thevar1able/ClickHouse
1
0
Fork 0
mirror of https://github.com/ClickHouse/ClickHouse.git synced 2024-11-21 23:21:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'ClickHouse/master' into bump-s2geometry

Browse Source
This commit is contained in:
Robert Schulze 2024-07-04 15:43:47 +00:00
commit 1bd515cc1f
No known key found for this signature in database
GPG Key ID: 26703B55FB13728A
249 changed files with 9507 additions and 2393 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
contrib/openssl vendored

@ -1 +1 @@
Subproject commit 5d81fa7068fc8c07f4d0997d5b703f3c541a637c
Subproject commit ee2bb8513b28bf86b35404dd17a0e29305ca9e08

1
docker/packager/binary-builder/build.sh
View File

@ -111,6 +111,7 @@ fi
mv ./programs/clickhouse* /output || mv ./programs/*_fuzzer /output
[ -x ./programs/self-extracting/clickhouse ] && mv ./programs/self-extracting/clickhouse /output
[ -x ./programs/self-extracting/clickhouse-stripped ] && mv ./programs/self-extracting/clickhouse-stripped /output
[ -x ./programs/self-extracting/clickhouse-keeper ] && mv ./programs/self-extracting/clickhouse-keeper /output
mv ./src/unit_tests_dbms /output ||: # may not exist for some binary builds
mv ./programs/*.dict ./programs/*.options ./programs/*_seed_corpus.zip /output ||: # libFuzzer oss-fuzz compatible infrastructure

5
docker/packager/packager
View File

@ -276,10 +276,7 @@ def parse_env_variables(
if is_release_build(debug_build, package_type, sanitizer, coverage):
cmake_flags.append("-DSPLIT_DEBUG_SYMBOLS=ON")
result.append("WITH_PERFORMANCE=1")
if is_cross_arm:
cmake_flags.append("-DBUILD_STANDALONE_KEEPER=1")
else:
result.append("BUILD_MUSL_KEEPER=1")
cmake_flags.append("-DBUILD_STANDALONE_KEEPER=1")
elif package_type == "fuzzers":
cmake_flags.append("-DENABLE_FUZZING=1")
cmake_flags.append("-DENABLE_PROTOBUF=1")

4
docker/test/stateful/run.sh
View File

@ -213,6 +213,10 @@ function run_tests()
ADDITIONAL_OPTIONS+=('--s3-storage')
fi
if [[ -n "$USE_AZURE_STORAGE_FOR_MERGE_TREE" ]] && [[ "$USE_AZURE_STORAGE_FOR_MERGE_TREE" -eq 1 ]]; then
ADDITIONAL_OPTIONS+=('--azure-blob-storage')
fi
if [[ -n "$USE_DATABASE_ORDINARY" ]] && [[ "$USE_DATABASE_ORDINARY" -eq 1 ]]; then
ADDITIONAL_OPTIONS+=('--db-engine=Ordinary')
fi

2
docker/test/stateless/run.sh
View File

@ -207,7 +207,7 @@ function run_tests()
if [[ -n "$USE_AZURE_STORAGE_FOR_MERGE_TREE" ]] && [[ "$USE_AZURE_STORAGE_FOR_MERGE_TREE" -eq 1 ]]; then
# to disable the same tests
ADDITIONAL_OPTIONS+=('--s3-storage')
ADDITIONAL_OPTIONS+=('--azure-blob-storage')
# azurite is slow, but with these two settings it can be super slow
ADDITIONAL_OPTIONS+=('--no-random-settings')
ADDITIONAL_OPTIONS+=('--no-random-merge-tree-settings')

