    auto metadata = readMetadata(path);
-
-    LOG_TEST(log, "Read from file by path: {}", backQuote(metadata_disk->getPath() + path));
-
-    auto reader_impl = std::make_unique<ReadBufferFromAzureBlobStorageGather>(
-        blob_container_client, metadata.remote_fs_root_path, metadata.remote_fs_objects,
-        settings->max_single_read_retries, settings->max_single_download_retries, read_settings);
-
-    if (read_settings.remote_fs_method == RemoteFSReadMethod::threadpool)
-    {
-        auto reader = getThreadPoolReader();
-        return std::make_unique<AsynchronousReadIndirectBufferFromRemoteFS>(reader, read_settings, std::move(reader_impl));
-    }
-    else
-    {
-        auto buf = std::make_unique<ReadIndirectBufferFromRemoteFS>(std::move(reader_impl));
-        return std::make_unique<SeekAvoidingReadBuffer>(std::move(buf), current_settings.get()->min_bytes_for_seek);
-    }
-}
-
-
-std::unique_ptr<WriteBufferFromFileBase> DiskAzureBlobStorage::writeFile(
-    const String & path,
-    size_t buf_size,
-    WriteMode mode,
-    const WriteSettings &)
-{
-    auto blob_path = path + "_" + getRandomASCIIString(8); /// NOTE: path contains the tmp_* prefix in the blob name
-
-    LOG_TRACE(log, "{} to file by path: {}. AzureBlob Storage path: {}",
-        mode == WriteMode::Rewrite ? "Write" : "Append", backQuote(metadata_disk->getPath() + path), blob_path);
-
-    auto buffer = std::make_unique<WriteBufferFromAzureBlobStorage>(
-        blob_container_client,
-        blob_path,
-        current_settings.get()->max_single_part_upload_size,
-        buf_size);
-
-    auto create_metadata_callback = [this, path, mode, blob_path] (size_t count)
-    {
-        readOrCreateUpdateAndStoreMetadata(path, mode, false, [blob_path, count] (Metadata & metadata) { metadata.addObject(blob_path, count); return true; });
-    };
-
-    return std::make_unique<WriteIndirectBufferFromRemoteFS>(std::move(buffer), std::move(create_metadata_callback), blob_path);
-}
-
-
-DiskType DiskAzureBlobStorage::getType() const
-{
-    return DiskType::AzureBlobStorage;
-}
-
-
-bool DiskAzureBlobStorage::isRemote() const
-{
-    return true;
-}
-
-
-bool DiskAzureBlobStorage::supportZeroCopyReplication() const
-{
-    return true;
-}
-
-
-bool DiskAzureBlobStorage::checkUniqueId(const String & id) const
-{
-    Azure::Storage::Blobs::ListBlobsOptions blobs_list_options;
-    blobs_list_options.Prefix = id;
-    blobs_list_options.PageSizeHint = 1;
-
-    auto blobs_list_response = blob_container_client->ListBlobs(blobs_list_options);
-    auto blobs_list = blobs_list_response.Blobs;
-
-    for (const auto & blob : blobs_list)
-    {
-        if (id == blob.Name)
-            return true;
-    }
-
-    return false;
-}
-
-
-void DiskAzureBlobStorage::removeFromRemoteFS(const std::vector<String> & paths)
-{
-    for (const auto & path : paths)
-    {
-        try
-        {
-            auto delete_info = blob_container_client->DeleteBlob(path);
-            if (!delete_info.Value.Deleted)
-                throw Exception(ErrorCodes::AZURE_BLOB_STORAGE_ERROR, "Failed to delete file in AzureBlob Storage: {}", path);
-        }
-        catch (const Azure::Storage::StorageException & e)
-        {
-            LOG_INFO(log, "Caught an error while deleting file {} : {}", path, e.Message);
-            throw;
-        }
-    }
-}
-
-void DiskAzureBlobStorage::applyNewSettings(const Poco::Util::AbstractConfiguration & config, ContextPtr context, const String &, const DisksMap &)
-{
-    auto new_settings = settings_getter(config, "storage_configuration.disks." + name, context);
-
-    current_settings.set(std::move(new_settings));
-
-    if (AsyncExecutor * exec = dynamic_cast<AsyncExecutor*>(&getExecutor()))
-        exec->setMaxThreads(current_settings.get()->thread_pool_size);
-}
-
-}
-
-#endif

src/Disks/AzureBlobStorage/DiskAzureBlobStorage.h

@@ -1,86 +0,0 @@
-#pragma once
-
-#include <Common/config.h>
-
-#if USE_AZURE_BLOB_STORAGE
-
-#include <Disks/IDiskRemote.h>
-#include <IO/ReadBufferFromAzureBlobStorage.h>
-#include <IO/WriteBufferFromAzureBlobStorage.h>
-#include <IO/SeekAvoidingReadBuffer.h>
-
-#include <azure/identity/managed_identity_credential.hpp>
-#include <azure/storage/blobs.hpp>
-
-
-namespace DB
-{
-
-struct DiskAzureBlobStorageSettings final
-{
-    DiskAzureBlobStorageSettings(
-        UInt64 max_single_part_upload_size_,
-        UInt64 min_bytes_for_seek_,
-        int max_single_read_retries,
-        int max_single_download_retries,
-        int thread_pool_size_);
-
-    size_t max_single_part_upload_size; /// NOTE: on 32-bit machines it will be at most 4GB, but size_t is also used in BufferBase for offset
-    UInt64 min_bytes_for_seek;
-    size_t max_single_read_retries;
-    size_t max_single_download_retries;
-    size_t thread_pool_size;
-};
-
-
-class DiskAzureBlobStorage final : public IDiskRemote
-{
-public:
-
-    using SettingsPtr = std::unique_ptr<DiskAzureBlobStorageSettings>;
-    using GetDiskSettings = std::function<SettingsPtr(const Poco::Util::AbstractConfiguration &, const String, ContextPtr)>;
-
-    DiskAzureBlobStorage(
-        const String & name_,
-        DiskPtr metadata_disk_,
-        std::shared_ptr<Azure::Storage::Blobs::BlobContainerClient> blob_container_client_,
-        SettingsPtr settings_,
-        GetDiskSettings settings_getter_);
-
-    std::unique_ptr<ReadBufferFromFileBase> readFile(
-        const String & path,
-        const ReadSettings & settings,
-        std::optional<size_t> read_hint,
-        std::optional<size_t> file_size) const override;
-
-    std::unique_ptr<WriteBufferFromFileBase> writeFile(
-        const String & path,
-        size_t buf_size,
-        WriteMode mode,
-        const WriteSettings & settings) override;
-
-    DiskType getType() const override;
-
-    bool isRemote() const override;
-
-    bool supportZeroCopyReplication() const override;
-
-    bool checkUniqueId(const String & id) const override;
-
-    void removeFromRemoteFS(const std::vector<String> & paths) override;
-
-    void applyNewSettings(const Poco::Util::AbstractConfiguration & config, ContextPtr context, const String &, const DisksMap &) override;
-
-private:
-
-    /// client used to access the files in the Blob Storage cloud
-    std::shared_ptr<Azure::Storage::Blobs::BlobContainerClient> blob_container_client;
-
-    MultiVersion<DiskAzureBlobStorageSettings> current_settings;
-    /// Gets disk settings from context.
-    GetDiskSettings settings_getter;
-};
-
-}
-
-#endif

src/Disks/DiskDecorator.cpp

@@ -211,9 +211,9 @@ void DiskDecorator::shutdown()
     delegate->shutdown();
 }
 
-void DiskDecorator::startup()
+void DiskDecorator::startup(ContextPtr context)
 {
-    delegate->startup();
+    delegate->startup(context);
 }
 
 void DiskDecorator::applyNewSettings(const Poco::Util::AbstractConfiguration & config, ContextPtr context, const String & config_prefix, const DisksMap & map)

src/Disks/DiskDecorator.h

@@ -71,7 +71,7 @@ public:
     void onFreeze(const String & path) override;
     SyncGuardPtr getDirectorySyncGuard(const String & path) const override;
     void shutdown() override;
-    void startup() override;
+    void startup(ContextPtr context) override;
     void applyNewSettings(const Poco::Util::AbstractConfiguration & config, ContextPtr context, const String & config_prefix, const DisksMap & map) override;
     String getCacheBasePath() const override { return delegate->getCacheBasePath(); }
     std::vector<String> getRemotePaths(const String & path) const override { return delegate->getRemotePaths(path); }

src/Disks/DiskLocal.cpp

@@ -494,7 +494,7 @@ DiskLocal::DiskLocal(
         disk_checker = std::make_unique<DiskLocalCheckThread>(this, context, local_disk_check_period_ms);
 }
 
-void DiskLocal::startup()
+void DiskLocal::startup(ContextPtr)
 {
     try
     {
@@ -682,7 +682,7 @@ void registerDiskLocal(DiskFactory & factory)
 
         std::shared_ptr<IDisk> disk
             = std::make_shared<DiskLocal>(name, path, keep_free_space_bytes, context, config.getUInt("local_disk_check_period_ms", 0));
-        disk->startup();
+        disk->startup(context);
         return std::make_shared<DiskRestartProxy>(disk);
     };
     factory.registerDiskType("local", creator);

src/Disks/DiskLocal.h

@@ -110,7 +110,7 @@ public:
 
     bool isBroken() const override { return broken; }
 
-    void startup() override;
+    void startup(ContextPtr) override;
 
     void shutdown() override;
 

src/Disks/DiskRestartProxy.cpp

@@ -5,6 +5,7 @@
 
 namespace DB
 {
+
 namespace ErrorCodes
 {
     extern const int DEADLOCK_AVOIDED;
@@ -329,7 +330,7 @@ void DiskRestartProxy::getRemotePathsRecursive(const String & path, std::vector<
     return DiskDecorator::getRemotePathsRecursive(path, paths_map);
 }
 
-void DiskRestartProxy::restart()
+void DiskRestartProxy::restart(ContextPtr context)
 {
     /// Speed up processing unhealthy requests.
     DiskDecorator::shutdown();
@@ -352,7 +353,7 @@ void DiskRestartProxy::restart()
 
     LOG_INFO(log, "Restart lock acquired. Restarting disk {}", DiskDecorator::getName());
 
-    DiskDecorator::startup();
+    DiskDecorator::startup(context);
 
     LOG_INFO(log, "Disk restarted {}", DiskDecorator::getName());
 }

src/Disks/DiskRestartProxy.h

@@ -68,7 +68,7 @@ public:
     std::vector<String> getRemotePaths(const String & path) const override;
     void getRemotePathsRecursive(const String & path, std::vector<LocalPathWithRemotePaths> & paths_map) override;
 
-    void restart();
+    void restart(ContextPtr context);
 
 private:
     friend class RestartAwareReadBuffer;

src/Disks/DiskWebServer.cpp

@@ -9,8 +9,12 @@
 #include <IO/ReadHelpers.h>
 #include <IO/WriteHelpers.h>
 
-#include <Disks/IDiskRemote.h>
+#include <Disks/IDisk.h>
+#include <Disks/ObjectStorages/IObjectStorage.h>
+#include <IO/ReadBufferFromFile.h>
+
 #include <Disks/IO/AsynchronousReadIndirectBufferFromRemoteFS.h>
+
 #include <Disks/IO/ReadIndirectBufferFromRemoteFS.h>
 #include <Disks/IO/WriteIndirectBufferFromRemoteFS.h>
 #include <Disks/IO/ReadBufferFromRemoteFSGather.h>
@@ -173,7 +177,7 @@ std::unique_ptr<ReadBufferFromFileBase> DiskWebServer::readFile(const String & p
 
     if (read_settings.remote_fs_method == RemoteFSReadMethod::threadpool)
     {
-        auto reader = IDiskRemote::getThreadPoolReader();
+        auto reader = IObjectStorage::getThreadPoolReader();
         return std::make_unique<AsynchronousReadIndirectBufferFromRemoteFS>(reader, read_settings, std::move(web_impl), min_bytes_for_seek);
     }
     else

src/Disks/DiskWebServer.h

@@ -1,10 +1,13 @@
 #pragma once
 
-#include <Disks/IDiskRemote.h>
 #include <IO/WriteBufferFromFile.h>
 #include <Core/UUID.h>
 #include <set>
 
+#include <Interpreters/Context_fwd.h>
+#include <Disks/IDisk.h>
+#include <IO/ReadBufferFromFile.h>
+
 
 namespace DB
 {

src/Disks/HDFS/DiskHDFS.cpp

@@ -1,143 +0,0 @@
-#include <Disks/HDFS/DiskHDFS.h>
-
-#if USE_HDFS
-
-#include <Disks/DiskLocal.h>
-#include <Disks/RemoteDisksCommon.h>
-
-#include <IO/SeekAvoidingReadBuffer.h>
-#include <Storages/HDFS/WriteBufferFromHDFS.h>
-#include <Storages/HDFS/HDFSCommon.h>
-
-#include <Disks/IO/AsynchronousReadIndirectBufferFromRemoteFS.h>
-#include <Disks/IO/ReadIndirectBufferFromRemoteFS.h>
-#include <Disks/IO/WriteIndirectBufferFromRemoteFS.h>
-#include <Disks/IO/ReadBufferFromRemoteFSGather.h>
-
-#include <boost/algorithm/string/predicate.hpp>
-
-#include <Common/logger_useful.h>
-#include <base/FnTraits.h>
-
-
-namespace DB
-{
-
-namespace ErrorCodes
-{
-    extern const int BAD_ARGUMENTS;
-    extern const int LOGICAL_ERROR;
-}
-
-
-DiskHDFS::DiskHDFS(
-    const String & disk_name_,
-    const String & hdfs_root_path_,
-    SettingsPtr settings_,
-    DiskPtr metadata_disk_,
-    const Poco::Util::AbstractConfiguration & config_)
-    : IDiskRemote(disk_name_, hdfs_root_path_, metadata_disk_, nullptr, "DiskHDFS", settings_->thread_pool_size)
-    , config(config_)
-    , hdfs_builder(createHDFSBuilder(hdfs_root_path_, config))
-    , hdfs_fs(createHDFSFS(hdfs_builder.get()))
-    , settings(std::move(settings_))
-{
-}
-
-
-std::unique_ptr<ReadBufferFromFileBase> DiskHDFS::readFile(const String & path, const ReadSettings & read_settings, std::optional<size_t>, std::optional<size_t>) const
-{
-    auto metadata = readMetadata(path);
-
-    LOG_TEST(log,
-        "Read from file by path: {}. Existing HDFS objects: {}",
-        backQuote(metadata_disk->getPath() + path), metadata.remote_fs_objects.size());
-
-    auto hdfs_impl = std::make_unique<ReadBufferFromHDFSGather>(config, remote_fs_root_path, remote_fs_root_path, metadata.remote_fs_objects, read_settings);
-    auto buf = std::make_unique<ReadIndirectBufferFromRemoteFS>(std::move(hdfs_impl));
-    return std::make_unique<SeekAvoidingReadBuffer>(std::move(buf), settings->min_bytes_for_seek);
-}
-
-
-std::unique_ptr<WriteBufferFromFileBase> DiskHDFS::writeFile(const String & path, size_t buf_size, WriteMode mode, const WriteSettings &)
-{
-    /// Path to store new HDFS object.
-    std::string file_name = getRandomName();
-    std::string hdfs_path = fs::path(remote_fs_root_path) / file_name;
-
-    LOG_TRACE(log, "{} to file by path: {}. HDFS path: {}", mode == WriteMode::Rewrite ? "Write" : "Append",
-              backQuote(metadata_disk->getPath() + path), hdfs_path);
-
-    /// Single O_WRONLY in libhdfs adds O_TRUNC
-    auto hdfs_buffer = std::make_unique<WriteBufferFromHDFS>(hdfs_path,
-                                                             config, settings->replication, buf_size,
-                                                             mode == WriteMode::Rewrite ? O_WRONLY :  O_WRONLY | O_APPEND);
-    auto create_metadata_callback = [this, path, mode, file_name] (size_t count)
-    {
-        readOrCreateUpdateAndStoreMetadata(path, mode, false, [file_name, count] (Metadata & metadata) { metadata.addObject(file_name, count); return true; });
-    };
-
-    return std::make_unique<WriteIndirectBufferFromRemoteFS>(std::move(hdfs_buffer), std::move(create_metadata_callback), hdfs_path);
-}
-
-void DiskHDFS::removeFromRemoteFS(const std::vector<String> & paths)
-{
-    for (const auto & hdfs_path : paths)
-    {
-        const size_t begin_of_path = hdfs_path.find('/', hdfs_path.find("//") + 2);
-
-        /// Add path from root to file name
-        int res = hdfsDelete(hdfs_fs.get(), hdfs_path.substr(begin_of_path).c_str(), 0);
-        if (res == -1)
-            throw Exception(ErrorCodes::LOGICAL_ERROR, "HDFSDelete failed with path: " + hdfs_path);
-    }
-}
-
-bool DiskHDFS::checkUniqueId(const String & hdfs_uri) const
-{
-    if (!boost::algorithm::starts_with(hdfs_uri, remote_fs_root_path))
-        return false;
-    const size_t begin_of_path = hdfs_uri.find('/', hdfs_uri.find("//") + 2);
-    const String remote_fs_object_path = hdfs_uri.substr(begin_of_path);
-    return (0 == hdfsExists(hdfs_fs.get(), remote_fs_object_path.c_str()));
-}
-
-namespace
-{
-std::unique_ptr<DiskHDFSSettings> getSettings(const Poco::Util::AbstractConfiguration & config, const String & config_prefix, const Settings & settings)
-{
-    return std::make_unique<DiskHDFSSettings>(
-        config.getUInt64(config_prefix + ".min_bytes_for_seek", 1024 * 1024),
-        config.getInt(config_prefix + ".thread_pool_size", 16),
-        config.getInt(config_prefix + ".objects_chunk_size_to_delete", 1000),
-        settings.hdfs_replication);
-}
-}
-
-void registerDiskHDFS(DiskFactory & factory)
-{
-    auto creator = [](const String & name,
-                      const Poco::Util::AbstractConfiguration & config,
-                      const String & config_prefix,
-                      ContextPtr context_,
-                      const DisksMap & /*map*/) -> DiskPtr
-    {
-        String uri{config.getString(config_prefix + ".endpoint")};
-        checkHDFSURL(uri);
-
-        if (uri.back() != '/')
-            throw Exception(ErrorCodes::BAD_ARGUMENTS, "HDFS path must ends with '/', but '{}' doesn't.", uri);
-
-        auto metadata_disk = prepareForLocalMetadata(name, config, config_prefix, context_).second;
-
-        return std::make_shared<DiskHDFS>(
-            name, uri,
-            getSettings(config, config_prefix, context_->getSettingsRef()),
-            metadata_disk, config);
-    };
-
-    factory.registerDiskType("hdfs", creator);
-}
-
-}
-#endif

src/Disks/HDFS/DiskHDFS.h

@@ -1,84 +0,0 @@
-#pragma once
-
-#include <Common/config.h>
-
-#if USE_HDFS
-
-#include <Disks/IDiskRemote.h>
-#include <Storages/HDFS/HDFSCommon.h>
-#include <Core/UUID.h>
-#include <memory>
-
-
-namespace DB
-{
-
-struct DiskHDFSSettings
-{
-    size_t min_bytes_for_seek;
-    int thread_pool_size;
-    int objects_chunk_size_to_delete;
-    int replication;
-
-    DiskHDFSSettings(
-            int min_bytes_for_seek_,
-            int thread_pool_size_,
-            int objects_chunk_size_to_delete_,
-            int replication_)
-        : min_bytes_for_seek(min_bytes_for_seek_)
-        , thread_pool_size(thread_pool_size_)
-        , objects_chunk_size_to_delete(objects_chunk_size_to_delete_)
-        , replication(replication_) {}
-};
-
-
-/**
- * Storage for persisting data in HDFS and metadata on the local disk.
- * Files are represented by file in local filesystem (clickhouse_root/disks/disk_name/path/to/file)
- * that contains HDFS object key with actual data.
- */
-class DiskHDFS final : public IDiskRemote
-{
-public:
-    using SettingsPtr = std::unique_ptr<DiskHDFSSettings>;
-
-    DiskHDFS(
-        const String & disk_name_,
-        const String & hdfs_root_path_,
-        SettingsPtr settings_,
-        DiskPtr metadata_disk_,
-        const Poco::Util::AbstractConfiguration & config_);
-
-    DiskType getType() const override { return DiskType::HDFS; }
-    bool isRemote() const override { return true; }
-
-    bool supportZeroCopyReplication() const override { return true; }
-
-    std::unique_ptr<ReadBufferFromFileBase> readFile(
-        const String & path,
-        const ReadSettings & settings,
-        std::optional<size_t> read_hint,
-        std::optional<size_t> file_size) const override;
-
-    std::unique_ptr<WriteBufferFromFileBase> writeFile(const String & path, size_t buf_size, WriteMode mode, const WriteSettings & settings) override;
-
-    void removeFromRemoteFS(const std::vector<String> & paths) override;
-
-    /// Check file exists and ClickHouse has an access to it
-    /// Overrode in remote disk
-    /// Required for remote disk to ensure that replica has access to data written by other node
-    bool checkUniqueId(const String & hdfs_uri) const override;
-
-private:
-    String getRandomName() { return toString(UUIDHelpers::generateV4()); }
-
-    const Poco::Util::AbstractConfiguration & config;
-
-    HDFSBuilderWrapper hdfs_builder;
-    HDFSFSPtr hdfs_fs;
-
-    SettingsPtr settings;
-};
-
-}
-#endif

src/Disks/IDisk.h

@@ -10,6 +10,8 @@
 #include <Disks/DiskType.h>
 #include <IO/ReadSettings.h>
 #include <IO/WriteSettings.h>
+#include <Disks/ObjectStorages/IObjectStorage.h>
+#include <Disks/WriteMode.h>
 
 #include <memory>
 #include <mutex>
@@ -48,14 +50,6 @@ class ReadBufferFromFileBase;
 class WriteBufferFromFileBase;
 class MMappedFileCache;
 
-/**
- * Mode of opening a file for write.
- */
-enum class WriteMode
-{
-    Rewrite,
-    Append
-};
 
 /**
  * Provide interface for reservation.
@@ -289,14 +283,14 @@ public:
 
     virtual bool isReadOnly() const { return false; }
 
-    /// Check if disk is broken. Broken disks will have 0 space and not be used.
+    /// Check if disk is broken. Broken disks will have 0 space and cannot be used.
     virtual bool isBroken() const { return false; }
 
     /// Invoked when Global Context is shutdown.
     virtual void shutdown() {}
 
     /// Performs action on disk startup.
-    virtual void startup() {}
+    virtual void startup(ContextPtr) {}
 
     /// Return some uniq string for file, overrode for IDiskRemote
     /// Required for distinguish different copies of the same part on remote disk

src/Disks/IDiskRemote.cpp

@@ -1,708 +0,0 @@
-#include <Disks/IDiskRemote.h>
-
-#include "Disks/DiskFactory.h"
-#include <IO/ReadBufferFromFile.h>
-#include <IO/ReadHelpers.h>
-#include <IO/WriteBufferFromFile.h>
-#include <IO/WriteHelpers.h>
-#include <Common/createHardLink.h>
-#include <Common/quoteString.h>
-#include <Common/logger_useful.h>
-#include <Common/checkStackSize.h>
-#include <boost/algorithm/string.hpp>
-#include <Common/filesystemHelpers.h>
-#include <Disks/IO/ThreadPoolRemoteFSReader.h>
-#include <Common/FileCache.h>
-
-
-namespace DB
-{
-
-namespace ErrorCodes
-{
-    extern const int INCORRECT_DISK_INDEX;
-    extern const int UNKNOWN_FORMAT;
-    extern const int FILE_ALREADY_EXISTS;
-    extern const int PATH_ACCESS_DENIED;
-    extern const int FILE_DOESNT_EXIST;
-    extern const int BAD_FILE_TYPE;
-}
-
-
-IDiskRemote::Metadata IDiskRemote::Metadata::readMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_)
-{
-    Metadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
-    result.load();
-    return result;
-}
-
-
-IDiskRemote::Metadata IDiskRemote::Metadata::createAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync)
-{
-    Metadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
-    result.save(sync);
-    return result;
-}
-
-IDiskRemote::Metadata IDiskRemote::Metadata::readUpdateAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, IDiskRemote::MetadataUpdater updater)
-{
-    Metadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
-    result.load();
-    if (updater(result))
-        result.save(sync);
-    return result;
-}
-
-IDiskRemote::Metadata IDiskRemote::Metadata::createUpdateAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, IDiskRemote::MetadataUpdater updater)
-{
-    Metadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
-    updater(result);
-    result.save(sync);
-    return result;
-}
-
-IDiskRemote::Metadata IDiskRemote::Metadata::readUpdateStoreMetadataAndRemove(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, IDiskRemote::MetadataUpdater updater)
-{
-    Metadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
-    result.load();
-    if (updater(result))
-        result.save(sync);
-    metadata_disk_->removeFile(metadata_file_path_);
-
-    return result;
-
-}
-
-IDiskRemote::Metadata IDiskRemote::Metadata::createAndStoreMetadataIfNotExists(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, bool overwrite)
-{
-    if (overwrite || !metadata_disk_->exists(metadata_file_path_))
-    {
-        return createAndStoreMetadata(remote_fs_root_path_, metadata_disk_, metadata_file_path_, sync);
-    }
-    else
-    {
-        auto result = readMetadata(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
-        if (result.read_only)
-            throw Exception("File is read-only: " + metadata_file_path_, ErrorCodes::PATH_ACCESS_DENIED);
-        return result;
-    }
-}
-
-void IDiskRemote::Metadata::load()
-{
-    const ReadSettings read_settings;
-    auto buf = metadata_disk->readFile(metadata_file_path, read_settings, 1024);  /* reasonable buffer size for small file */
-
-    UInt32 version;
-    readIntText(version, *buf);
-
-    if (version < VERSION_ABSOLUTE_PATHS || version > VERSION_READ_ONLY_FLAG)
-        throw Exception(
-            ErrorCodes::UNKNOWN_FORMAT,
-            "Unknown metadata file version. Path: {}. Version: {}. Maximum expected version: {}",
-            metadata_disk->getPath() + metadata_file_path, toString(version), toString(VERSION_READ_ONLY_FLAG));
-
-    assertChar('\n', *buf);
-
-    UInt32 remote_fs_objects_count;
-    readIntText(remote_fs_objects_count, *buf);
-    assertChar('\t', *buf);
-    readIntText(total_size, *buf);
-    assertChar('\n', *buf);
-    remote_fs_objects.resize(remote_fs_objects_count);
-
-    for (size_t i = 0; i < remote_fs_objects_count; ++i)
-    {
-        String remote_fs_object_path;
-        size_t remote_fs_object_size;
-        readIntText(remote_fs_object_size, *buf);
-        assertChar('\t', *buf);
-        readEscapedString(remote_fs_object_path, *buf);
-        if (version == VERSION_ABSOLUTE_PATHS)
-        {
-            if (!remote_fs_object_path.starts_with(remote_fs_root_path))
-                throw Exception(ErrorCodes::UNKNOWN_FORMAT,
-                    "Path in metadata does not correspond to root path. Path: {}, root path: {}, disk path: {}",
-                    remote_fs_object_path, remote_fs_root_path, metadata_disk->getPath());
-
-            remote_fs_object_path = remote_fs_object_path.substr(remote_fs_root_path.size());
-        }
-        assertChar('\n', *buf);
-        remote_fs_objects[i].relative_path = remote_fs_object_path;
-        remote_fs_objects[i].bytes_size = remote_fs_object_size;
-    }
-
-    readIntText(ref_count, *buf);
-    assertChar('\n', *buf);
-
-    if (version >= VERSION_READ_ONLY_FLAG)
-    {
-        readBoolText(read_only, *buf);
-        assertChar('\n', *buf);
-    }
-}
-
-/// Load metadata by path or create empty if `create` flag is set.
-IDiskRemote::Metadata::Metadata(
-        const String & remote_fs_root_path_,
-        DiskPtr metadata_disk_,
-        const String & metadata_file_path_)
-    : remote_fs_root_path(remote_fs_root_path_)
-    , metadata_file_path(metadata_file_path_)
-    , metadata_disk(metadata_disk_)
-{
-}
-
-void IDiskRemote::Metadata::addObject(const String & path, size_t size)
-{
-    total_size += size;
-    remote_fs_objects.emplace_back(path, size);
-}
-
-
-void IDiskRemote::Metadata::saveToBuffer(WriteBuffer & buf, bool sync)
-{
-    writeIntText(VERSION_RELATIVE_PATHS, buf);
-    writeChar('\n', buf);
-
-    writeIntText(remote_fs_objects.size(), buf);
-    writeChar('\t', buf);
-    writeIntText(total_size, buf);
-    writeChar('\n', buf);
-
-    for (const auto & [remote_fs_object_path, remote_fs_object_size] : remote_fs_objects)
-    {
-        writeIntText(remote_fs_object_size, buf);
-        writeChar('\t', buf);
-        writeEscapedString(remote_fs_object_path, buf);
-        writeChar('\n', buf);
-    }
-
-    writeIntText(ref_count, buf);
-    writeChar('\n', buf);
-
-    writeBoolText(read_only, buf);
-    writeChar('\n', buf);
-
-    buf.finalize();
-    if (sync)
-        buf.sync();
-
-}
-
-/// Fsync metadata file if 'sync' flag is set.
-void IDiskRemote::Metadata::save(bool sync)
-{
-    auto buf = metadata_disk->writeFile(metadata_file_path, 1024);
-    saveToBuffer(*buf, sync);
-}
-
-std::string IDiskRemote::Metadata::serializeToString()
-{
-    WriteBufferFromOwnString write_buf;
-    saveToBuffer(write_buf, false);
-    return write_buf.str();
-}
-
-IDiskRemote::Metadata IDiskRemote::readMetadataUnlocked(const String & path, std::shared_lock<std::shared_mutex> &) const
-{
-    return Metadata::readMetadata(remote_fs_root_path, metadata_disk, path);
-}
-
-
-IDiskRemote::Metadata IDiskRemote::readMetadata(const String & path) const
-{
-    std::shared_lock lock(metadata_mutex);
-    return readMetadataUnlocked(path, lock);
-}
-
-IDiskRemote::Metadata IDiskRemote::readUpdateAndStoreMetadata(const String & path, bool sync, IDiskRemote::MetadataUpdater updater)
-{
-    std::unique_lock lock(metadata_mutex);
-    return Metadata::readUpdateAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync, updater);
-}
-
-
-IDiskRemote::Metadata IDiskRemote::readUpdateStoreMetadataAndRemove(const String & path, bool sync, IDiskRemote::MetadataUpdater updater)
-{
-    std::unique_lock lock(metadata_mutex);
-    return Metadata::readUpdateStoreMetadataAndRemove(remote_fs_root_path, metadata_disk, path, sync, updater);
-}
-
-IDiskRemote::Metadata IDiskRemote::readOrCreateUpdateAndStoreMetadata(const String & path, WriteMode mode, bool sync, IDiskRemote::MetadataUpdater updater)
-{
-    if (mode == WriteMode::Rewrite || !metadata_disk->exists(path))
-    {
-        std::unique_lock lock(metadata_mutex);
-        return Metadata::createUpdateAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync, updater);
-    }
-    else
-    {
-        return Metadata::readUpdateAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync, updater);
-    }
-}
-
-IDiskRemote::Metadata IDiskRemote::createAndStoreMetadata(const String & path, bool sync)
-{
-    return Metadata::createAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync);
-}
-
-IDiskRemote::Metadata IDiskRemote::createUpdateAndStoreMetadata(const String & path, bool sync, IDiskRemote::MetadataUpdater updater)
-{
-    return Metadata::createUpdateAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync, updater);
-}
-
-
-std::unordered_map<String, String> IDiskRemote::getSerializedMetadata(const std::vector<std::string> & file_paths) const
-{
-    std::unordered_map<String, String> metadatas;
-
-    std::shared_lock lock(metadata_mutex);
-
-    for (const auto & path : file_paths)
-    {
-        IDiskRemote::Metadata metadata = readMetadataUnlocked(path, lock);
-        metadata.ref_count = 0;
-        metadatas[path] = metadata.serializeToString();
-    }
-
-    return metadatas;
-}
-
-void IDiskRemote::removeMetadata(const String & path, std::vector<String> & paths_to_remove)
-{
-    LOG_TRACE(log, "Remove file by path: {}", backQuote(metadata_disk->getPath() + path));
-
-    if (!metadata_disk->exists(path))
-        throw Exception(ErrorCodes::FILE_DOESNT_EXIST, "Metadata path '{}' doesn't exist", path);
-
-    if (!metadata_disk->isFile(path))
-        throw Exception(ErrorCodes::BAD_FILE_TYPE, "Path '{}' is not a regular file", path);
-
-    try
-    {
-        auto metadata_updater = [&paths_to_remove, this] (Metadata & metadata)
-        {
-            if (metadata.ref_count == 0)
-            {
-                for (const auto & [remote_fs_object_path, _] : metadata.remote_fs_objects)
-                {
-
-                    paths_to_remove.push_back(remote_fs_root_path + remote_fs_object_path);
-
-                    if (cache)
-                    {
-                        auto key = cache->hash(remote_fs_object_path);
-                        cache->remove(key);
-                    }
-                }
-
-                return false;
-            }
-            else /// In other case decrement number of references, save metadata and delete hardlink.
-            {
-                --metadata.ref_count;
-            }
-
-            return true;
-        };
-
-        readUpdateStoreMetadataAndRemove(path, false, metadata_updater);
-        /// If there is no references - delete content from remote FS.
-    }
-    catch (const Exception & e)
-    {
-        /// If it's impossible to read meta - just remove it from FS.
-        if (e.code() == ErrorCodes::UNKNOWN_FORMAT)
-        {
-            LOG_WARNING(log,
-                "Metadata file {} can't be read by reason: {}. Removing it forcibly.",
-                backQuote(path), e.nested() ? e.nested()->message() : e.message());
-            metadata_disk->removeFile(path);
-        }
-        else
-            throw;
-    }
-}
-
-
-void IDiskRemote::removeMetadataRecursive(const String & path, std::unordered_map<String, std::vector<String>> & paths_to_remove)
-{
-    checkStackSize(); /// This is needed to prevent stack overflow in case of cyclic symlinks.
-
-    if (metadata_disk->isFile(path))
-    {
-        removeMetadata(path, paths_to_remove[path]);
-    }
-    else
-    {
-        for (auto it = iterateDirectory(path); it->isValid(); it->next())
-            removeMetadataRecursive(it->path(), paths_to_remove);
-
-        metadata_disk->removeDirectory(path);
-    }
-}
-
-std::vector<String> IDiskRemote::getRemotePaths(const String & local_path) const
-{
-    auto metadata = readMetadata(local_path);
-
-    std::vector<String> remote_paths;
-    for (const auto & [remote_path, _] : metadata.remote_fs_objects)
-        remote_paths.push_back(fs::path(metadata.remote_fs_root_path) / remote_path);
-
-    return remote_paths;
-}
-
-void IDiskRemote::getRemotePathsRecursive(const String & local_path, std::vector<LocalPathWithRemotePaths> & paths_map)
-{
-    /// Protect against concurrent delition of files (for example because of a merge).
-    if (metadata_disk->isFile(local_path))
-    {
-        try
-        {
-            paths_map.emplace_back(local_path, getRemotePaths(local_path));
-        }
-        catch (const Exception & e)
-        {
-            if (e.code() == ErrorCodes::FILE_DOESNT_EXIST)
-                return;
-            throw;
-        }
-    }
-    else
-    {
-        DiskDirectoryIteratorPtr it;
-        try
-        {
-            it = iterateDirectory(local_path);
-        }
-        catch (const fs::filesystem_error & e)
-        {
-            if (e.code() == std::errc::no_such_file_or_directory)
-                return;
-            throw;
-        }
-
-        for (; it->isValid(); it->next())
-            IDiskRemote::getRemotePathsRecursive(fs::path(local_path) / it->name(), paths_map);
-    }
-}
-
-DiskPtr DiskRemoteReservation::getDisk(size_t i) const
-{
-    if (i != 0)
-        throw Exception("Can't use i != 0 with single disk reservation", ErrorCodes::INCORRECT_DISK_INDEX);
-    return disk;
-}
-
-void DiskRemoteReservation::update(UInt64 new_size)
-{
-    std::lock_guard lock(disk->reservation_mutex);
-    disk->reserved_bytes -= size;
-    size = new_size;
-    disk->reserved_bytes += size;
-}
-
-
-DiskRemoteReservation::~DiskRemoteReservation()
-{
-    try
-    {
-        std::lock_guard lock(disk->reservation_mutex);
-        if (disk->reserved_bytes < size)
-        {
-            disk->reserved_bytes = 0;
-            LOG_ERROR(disk->log, "Unbalanced reservations size for disk '{}'.", disk->getName());
-        }
-        else
-        {
-            disk->reserved_bytes -= size;
-        }
-
-        if (disk->reservation_count == 0)
-            LOG_ERROR(disk->log, "Unbalanced reservation count for disk '{}'.", disk->getName());
-        else
-            --disk->reservation_count;
-    }
-    catch (...)
-    {
-        tryLogCurrentException(__PRETTY_FUNCTION__);
-    }
-}
-
-
-IDiskRemote::IDiskRemote(
-    const String & name_,
-    const String & remote_fs_root_path_,
-    DiskPtr metadata_disk_,
-    FileCachePtr cache_,
-    const String & log_name_,
-    size_t thread_pool_size)
-    : IDisk(std::make_unique<AsyncExecutor>(log_name_, thread_pool_size))
-    , log(&Poco::Logger::get(log_name_))
-    , name(name_)
-    , remote_fs_root_path(remote_fs_root_path_)
-    , metadata_disk(metadata_disk_)
-    , cache(cache_)
-{
-}
-
-
-String IDiskRemote::getCacheBasePath() const
-{
-    return cache ? cache->getBasePath() : "";
-}
-
-
-bool IDiskRemote::exists(const String & path) const
-{
-    return metadata_disk->exists(path);
-}
-
-
-bool IDiskRemote::isFile(const String & path) const
-{
-    return metadata_disk->isFile(path);
-}
-
-
-void IDiskRemote::createFile(const String & path)
-{
-    createAndStoreMetadata(path, false);
-}
-
-
-size_t IDiskRemote::getFileSize(const String & path) const
-{
-    return readMetadata(path).total_size;
-}
-
-
-void IDiskRemote::moveFile(const String & from_path, const String & to_path)
-{
-    if (exists(to_path))
-        throw Exception("File already exists: " + to_path, ErrorCodes::FILE_ALREADY_EXISTS);
-
-    metadata_disk->moveFile(from_path, to_path);
-}
-
-
-void IDiskRemote::replaceFile(const String & from_path, const String & to_path)
-{
-    if (exists(to_path))
-    {
-        const String tmp_path = to_path + ".old";
-        moveFile(to_path, tmp_path);
-        moveFile(from_path, to_path);
-        removeFile(tmp_path);
-    }
-    else
-        moveFile(from_path, to_path);
-}
-
-void IDiskRemote::removeSharedFile(const String & path, bool delete_metadata_only)
-{
-    std::vector<String> paths_to_remove;
-    removeMetadata(path, paths_to_remove);
-
-    if (!delete_metadata_only)
-        removeFromRemoteFS(paths_to_remove);
-}
-
-void IDiskRemote::removeSharedFileIfExists(const String & path, bool delete_metadata_only)
-{
-    std::vector<String> paths_to_remove;
-    if (metadata_disk->exists(path))
-    {
-        removeMetadata(path, paths_to_remove);
-        if (!delete_metadata_only)
-            removeFromRemoteFS(paths_to_remove);
-    }
-}
-
-void IDiskRemote::removeSharedFiles(const RemoveBatchRequest & files, bool keep_all_batch_data, const NameSet & file_names_remove_metadata_only)
-{
-    std::unordered_map<String, std::vector<String>> paths_to_remove;
-    for (const auto & file : files)
-    {
-        bool skip = file.if_exists && !metadata_disk->exists(file.path);
-        if (!skip)
-            removeMetadata(file.path, paths_to_remove[file.path]);
-    }
-
-    if (!keep_all_batch_data)
-    {
-        std::vector<String> remove_from_remote;
-        for (auto && [path, remote_paths] : paths_to_remove)
-        {
-            if (!file_names_remove_metadata_only.contains(fs::path(path).filename()))
-                remove_from_remote.insert(remove_from_remote.end(), remote_paths.begin(), remote_paths.end());
-        }
-        removeFromRemoteFS(remove_from_remote);
-    }
-}
-
-void IDiskRemote::removeSharedRecursive(const String & path, bool keep_all_batch_data, const NameSet & file_names_remove_metadata_only)
-{
-    std::unordered_map<String, std::vector<String>> paths_to_remove;
-    removeMetadataRecursive(path, paths_to_remove);
-
-    if (!keep_all_batch_data)
-    {
-        std::vector<String> remove_from_remote;
-        for (auto && [local_path, remote_paths] : paths_to_remove)
-        {
-            if (!file_names_remove_metadata_only.contains(fs::path(local_path).filename()))
-                remove_from_remote.insert(remove_from_remote.end(), remote_paths.begin(), remote_paths.end());
-        }
-        removeFromRemoteFS(remove_from_remote);
-    }
-}
-
-
-void IDiskRemote::setReadOnly(const String & path)
-{
-    /// We should store read only flag inside metadata file (instead of using FS flag),
-    /// because we modify metadata file when create hard-links from it.
-    readUpdateAndStoreMetadata(path, false, [] (Metadata & metadata) { metadata.read_only = true; return true; });
-}
-
-
-bool IDiskRemote::isDirectory(const String & path) const
-{
-    return metadata_disk->isDirectory(path);
-}
-
-
-void IDiskRemote::createDirectory(const String & path)
-{
-    metadata_disk->createDirectory(path);
-}
-
-
-void IDiskRemote::createDirectories(const String & path)
-{
-    metadata_disk->createDirectories(path);
-}
-
-
-void IDiskRemote::clearDirectory(const String & path)
-{
-    for (auto it = iterateDirectory(path); it->isValid(); it->next())
-        if (isFile(it->path()))
-            removeFile(it->path());
-}
-
-
-void IDiskRemote::removeDirectory(const String & path)
-{
-    metadata_disk->removeDirectory(path);
-}
-
-
-DiskDirectoryIteratorPtr IDiskRemote::iterateDirectory(const String & path)
-{
-    return metadata_disk->iterateDirectory(path);
-}
-
-
-void IDiskRemote::listFiles(const String & path, std::vector<String> & file_names)
-{
-    for (auto it = iterateDirectory(path); it->isValid(); it->next())
-        file_names.push_back(it->name());
-}
-
-
-void IDiskRemote::setLastModified(const String & path, const Poco::Timestamp & timestamp)
-{
-    metadata_disk->setLastModified(path, timestamp);
-}
-
-
-Poco::Timestamp IDiskRemote::getLastModified(const String & path)
-{
-    return metadata_disk->getLastModified(path);
-}
-
-
-void IDiskRemote::createHardLink(const String & src_path, const String & dst_path)
-{
-    readUpdateAndStoreMetadata(src_path, false, [] (Metadata & metadata) { metadata.ref_count++; return true; });
-
-    /// Create FS hardlink to metadata file.
-    metadata_disk->createHardLink(src_path, dst_path);
-}
-
-
-ReservationPtr IDiskRemote::reserve(UInt64 bytes)
-{
-    auto unreserved_space = tryReserve(bytes);
-    if (!unreserved_space.has_value())
-        return {};
-
-    return std::make_unique<DiskRemoteReservation>(
-        std::static_pointer_cast<IDiskRemote>(shared_from_this()),
-        bytes, unreserved_space.value());
-}
-
-
-std::optional<UInt64> IDiskRemote::tryReserve(UInt64 bytes)
-{
-    std::lock_guard lock(reservation_mutex);
-
-    auto available_space = getAvailableSpace();
-    UInt64 unreserved_space = available_space - std::min(available_space, reserved_bytes);
-
-    if (bytes == 0)
-    {
-        LOG_TRACE(log, "Reserving 0 bytes on remote_fs disk {}", backQuote(name));
-        ++reservation_count;
-        return {unreserved_space};
-    }
-
-    if (unreserved_space >= bytes)
-    {
-        LOG_TRACE(log, "Reserving {} on disk {}, having unreserved {}.",
-            ReadableSize(bytes), backQuote(name), ReadableSize(unreserved_space));
-        ++reservation_count;
-        reserved_bytes += bytes;
-        return {unreserved_space - bytes};
-    }
-
-    return {};
-}
-
-String IDiskRemote::getUniqueId(const String & path) const
-{
-    LOG_TRACE(log, "Remote path: {}, Path: {}", remote_fs_root_path, path);
-    auto metadata = readMetadata(path);
-    String id;
-    if (!metadata.remote_fs_objects.empty())
-        id = metadata.remote_fs_root_path + metadata.remote_fs_objects[0].relative_path;
-    return id;
-}
-
-
-AsynchronousReaderPtr IDiskRemote::getThreadPoolReader()
-{
-    constexpr size_t pool_size = 50;
-    constexpr size_t queue_size = 1000000;
-    static AsynchronousReaderPtr reader = std::make_shared<ThreadPoolRemoteFSReader>(pool_size, queue_size);
-    return reader;
-}
-
-UInt32 IDiskRemote::getRefCount(const String & path) const
-{
-    return readMetadata(path).ref_count;
-}
-
-ThreadPool & IDiskRemote::getThreadPoolWriter()
-{
-    constexpr size_t pool_size = 100;
-    constexpr size_t queue_size = 1000000;
-    static ThreadPool writer(pool_size, pool_size, queue_size);
-    return writer;
-}
-
-}

src/Disks/IDiskRemote.h

@@ -1,325 +0,0 @@
-#pragma once
-
-#include <Common/config.h>
-
-#include <atomic>
-#include <Common/FileCache_fwd.h>
-#include <Disks/DiskFactory.h>
-#include <Disks/Executor.h>
-#include <utility>
-#include <mutex>
-#include <shared_mutex>
-#include <Common/MultiVersion.h>
-#include <Common/ThreadPool.h>
-#include <filesystem>
-
-namespace CurrentMetrics
-{
-    extern const Metric DiskSpaceReservedForMerge;
-}
-
-namespace DB
-{
-
-/// Path to blob with it's size
-struct BlobPathWithSize
-{
-    std::string relative_path;
-    uint64_t bytes_size;
-
-    BlobPathWithSize() = default;
-    BlobPathWithSize(const BlobPathWithSize & other) = default;
-
-    BlobPathWithSize(const std::string & relative_path_, uint64_t bytes_size_)
-        : relative_path(relative_path_)
-        , bytes_size(bytes_size_)
-    {}
-};
-
-/// List of blobs with their sizes
-using BlobsPathToSize = std::vector<BlobPathWithSize>;
-
-class IAsynchronousReader;
-using AsynchronousReaderPtr = std::shared_ptr<IAsynchronousReader>;
-
-
-/// Base Disk class for remote FS's, which are not posix-compatible (e.g. DiskS3, DiskHDFS, DiskBlobStorage)
-class IDiskRemote : public IDisk
-{
-
-friend class DiskRemoteReservation;
-
-public:
-    IDiskRemote(
-        const String & name_,
-        const String & remote_fs_root_path_,
-        DiskPtr metadata_disk_,
-        FileCachePtr cache_,
-        const String & log_name_,
-        size_t thread_pool_size);
-
-    struct Metadata;
-    using MetadataUpdater = std::function<bool(Metadata & metadata)>;
-
-    const String & getName() const final override { return name; }
-
-    const String & getPath() const final override { return metadata_disk->getPath(); }
-
-    String getCacheBasePath() const final override;
-
-    std::vector<String> getRemotePaths(const String & local_path) const final override;
-
-    void getRemotePathsRecursive(const String & local_path, std::vector<LocalPathWithRemotePaths> & paths_map) override;
-
-    /// Methods for working with metadata. For some operations (like hardlink
-    /// creation) metadata can be updated concurrently from multiple threads
-    /// (file actually rewritten on disk). So additional RW lock is required for
-    /// metadata read and write, but not for create new metadata.
-    Metadata readMetadata(const String & path) const;
-    Metadata readMetadataUnlocked(const String & path, std::shared_lock<std::shared_mutex> &) const;
-    Metadata readUpdateAndStoreMetadata(const String & path, bool sync, MetadataUpdater updater);
-    Metadata readUpdateStoreMetadataAndRemove(const String & path, bool sync, MetadataUpdater updater);
-
-    Metadata readOrCreateUpdateAndStoreMetadata(const String & path, WriteMode mode, bool sync, MetadataUpdater updater);
-
-    Metadata createAndStoreMetadata(const String & path, bool sync);
-    Metadata createUpdateAndStoreMetadata(const String & path, bool sync, MetadataUpdater updater);
-
-    UInt64 getTotalSpace() const override { return std::numeric_limits<UInt64>::max(); }
-
-    UInt64 getAvailableSpace() const override { return std::numeric_limits<UInt64>::max(); }
-
-    UInt64 getUnreservedSpace() const override { return std::numeric_limits<UInt64>::max(); }
-
-    UInt64 getKeepingFreeSpace() const override { return 0; }
-
-    bool exists(const String & path) const override;
-
-    bool isFile(const String & path) const override;
-
-    void createFile(const String & path) override;
-
-    size_t getFileSize(const String & path) const override;
-
-    void moveFile(const String & from_path, const String & to_path) override;
-
-    void replaceFile(const String & from_path, const String & to_path) override;
-
-    void removeFile(const String & path) override { removeSharedFile(path, false); }
-
-    void removeFileIfExists(const String & path) override { removeSharedFileIfExists(path, false); }
-
-    void removeRecursive(const String & path) override { removeSharedRecursive(path, false, {}); }
-
-
-    void removeSharedFile(const String & path, bool delete_metadata_only) override;
-
-    void removeSharedFileIfExists(const String & path, bool delete_metadata_only) override;
-
-    void removeSharedFiles(const RemoveBatchRequest & files, bool keep_all_batch_data, const NameSet & file_names_remove_metadata_only) override;
-
-    void removeSharedRecursive(const String & path, bool keep_all_batch_data, const NameSet & file_names_remove_metadata_only) override;
-
-    void listFiles(const String & path, std::vector<String> & file_names) override;
-
-    void setReadOnly(const String & path) override;
-
-    bool isDirectory(const String & path) const override;
-
-    void createDirectory(const String & path) override;
-
-    void createDirectories(const String & path) override;
-
-    void clearDirectory(const String & path) override;
-
-    void moveDirectory(const String & from_path, const String & to_path) override { moveFile(from_path, to_path); }
-
-    void removeDirectory(const String & path) override;
-
-    DiskDirectoryIteratorPtr iterateDirectory(const String & path) override;
-
-    void setLastModified(const String & path, const Poco::Timestamp & timestamp) override;
-
-    Poco::Timestamp getLastModified(const String & path) override;
-
-    void createHardLink(const String & src_path, const String & dst_path) override;
-
-    ReservationPtr reserve(UInt64 bytes) override;
-
-    String getUniqueId(const String & path) const override;
-
-    bool checkUniqueId(const String & id) const override = 0;
-
-    virtual void removeFromRemoteFS(const std::vector<String> & paths) = 0;
-
-    static AsynchronousReaderPtr getThreadPoolReader();
-    static ThreadPool & getThreadPoolWriter();
-
-    DiskPtr getMetadataDiskIfExistsOrSelf() override { return metadata_disk; }
-
-    UInt32 getRefCount(const String & path) const override;
-
-    /// Return metadata for each file path. Also, before serialization reset
-    /// ref_count for each metadata to zero. This function used only for remote
-    /// fetches/sends in replicated engines. That's why we reset ref_count to zero.
-    std::unordered_map<String, String> getSerializedMetadata(const std::vector<String> & file_paths) const override;
-protected:
-    Poco::Logger * log;
-    const String name;
-    const String remote_fs_root_path;
-
-    DiskPtr metadata_disk;
-
-    FileCachePtr cache;
-
-private:
-    void removeMetadata(const String & path, std::vector<String> & paths_to_remove);
-
-    void removeMetadataRecursive(const String & path, std::unordered_map<String, std::vector<String>> & paths_to_remove);
-
-    std::optional<UInt64> tryReserve(UInt64 bytes);
-
-    UInt64 reserved_bytes = 0;
-    UInt64 reservation_count = 0;
-    std::mutex reservation_mutex;
-    mutable std::shared_mutex metadata_mutex;
-};
-
-using RemoteDiskPtr = std::shared_ptr<IDiskRemote>;
-
-/// Remote FS (S3, HDFS) metadata file layout:
-/// FS objects, their number and total size of all FS objects.
-/// Each FS object represents a file path in remote FS and its size.
-
-struct IDiskRemote::Metadata
-{
-    using Updater = std::function<bool(IDiskRemote::Metadata & metadata)>;
-    /// Metadata file version.
-    static constexpr UInt32 VERSION_ABSOLUTE_PATHS = 1;
-    static constexpr UInt32 VERSION_RELATIVE_PATHS = 2;
-    static constexpr UInt32 VERSION_READ_ONLY_FLAG = 3;
-
-    /// Remote FS objects paths and their sizes.
-    std::vector<BlobPathWithSize> remote_fs_objects;
-
-    /// URI
-    const String & remote_fs_root_path;
-
-    /// Relative path to metadata file on local FS.
-    const String metadata_file_path;
-
-    DiskPtr metadata_disk;
-
-    /// Total size of all remote FS (S3, HDFS) objects.
-    size_t total_size = 0;
-
-    /// Number of references (hardlinks) to this metadata file.
-    ///
-    /// FIXME: Why we are tracking it explicetly, without
-    /// info from filesystem????
-    UInt32 ref_count = 0;
-
-    /// Flag indicates that file is read only.
-    bool read_only = false;
-
-    Metadata(
-        const String & remote_fs_root_path_,
-        DiskPtr metadata_disk_,
-        const String & metadata_file_path_);
-
-    void addObject(const String & path, size_t size);
-
-    static Metadata readMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_);
-    static Metadata readUpdateAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, Updater updater);
-    static Metadata readUpdateStoreMetadataAndRemove(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, Updater updater);
-
-    static Metadata createAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync);
-    static Metadata createUpdateAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, Updater updater);
-    static Metadata createAndStoreMetadataIfNotExists(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, bool overwrite);
-
-    /// Serialize metadata to string (very same with saveToBuffer)
-    std::string serializeToString();
-
-private:
-    /// Fsync metadata file if 'sync' flag is set.
-    void save(bool sync = false);
-    void saveToBuffer(WriteBuffer & buffer, bool sync);
-    void load();
-};
-
-class DiskRemoteReservation final : public IReservation
-{
-public:
-    DiskRemoteReservation(const RemoteDiskPtr & disk_, UInt64 size_, UInt64 unreserved_space_)
-        : disk(disk_)
-        , size(size_)
-        , unreserved_space(unreserved_space_)
-        , metric_increment(CurrentMetrics::DiskSpaceReservedForMerge, size_)
-    {
-    }
-
-    UInt64 getSize() const override { return size; }
-
-    UInt64 getUnreservedSpace() const override { return unreserved_space; }
-
-    DiskPtr getDisk(size_t i) const override;
-
-    Disks getDisks() const override { return {disk}; }
-
-    void update(UInt64 new_size) override;
-
-    ~DiskRemoteReservation() override;
-
-private:
-    RemoteDiskPtr disk;
-    UInt64 size;
-    UInt64 unreserved_space;
-    CurrentMetrics::Increment metric_increment;
-};
-
-
-/// Runs tasks asynchronously using thread pool.
-class AsyncExecutor : public Executor
-{
-public:
-    explicit AsyncExecutor(const String & name_, int thread_pool_size)
-        : name(name_)
-        , pool(ThreadPool(thread_pool_size)) {}
-
-    std::future<void> execute(std::function<void()> task) override
-    {
-        auto promise = std::make_shared<std::promise<void>>();
-        pool.scheduleOrThrowOnError(
-            [promise, task]()
-            {
-                try
-                {
-                    task();
-                    promise->set_value();
-                }
-                catch (...)
-                {
-                    tryLogCurrentException("Failed to run async task");
-
-                    try
-                    {
-                        promise->set_exception(std::current_exception());
-                    }
-                    catch (...) {}
-                }
-            });
-
-        return promise->get_future();
-    }
-
-    void setMaxThreads(size_t threads)
-    {
-        pool.setMaxThreads(threads);
-    }
-
-private:
-    String name;
-    ThreadPool pool;
-};
-
-}

src/Disks/IO/CachedReadBufferFromRemoteFS.cpp

@@ -212,38 +212,6 @@ SeekableReadBufferPtr CachedReadBufferFromRemoteFS::getReadBufferForFileSegment(
                 read_type = ReadType::REMOTE_FS_READ_BYPASS_CACHE;
                 return getRemoteFSReadBuffer(file_segment, read_type);
             }
-            case FileSegment::State::EMPTY:
-            {
-                auto downloader_id = file_segment->getOrSetDownloader();
-                if (downloader_id == file_segment->getCallerId())
-                {
-                    if (file_offset_of_buffer_end == file_segment->getDownloadOffset())
-                    {
-                        read_type = ReadType::REMOTE_FS_READ_AND_PUT_IN_CACHE;
-                        return getRemoteFSReadBuffer(file_segment, read_type);
-                    }
-                    else
-                    {
-                        ///                      segment{k}
-                        /// cache:           [______|___________
-                        ///                         ^
-                        ///                         download_offset
-                        /// requested_range:            [__________]
-                        ///                             ^
-                        ///                             file_offset_of_buffer_end
-                        assert(file_offset_of_buffer_end > file_segment->getDownloadOffset());
-                        bytes_to_predownload = file_offset_of_buffer_end - file_segment->getDownloadOffset();
-
-                        read_type = ReadType::REMOTE_FS_READ_AND_PUT_IN_CACHE;
-                        return getRemoteFSReadBuffer(file_segment, read_type);
-                    }
-                }
-                else
-                {
-                    download_state = file_segment->state();
-                    continue;
-                }
-            }
             case FileSegment::State::DOWNLOADING:
             {
                 size_t download_offset = file_segment->getDownloadOffset();
@@ -280,6 +248,7 @@ SeekableReadBufferPtr CachedReadBufferFromRemoteFS::getReadBufferForFileSegment(
                 read_type = ReadType::CACHED;
                 return getCacheReadBuffer(range.left);
             }
+            case FileSegment::State::EMPTY:
             case FileSegment::State::PARTIALLY_DOWNLOADED:
             {
                 if (file_segment->getDownloadOffset() > file_offset_of_buffer_end)
@@ -491,7 +460,10 @@ bool CachedReadBufferFromRemoteFS::completeFileSegmentAndGetNext()
 
     /// Do not hold pointer to file segment if it is not needed anymore
     /// so can become releasable and can be evicted from cache.
-    file_segments_holder->file_segments.erase(file_segment_it);
+    /// If the status of filesegment state is SKIP_CACHE, it will not be deleted.
+    /// It will be deleted from the cache when the holder is destructed.
+    if ((*file_segment_it)->state() != FileSegment::State::SKIP_CACHE)
+        file_segments_holder->file_segments.erase(file_segment_it);
 
     if (current_file_segment_it == file_segments_holder->file_segments.end())
         return false;

src/Disks/IO/ReadBufferFromRemoteFSGather.cpp

@@ -1,6 +1,5 @@
 #include "ReadBufferFromRemoteFSGather.h"
 
-#include <Disks/IDiskRemote.h>
 #include <IO/SeekableReadBuffer.h>
 #include <Disks/IO/ReadBufferFromWebServer.h>
 

src/Disks/IO/ReadBufferFromRemoteFSGather.h

@@ -1,9 +1,9 @@
 #pragma once
 
 #include <Common/config.h>
-#include <Disks/IDiskRemote.h>
 #include <IO/ReadBufferFromFile.h>
 #include <IO/ReadSettings.h>
+#include <Disks/ObjectStorages/IObjectStorage.h>
 
 #if USE_AZURE_BLOB_STORAGE
 #include <azure/storage/blobs.hpp>
@@ -114,7 +114,7 @@ class ReadBufferFromS3Gather final : public ReadBufferFromRemoteFSGather
 {
 public:
     ReadBufferFromS3Gather(
-        std::shared_ptr<Aws::S3::S3Client> client_ptr_,
+        std::shared_ptr<const Aws::S3::S3Client> client_ptr_,
         const String & bucket_,
         const String & version_id_,
         const std::string & common_path_prefix_,
@@ -132,7 +132,7 @@ public:
     SeekableReadBufferPtr createImplementationBufferImpl(const String & path, size_t file_size) override;
 
 private:
-    std::shared_ptr<Aws::S3::S3Client> client_ptr;
+    std::shared_ptr<const Aws::S3::S3Client> client_ptr;
     String bucket;
     String version_id;
     UInt64 max_single_read_retries;
@@ -146,7 +146,7 @@ class ReadBufferFromAzureBlobStorageGather final : public ReadBufferFromRemoteFS
 {
 public:
     ReadBufferFromAzureBlobStorageGather(
-        std::shared_ptr<Azure::Storage::Blobs::BlobContainerClient> blob_container_client_,
+        std::shared_ptr<const Azure::Storage::Blobs::BlobContainerClient> blob_container_client_,
         const std::string & common_path_prefix_,
         const BlobsPathToSize & blobs_to_read_,
         size_t max_single_read_retries_,
@@ -162,7 +162,7 @@ public:
     SeekableReadBufferPtr createImplementationBufferImpl(const String & path, size_t file_size) override;
 
 private:
-    std::shared_ptr<Azure::Storage::Blobs::BlobContainerClient> blob_container_client;
+    std::shared_ptr<const Azure::Storage::Blobs::BlobContainerClient> blob_container_client;
     size_t max_single_read_retries;
     size_t max_single_download_retries;
 };

src/Disks/IO/ReadIndirectBufferFromRemoteFS.h

@@ -2,7 +2,6 @@
 
 #include <Common/config.h>
 #include <IO/ReadBufferFromFile.h>
-#include <Disks/IDiskRemote.h>
 #include <utility>
 
 

src/Disks/IO/ThreadPoolRemoteFSReader.h

@@ -4,7 +4,6 @@
 #include <IO/SeekableReadBuffer.h>
 #include <Common/ThreadPool.h>
 #include <Disks/IO/ReadBufferFromRemoteFSGather.h>
-#include <Disks/IDiskRemote.h>
 
 
 namespace DB

src/Disks/IO/WriteIndirectBufferFromRemoteFS.cpp

@@ -36,7 +36,8 @@ WriteIndirectBufferFromRemoteFS::~WriteIndirectBufferFromRemoteFS()
 void WriteIndirectBufferFromRemoteFS::finalizeImpl()
 {
     WriteBufferFromFileDecorator::finalizeImpl();
-    create_metadata_callback(count());
+    if (create_metadata_callback)
+        create_metadata_callback(count());
 }
 
 

src/Disks/IO/WriteIndirectBufferFromRemoteFS.h

@@ -2,7 +2,6 @@
 
 #include <Common/config.h>
 
-#include <Disks/IDiskRemote.h>
 #include <IO/WriteBufferFromFile.h>
 #include <IO/WriteBufferFromFileDecorator.h>
 

src/Disks/ObjectStorages/AzureBlobStorage/AzureBlobStorageAuth.cpp

@@ -1,4 +1,4 @@
-#include <Disks/AzureBlobStorage/AzureBlobStorageAuth.h>
+#include <Disks/ObjectStorages/AzureBlobStorage/AzureBlobStorageAuth.h>
 
 #if USE_AZURE_BLOB_STORAGE
 
@@ -66,27 +66,27 @@ AzureBlobStorageEndpoint processAzureBlobStorageEndpoint(const Poco::Util::Abstr
 
 
 template <class T>
-std::shared_ptr<T> getClientWithConnectionString(const String & connection_str, const String & container_name) = delete;
+std::unique_ptr<T> getClientWithConnectionString(const String & connection_str, const String & container_name) = delete;
 
 
 template<>
-std::shared_ptr<BlobServiceClient> getClientWithConnectionString(
+std::unique_ptr<BlobServiceClient> getClientWithConnectionString(
     const String & connection_str, const String & /*container_name*/)
 {
-    return std::make_shared<BlobServiceClient>(BlobServiceClient::CreateFromConnectionString(connection_str));
+    return std::make_unique<BlobServiceClient>(BlobServiceClient::CreateFromConnectionString(connection_str));
 }
 
 
 template<>
-std::shared_ptr<BlobContainerClient> getClientWithConnectionString(
+std::unique_ptr<BlobContainerClient> getClientWithConnectionString(
     const String & connection_str, const String & container_name)
 {
-    return std::make_shared<BlobContainerClient>(BlobContainerClient::CreateFromConnectionString(connection_str, container_name));
+    return std::make_unique<BlobContainerClient>(BlobContainerClient::CreateFromConnectionString(connection_str, container_name));
 }
 
 
 template <class T>
-std::shared_ptr<T> getAzureBlobStorageClientWithAuth(
+std::unique_ptr<T> getAzureBlobStorageClientWithAuth(
     const String & url, const String & container_name, const Poco::Util::AbstractConfiguration & config, const String & config_prefix)
 {
     if (config.has(config_prefix + ".connection_string"))
@@ -101,15 +101,15 @@ std::shared_ptr<T> getAzureBlobStorageClientWithAuth(
             config.getString(config_prefix + ".account_name"),
             config.getString(config_prefix + ".account_key")
         );
-        return std::make_shared<T>(url, storage_shared_key_credential);
+        return std::make_unique<T>(url, storage_shared_key_credential);
     }
 
     auto managed_identity_credential = std::make_shared<Azure::Identity::ManagedIdentityCredential>();
-    return std::make_shared<T>(url, managed_identity_credential);
+    return std::make_unique<T>(url, managed_identity_credential);
 }
 
 
-std::shared_ptr<BlobContainerClient> getAzureBlobContainerClient(
+std::unique_ptr<BlobContainerClient> getAzureBlobContainerClient(
     const Poco::Util::AbstractConfiguration & config, const String & config_prefix)
 {
     auto endpoint = processAzureBlobStorageEndpoint(config, config_prefix);
@@ -136,10 +136,20 @@ std::shared_ptr<BlobContainerClient> getAzureBlobContainerClient(
         }
     }
 
-    return std::make_shared<BlobContainerClient>(
+    return std::make_unique<BlobContainerClient>(
         blob_service_client->CreateBlobContainer(container_name).Value);
 }
 
+std::unique_ptr<AzureObjectStorageSettings> getAzureBlobStorageSettings(const Poco::Util::AbstractConfiguration & config, const String & config_prefix, ContextPtr /*context*/)
+{
+    return std::make_unique<AzureObjectStorageSettings>(
+        config.getUInt64(config_prefix + ".max_single_part_upload_size", 100 * 1024 * 1024),
+        config.getUInt64(config_prefix + ".min_bytes_for_seek", 1024 * 1024),
+        config.getInt(config_prefix + ".max_single_read_retries", 3),
+        config.getInt(config_prefix + ".max_single_download_retries", 3)
+    );
+}
+
 }
 
 #endif

src/Disks/ObjectStorages/AzureBlobStorage/AzureBlobStorageAuth.h

@@ -0,0 +1,20 @@
+#pragma once
+
+#include <Common/config.h>
+
+#if USE_AZURE_BLOB_STORAGE
+
+#include <azure/storage/blobs.hpp>
+#include <Disks/ObjectStorages/AzureBlobStorage/AzureObjectStorage.h>
+
+namespace DB
+{
+
+std::unique_ptr<Azure::Storage::Blobs::BlobContainerClient> getAzureBlobContainerClient(
+    const Poco::Util::AbstractConfiguration & config, const String & config_prefix);
+
+std::unique_ptr<AzureObjectStorageSettings> getAzureBlobStorageSettings(const Poco::Util::AbstractConfiguration & config, const String & config_prefix, ContextPtr /*context*/);
+
+}
+
+#endif

src/Disks/ObjectStorages/AzureBlobStorage/AzureObjectStorage.cpp

@@ -0,0 +1,218 @@
+#include <Disks/ObjectStorages/AzureBlobStorage/AzureObjectStorage.h>
+
+#if USE_AZURE_BLOB_STORAGE
+
+#include <IO/ReadBufferFromAzureBlobStorage.h>
+#include <IO/WriteBufferFromAzureBlobStorage.h>
+#include <IO/SeekAvoidingReadBuffer.h>
+#include <Disks/IO/ReadBufferFromRemoteFSGather.h>
+
+#include <Disks/ObjectStorages/AzureBlobStorage/AzureBlobStorageAuth.h>
+
+namespace DB
+{
+
+
+namespace ErrorCodes
+{
+    extern const int AZURE_BLOB_STORAGE_ERROR;
+    extern const int UNSUPPORTED_METHOD;
+}
+
+
+AzureObjectStorage::AzureObjectStorage(
+    FileCachePtr && cache_,
+    const String & name_,
+    AzureClientPtr && client_,
+    SettingsPtr && settings_)
+    : IObjectStorage(std::move(cache_))
+    , name(name_)
+    , client(std::move(client_))
+    , settings(std::move(settings_))
+{
+}
+
+bool AzureObjectStorage::exists(const std::string & uri) const
+{
+    auto client_ptr = client.get();
+
+    /// What a shame, no Exists method...
+    Azure::Storage::Blobs::ListBlobsOptions options;
+    options.Prefix = uri;
+    options.PageSizeHint = 1;
+
+    auto blobs_list_response = client_ptr->ListBlobs(options);
+    auto blobs_list = blobs_list_response.Blobs;
+
+    for (const auto & blob : blobs_list)
+    {
+        if (uri == blob.Name)
+            return true;
+    }
+
+    return false;
+}
+
+std::unique_ptr<SeekableReadBuffer> AzureObjectStorage::readObject( /// NOLINT
+    const std::string & path,
+    const ReadSettings & read_settings,
+    std::optional<size_t>,
+    std::optional<size_t>) const
+{
+    auto settings_ptr = settings.get();
+
+    return std::make_unique<ReadBufferFromAzureBlobStorage>(
+        client.get(), path, settings_ptr->max_single_read_retries,
+        settings_ptr->max_single_download_retries, read_settings.remote_fs_buffer_size);
+}
+
+std::unique_ptr<ReadBufferFromFileBase> AzureObjectStorage::readObjects( /// NOLINT
+    const std::string & common_path_prefix,
+    const BlobsPathToSize & blobs_to_read,
+    const ReadSettings & read_settings,
+    std::optional<size_t>,
+    std::optional<size_t>) const
+{
+    auto settings_ptr = settings.get();
+    auto reader_impl = std::make_unique<ReadBufferFromAzureBlobStorageGather>(
+        client.get(), common_path_prefix, blobs_to_read,
+        settings_ptr->max_single_read_retries, settings_ptr->max_single_download_retries, read_settings);
+
+    if (read_settings.remote_fs_method == RemoteFSReadMethod::threadpool)
+    {
+        auto reader = getThreadPoolReader();
+        return std::make_unique<AsynchronousReadIndirectBufferFromRemoteFS>(reader, read_settings, std::move(reader_impl));
+    }
+    else
+    {
+        auto buf = std::make_unique<ReadIndirectBufferFromRemoteFS>(std::move(reader_impl));
+        return std::make_unique<SeekAvoidingReadBuffer>(std::move(buf), settings_ptr->min_bytes_for_seek);
+    }
+}
+
+/// Open the file for write and return WriteBufferFromFileBase object.
+std::unique_ptr<WriteBufferFromFileBase> AzureObjectStorage::writeObject( /// NOLINT
+    const std::string & path,
+    WriteMode mode,
+    std::optional<ObjectAttributes>,
+    FinalizeCallback && finalize_callback,
+    size_t buf_size,
+    const WriteSettings &)
+{
+    if (mode != WriteMode::Rewrite)
+        throw Exception("Azure storage doesn't support append", ErrorCodes::UNSUPPORTED_METHOD);
+
+    auto buffer = std::make_unique<WriteBufferFromAzureBlobStorage>(
+        client.get(),
+        path,
+        settings.get()->max_single_part_upload_size,
+        buf_size);
+
+    return std::make_unique<WriteIndirectBufferFromRemoteFS>(std::move(buffer), std::move(finalize_callback), path);
+}
+
+void AzureObjectStorage::listPrefix(const std::string & path, BlobsPathToSize & children) const
+{
+    auto client_ptr = client.get();
+
+    Azure::Storage::Blobs::ListBlobsOptions blobs_list_options;
+    blobs_list_options.Prefix = path;
+
+    auto blobs_list_response = client_ptr->ListBlobs(blobs_list_options);
+    auto blobs_list = blobs_list_response.Blobs;
+
+    for (const auto & blob : blobs_list)
+        children.emplace_back(blob.Name, blob.BlobSize);
+}
+
+/// Remove file. Throws exception if file doesn't exists or it's a directory.
+void AzureObjectStorage::removeObject(const std::string & path)
+{
+    auto client_ptr = client.get();
+    auto delete_info = client_ptr->DeleteBlob(path);
+    if (!delete_info.Value.Deleted)
+        throw Exception(ErrorCodes::AZURE_BLOB_STORAGE_ERROR, "Failed to delete file in AzureBlob Storage: {}", path);
+}
+
+void AzureObjectStorage::removeObjects(const std::vector<std::string> & paths)
+{
+    auto client_ptr = client.get();
+    for (const auto & path : paths)
+    {
+        auto delete_info = client_ptr->DeleteBlob(path);
+        if (!delete_info.Value.Deleted)
+            throw Exception(ErrorCodes::AZURE_BLOB_STORAGE_ERROR, "Failed to delete file in AzureBlob Storage: {}", path);
+    }
+}
+
+void AzureObjectStorage::removeObjectIfExists(const std::string & path)
+{
+    auto client_ptr = client.get();
+    auto delete_info = client_ptr->DeleteBlob(path);
+}
+
+void AzureObjectStorage::removeObjectsIfExist(const std::vector<std::string> & paths)
+{
+    auto client_ptr = client.get();
+    for (const auto & path : paths)
+        auto delete_info = client_ptr->DeleteBlob(path);
+}
+
+
+ObjectMetadata AzureObjectStorage::getObjectMetadata(const std::string & path) const
+{
+    auto client_ptr = client.get();
+    auto blob_client = client_ptr->GetBlobClient(path);
+    auto properties = blob_client.GetProperties().Value;
+    ObjectMetadata result;
+    result.size_bytes = properties.BlobSize;
+    if (!properties.Metadata.empty())
+    {
+        result.attributes.emplace();
+        for (const auto & [key, value] : properties.Metadata)
+            (*result.attributes)[key] = value;
+    }
+    result.last_modified.emplace(properties.LastModified.time_since_epoch().count());
+    return result;
+}
+
+void AzureObjectStorage::copyObject( /// NOLINT
+    const std::string & object_from,
+    const std::string & object_to,
+    std::optional<ObjectAttributes> object_to_attributes)
+{
+    auto client_ptr = client.get();
+    auto dest_blob_client = client_ptr->GetBlobClient(object_to);
+    auto source_blob_client = client_ptr->GetBlobClient(object_from);
+    Azure::Storage::Blobs::CopyBlobFromUriOptions copy_options;
+    if (object_to_attributes.has_value())
+    {
+        for (const auto & [key, value] : *object_to_attributes)
+            copy_options.Metadata[key] = value;
+    }
+
+    dest_blob_client.CopyFromUri(source_blob_client.GetUrl(), copy_options);
+}
+
+void AzureObjectStorage::applyNewSettings(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context)
+{
+    auto new_settings = getAzureBlobStorageSettings(config, config_prefix, context);
+    settings.set(std::move(new_settings));
+
+    /// We don't update client
+}
+
+
+std::unique_ptr<IObjectStorage> AzureObjectStorage::cloneObjectStorage(const std::string &, const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context)
+{
+    return std::make_unique<AzureObjectStorage>(
+        nullptr,
+        name,
+        getAzureBlobContainerClient(config, config_prefix),
+        getAzureBlobStorageSettings(config, config_prefix, context)
+    );
+}
+
+}
+
+#endif

src/Disks/ObjectStorages/AzureBlobStorage/AzureObjectStorage.h

@@ -0,0 +1,113 @@
+#pragma once
+#include <Common/config.h>
+
+#if USE_AZURE_BLOB_STORAGE
+
+#include <Disks/ObjectStorages/DiskObjectStorageCommon.h>
+#include <Disks/IO/ReadBufferFromRemoteFSGather.h>
+#include <Disks/IO/AsynchronousReadIndirectBufferFromRemoteFS.h>
+#include <Disks/IO/ReadIndirectBufferFromRemoteFS.h>
+#include <Disks/IO/WriteIndirectBufferFromRemoteFS.h>
+#include <Disks/ObjectStorages/IObjectStorage.h>
+#include <Common/getRandomASCIIString.h>
+
+
+namespace DB
+{
+
+struct AzureObjectStorageSettings
+{
+    AzureObjectStorageSettings(
+        uint64_t max_single_part_upload_size_,
+        uint64_t min_bytes_for_seek_,
+        int max_single_read_retries_,
+        int max_single_download_retries_)
+        : max_single_part_upload_size(max_single_part_upload_size_)
+        , min_bytes_for_seek(min_bytes_for_seek_)
+        , max_single_read_retries(max_single_read_retries_)
+        , max_single_download_retries(max_single_download_retries_)
+    {
+    }
+
+    size_t max_single_part_upload_size; /// NOTE: on 32-bit machines it will be at most 4GB, but size_t is also used in BufferBase for offset
+    uint64_t min_bytes_for_seek;
+    size_t max_single_read_retries;
+    size_t max_single_download_retries;
+};
+
+using AzureClient = Azure::Storage::Blobs::BlobContainerClient;
+using AzureClientPtr = std::unique_ptr<Azure::Storage::Blobs::BlobContainerClient>;
+
+class AzureObjectStorage : public IObjectStorage
+{
+public:
+
+    using SettingsPtr = std::unique_ptr<AzureObjectStorageSettings>;
+
+    AzureObjectStorage(
+        FileCachePtr && cache_,
+        const String & name_,
+        AzureClientPtr && client_,
+        SettingsPtr && settings_);
+
+    bool exists(const std::string & uri) const override;
+
+    std::unique_ptr<SeekableReadBuffer> readObject( /// NOLINT
+        const std::string & path,
+        const ReadSettings & read_settings = ReadSettings{},
+        std::optional<size_t> read_hint = {},
+        std::optional<size_t> file_size = {}) const override;
+
+    std::unique_ptr<ReadBufferFromFileBase> readObjects( /// NOLINT
+        const std::string & common_path_prefix,
+        const BlobsPathToSize & blobs_to_read,
+        const ReadSettings & read_settings = ReadSettings{},
+        std::optional<size_t> read_hint = {},
+        std::optional<size_t> file_size = {}) const override;
+
+    /// Open the file for write and return WriteBufferFromFileBase object.
+    std::unique_ptr<WriteBufferFromFileBase> writeObject( /// NOLINT
+        const std::string & path,
+        WriteMode mode,
+        std::optional<ObjectAttributes> attributes = {},
+        FinalizeCallback && finalize_callback = {},
+        size_t buf_size = DBMS_DEFAULT_BUFFER_SIZE,
+        const WriteSettings & write_settings = {}) override;
+
+    void listPrefix(const std::string & path, BlobsPathToSize & children) const override;
+    /// Remove file. Throws exception if file doesn't exists or it's a directory.
+    void removeObject(const std::string & path) override;
+
+    void removeObjects(const std::vector<std::string> & paths) override;
+
+    void removeObjectIfExists(const std::string & path) override;
+
+    void removeObjectsIfExist(const std::vector<std::string> & paths) override;
+
+    ObjectMetadata getObjectMetadata(const std::string & path) const override;
+
+    void copyObject( /// NOLINT
+        const std::string & object_from,
+        const std::string & object_to,
+        std::optional<ObjectAttributes> object_to_attributes = {}) override;
+
+    void shutdown() override {}
+
+    void startup() override {}
+
+    void applyNewSettings(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context) override;
+
+    String getObjectsNamespace() const override { return ""; }
+
+    std::unique_ptr<IObjectStorage> cloneObjectStorage(const std::string & new_namespace, const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context) override;
+
+private:
+    const String name;
+    /// client used to access the files in the Blob Storage cloud
+    MultiVersion<Azure::Storage::Blobs::BlobContainerClient> client;
+    MultiVersion<AzureObjectStorageSettings> settings;
+};
+
+}
+
+#endif

src/Disks/ObjectStorages/AzureBlobStorage/registerDiskAzureBlobStorage.cpp

@@ -6,10 +6,11 @@
 
 #include <Disks/DiskRestartProxy.h>
 #include <Disks/DiskCacheWrapper.h>
-#include <Disks/RemoteDisksCommon.h>
-#include <Disks/AzureBlobStorage/DiskAzureBlobStorage.h>
-#include <Disks/AzureBlobStorage/AzureBlobStorageAuth.h>
+#include <Disks/ObjectStorages/DiskObjectStorageCommon.h>
+#include <Disks/ObjectStorages/DiskObjectStorage.h>
 
+#include <Disks/ObjectStorages/AzureBlobStorage/AzureBlobStorageAuth.h>
+#include <Disks/ObjectStorages/AzureBlobStorage/AzureObjectStorage.h>
 
 namespace DB
 {
@@ -19,18 +20,19 @@ namespace ErrorCodes
     extern const int PATH_ACCESS_DENIED;
 }
 
+namespace
+{
+
 constexpr char test_file[] = "test.txt";
 constexpr char test_str[] = "test";
 constexpr size_t test_str_size = 4;
 
-
 void checkWriteAccess(IDisk & disk)
 {
     auto file = disk.writeFile(test_file, DBMS_DEFAULT_BUFFER_SIZE, WriteMode::Rewrite);
     file->write(test_str, test_str_size);
 }
 
-
 void checkReadAccess(IDisk & disk)
 {
     auto file = disk.readFile(test_file);
@@ -40,7 +42,6 @@ void checkReadAccess(IDisk & disk)
         throw Exception("No read access to disk", ErrorCodes::PATH_ACCESS_DENIED);
 }
 
-
 void checkReadWithOffset(IDisk & disk)
 {
     auto file = disk.readFile(test_file);
@@ -53,25 +54,13 @@ void checkReadWithOffset(IDisk & disk)
         throw Exception("Failed to read file with offset", ErrorCodes::PATH_ACCESS_DENIED);
 }
 
-
 void checkRemoveAccess(IDisk & disk)
 {
     disk.removeFile(test_file);
 }
 
-
-std::unique_ptr<DiskAzureBlobStorageSettings> getSettings(const Poco::Util::AbstractConfiguration & config, const String & config_prefix, ContextPtr /*context*/)
-{
-    return std::make_unique<DiskAzureBlobStorageSettings>(
-        config.getUInt64(config_prefix + ".max_single_part_upload_size", 100 * 1024 * 1024),
-        config.getUInt64(config_prefix + ".min_bytes_for_seek", 1024 * 1024),
-        config.getInt(config_prefix + ".max_single_read_retries", 3),
-        config.getInt(config_prefix + ".max_single_download_retries", 3),
-        config.getInt(config_prefix + ".thread_pool_size", 16)
-    );
 }
 
-
 void registerDiskAzureBlobStorage(DiskFactory & factory)
 {
     auto creator = [](
@@ -83,12 +72,25 @@ void registerDiskAzureBlobStorage(DiskFactory & factory)
     {
         auto [metadata_path, metadata_disk] = prepareForLocalMetadata(name, config, config_prefix, context);
 
-        std::shared_ptr<IDisk> azure_blob_storage_disk = std::make_shared<DiskAzureBlobStorage>(
+        /// FIXME Cache currently unsupported :(
+        ObjectStoragePtr azure_object_storage = std::make_unique<AzureObjectStorage>(
+            nullptr,
             name,
-            metadata_disk,
             getAzureBlobContainerClient(config, config_prefix),
-            getSettings(config, config_prefix, context),
-            getSettings
+            getAzureBlobStorageSettings(config, config_prefix, context));
+
+        uint64_t copy_thread_pool_size = config.getUInt(config_prefix + ".thread_pool_size", 16);
+        bool send_metadata = config.getBool(config_prefix + ".send_metadata", false);
+
+        std::shared_ptr<IDisk> azure_blob_storage_disk = std::make_shared<DiskObjectStorage>(
+            name,
+            /* no namespaces */"",
+            "DiskAzureBlobStorage",
+            metadata_disk,
+            std::move(azure_object_storage),
+            DiskType::AzureBlobStorage,
+            send_metadata,
+            copy_thread_pool_size
         );
 
         if (!config.getBool(config_prefix + ".skip_access_check", false))
@@ -99,9 +101,17 @@ void registerDiskAzureBlobStorage(DiskFactory & factory)
             checkRemoveAccess(*azure_blob_storage_disk);
         }
 
-        azure_blob_storage_disk->startup();
+#ifdef NDEBUG
+        bool use_cache = true;
+#else
+        /// Current cache implementation lead to allocations in destructor of
+        /// read buffer.
+        bool use_cache = false;
+#endif
 
-        if (config.getBool(config_prefix + ".cache_enabled", true))
+        azure_blob_storage_disk->startup(context);
+
+        if (config.getBool(config_prefix + ".cache_enabled", use_cache))
         {
             String cache_path = config.getString(config_prefix + ".cache_path", context->getPath() + "disks/" + name + "/cache/");
             azure_blob_storage_disk = wrapWithCache(azure_blob_storage_disk, "azure-blob-storage-cache", cache_path, metadata_path);

src/Disks/ObjectStorages/DiskObjectStorage.cpp

@@ -0,0 +1,678 @@
+#include <Disks/ObjectStorages/DiskObjectStorage.h>
+
+#include <IO/ReadBufferFromString.h>
+#include <IO/ReadBufferFromFile.h>
+#include <IO/ReadHelpers.h>
+#include <IO/WriteBufferFromFile.h>
+#include <IO/WriteHelpers.h>
+#include <Common/createHardLink.h>
+#include <Common/quoteString.h>
+#include <Common/logger_useful.h>
+#include <Common/checkStackSize.h>
+#include <Common/getRandomASCIIString.h>
+#include <boost/algorithm/string.hpp>
+#include <Common/filesystemHelpers.h>
+#include <Disks/IO/ThreadPoolRemoteFSReader.h>
+#include <Common/FileCache.h>
+#include <Disks/ObjectStorages/DiskObjectStorageMetadataHelper.h>
+#include <Poco/Util/AbstractConfiguration.h>
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int INCORRECT_DISK_INDEX;
+    extern const int UNKNOWN_FORMAT;
+    extern const int FILE_ALREADY_EXISTS;
+    extern const int FILE_DOESNT_EXIST;
+    extern const int BAD_FILE_TYPE;
+}
+
+static String revisionToString(UInt64 revision)
+{
+    return std::bitset<64>(revision).to_string();
+}
+
+namespace
+{
+
+/// Runs tasks asynchronously using thread pool.
+class AsyncThreadPoolExecutor : public Executor
+{
+public:
+    AsyncThreadPoolExecutor(const String & name_, int thread_pool_size)
+        : name(name_)
+        , pool(ThreadPool(thread_pool_size)) {}
+
+    std::future<void> execute(std::function<void()> task) override
+    {
+        auto promise = std::make_shared<std::promise<void>>();
+        pool.scheduleOrThrowOnError(
+            [promise, task]()
+            {
+                try
+                {
+                    task();
+                    promise->set_value();
+                }
+                catch (...)
+                {
+                    tryLogCurrentException("Failed to run async task");
+
+                    try
+                    {
+                        promise->set_exception(std::current_exception());
+                    }
+                    catch (...) {}
+                }
+            });
+
+        return promise->get_future();
+    }
+
+    void setMaxThreads(size_t threads)
+    {
+        pool.setMaxThreads(threads);
+    }
+
+private:
+    String name;
+    ThreadPool pool;
+};
+
+}
+
+DiskObjectStorage::DiskObjectStorage(
+    const String & name_,
+    const String & remote_fs_root_path_,
+    const String & log_name,
+    DiskPtr metadata_disk_,
+    ObjectStoragePtr && object_storage_,
+    DiskType disk_type_,
+    bool send_metadata_,
+    uint64_t thread_pool_size)
+    : IDisk(std::make_unique<AsyncThreadPoolExecutor>(log_name, thread_pool_size))
+    , name(name_)
+    , remote_fs_root_path(remote_fs_root_path_)
+    , log (&Poco::Logger::get(log_name))
+    , metadata_disk(metadata_disk_)
+    , disk_type(disk_type_)
+    , object_storage(std::move(object_storage_))
+    , send_metadata(send_metadata_)
+    , metadata_helper(std::make_unique<DiskObjectStorageMetadataHelper>(this, ReadSettings{}))
+{}
+
+DiskObjectStorage::Metadata DiskObjectStorage::readMetadataUnlocked(const String & path, std::shared_lock<std::shared_mutex> &) const
+{
+    return Metadata::readMetadata(remote_fs_root_path, metadata_disk, path);
+}
+
+
+DiskObjectStorage::Metadata DiskObjectStorage::readMetadata(const String & path) const
+{
+    std::shared_lock lock(metadata_mutex);
+    return readMetadataUnlocked(path, lock);
+}
+
+DiskObjectStorage::Metadata DiskObjectStorage::readUpdateAndStoreMetadata(const String & path, bool sync, DiskObjectStorage::MetadataUpdater updater)
+{
+    std::unique_lock lock(metadata_mutex);
+    return Metadata::readUpdateAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync, updater);
+}
+
+
+DiskObjectStorage::Metadata DiskObjectStorage::readUpdateStoreMetadataAndRemove(const String & path, bool sync, DiskObjectStorage::MetadataUpdater updater)
+{
+    std::unique_lock lock(metadata_mutex);
+    return Metadata::readUpdateStoreMetadataAndRemove(remote_fs_root_path, metadata_disk, path, sync, updater);
+}
+
+DiskObjectStorage::Metadata DiskObjectStorage::readOrCreateUpdateAndStoreMetadata(const String & path, WriteMode mode, bool sync, DiskObjectStorage::MetadataUpdater updater)
+{
+    if (mode == WriteMode::Rewrite || !metadata_disk->exists(path))
+    {
+        std::unique_lock lock(metadata_mutex);
+        return Metadata::createUpdateAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync, updater);
+    }
+    else
+    {
+        return Metadata::readUpdateAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync, updater);
+    }
+}
+
+DiskObjectStorage::Metadata DiskObjectStorage::createAndStoreMetadata(const String & path, bool sync)
+{
+    return Metadata::createAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync);
+}
+
+DiskObjectStorage::Metadata DiskObjectStorage::createUpdateAndStoreMetadata(const String & path, bool sync, DiskObjectStorage::MetadataUpdater updater)
+{
+    return Metadata::createUpdateAndStoreMetadata(remote_fs_root_path, metadata_disk, path, sync, updater);
+}
+
+std::vector<String> DiskObjectStorage::getRemotePaths(const String & local_path) const
+{
+    auto metadata = readMetadata(local_path);
+
+    std::vector<String> remote_paths;
+    for (const auto & [remote_path, _] : metadata.remote_fs_objects)
+        remote_paths.push_back(fs::path(metadata.remote_fs_root_path) / remote_path);
+
+    return remote_paths;
+
+}
+
+void DiskObjectStorage::getRemotePathsRecursive(const String & local_path, std::vector<LocalPathWithRemotePaths> & paths_map)
+{
+    /// Protect against concurrent delition of files (for example because of a merge).
+    if (metadata_disk->isFile(local_path))
+    {
+        try
+        {
+            paths_map.emplace_back(local_path, getRemotePaths(local_path));
+        }
+        catch (const Exception & e)
+        {
+            if (e.code() == ErrorCodes::FILE_DOESNT_EXIST)
+                return;
+            throw;
+        }
+    }
+    else
+    {
+        DiskDirectoryIteratorPtr it;
+        try
+        {
+            it = iterateDirectory(local_path);
+        }
+        catch (const fs::filesystem_error & e)
+        {
+            if (e.code() == std::errc::no_such_file_or_directory)
+                return;
+            throw;
+        }
+
+        for (; it->isValid(); it->next())
+            DiskObjectStorage::getRemotePathsRecursive(fs::path(local_path) / it->name(), paths_map);
+    }
+}
+
+bool DiskObjectStorage::exists(const String & path) const
+{
+    return metadata_disk->exists(path);
+}
+
+
+bool DiskObjectStorage::isFile(const String & path) const
+{
+    return metadata_disk->isFile(path);
+}
+
+
+void DiskObjectStorage::createFile(const String & path)
+{
+    createAndStoreMetadata(path, false);
+}
+
+size_t DiskObjectStorage::getFileSize(const String & path) const
+{
+    return readMetadata(path).total_size;
+}
+
+void DiskObjectStorage::moveFile(const String & from_path, const String & to_path, bool should_send_metadata)
+{
+    if (exists(to_path))
+        throw Exception("File already exists: " + to_path, ErrorCodes::FILE_ALREADY_EXISTS);
+
+    if (should_send_metadata)
+    {
+        auto revision = metadata_helper->revision_counter + 1;
+        metadata_helper->revision_counter += 1;
+
+        const ObjectAttributes object_metadata {
+            {"from_path", from_path},
+            {"to_path", to_path}
+        };
+        metadata_helper->createFileOperationObject("rename", revision, object_metadata);
+    }
+
+    metadata_disk->moveFile(from_path, to_path);
+}
+
+void DiskObjectStorage::moveFile(const String & from_path, const String & to_path)
+{
+    moveFile(from_path, to_path, send_metadata);
+}
+
+void DiskObjectStorage::replaceFile(const String & from_path, const String & to_path)
+{
+    if (exists(to_path))
+    {
+        const String tmp_path = to_path + ".old";
+        moveFile(to_path, tmp_path);
+        moveFile(from_path, to_path);
+        removeFile(tmp_path);
+    }
+    else
+        moveFile(from_path, to_path);
+}
+
+void DiskObjectStorage::removeSharedFile(const String & path, bool delete_metadata_only)
+{
+    std::vector<String> paths_to_remove;
+    removeMetadata(path, paths_to_remove);
+
+    if (!delete_metadata_only)
+        removeFromRemoteFS(paths_to_remove);
+}
+
+void DiskObjectStorage::removeFromRemoteFS(const std::vector<String> & paths)
+{
+    object_storage->removeObjects(paths);
+}
+
+UInt32 DiskObjectStorage::getRefCount(const String & path) const
+{
+    return readMetadata(path).ref_count;
+}
+
+std::unordered_map<String, String> DiskObjectStorage::getSerializedMetadata(const std::vector<String> & file_paths) const
+{
+    std::unordered_map<String, String> metadatas;
+
+    std::shared_lock lock(metadata_mutex);
+
+    for (const auto & path : file_paths)
+    {
+        DiskObjectStorage::Metadata metadata = readMetadataUnlocked(path, lock);
+        metadata.ref_count = 0;
+        metadatas[path] = metadata.serializeToString();
+    }
+
+    return metadatas;
+}
+
+String DiskObjectStorage::getUniqueId(const String & path) const
+{
+    LOG_TRACE(log, "Remote path: {}, Path: {}", remote_fs_root_path, path);
+    auto metadata = readMetadata(path);
+    String id;
+    if (!metadata.remote_fs_objects.empty())
+        id = metadata.remote_fs_root_path + metadata.remote_fs_objects[0].relative_path;
+    return id;
+}
+
+bool DiskObjectStorage::checkObjectExists(const String & path) const
+{
+    if (!path.starts_with(remote_fs_root_path))
+        return false;
+
+    return object_storage->exists(path);
+}
+
+bool DiskObjectStorage::checkUniqueId(const String & id) const
+{
+    return checkObjectExists(id);
+}
+
+void DiskObjectStorage::createHardLink(const String & src_path, const String & dst_path, bool should_send_metadata)
+{
+    readUpdateAndStoreMetadata(src_path, false, [](Metadata & metadata) { metadata.ref_count++; return true; });
+
+    if (should_send_metadata && !dst_path.starts_with("shadow/"))
+    {
+        auto revision = metadata_helper->revision_counter + 1;
+        metadata_helper->revision_counter += 1;
+        const ObjectAttributes object_metadata {
+            {"src_path", src_path},
+            {"dst_path", dst_path}
+        };
+        metadata_helper->createFileOperationObject("hardlink", revision, object_metadata);
+    }
+
+    /// Create FS hardlink to metadata file.
+    metadata_disk->createHardLink(src_path, dst_path);
+}
+
+void DiskObjectStorage::createHardLink(const String & src_path, const String & dst_path)
+{
+    createHardLink(src_path, dst_path, send_metadata);
+}
+
+
+void DiskObjectStorage::setReadOnly(const String & path)
+{
+    /// We should store read only flag inside metadata file (instead of using FS flag),
+    /// because we modify metadata file when create hard-links from it.
+    readUpdateAndStoreMetadata(path, false, [](Metadata & metadata) { metadata.read_only = true; return true; });
+}
+
+
+bool DiskObjectStorage::isDirectory(const String & path) const
+{
+    return metadata_disk->isDirectory(path);
+}
+
+
+void DiskObjectStorage::createDirectory(const String & path)
+{
+    metadata_disk->createDirectory(path);
+}
+
+
+void DiskObjectStorage::createDirectories(const String & path)
+{
+    metadata_disk->createDirectories(path);
+}
+
+
+void DiskObjectStorage::clearDirectory(const String & path)
+{
+    for (auto it = iterateDirectory(path); it->isValid(); it->next())
+        if (isFile(it->path()))
+            removeFile(it->path());
+}
+
+
+void DiskObjectStorage::removeDirectory(const String & path)
+{
+    metadata_disk->removeDirectory(path);
+}
+
+
+DiskDirectoryIteratorPtr DiskObjectStorage::iterateDirectory(const String & path)
+{
+    return metadata_disk->iterateDirectory(path);
+}
+
+
+void DiskObjectStorage::listFiles(const String & path, std::vector<String> & file_names)
+{
+    for (auto it = iterateDirectory(path); it->isValid(); it->next())
+        file_names.push_back(it->name());
+}
+
+
+void DiskObjectStorage::setLastModified(const String & path, const Poco::Timestamp & timestamp)
+{
+    metadata_disk->setLastModified(path, timestamp);
+}
+
+
+Poco::Timestamp DiskObjectStorage::getLastModified(const String & path)
+{
+    return metadata_disk->getLastModified(path);
+}
+
+void DiskObjectStorage::removeMetadata(const String & path, std::vector<String> & paths_to_remove)
+{
+    LOG_TRACE(log, "Remove file by path: {}", backQuote(metadata_disk->getPath() + path));
+
+    if (!metadata_disk->exists(path))
+        throw Exception(ErrorCodes::FILE_DOESNT_EXIST, "Metadata path '{}' doesn't exist", path);
+
+    if (!metadata_disk->isFile(path))
+        throw Exception(ErrorCodes::BAD_FILE_TYPE, "Path '{}' is not a regular file", path);
+
+    try
+    {
+        auto metadata_updater = [&paths_to_remove, this] (Metadata & metadata)
+        {
+            if (metadata.ref_count == 0)
+            {
+                for (const auto & [remote_fs_object_path, _] : metadata.remote_fs_objects)
+                {
+                    String object_path = fs::path(remote_fs_root_path) / remote_fs_object_path;
+                    paths_to_remove.push_back(object_path);
+                    object_storage->removeFromCache(object_path);
+                }
+
+                return false;
+            }
+            else /// In other case decrement number of references, save metadata and delete hardlink.
+            {
+                --metadata.ref_count;
+            }
+
+            return true;
+        };
+
+        readUpdateStoreMetadataAndRemove(path, false, metadata_updater);
+        /// If there is no references - delete content from remote FS.
+    }
+    catch (const Exception & e)
+    {
+        /// If it's impossible to read meta - just remove it from FS.
+        if (e.code() == ErrorCodes::UNKNOWN_FORMAT)
+        {
+            LOG_WARNING(log,
+                "Metadata file {} can't be read by reason: {}. Removing it forcibly.",
+                backQuote(path), e.nested() ? e.nested()->message() : e.message());
+            metadata_disk->removeFile(path);
+        }
+        else
+            throw;
+    }
+}
+
+
+void DiskObjectStorage::removeMetadataRecursive(const String & path, std::unordered_map<String, std::vector<String>> & paths_to_remove)
+{
+    checkStackSize(); /// This is needed to prevent stack overflow in case of cyclic symlinks.
+
+    if (metadata_disk->isFile(path))
+    {
+        removeMetadata(path, paths_to_remove[path]);
+    }
+    else
+    {
+        for (auto it = iterateDirectory(path); it->isValid(); it->next())
+            removeMetadataRecursive(it->path(), paths_to_remove);
+
+        metadata_disk->removeDirectory(path);
+    }
+}
+
+
+void DiskObjectStorage::shutdown()
+{
+    LOG_INFO(log, "Shutting down disk {}", name);
+    object_storage->shutdown();
+    LOG_INFO(log, "Disk {} shut down", name);
+}
+
+void DiskObjectStorage::startup(ContextPtr context)
+{
+
+    LOG_INFO(log, "Starting up disk {}", name);
+    object_storage->startup();
+
+    restoreMetadataIfNeeded(context->getConfigRef(), "storage_configuration.disks." + name, context);
+
+    LOG_INFO(log, "Disk {} started up", name);
+}
+
+ReservationPtr DiskObjectStorage::reserve(UInt64 bytes)
+{
+    if (!tryReserve(bytes))
+        return {};
+
+    return std::make_unique<DiskObjectStorageReservation>(std::static_pointer_cast<DiskObjectStorage>(shared_from_this()), bytes);
+}
+
+void DiskObjectStorage::removeSharedFileIfExists(const String & path, bool delete_metadata_only)
+{
+    std::vector<String> paths_to_remove;
+    if (metadata_disk->exists(path))
+    {
+        removeMetadata(path, paths_to_remove);
+        if (!delete_metadata_only)
+            removeFromRemoteFS(paths_to_remove);
+    }
+}
+
+void DiskObjectStorage::removeSharedRecursive(const String & path, bool keep_all_batch_data, const NameSet & file_names_remove_metadata_only)
+{
+    std::unordered_map<String, std::vector<String>> paths_to_remove;
+    removeMetadataRecursive(path, paths_to_remove);
+
+    if (!keep_all_batch_data)
+    {
+        std::vector<String> remove_from_remote;
+        for (auto && [local_path, remote_paths] : paths_to_remove)
+        {
+            if (!file_names_remove_metadata_only.contains(fs::path(local_path).filename()))
+            {
+                remove_from_remote.insert(remove_from_remote.end(), remote_paths.begin(), remote_paths.end());
+            }
+        }
+        removeFromRemoteFS(remove_from_remote);
+    }
+}
+
+std::optional<UInt64>  DiskObjectStorage::tryReserve(UInt64 bytes)
+{
+    std::lock_guard lock(reservation_mutex);
+
+    auto available_space = getAvailableSpace();
+    UInt64 unreserved_space = available_space - std::min(available_space, reserved_bytes);
+
+    if (bytes == 0)
+    {
+        LOG_TRACE(log, "Reserving 0 bytes on remote_fs disk {}", backQuote(name));
+        ++reservation_count;
+        return {unreserved_space};
+    }
+
+    if (unreserved_space >= bytes)
+    {
+        LOG_TRACE(log, "Reserving {} on disk {}, having unreserved {}.",
+            ReadableSize(bytes), backQuote(name), ReadableSize(unreserved_space));
+        ++reservation_count;
+        reserved_bytes += bytes;
+        return {unreserved_space - bytes};
+    }
+
+    return {};
+}
+
+std::unique_ptr<ReadBufferFromFileBase> DiskObjectStorage::readFile(
+    const String & path,
+    const ReadSettings & settings,
+    std::optional<size_t> read_hint,
+    std::optional<size_t> file_size) const
+{
+    auto metadata = readMetadata(path);
+    return object_storage->readObjects(remote_fs_root_path, metadata.remote_fs_objects, settings, read_hint, file_size);
+}
+
+std::unique_ptr<WriteBufferFromFileBase> DiskObjectStorage::writeFile(
+    const String & path,
+    size_t buf_size,
+    WriteMode mode,
+    const WriteSettings & settings)
+{
+    auto blob_name = getRandomASCIIString();
+
+    std::optional<ObjectAttributes> object_attributes;
+    if (send_metadata)
+    {
+        auto revision = metadata_helper->revision_counter + 1;
+        metadata_helper->revision_counter++;
+        object_attributes = {
+            {"path", path}
+        };
+        blob_name = "r" + revisionToString(revision) + "-file-" + blob_name;
+    }
+
+    auto create_metadata_callback = [this, path, blob_name, mode] (size_t count)
+    {
+        readOrCreateUpdateAndStoreMetadata(path, mode, false,
+            [blob_name, count] (DiskObjectStorage::Metadata & metadata) { metadata.addObject(blob_name, count); return true; });
+    };
+
+    /// We always use mode Rewrite because we simulate append using metadata and different files
+    return object_storage->writeObject(
+        fs::path(remote_fs_root_path) / blob_name, WriteMode::Rewrite, object_attributes,
+        std::move(create_metadata_callback),
+        buf_size, settings);
+}
+
+
+void DiskObjectStorage::applyNewSettings(const Poco::Util::AbstractConfiguration & config, ContextPtr context_, const String &, const DisksMap &)
+{
+    const auto config_prefix = "storage_configuration.disks." + name;
+    object_storage->applyNewSettings(config, config_prefix, context_);
+
+    if (AsyncThreadPoolExecutor * exec = dynamic_cast<AsyncThreadPoolExecutor *>(&getExecutor()))
+        exec->setMaxThreads(config.getInt(config_prefix + ".thread_pool_size", 16));
+}
+
+void DiskObjectStorage::restoreMetadataIfNeeded(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context)
+{
+    if (send_metadata)
+    {
+        metadata_helper->restore(config, config_prefix, context);
+
+        if (metadata_helper->readSchemaVersion(object_storage.get(), remote_fs_root_path) < DiskObjectStorageMetadataHelper::RESTORABLE_SCHEMA_VERSION)
+            metadata_helper->migrateToRestorableSchema();
+
+        metadata_helper->findLastRevision();
+    }
+}
+
+void DiskObjectStorage::syncRevision(UInt64 revision)
+{
+    metadata_helper->syncRevision(revision);
+}
+
+UInt64 DiskObjectStorage::getRevision() const
+{
+    return metadata_helper->getRevision();
+}
+
+
+DiskPtr DiskObjectStorageReservation::getDisk(size_t i) const
+{
+    if (i != 0)
+        throw Exception("Can't use i != 0 with single disk reservation", ErrorCodes::INCORRECT_DISK_INDEX);
+    return disk;
+}
+
+void DiskObjectStorageReservation::update(UInt64 new_size)
+{
+    std::lock_guard lock(disk->reservation_mutex);
+    disk->reserved_bytes -= size;
+    size = new_size;
+    disk->reserved_bytes += size;
+}
+
+DiskObjectStorageReservation::~DiskObjectStorageReservation()
+{
+    try
+    {
+        std::lock_guard lock(disk->reservation_mutex);
+        if (disk->reserved_bytes < size)
+        {
+            disk->reserved_bytes = 0;
+            LOG_ERROR(disk->log, "Unbalanced reservations size for disk '{}'.", disk->getName());
+        }
+        else
+        {
+            disk->reserved_bytes -= size;
+        }
+
+        if (disk->reservation_count == 0)
+            LOG_ERROR(disk->log, "Unbalanced reservation count for disk '{}'.", disk->getName());
+        else
+            --disk->reservation_count;
+    }
+    catch (...)
+    {
+        tryLogCurrentException(__PRETTY_FUNCTION__);
+    }
+}
+
+
+}

src/Disks/ObjectStorages/DiskObjectStorage.h

@@ -0,0 +1,232 @@
+#pragma once
+
+#include <Disks/IDisk.h>
+#include <Disks/ObjectStorages/IObjectStorage.h>
+#include <Disks/ObjectStorages/DiskObjectStorageMetadataHelper.h>
+#include <Disks/ObjectStorages/DiskObjectStorageMetadata.h>
+#include <re2/re2.h>
+
+namespace CurrentMetrics
+{
+    extern const Metric DiskSpaceReservedForMerge;
+}
+
+namespace DB
+{
+
+/// Disk build on top of IObjectStorage. Use additional disk (local for example)
+/// for metadata storage. Metadata is a small files with mapping from local paths to
+/// objects in object storage, like:
+/// "/var/lib/clickhouse/data/db/table/all_0_0_0/columns.txt" -> /xxxxxxxxxxxxxxxxxxxx
+///                                                           -> /yyyyyyyyyyyyyyyyyyyy
+class DiskObjectStorage : public IDisk
+{
+
+friend class DiskObjectStorageReservation;
+friend class DiskObjectStorageMetadataHelper;
+
+public:
+    DiskObjectStorage(
+        const String & name_,
+        const String & remote_fs_root_path_,
+        const String & log_name,
+        DiskPtr metadata_disk_,
+        ObjectStoragePtr && object_storage_,
+        DiskType disk_type_,
+        bool send_metadata_,
+        uint64_t thread_pool_size);
+
+    DiskType getType() const override { return disk_type; }
+
+    bool supportZeroCopyReplication() const override { return true; }
+
+    bool supportParallelWrite() const override { return true; }
+
+    using Metadata = DiskObjectStorageMetadata;
+    using MetadataUpdater = std::function<bool(Metadata & metadata)>;
+
+    const String & getName() const override { return name; }
+
+    const String & getPath() const override { return metadata_disk->getPath(); }
+
+    std::vector<String> getRemotePaths(const String & local_path) const override;
+
+    void getRemotePathsRecursive(const String & local_path, std::vector<LocalPathWithRemotePaths> & paths_map) override;
+
+    std::string getCacheBasePath() const override
+    {
+        return object_storage->getCacheBasePath();
+    }
+
+    /// Methods for working with metadata. For some operations (like hardlink
+    /// creation) metadata can be updated concurrently from multiple threads
+    /// (file actually rewritten on disk). So additional RW lock is required for
+    /// metadata read and write, but not for create new metadata.
+    Metadata readMetadata(const String & path) const;
+    Metadata readMetadataUnlocked(const String & path, std::shared_lock<std::shared_mutex> &) const;
+    Metadata readUpdateAndStoreMetadata(const String & path, bool sync, MetadataUpdater updater);
+    Metadata readUpdateStoreMetadataAndRemove(const String & path, bool sync, MetadataUpdater updater);
+
+    Metadata readOrCreateUpdateAndStoreMetadata(const String & path, WriteMode mode, bool sync, MetadataUpdater updater);
+
+    Metadata createAndStoreMetadata(const String & path, bool sync);
+    Metadata createUpdateAndStoreMetadata(const String & path, bool sync, MetadataUpdater updater);
+
+    UInt64 getTotalSpace() const override { return std::numeric_limits<UInt64>::max(); }
+
+    UInt64 getAvailableSpace() const override { return std::numeric_limits<UInt64>::max(); }
+
+    UInt64 getUnreservedSpace() const override { return std::numeric_limits<UInt64>::max(); }
+
+    UInt64 getKeepingFreeSpace() const override { return 0; }
+
+    bool exists(const String & path) const override;
+
+    bool isFile(const String & path) const override;
+
+    void createFile(const String & path) override;
+
+    size_t getFileSize(const String & path) const override;
+
+    void moveFile(const String & from_path, const String & to_path) override;
+
+    void moveFile(const String & from_path, const String & to_path, bool should_send_metadata);
+
+    void replaceFile(const String & from_path, const String & to_path) override;
+
+    void removeFile(const String & path) override { removeSharedFile(path, false); }
+
+    void removeFileIfExists(const String & path) override { removeSharedFileIfExists(path, false); }
+
+    void removeRecursive(const String & path) override { removeSharedRecursive(path, false, {}); }
+
+    void removeSharedFile(const String & path, bool delete_metadata_only) override;
+
+    void removeSharedFileIfExists(const String & path, bool delete_metadata_only) override;
+
+    void removeSharedRecursive(const String & path, bool keep_all_batch_data, const NameSet & file_names_remove_metadata_only) override;
+
+    void removeFromRemoteFS(const std::vector<String> & paths);
+
+    DiskPtr getMetadataDiskIfExistsOrSelf() override { return metadata_disk; }
+
+    UInt32 getRefCount(const String & path) const override;
+
+    /// Return metadata for each file path. Also, before serialization reset
+    /// ref_count for each metadata to zero. This function used only for remote
+    /// fetches/sends in replicated engines. That's why we reset ref_count to zero.
+    std::unordered_map<String, String> getSerializedMetadata(const std::vector<String> & file_paths) const override;
+
+    String getUniqueId(const String & path) const override;
+
+    bool checkObjectExists(const String & path) const;
+    bool checkUniqueId(const String & id) const override;
+
+    void createHardLink(const String & src_path, const String & dst_path) override;
+    void createHardLink(const String & src_path, const String & dst_path, bool should_send_metadata);
+
+    void listFiles(const String & path, std::vector<String> & file_names) override;
+
+    void setReadOnly(const String & path) override;
+
+    bool isDirectory(const String & path) const override;
+
+    void createDirectory(const String & path) override;
+
+    void createDirectories(const String & path) override;
+
+    void clearDirectory(const String & path) override;
+
+    void moveDirectory(const String & from_path, const String & to_path) override { moveFile(from_path, to_path); }
+
+    void removeDirectory(const String & path) override;
+
+    DiskDirectoryIteratorPtr iterateDirectory(const String & path) override;
+
+    void setLastModified(const String & path, const Poco::Timestamp & timestamp) override;
+
+    Poco::Timestamp getLastModified(const String & path) override;
+
+    bool isRemote() const override { return true; }
+
+    void shutdown() override;
+
+    void startup(ContextPtr context) override;
+
+    ReservationPtr reserve(UInt64 bytes) override;
+
+    std::unique_ptr<ReadBufferFromFileBase> readFile(
+        const String & path,
+        const ReadSettings & settings,
+        std::optional<size_t> read_hint,
+        std::optional<size_t> file_size) const override;
+
+    std::unique_ptr<WriteBufferFromFileBase> writeFile(
+        const String & path,
+        size_t buf_size,
+        WriteMode mode,
+        const WriteSettings & settings) override;
+
+    void applyNewSettings(const Poco::Util::AbstractConfiguration & config, ContextPtr context_, const String &, const DisksMap &) override;
+
+    void restoreMetadataIfNeeded(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context);
+
+    void onFreeze(const String & path) override;
+
+    void syncRevision(UInt64 revision) override;
+
+    UInt64 getRevision() const override;
+private:
+    const String name;
+    const String remote_fs_root_path;
+    Poco::Logger * log;
+    DiskPtr metadata_disk;
+
+    const DiskType disk_type;
+    ObjectStoragePtr object_storage;
+
+    UInt64 reserved_bytes = 0;
+    UInt64 reservation_count = 0;
+    std::mutex reservation_mutex;
+
+    mutable std::shared_mutex metadata_mutex;
+    void removeMetadata(const String & path, std::vector<String> & paths_to_remove);
+
+    void removeMetadataRecursive(const String & path, std::unordered_map<String, std::vector<String>> & paths_to_remove);
+
+    std::optional<UInt64> tryReserve(UInt64 bytes);
+
+    bool send_metadata;
+
+    std::unique_ptr<DiskObjectStorageMetadataHelper> metadata_helper;
+};
+
+class DiskObjectStorageReservation final : public IReservation
+{
+public:
+    DiskObjectStorageReservation(const std::shared_ptr<DiskObjectStorage> & disk_, UInt64 size_)
+        : disk(disk_)
+        , size(size_)
+        , metric_increment(CurrentMetrics::DiskSpaceReservedForMerge, size_)
+    {}
+
+    UInt64 getSize() const override { return size; }
+
+    UInt64 getUnreservedSpace() const override { return unreserved_space; }
+
+    DiskPtr getDisk(size_t i) const override;
+
+    Disks getDisks() const override { return {disk}; }
+
+    void update(UInt64 new_size) override;
+
+    ~DiskObjectStorageReservation() override;
+
+private:
+    std::shared_ptr<DiskObjectStorage> disk;
+    UInt64 size;
+    UInt64 unreserved_space;
+    CurrentMetrics::Increment metric_increment;
+};
+
+}

src/Disks/ObjectStorages/DiskObjectStorageCommon.cpp

@@ -1,4 +1,4 @@
-#include <Disks/RemoteDisksCommon.h>
+#include <Disks/ObjectStorages/DiskObjectStorageCommon.h>
 #include <Common/getRandomASCIIString.h>
 #include <Common/FileCacheFactory.h>
 #include <Common/FileCache.h>
@@ -8,7 +8,8 @@ namespace DB
 {
 
 namespace ErrorCodes
-{extern const int BAD_ARGUMENTS;
+{
+    extern const int BAD_ARGUMENTS;
 }
 
 std::shared_ptr<DiskCacheWrapper> wrapWithCache(

src/Disks/ObjectStorages/DiskObjectStorageCommon.h

@@ -2,18 +2,22 @@
 
 #include <random>
 #include <utility>
+
 #include <Core/Types.h>
 #include <Common/thread_local_rng.h>
-#include <Disks/IDisk.h>
-#include <Disks/DiskCacheWrapper.h>
 #include <Common/getRandomASCIIString.h>
 
+#include <Disks/IDisk.h>
+#include <Disks/DiskCacheWrapper.h>
 
 namespace DB
 {
 
 std::shared_ptr<DiskCacheWrapper> wrapWithCache(
-    std::shared_ptr<IDisk> disk, String cache_name, String cache_path, String metadata_path);
+    std::shared_ptr<IDisk> disk,
+    String cache_name,
+    String cache_path,
+    String metadata_path);
 
 std::pair<String, DiskPtr> prepareForLocalMetadata(
     const String & name,

src/Disks/ObjectStorages/DiskObjectStorageMetadata.cpp

@@ -0,0 +1,208 @@
+#include <Disks/ObjectStorages/DiskObjectStorageMetadata.h>
+
+#include <IO/ReadHelpers.h>
+#include <IO/WriteHelpers.h>
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int UNKNOWN_FORMAT;
+    extern const int PATH_ACCESS_DENIED;
+    extern const int MEMORY_LIMIT_EXCEEDED;
+}
+
+DiskObjectStorageMetadata DiskObjectStorageMetadata::readMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_)
+{
+    DiskObjectStorageMetadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
+    result.load();
+    return result;
+}
+
+
+DiskObjectStorageMetadata DiskObjectStorageMetadata::createAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync)
+{
+    DiskObjectStorageMetadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
+    result.save(sync);
+    return result;
+}
+
+DiskObjectStorageMetadata DiskObjectStorageMetadata::readUpdateAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, DiskObjectStorageMetadataUpdater updater)
+{
+    DiskObjectStorageMetadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
+    result.load();
+    if (updater(result))
+        result.save(sync);
+    return result;
+}
+
+DiskObjectStorageMetadata DiskObjectStorageMetadata::createUpdateAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, DiskObjectStorageMetadataUpdater updater)
+{
+    DiskObjectStorageMetadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
+    updater(result);
+    result.save(sync);
+    return result;
+}
+
+DiskObjectStorageMetadata DiskObjectStorageMetadata::readUpdateStoreMetadataAndRemove(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, DiskObjectStorageMetadataUpdater updater)
+{
+    DiskObjectStorageMetadata result(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
+    result.load();
+    if (updater(result))
+        result.save(sync);
+    metadata_disk_->removeFile(metadata_file_path_);
+
+    return result;
+
+}
+
+DiskObjectStorageMetadata DiskObjectStorageMetadata::createAndStoreMetadataIfNotExists(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, bool overwrite)
+{
+    if (overwrite || !metadata_disk_->exists(metadata_file_path_))
+    {
+        return createAndStoreMetadata(remote_fs_root_path_, metadata_disk_, metadata_file_path_, sync);
+    }
+    else
+    {
+        auto result = readMetadata(remote_fs_root_path_, metadata_disk_, metadata_file_path_);
+        if (result.read_only)
+            throw Exception("File is read-only: " + metadata_file_path_, ErrorCodes::PATH_ACCESS_DENIED);
+        return result;
+    }
+}
+
+void DiskObjectStorageMetadata::load()
+{
+    try
+    {
+        const ReadSettings read_settings;
+        auto buf = metadata_disk->readFile(metadata_file_path, read_settings, 1024);  /* reasonable buffer size for small file */
+
+        UInt32 version;
+        readIntText(version, *buf);
+
+        if (version < VERSION_ABSOLUTE_PATHS || version > VERSION_READ_ONLY_FLAG)
+            throw Exception(
+                ErrorCodes::UNKNOWN_FORMAT,
+                "Unknown metadata file version. Path: {}. Version: {}. Maximum expected version: {}",
+                metadata_disk->getPath() + metadata_file_path, toString(version), toString(VERSION_READ_ONLY_FLAG));
+
+        assertChar('\n', *buf);
+
+        UInt32 remote_fs_objects_count;
+        readIntText(remote_fs_objects_count, *buf);
+        assertChar('\t', *buf);
+        readIntText(total_size, *buf);
+        assertChar('\n', *buf);
+        remote_fs_objects.resize(remote_fs_objects_count);
+
+        for (size_t i = 0; i < remote_fs_objects_count; ++i)
+        {
+            String remote_fs_object_path;
+            size_t remote_fs_object_size;
+            readIntText(remote_fs_object_size, *buf);
+            assertChar('\t', *buf);
+            readEscapedString(remote_fs_object_path, *buf);
+            if (version == VERSION_ABSOLUTE_PATHS)
+            {
+                if (!remote_fs_object_path.starts_with(remote_fs_root_path))
+                    throw Exception(ErrorCodes::UNKNOWN_FORMAT,
+                        "Path in metadata does not correspond to root path. Path: {}, root path: {}, disk path: {}",
+                        remote_fs_object_path, remote_fs_root_path, metadata_disk->getPath());
+
+                remote_fs_object_path = remote_fs_object_path.substr(remote_fs_root_path.size());
+            }
+            assertChar('\n', *buf);
+            remote_fs_objects[i].relative_path = remote_fs_object_path;
+            remote_fs_objects[i].bytes_size = remote_fs_object_size;
+        }
+
+        readIntText(ref_count, *buf);
+        assertChar('\n', *buf);
+
+        if (version >= VERSION_READ_ONLY_FLAG)
+        {
+            readBoolText(read_only, *buf);
+            assertChar('\n', *buf);
+        }
+    }
+    catch (Exception & e)
+    {
+        tryLogCurrentException(__PRETTY_FUNCTION__);
+
+        if (e.code() == ErrorCodes::UNKNOWN_FORMAT)
+            throw;
+
+        if (e.code() == ErrorCodes::MEMORY_LIMIT_EXCEEDED)
+            throw;
+
+        throw Exception("Failed to read metadata file: " + metadata_file_path, ErrorCodes::UNKNOWN_FORMAT);
+    }
+}
+
+/// Load metadata by path or create empty if `create` flag is set.
+DiskObjectStorageMetadata::DiskObjectStorageMetadata(
+        const String & remote_fs_root_path_,
+        DiskPtr metadata_disk_,
+        const String & metadata_file_path_)
+    : remote_fs_root_path(remote_fs_root_path_)
+    , metadata_file_path(metadata_file_path_)
+    , metadata_disk(metadata_disk_)
+    , total_size(0), ref_count(0)
+{
+}
+
+void DiskObjectStorageMetadata::addObject(const String & path, size_t size)
+{
+    total_size += size;
+    remote_fs_objects.emplace_back(path, size);
+}
+
+
+void DiskObjectStorageMetadata::saveToBuffer(WriteBuffer & buf, bool sync)
+{
+    writeIntText(VERSION_RELATIVE_PATHS, buf);
+    writeChar('\n', buf);
+
+    writeIntText(remote_fs_objects.size(), buf);
+    writeChar('\t', buf);
+    writeIntText(total_size, buf);
+    writeChar('\n', buf);
+
+    for (const auto & [remote_fs_object_path, remote_fs_object_size] : remote_fs_objects)
+    {
+        writeIntText(remote_fs_object_size, buf);
+        writeChar('\t', buf);
+        writeEscapedString(remote_fs_object_path, buf);
+        writeChar('\n', buf);
+    }
+
+    writeIntText(ref_count, buf);
+    writeChar('\n', buf);
+
+    writeBoolText(read_only, buf);
+    writeChar('\n', buf);
+
+    buf.finalize();
+    if (sync)
+        buf.sync();
+
+}
+
+/// Fsync metadata file if 'sync' flag is set.
+void DiskObjectStorageMetadata::save(bool sync)
+{
+    auto buf = metadata_disk->writeFile(metadata_file_path, 1024);
+    saveToBuffer(*buf, sync);
+}
+
+std::string DiskObjectStorageMetadata::serializeToString()
+{
+    WriteBufferFromOwnString write_buf;
+    saveToBuffer(write_buf, false);
+    return write_buf.str();
+}
+
+
+}

src/Disks/ObjectStorages/DiskObjectStorageMetadata.h

@@ -0,0 +1,68 @@
+#pragma once
+
+#include <Disks/IDisk.h>
+#include <Core/Types.h>
+
+namespace DB
+{
+
+/// Metadata for DiskObjectStorage, stored on local disk
+struct DiskObjectStorageMetadata
+{
+    using Updater = std::function<bool(DiskObjectStorageMetadata & metadata)>;
+    /// Metadata file version.
+    static constexpr UInt32 VERSION_ABSOLUTE_PATHS = 1;
+    static constexpr UInt32 VERSION_RELATIVE_PATHS = 2;
+    static constexpr UInt32 VERSION_READ_ONLY_FLAG = 3;
+
+    /// Remote FS objects paths and their sizes.
+    std::vector<BlobPathWithSize> remote_fs_objects;
+
+    /// URI
+    const String & remote_fs_root_path;
+
+    /// Relative path to metadata file on local FS.
+    const String metadata_file_path;
+
+    DiskPtr metadata_disk;
+
+    /// Total size of all remote FS (S3, HDFS) objects.
+    size_t total_size = 0;
+
+    /// Number of references (hardlinks) to this metadata file.
+    ///
+    /// FIXME: Why we are tracking it explicetly, without
+    /// info from filesystem????
+    UInt32 ref_count = 0;
+
+    /// Flag indicates that file is read only.
+    bool read_only = false;
+
+    DiskObjectStorageMetadata(
+        const String & remote_fs_root_path_,
+        DiskPtr metadata_disk_,
+        const String & metadata_file_path_);
+
+    void addObject(const String & path, size_t size);
+
+    static DiskObjectStorageMetadata readMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_);
+    static DiskObjectStorageMetadata readUpdateAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, Updater updater);
+    static DiskObjectStorageMetadata readUpdateStoreMetadataAndRemove(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, Updater updater);
+
+    static DiskObjectStorageMetadata createAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync);
+    static DiskObjectStorageMetadata createUpdateAndStoreMetadata(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, Updater updater);
+    static DiskObjectStorageMetadata createAndStoreMetadataIfNotExists(const String & remote_fs_root_path_, DiskPtr metadata_disk_, const String & metadata_file_path_, bool sync, bool overwrite);
+
+    /// Serialize metadata to string (very same with saveToBuffer)
+    std::string serializeToString();
+
+private:
+    /// Fsync metadata file if 'sync' flag is set.
+    void save(bool sync = false);
+    void saveToBuffer(WriteBuffer & buffer, bool sync);
+    void load();
+};
+
+using DiskObjectStorageMetadataUpdater = std::function<bool(DiskObjectStorageMetadata & metadata)>;
+
+}

src/Disks/ObjectStorages/DiskObjectStorageMetadataHelper.cpp

@@ -0,0 +1,571 @@
+#include <Disks/ObjectStorages/DiskObjectStorageMetadataHelper.h>
+#include <Disks/ObjectStorages/DiskObjectStorage.h>
+#include <IO/WriteHelpers.h>
+#include <IO/ReadHelpers.h>
+#include <IO/ReadBufferFromFile.h>
+#include <IO/ReadBufferFromString.h>
+#include <Common/checkStackSize.h>
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int UNKNOWN_FORMAT;
+    extern const int LOGICAL_ERROR;
+    extern const int BAD_ARGUMENTS;
+}
+
+static String revisionToString(UInt64 revision)
+{
+    return std::bitset<64>(revision).to_string();
+}
+
+void DiskObjectStorageMetadataHelper::createFileOperationObject(const String & operation_name, UInt64 revision, const ObjectAttributes & metadata) const
+{
+    const String path = disk->remote_fs_root_path + "operations/r" + revisionToString(revision) + operation_log_suffix + "-" + operation_name;
+    auto buf = disk->object_storage->writeObject(path, WriteMode::Rewrite, metadata);
+    buf->write('0');
+    buf->finalize();
+}
+
+void DiskObjectStorageMetadataHelper::findLastRevision()
+{
+    /// Construct revision number from high to low bits.
+    String revision;
+    revision.reserve(64);
+    for (int bit = 0; bit < 64; ++bit)
+    {
+        auto revision_prefix = revision + "1";
+
+        LOG_TRACE(disk->log, "Check object exists with revision prefix {}", revision_prefix);
+
+        /// Check file or operation with such revision prefix exists.
+        if (disk->object_storage->exists(disk->remote_fs_root_path + "r" + revision_prefix)
+            || disk->object_storage->exists(disk->remote_fs_root_path + "operations/r" + revision_prefix))
+            revision += "1";
+        else
+            revision += "0";
+    }
+    revision_counter = static_cast<UInt64>(std::bitset<64>(revision).to_ullong());
+    LOG_INFO(disk->log, "Found last revision number {} for disk {}", revision_counter, disk->name);
+}
+
+int DiskObjectStorageMetadataHelper::readSchemaVersion(IObjectStorage * object_storage, const String & source_path)
+{
+    const std::string path = source_path + SCHEMA_VERSION_OBJECT;
+    int version = 0;
+    if (!object_storage->exists(path))
+        return version;
+
+    auto buf = object_storage->readObject(path);
+    readIntText(version, *buf);
+
+    return version;
+}
+
+void DiskObjectStorageMetadataHelper::saveSchemaVersion(const int & version) const
+{
+    auto path = disk->remote_fs_root_path + SCHEMA_VERSION_OBJECT;
+
+    auto buf = disk->object_storage->writeObject(path, WriteMode::Rewrite);
+    writeIntText(version, *buf);
+    buf->finalize();
+
+}
+
+void DiskObjectStorageMetadataHelper::updateObjectMetadata(const String & key, const ObjectAttributes & metadata) const
+{
+    disk->object_storage->copyObject(key, key, metadata);
+}
+
+void DiskObjectStorageMetadataHelper::migrateFileToRestorableSchema(const String & path) const
+{
+    LOG_TRACE(disk->log, "Migrate file {} to restorable schema", disk->metadata_disk->getPath() + path);
+
+    auto meta = disk->readMetadata(path);
+
+    for (const auto & [key, _] : meta.remote_fs_objects)
+    {
+        ObjectAttributes metadata {
+            {"path", path}
+        };
+        updateObjectMetadata(disk->remote_fs_root_path + key, metadata);
+    }
+}
+void DiskObjectStorageMetadataHelper::migrateToRestorableSchemaRecursive(const String & path, Futures & results)
+{
+    checkStackSize(); /// This is needed to prevent stack overflow in case of cyclic symlinks.
+
+    LOG_TRACE(disk->log, "Migrate directory {} to restorable schema", disk->metadata_disk->getPath() + path);
+
+    bool dir_contains_only_files = true;
+    for (auto it = disk->iterateDirectory(path); it->isValid(); it->next())
+    {
+        if (disk->isDirectory(it->path()))
+        {
+            dir_contains_only_files = false;
+            break;
+        }
+    }
+
+    /// The whole directory can be migrated asynchronously.
+    if (dir_contains_only_files)
+    {
+        auto result = disk->getExecutor().execute([this, path]
+        {
+            for (auto it = disk->iterateDirectory(path); it->isValid(); it->next())
+                migrateFileToRestorableSchema(it->path());
+        });
+
+        results.push_back(std::move(result));
+    }
+    else
+    {
+        for (auto it = disk->iterateDirectory(path); it->isValid(); it->next())
+            if (!disk->isDirectory(it->path()))
+            {
+                auto source_path = it->path();
+                auto result = disk->getExecutor().execute([this, source_path]
+                    {
+                        migrateFileToRestorableSchema(source_path);
+                    });
+
+                results.push_back(std::move(result));
+            }
+            else
+                migrateToRestorableSchemaRecursive(it->path(), results);
+    }
+
+}
+
+void DiskObjectStorageMetadataHelper::migrateToRestorableSchema()
+{
+    try
+    {
+        LOG_INFO(disk->log, "Start migration to restorable schema for disk {}", disk->name);
+
+        Futures results;
+
+        for (const auto & root : data_roots)
+            if (disk->exists(root))
+                migrateToRestorableSchemaRecursive(root + '/', results);
+
+        for (auto & result : results)
+            result.wait();
+        for (auto & result : results)
+            result.get();
+
+        saveSchemaVersion(RESTORABLE_SCHEMA_VERSION);
+    }
+    catch (const Exception &)
+    {
+        tryLogCurrentException(disk->log, fmt::format("Failed to migrate to restorable schema for disk {}", disk->name));
+
+        throw;
+    }
+}
+
+void DiskObjectStorageMetadataHelper::restore(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context)
+{
+    LOG_INFO(disk->log, "Restore operation for disk {} called", disk->name);
+
+    if (!disk->exists(RESTORE_FILE_NAME))
+    {
+        LOG_INFO(disk->log, "No restore file '{}' exists, finishing restore", RESTORE_FILE_NAME);
+        return;
+    }
+
+    try
+    {
+        RestoreInformation information;
+        information.source_path = disk->remote_fs_root_path;
+        information.source_namespace = disk->object_storage->getObjectsNamespace();
+
+        readRestoreInformation(information);
+        if (information.revision == 0)
+            information.revision = LATEST_REVISION;
+        if (!information.source_path.ends_with('/'))
+            information.source_path += '/';
+
+        IObjectStorage * source_object_storage = disk->object_storage.get();
+        if (information.source_namespace == disk->object_storage->getObjectsNamespace())
+        {
+            /// In this case we need to additionally cleanup S3 from objects with later revision.
+            /// Will be simply just restore to different path.
+            if (information.source_path == disk->remote_fs_root_path && information.revision != LATEST_REVISION)
+                throw Exception("Restoring to the same bucket and path is allowed if revision is latest (0)", ErrorCodes::BAD_ARGUMENTS);
+
+            /// This case complicates S3 cleanup in case of unsuccessful restore.
+            if (information.source_path != disk->remote_fs_root_path && disk->remote_fs_root_path.starts_with(information.source_path))
+                throw Exception("Restoring to the same bucket is allowed only if source path is not a sub-path of configured path in S3 disk", ErrorCodes::BAD_ARGUMENTS);
+        }
+        else
+        {
+            object_storage_from_another_namespace = disk->object_storage->cloneObjectStorage(information.source_namespace, config, config_prefix, context);
+            source_object_storage = object_storage_from_another_namespace.get();
+        }
+
+        LOG_INFO(disk->log, "Starting to restore disk {}. Revision: {}, Source path: {}",
+                 disk->name, information.revision, information.source_path);
+
+        if (readSchemaVersion(source_object_storage, information.source_path) < RESTORABLE_SCHEMA_VERSION)
+            throw Exception("Source bucket doesn't have restorable schema.", ErrorCodes::BAD_ARGUMENTS);
+
+        LOG_INFO(disk->log, "Removing old metadata...");
+
+        bool cleanup_s3 = information.source_path != disk->remote_fs_root_path;
+        for (const auto & root : data_roots)
+            if (disk->exists(root))
+                disk->removeSharedRecursive(root + '/', !cleanup_s3, {});
+
+        LOG_INFO(disk->log, "Old metadata removed, restoring new one");
+        restoreFiles(source_object_storage, information);
+        restoreFileOperations(source_object_storage, information);
+
+        disk->metadata_disk->removeFile(RESTORE_FILE_NAME);
+
+        saveSchemaVersion(RESTORABLE_SCHEMA_VERSION);
+
+        LOG_INFO(disk->log, "Restore disk {} finished", disk->name);
+    }
+    catch (const Exception &)
+    {
+        tryLogCurrentException(disk->log, fmt::format("Failed to restore disk {}", disk->name));
+
+        throw;
+    }
+}
+
+void DiskObjectStorageMetadataHelper::readRestoreInformation(RestoreInformation & restore_information) /// NOLINT
+{
+    auto buffer = disk->metadata_disk->readFile(RESTORE_FILE_NAME, ReadSettings{}, 512);
+    buffer->next();
+
+    try
+    {
+        std::map<String, String> properties;
+
+        while (buffer->hasPendingData())
+        {
+            String property;
+            readText(property, *buffer);
+            assertChar('\n', *buffer);
+
+            auto pos = property.find('=');
+            if (pos == std::string::npos || pos == 0 || pos == property.length())
+                throw Exception(fmt::format("Invalid property {} in restore file", property), ErrorCodes::UNKNOWN_FORMAT);
+
+            auto key = property.substr(0, pos);
+            auto value = property.substr(pos + 1);
+
+            auto it = properties.find(key);
+            if (it != properties.end())
+                throw Exception(fmt::format("Property key duplication {} in restore file", key), ErrorCodes::UNKNOWN_FORMAT);
+
+            properties[key] = value;
+        }
+
+        for (const auto & [key, value] : properties)
+        {
+            ReadBufferFromString value_buffer(value);
+
+            if (key == "revision")
+                readIntText(restore_information.revision, value_buffer);
+            else if (key == "source_bucket" || key == "source_namespace")
+                readText(restore_information.source_namespace, value_buffer);
+            else if (key == "source_path")
+                readText(restore_information.source_path, value_buffer);
+            else if (key == "detached")
+                readBoolTextWord(restore_information.detached, value_buffer);
+            else
+                throw Exception(fmt::format("Unknown key {} in restore file", key), ErrorCodes::UNKNOWN_FORMAT);
+        }
+    }
+    catch (const Exception &)
+    {
+        tryLogCurrentException(disk->log, "Failed to read restore information");
+        throw;
+    }
+}
+
+static String shrinkKey(const String & path, const String & key)
+{
+    if (!key.starts_with(path))
+        throw Exception("The key " + key + " prefix mismatch with given " + path, ErrorCodes::LOGICAL_ERROR);
+
+    return key.substr(path.length());
+}
+
+static std::tuple<UInt64, String> extractRevisionAndOperationFromKey(const String & key)
+{
+    String revision_str;
+    String suffix;
+    String operation;
+    /// Key has format: ../../r{revision}(-{hostname})-{operation}
+    static const re2::RE2 key_regexp{R"(.*/r(\d+)(-[\w\d\-\.]+)?-(\w+)$)"};
+
+    re2::RE2::FullMatch(key, key_regexp, &revision_str, &suffix, &operation);
+
+    return {(revision_str.empty() ? 0 : static_cast<UInt64>(std::bitset<64>(revision_str).to_ullong())), operation};
+}
+
+void DiskObjectStorageMetadataHelper::moveRecursiveOrRemove(const String & from_path, const String & to_path, bool send_metadata)
+{
+    if (disk->exists(to_path))
+    {
+        if (send_metadata)
+        {
+            auto revision = ++revision_counter;
+            const ObjectAttributes object_metadata {
+                {"from_path", from_path},
+                {"to_path", to_path}
+            };
+            createFileOperationObject("rename", revision, object_metadata);
+        }
+        if (disk->isDirectory(from_path))
+        {
+            for (auto it = disk->iterateDirectory(from_path); it->isValid(); it->next())
+                moveRecursiveOrRemove(it->path(), fs::path(to_path) / it->name(), false);
+        }
+        else
+        {
+            disk->removeFile(from_path);
+        }
+    }
+    else
+    {
+        disk->moveFile(from_path, to_path, send_metadata);
+    }
+}
+
+void DiskObjectStorageMetadataHelper::restoreFiles(IObjectStorage * source_object_storage, const RestoreInformation & restore_information)
+{
+    LOG_INFO(disk->log, "Starting restore files for disk {}", disk->name);
+
+    std::vector<std::future<void>> results;
+    auto restore_files = [this, &source_object_storage, &restore_information, &results](const BlobsPathToSize & keys)
+    {
+        std::vector<String> keys_names;
+        for (const auto & [key, size] : keys)
+        {
+
+            LOG_INFO(disk->log, "Calling restore for key for disk {}", key);
+
+            /// Skip file operations objects. They will be processed separately.
+            if (key.find("/operations/") != String::npos)
+                continue;
+
+            const auto [revision, _] = extractRevisionAndOperationFromKey(key);
+            /// Filter early if it's possible to get revision from key.
+            if (revision > restore_information.revision)
+                continue;
+
+            keys_names.push_back(key);
+        }
+
+        if (!keys_names.empty())
+        {
+            auto result = disk->getExecutor().execute([this, &source_object_storage, &restore_information, keys_names]()
+            {
+                processRestoreFiles(source_object_storage, restore_information.source_path, keys_names);
+            });
+
+            results.push_back(std::move(result));
+        }
+
+        return true;
+    };
+
+    BlobsPathToSize children;
+    source_object_storage->listPrefix(restore_information.source_path, children);
+
+    restore_files(children);
+
+    for (auto & result : results)
+        result.wait();
+    for (auto & result : results)
+        result.get();
+
+    LOG_INFO(disk->log, "Files are restored for disk {}", disk->name);
+
+}
+
+void DiskObjectStorageMetadataHelper::processRestoreFiles(IObjectStorage * source_object_storage, const String & source_path, const std::vector<String> & keys) const
+{
+    for (const auto & key : keys)
+    {
+        auto meta = source_object_storage->getObjectMetadata(key);
+        auto object_attributes = meta.attributes;
+
+        String path;
+        if (object_attributes.has_value())
+        {
+            /// Restore file if object has 'path' in metadata.
+            auto path_entry = object_attributes->find("path");
+            if (path_entry == object_attributes->end())
+            {
+                /// Such keys can remain after migration, we can skip them.
+                LOG_WARNING(disk->log, "Skip key {} because it doesn't have 'path' in metadata", key);
+                continue;
+            }
+
+            path = path_entry->second;
+        }
+        else
+            continue;
+
+        disk->createDirectories(directoryPath(path));
+        auto relative_key = shrinkKey(source_path, key);
+
+        /// Copy object if we restore to different bucket / path.
+        if (source_object_storage->getObjectsNamespace() != disk->object_storage->getObjectsNamespace() || disk->remote_fs_root_path != source_path)
+            source_object_storage->copyObjectToAnotherObjectStorage(key, disk->remote_fs_root_path + relative_key, *disk->object_storage);
+
+        auto updater = [relative_key, meta] (DiskObjectStorage::Metadata & metadata)
+        {
+            metadata.addObject(relative_key, meta.size_bytes);
+            return true;
+        };
+
+        disk->createUpdateAndStoreMetadata(path, false, updater);
+
+        LOG_TRACE(disk->log, "Restored file {}", path);
+    }
+
+}
+
+void DiskObjectStorage::onFreeze(const String & path)
+{
+    createDirectories(path);
+    auto revision_file_buf = metadata_disk->writeFile(path + "revision.txt", 32);
+    writeIntText(metadata_helper->revision_counter.load(), *revision_file_buf);
+    revision_file_buf->finalize();
+}
+
+static String pathToDetached(const String & source_path)
+{
+    if (source_path.ends_with('/'))
+        return fs::path(source_path).parent_path().parent_path() / "detached/";
+    return fs::path(source_path).parent_path() / "detached/";
+}
+
+void DiskObjectStorageMetadataHelper::restoreFileOperations(IObjectStorage * source_object_storage, const RestoreInformation & restore_information)
+{
+    /// Enable recording file operations if we restore to different bucket / path.
+    bool send_metadata = source_object_storage->getObjectsNamespace() != disk->object_storage->getObjectsNamespace() || disk->remote_fs_root_path != restore_information.source_path;
+
+    std::set<String> renames;
+    auto restore_file_operations = [this, &source_object_storage, &restore_information, &renames, &send_metadata](const BlobsPathToSize & keys)
+    {
+        const String rename = "rename";
+        const String hardlink = "hardlink";
+
+        for (const auto & [key, _]: keys)
+        {
+            const auto [revision, operation] = extractRevisionAndOperationFromKey(key);
+            if (revision == UNKNOWN_REVISION)
+            {
+                LOG_WARNING(disk->log, "Skip key {} with unknown revision", key);
+                continue;
+            }
+
+            /// S3 ensures that keys will be listed in ascending UTF-8 bytes order (revision order).
+            /// We can stop processing if revision of the object is already more than required.
+            if (revision > restore_information.revision)
+                return false;
+
+            /// Keep original revision if restore to different bucket / path.
+            if (send_metadata)
+                revision_counter = revision - 1;
+
+            auto object_attributes = *(source_object_storage->getObjectMetadata(key).attributes);
+            if (operation == rename)
+            {
+                auto from_path = object_attributes["from_path"];
+                auto to_path = object_attributes["to_path"];
+                if (disk->exists(from_path))
+                {
+                    moveRecursiveOrRemove(from_path, to_path, send_metadata);
+
+                    LOG_TRACE(disk->log, "Revision {}. Restored rename {} -> {}", revision, from_path, to_path);
+
+                    if (restore_information.detached && disk->isDirectory(to_path))
+                    {
+                        /// Sometimes directory paths are passed without trailing '/'. We should keep them in one consistent way.
+                        if (!from_path.ends_with('/'))
+                            from_path += '/';
+                        if (!to_path.ends_with('/'))
+                            to_path += '/';
+
+                        /// Always keep latest actual directory path to avoid 'detaching' not existing paths.
+                        auto it = renames.find(from_path);
+                        if (it != renames.end())
+                            renames.erase(it);
+
+                        renames.insert(to_path);
+                    }
+                }
+            }
+            else if (operation == hardlink)
+            {
+                auto src_path = object_attributes["src_path"];
+                auto dst_path = object_attributes["dst_path"];
+                if (disk->exists(src_path))
+                {
+                    disk->createDirectories(directoryPath(dst_path));
+                    disk->createHardLink(src_path, dst_path, send_metadata);
+                    LOG_TRACE(disk->log, "Revision {}. Restored hardlink {} -> {}", revision, src_path, dst_path);
+                }
+            }
+        }
+
+        return true;
+    };
+
+    BlobsPathToSize children;
+    source_object_storage->listPrefix(restore_information.source_path + "operations/", children);
+    restore_file_operations(children);
+
+    if (restore_information.detached)
+    {
+        Strings not_finished_prefixes{"tmp_", "delete_tmp_", "attaching_", "deleting_"};
+
+        for (const auto & path : renames)
+        {
+            /// Skip already detached parts.
+            if (path.find("/detached/") != std::string::npos)
+                continue;
+
+            /// Skip not finished parts. They shouldn't be in 'detached' directory, because CH wouldn't be able to finish processing them.
+            fs::path directory_path(path);
+            auto directory_name = directory_path.parent_path().filename().string();
+
+            auto predicate = [&directory_name](String & prefix) { return directory_name.starts_with(prefix); };
+            if (std::any_of(not_finished_prefixes.begin(), not_finished_prefixes.end(), predicate))
+                continue;
+
+            auto detached_path = pathToDetached(path);
+
+            LOG_TRACE(disk->log, "Move directory to 'detached' {} -> {}", path, detached_path);
+
+            fs::path from_path = fs::path(path);
+            fs::path to_path = fs::path(detached_path);
+            if (path.ends_with('/'))
+                to_path /= from_path.parent_path().filename();
+            else
+                to_path /= from_path.filename();
+
+            /// to_path may exist and non-empty in case for example abrupt restart, so remove it before rename
+            if (disk->metadata_disk->exists(to_path))
+                disk->metadata_disk->removeRecursive(to_path);
+
+            disk->createDirectories(directoryPath(to_path));
+            disk->metadata_disk->moveDirectory(from_path, to_path);
+        }
+    }
+
+    LOG_INFO(disk->log, "File operations restored for disk {}", disk->name);
+}
+
+}

src/Disks/ObjectStorages/DiskObjectStorageMetadataHelper.h

@@ -0,0 +1,100 @@
+#pragma once
+
+#include <Disks/ObjectStorages/IObjectStorage.h>
+#include <base/getFQDNOrHostName.h>
+
+namespace DB
+{
+
+class DiskObjectStorage;
+
+/// Class implements storage of ObjectStorage metadata inside object storage itself,
+/// so it's possible to recover from this remote information in case of local disk loss.
+///
+/// This mechanism can be enabled with `<send_metadata>true</send_metadata>` option inside
+/// disk configuration. Implemented only for S3 and Azure Blob storage. Other object storages
+/// don't support metadata for blobs.
+///
+/// FIXME: this class is very intrusive and use a lot of DiskObjectStorage internals.
+/// FIXME: it's very complex and unreliable, need to implement something better.
+class DiskObjectStorageMetadataHelper
+{
+public:
+    static constexpr UInt64 LATEST_REVISION = std::numeric_limits<UInt64>::max();
+    static constexpr UInt64 UNKNOWN_REVISION = 0;
+
+    DiskObjectStorageMetadataHelper(DiskObjectStorage * disk_, ReadSettings read_settings_)
+        : disk(disk_)
+        , read_settings(std::move(read_settings_))
+        , operation_log_suffix("-" + getFQDNOrHostName())
+    {
+    }
+
+    /// Most important method, called on DiskObjectStorage startup
+    void restore(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context);
+
+    void syncRevision(UInt64 revision)
+    {
+        UInt64 local_revision = revision_counter.load();
+        while ((revision > local_revision) && revision_counter.compare_exchange_weak(local_revision, revision));
+    }
+
+    UInt64 getRevision() const
+    {
+        return revision_counter.load();
+    }
+
+    static int readSchemaVersion(IObjectStorage * object_storage, const String & source_path);
+
+    void migrateToRestorableSchema();
+
+    void findLastRevision();
+
+    void createFileOperationObject(const String & operation_name, UInt64 revision, const ObjectAttributes & metadata) const;
+
+    /// Version with possibility to backup-restore metadata.
+    static constexpr int RESTORABLE_SCHEMA_VERSION = 1;
+
+    std::atomic<UInt64> revision_counter = 0;
+private:
+    struct RestoreInformation
+    {
+        UInt64 revision = LATEST_REVISION;
+        String source_namespace;
+        String source_path;
+        bool detached = false;
+    };
+
+    using Futures = std::vector<std::future<void>>;
+
+    /// Move file or files in directory when possible and remove files in other case
+    /// to restore by S3 operation log with same operations from different replicas
+    void moveRecursiveOrRemove(const String & from_path, const String & to_path, bool send_metadata);
+
+    void saveSchemaVersion(const int & version) const;
+    void updateObjectMetadata(const String & key, const ObjectAttributes & metadata) const;
+    void migrateFileToRestorableSchema(const String & path) const;
+    void migrateToRestorableSchemaRecursive(const String & path, Futures & results);
+
+    void readRestoreInformation(RestoreInformation & restore_information);
+    void restoreFiles(IObjectStorage * source_object_storage, const RestoreInformation & restore_information);
+    void processRestoreFiles(IObjectStorage * source_object_storage, const String & source_path, const std::vector<String> & keys) const;
+    void restoreFileOperations(IObjectStorage * source_object_storage, const RestoreInformation & restore_information);
+
+    inline static const String RESTORE_FILE_NAME = "restore";
+
+    /// Object contains information about schema version.
+    inline static const String SCHEMA_VERSION_OBJECT = ".SCHEMA_VERSION";
+    /// Directories with data.
+    const std::vector<String> data_roots {"data", "store"};
+
+    DiskObjectStorage * disk;
+
+    ObjectStoragePtr object_storage_from_another_namespace;
+
+    ReadSettings read_settings;
+
+    String operation_log_suffix;
+};
+
+}

src/Disks/ObjectStorages/HDFS/HDFSObjectStorage.cpp

@@ -0,0 +1,156 @@
+#include <Disks/ObjectStorages/HDFS/HDFSObjectStorage.h>
+
+#include <IO/SeekAvoidingReadBuffer.h>
+#include <IO/copyData.h>
+
+#include <Storages/HDFS/WriteBufferFromHDFS.h>
+#include <Storages/HDFS/HDFSCommon.h>
+
+#include <Storages/HDFS/ReadBufferFromHDFS.h>
+#include <Disks/IO/AsynchronousReadIndirectBufferFromRemoteFS.h>
+#include <Disks/IO/ReadIndirectBufferFromRemoteFS.h>
+#include <Disks/IO/WriteIndirectBufferFromRemoteFS.h>
+#include <Disks/IO/ReadBufferFromRemoteFSGather.h>
+
+
+#if USE_HDFS
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int UNSUPPORTED_METHOD;
+    extern const int HDFS_ERROR;
+}
+
+void HDFSObjectStorage::shutdown()
+{
+}
+
+void HDFSObjectStorage::startup()
+{
+}
+
+bool HDFSObjectStorage::exists(const std::string & hdfs_uri) const
+{
+    const size_t begin_of_path = hdfs_uri.find('/', hdfs_uri.find("//") + 2);
+    const String remote_fs_object_path = hdfs_uri.substr(begin_of_path);
+    return (0 == hdfsExists(hdfs_fs.get(), remote_fs_object_path.c_str()));
+}
+
+std::unique_ptr<SeekableReadBuffer> HDFSObjectStorage::readObject( /// NOLINT
+    const std::string & path,
+    const ReadSettings & read_settings,
+    std::optional<size_t>,
+    std::optional<size_t>) const
+{
+    return std::make_unique<ReadBufferFromHDFS>(path, path, config, read_settings.remote_fs_buffer_size);
+}
+
+std::unique_ptr<ReadBufferFromFileBase> HDFSObjectStorage::readObjects( /// NOLINT
+    const std::string & common_path_prefix,
+    const BlobsPathToSize & blobs_to_read,
+    const ReadSettings & read_settings,
+    std::optional<size_t>,
+    std::optional<size_t>) const
+{
+    auto hdfs_impl = std::make_unique<ReadBufferFromHDFSGather>(config, common_path_prefix, common_path_prefix, blobs_to_read, read_settings);
+    auto buf = std::make_unique<ReadIndirectBufferFromRemoteFS>(std::move(hdfs_impl));
+    return std::make_unique<SeekAvoidingReadBuffer>(std::move(buf), settings->min_bytes_for_seek);
+}
+
+std::unique_ptr<WriteBufferFromFileBase> HDFSObjectStorage::writeObject( /// NOLINT
+    const std::string & path,
+    WriteMode mode,
+    std::optional<ObjectAttributes> attributes,
+    FinalizeCallback && finalize_callback,
+    size_t buf_size,
+    const WriteSettings &)
+{
+    if (attributes.has_value())
+        throw Exception(ErrorCodes::UNSUPPORTED_METHOD, "HDFS API doesn't support custom attributes/metadata for stored objects");
+
+    /// Single O_WRONLY in libhdfs adds O_TRUNC
+    auto hdfs_buffer = std::make_unique<WriteBufferFromHDFS>(
+        path, config, settings->replication, buf_size,
+        mode == WriteMode::Rewrite ? O_WRONLY : O_WRONLY | O_APPEND);
+
+    return std::make_unique<WriteIndirectBufferFromRemoteFS>(std::move(hdfs_buffer), std::move(finalize_callback), path);
+}
+
+
+void HDFSObjectStorage::listPrefix(const std::string & path, BlobsPathToSize & children) const
+{
+    const size_t begin_of_path = path.find('/', path.find("//") + 2);
+    int32_t num_entries;
+    auto * files_list = hdfsListDirectory(hdfs_fs.get(), path.substr(begin_of_path).c_str(), &num_entries);
+    if (num_entries == -1)
+        throw Exception(ErrorCodes::HDFS_ERROR, "HDFSDelete failed with path: " + path);
+
+    for (int32_t i = 0; i < num_entries; ++i)
+        children.emplace_back(files_list[i].mName, files_list[i].mSize);
+}
+
+/// Remove file. Throws exception if file doesn't exists or it's a directory.
+void HDFSObjectStorage::removeObject(const std::string & path)
+{
+    const size_t begin_of_path = path.find('/', path.find("//") + 2);
+
+    /// Add path from root to file name
+    int res = hdfsDelete(hdfs_fs.get(), path.substr(begin_of_path).c_str(), 0);
+    if (res == -1)
+        throw Exception(ErrorCodes::HDFS_ERROR, "HDFSDelete failed with path: " + path);
+
+}
+
+void HDFSObjectStorage::removeObjects(const std::vector<std::string> & paths)
+{
+    for (const auto & hdfs_path : paths)
+        removeObject(hdfs_path);
+}
+
+void HDFSObjectStorage::removeObjectIfExists(const std::string & path)
+{
+    if (exists(path))
+        removeObject(path);
+}
+
+void HDFSObjectStorage::removeObjectsIfExist(const std::vector<std::string> & paths)
+{
+    for (const auto & hdfs_path : paths)
+        removeObjectIfExists(hdfs_path);
+}
+
+ObjectMetadata HDFSObjectStorage::getObjectMetadata(const std::string &) const
+{
+    throw Exception(ErrorCodes::UNSUPPORTED_METHOD, "HDFS API doesn't support custom attributes/metadata for stored objects");
+}
+
+void HDFSObjectStorage::copyObject( /// NOLINT
+    const std::string & object_from,
+    const std::string & object_to,
+    std::optional<ObjectAttributes> object_to_attributes)
+{
+    if (object_to_attributes.has_value())
+        throw Exception(ErrorCodes::UNSUPPORTED_METHOD, "HDFS API doesn't support custom attributes/metadata for stored objects");
+
+    auto in = readObject(object_from);
+    auto out = writeObject(object_to, WriteMode::Rewrite);
+    copyData(*in, *out);
+    out->finalize();
+}
+
+
+void HDFSObjectStorage::applyNewSettings(const Poco::Util::AbstractConfiguration &, const std::string &, ContextPtr)
+{
+}
+
+std::unique_ptr<IObjectStorage> HDFSObjectStorage::cloneObjectStorage(const std::string &, const Poco::Util::AbstractConfiguration &, const std::string &, ContextPtr)
+{
+    throw Exception(ErrorCodes::UNSUPPORTED_METHOD, "HDFS object storage doesn't support cloning");
+}
+
+}
+
+#endif

src/Disks/ObjectStorages/HDFS/HDFSObjectStorage.h

@@ -0,0 +1,119 @@
+#pragma once
+#include <Common/config.h>
+
+
+#if USE_HDFS
+
+#include <Disks/IDisk.h>
+#include <Disks/ObjectStorages/IObjectStorage.h>
+#include <Storages/HDFS/HDFSCommon.h>
+#include <Core/UUID.h>
+#include <memory>
+#include <Poco/Util/AbstractConfiguration.h>
+
+namespace DB
+{
+
+struct HDFSObjectStorageSettings
+{
+
+    HDFSObjectStorageSettings() = default;
+
+    size_t min_bytes_for_seek;
+    int objects_chunk_size_to_delete;
+    int replication;
+
+    HDFSObjectStorageSettings(
+            int min_bytes_for_seek_,
+            int objects_chunk_size_to_delete_,
+            int replication_)
+        : min_bytes_for_seek(min_bytes_for_seek_)
+        , objects_chunk_size_to_delete(objects_chunk_size_to_delete_)
+        , replication(replication_)
+    {}
+};
+
+
+class HDFSObjectStorage : public IObjectStorage
+{
+public:
+
+    using SettingsPtr = std::unique_ptr<HDFSObjectStorageSettings>;
+
+    HDFSObjectStorage(
+        FileCachePtr && cache_,
+        const String & hdfs_root_path_,
+        SettingsPtr settings_,
+        const Poco::Util::AbstractConfiguration & config_)
+        : IObjectStorage(std::move(cache_))
+        , config(config_)
+        , hdfs_builder(createHDFSBuilder(hdfs_root_path_, config))
+        , hdfs_fs(createHDFSFS(hdfs_builder.get()))
+        , settings(std::move(settings_))
+    {}
+
+    bool exists(const std::string & hdfs_uri) const override;
+
+    std::unique_ptr<SeekableReadBuffer> readObject( /// NOLINT
+        const std::string & path,
+        const ReadSettings & read_settings = ReadSettings{},
+        std::optional<size_t> read_hint = {},
+        std::optional<size_t> file_size = {}) const override;
+
+    std::unique_ptr<ReadBufferFromFileBase> readObjects( /// NOLINT
+        const std::string & common_path_prefix,
+        const BlobsPathToSize & blobs_to_read,
+        const ReadSettings & read_settings = ReadSettings{},
+        std::optional<size_t> read_hint = {},
+        std::optional<size_t> file_size = {}) const override;
+
+    /// Open the file for write and return WriteBufferFromFileBase object.
+    std::unique_ptr<WriteBufferFromFileBase> writeObject( /// NOLINT
+        const std::string & path,
+        WriteMode mode,
+        std::optional<ObjectAttributes> attributes = {},
+        FinalizeCallback && finalize_callback = {},
+        size_t buf_size = DBMS_DEFAULT_BUFFER_SIZE,
+        const WriteSettings & write_settings = {}) override;
+
+    void listPrefix(const std::string & path, BlobsPathToSize & children) const override;
+    /// Remove file. Throws exception if file doesn't exists or it's a directory.
+    void removeObject(const std::string & path) override;
+
+    void removeObjects(const std::vector<std::string> & paths) override;
+
+    void removeObjectIfExists(const std::string & path) override;
+
+    void removeObjectsIfExist(const std::vector<std::string> & paths) override;
+
+    ObjectMetadata getObjectMetadata(const std::string & path) const override;
+
+    void copyObject( /// NOLINT
+        const std::string & object_from,
+        const std::string & object_to,
+        std::optional<ObjectAttributes> object_to_attributes = {}) override;
+
+    void shutdown() override;
+
+    void startup() override;
+
+    void applyNewSettings(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context) override;
+
+    String getObjectsNamespace() const override { return ""; }
+
+    std::unique_ptr<IObjectStorage> cloneObjectStorage(const std::string & new_namespace, const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context) override;
+
+private:
+    const Poco::Util::AbstractConfiguration & config;
+
+    HDFSBuilderWrapper hdfs_builder;
+    HDFSFSPtr hdfs_fs;
+
+    SettingsPtr settings;
+
+
+};
+
+}
+
+#endif

src/Disks/ObjectStorages/HDFS/registerDiskHDFS.cpp

@@ -0,0 +1,54 @@
+#include <Disks/ObjectStorages/HDFS/HDFSObjectStorage.h>
+#include <Disks/ObjectStorages/DiskObjectStorageCommon.h>
+#include <Disks/ObjectStorages/DiskObjectStorage.h>
+#include <Disks/DiskFactory.h>
+#include <Storages/HDFS/HDFSCommon.h>
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int BAD_ARGUMENTS;
+}
+
+void registerDiskHDFS(DiskFactory & factory)
+{
+    auto creator = [](const String & name,
+                      const Poco::Util::AbstractConfiguration & config,
+                      const String & config_prefix,
+                      ContextPtr context_,
+                      const DisksMap & /*map*/) -> DiskPtr
+    {
+        String uri{config.getString(config_prefix + ".endpoint")};
+        checkHDFSURL(uri);
+
+        if (uri.back() != '/')
+            throw Exception(ErrorCodes::BAD_ARGUMENTS, "HDFS path must ends with '/', but '{}' doesn't.", uri);
+
+        std::unique_ptr<HDFSObjectStorageSettings> settings = std::make_unique<HDFSObjectStorageSettings>(
+            config.getUInt64(config_prefix + ".min_bytes_for_seek", 1024 * 1024),
+            config.getInt(config_prefix + ".objects_chunk_size_to_delete", 1000),
+            context_->getSettingsRef().hdfs_replication
+        );
+        /// FIXME Cache currently unsupported :(
+        ObjectStoragePtr hdfs_storage = std::make_unique<HDFSObjectStorage>(nullptr, uri, std::move(settings), config);
+
+        auto metadata_disk = prepareForLocalMetadata(name, config, config_prefix, context_).second;
+        uint64_t copy_thread_pool_size = config.getUInt(config_prefix + ".thread_pool_size", 16);
+
+        return std::make_shared<DiskObjectStorage>(
+            name,
+            uri,
+            "DiskHDFS",
+            metadata_disk,
+            std::move(hdfs_storage),
+            DiskType::HDFS,
+            /* send_metadata = */ false,
+            copy_thread_pool_size);
+    };
+
+    factory.registerDiskType("hdfs", creator);
+}
+
+}

src/Disks/ObjectStorages/IObjectStorage.cpp

@@ -0,0 +1,48 @@
+#include <Disks/ObjectStorages/IObjectStorage.h>
+#include <Disks/IO/ThreadPoolRemoteFSReader.h>
+#include <IO/copyData.h>
+
+namespace DB
+{
+AsynchronousReaderPtr IObjectStorage::getThreadPoolReader()
+{
+    constexpr size_t pool_size = 50;
+    constexpr size_t queue_size = 1000000;
+    static AsynchronousReaderPtr reader = std::make_shared<ThreadPoolRemoteFSReader>(pool_size, queue_size);
+    return reader;
+}
+
+ThreadPool & IObjectStorage::getThreadPoolWriter()
+{
+    constexpr size_t pool_size = 100;
+    constexpr size_t queue_size = 1000000;
+    static ThreadPool writer(pool_size, pool_size, queue_size);
+    return writer;
+}
+
+std::string IObjectStorage::getCacheBasePath() const
+{
+    return cache ? cache->getBasePath() : "";
+}
+
+void IObjectStorage::removeFromCache(const std::string & path)
+{
+    if (cache)
+    {
+        auto key = cache->hash(path);
+        cache->remove(key);
+    }
+}
+
+void IObjectStorage::copyObjectToAnotherObjectStorage(const std::string & object_from, const std::string & object_to, IObjectStorage & object_storage_to, std::optional<ObjectAttributes> object_to_attributes) // NOLINT
+{
+    if (&object_storage_to == this)
+        copyObject(object_from, object_to, object_to_attributes);
+
+    auto in = readObject(object_from);
+    auto out = object_storage_to.writeObject(object_to, WriteMode::Rewrite);
+    copyData(*in, *out);
+    out->finalize();
+}
+
+}

src/Disks/ObjectStorages/IObjectStorage.h

@@ -0,0 +1,159 @@
+#pragma once
+
+#include <filesystem>
+#include <string>
+#include <map>
+#include <optional>
+
+#include <Poco/Timestamp.h>
+#include <Core/Defines.h>
+#include <Common/Exception.h>
+#include <IO/ReadSettings.h>
+#include <IO/WriteSettings.h>
+
+#include <Disks/IO/AsynchronousReadIndirectBufferFromRemoteFS.h>
+#include <Common/ThreadPool.h>
+#include <Common/FileCache.h>
+#include <Disks/WriteMode.h>
+
+
+namespace DB
+{
+
+class ReadBufferFromFileBase;
+class WriteBufferFromFileBase;
+
+using ObjectAttributes = std::map<std::string, std::string>;
+
+/// Path to blob with it's size
+struct BlobPathWithSize
+{
+    std::string relative_path;
+    uint64_t bytes_size;
+
+    BlobPathWithSize() = default;
+    BlobPathWithSize(const BlobPathWithSize & other) = default;
+
+    BlobPathWithSize(const std::string & relative_path_, uint64_t bytes_size_)
+        : relative_path(relative_path_)
+        , bytes_size(bytes_size_)
+    {}
+};
+
+/// List of blobs with their sizes
+using BlobsPathToSize = std::vector<BlobPathWithSize>;
+
+struct ObjectMetadata
+{
+    uint64_t size_bytes;
+    std::optional<Poco::Timestamp> last_modified;
+    std::optional<ObjectAttributes> attributes;
+};
+
+using FinalizeCallback = std::function<void(size_t bytes_count)>;
+
+/// Base class for all object storages which implement some subset of ordinary filesystem operations.
+///
+/// Examples of object storages are S3, Azure Blob Storage, HDFS.
+class IObjectStorage
+{
+public:
+    explicit IObjectStorage(FileCachePtr && cache_)
+        : cache(std::move(cache_))
+    {}
+
+    /// Path exists or not
+    virtual bool exists(const std::string & path) const = 0;
+
+    /// List on prefix, return children with their sizes.
+    virtual void listPrefix(const std::string & path, BlobsPathToSize & children) const = 0;
+
+    /// Get object metadata if supported. It should be possible to receive
+    /// at least size of object
+    virtual ObjectMetadata getObjectMetadata(const std::string & path) const = 0;
+
+    /// Read single path from object storage
+    virtual std::unique_ptr<SeekableReadBuffer> readObject( /// NOLINT
+        const std::string & path,
+        const ReadSettings & read_settings = ReadSettings{},
+        std::optional<size_t> read_hint = {},
+        std::optional<size_t> file_size = {}) const = 0;
+
+    /// Read multiple objects with common prefix
+    virtual std::unique_ptr<ReadBufferFromFileBase> readObjects( /// NOLINT
+        const std::string & common_path_prefix,
+        const BlobsPathToSize & blobs_to_read,
+        const ReadSettings & read_settings = ReadSettings{},
+        std::optional<size_t> read_hint = {},
+        std::optional<size_t> file_size = {}) const = 0;
+
+    /// Open the file for write and return WriteBufferFromFileBase object.
+    virtual std::unique_ptr<WriteBufferFromFileBase> writeObject( /// NOLINT
+        const std::string & path,
+        WriteMode mode,
+        std::optional<ObjectAttributes> attributes = {},
+        FinalizeCallback && finalize_callback = {},
+        size_t buf_size = DBMS_DEFAULT_BUFFER_SIZE,
+        const WriteSettings & write_settings = {}) = 0;
+
+    /// Remove object. Throws exception if object doesn't exists.
+    virtual void removeObject(const std::string & path) = 0;
+
+    /// Remove multiple objects. Some object storages can do batch remove in a more
+    /// optimal way.
+    virtual void removeObjects(const std::vector<std::string> & paths) = 0;
+
+    /// Remove object on path if exists
+    virtual void removeObjectIfExists(const std::string & path) = 0;
+
+    /// Remove objects on path if exists
+    virtual void removeObjectsIfExist(const std::vector<std::string> & paths) = 0;
+
+    /// Copy object with different attributes if required
+    virtual void copyObject( /// NOLINT
+        const std::string & object_from,
+        const std::string & object_to,
+        std::optional<ObjectAttributes> object_to_attributes = {}) = 0;
+
+    /// Copy object to another instance of object storage
+    /// by default just read the object from source object storage and write
+    /// to destination through buffers.
+    virtual void copyObjectToAnotherObjectStorage( /// NOLINT
+        const std::string & object_from,
+        const std::string & object_to,
+        IObjectStorage & object_storage_to,
+        std::optional<ObjectAttributes> object_to_attributes = {});
+
+    virtual ~IObjectStorage() = default;
+
+    /// Path to directory with objects cache
+    std::string getCacheBasePath() const;
+
+    static AsynchronousReaderPtr getThreadPoolReader();
+
+    static ThreadPool & getThreadPoolWriter();
+
+    virtual void shutdown() = 0;
+
+    virtual void startup() = 0;
+
+    void removeFromCache(const std::string & path);
+
+    /// Apply new settings, in most cases reiniatilize client and some other staff
+    virtual void applyNewSettings(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context) = 0;
+
+    /// Sometimes object storages have something similar to chroot or namespace, for example
+    /// buckets in S3. If object storage doesn't have any namepaces return empty string.
+    virtual String getObjectsNamespace() const = 0;
+
+    /// FIXME: confusing function required for a very specific case. Create new instance of object storage
+    /// in different namespace.
+    virtual std::unique_ptr<IObjectStorage> cloneObjectStorage(const std::string & new_namespace, const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix, ContextPtr context) = 0;
+
+protected:
+    FileCachePtr cache;
+};
+
+using ObjectStoragePtr = std::unique_ptr<IObjectStorage>;
+
+}