41
docker/test/stress/run.sh
View File

@ -110,6 +110,15 @@ start_server
clickhouse-client --query "SHOW TABLES FROM datasets"
clickhouse-client --query "SHOW TABLES FROM test"
if [[ "$USE_S3_STORAGE_FOR_MERGE_TREE" == "1" ]]; then
TEMP_POLICY="s3_cache"
elif [[ "$USE_AZURE_STORAGE_FOR_MERGE_TREE" == "1" ]]; then
TEMP_POLICY="azure_cache"
else
TEMP_POLICY="default"
fi
clickhouse-client --query "CREATE TABLE test.hits_s3 (WatchID UInt64, JavaEnable UInt8, Title String, GoodEvent Int16,
EventTime DateTime, EventDate Date, CounterID UInt32, ClientIP UInt32, ClientIP6 FixedString(16), RegionID UInt32,
UserID UInt64, CounterClass Int8, OS UInt8, UserAgent UInt8, URL String, Referer String, URLDomain String, RefererDomain String,
@ -135,7 +144,7 @@ clickhouse-client --query "CREATE TABLE test.hits_s3 (WatchID UInt64, JavaEnabl
URLHash UInt64, CLID UInt32, YCLID UInt64, ShareService String, ShareURL String, ShareTitle String,
ParsedParams Nested(Key1 String, Key2 String, Key3 String, Key4 String, Key5 String, ValueDouble Float64),
IslandID FixedString(16), RequestNum UInt32, RequestTry UInt8) ENGINE = MergeTree() PARTITION BY toYYYYMM(EventDate)
ORDER BY (CounterID, EventDate, intHash32(UserID)) SAMPLE BY intHash32(UserID) SETTINGS index_granularity = 8192, storage_policy='s3_cache'"
ORDER BY (CounterID, EventDate, intHash32(UserID)) SAMPLE BY intHash32(UserID) SETTINGS index_granularity = 8192, storage_policy='$TEMP_POLICY'"
clickhouse-client --query "CREATE TABLE test.hits (WatchID UInt64, JavaEnable UInt8, Title String, GoodEvent Int16,
EventTime DateTime, EventDate Date, CounterID UInt32, ClientIP UInt32, ClientIP6 FixedString(16), RegionID UInt32,
UserID UInt64, CounterClass Int8, OS UInt8, UserAgent UInt8, URL String, Referer String, URLDomain String,
@ -161,7 +170,7 @@ clickhouse-client --query "CREATE TABLE test.hits (WatchID UInt64, JavaEnable U
URLHash UInt64, CLID UInt32, YCLID UInt64, ShareService String, ShareURL String, ShareTitle String,
ParsedParams Nested(Key1 String, Key2 String, Key3 String, Key4 String, Key5 String, ValueDouble Float64),
IslandID FixedString(16), RequestNum UInt32, RequestTry UInt8) ENGINE = MergeTree() PARTITION BY toYYYYMM(EventDate)
ORDER BY (CounterID, EventDate, intHash32(UserID)) SAMPLE BY intHash32(UserID) SETTINGS index_granularity = 8192, storage_policy='s3_cache'"
ORDER BY (CounterID, EventDate, intHash32(UserID)) SAMPLE BY intHash32(UserID) SETTINGS index_granularity = 8192, storage_policy='$TEMP_POLICY'"
clickhouse-client --query "CREATE TABLE test.visits (CounterID UInt32, StartDate Date, Sign Int8, IsNew UInt8,
VisitID UInt64, UserID UInt64, StartTime DateTime, Duration UInt32, UTCStartTime DateTime, PageViews Int32,
Hits Int32, IsBounce UInt8, Referer String, StartURL String, RefererDomain String, StartURLDomain String,
@ -195,7 +204,7 @@ clickhouse-client --query "CREATE TABLE test.visits (CounterID UInt32, StartDat
Market Nested(Type UInt8, GoalID UInt32, OrderID String, OrderPrice Int64, PP UInt32, DirectPlaceID UInt32, DirectOrderID UInt32,
DirectBannerID UInt32, GoodID String, GoodName String, GoodQuantity Int32, GoodPrice Int64), IslandID FixedString(16))
ENGINE = CollapsingMergeTree(Sign) PARTITION BY toYYYYMM(StartDate) ORDER BY (CounterID, StartDate, intHash32(UserID), VisitID)
SAMPLE BY intHash32(UserID) SETTINGS index_granularity = 8192, storage_policy='s3_cache'"
SAMPLE BY intHash32(UserID) SETTINGS index_granularity = 8192, storage_policy='$TEMP_POLICY'"
clickhouse-client --query "INSERT INTO test.hits_s3 SELECT * FROM datasets.hits_v1 SETTINGS enable_filesystem_cache_on_write_operations=0"
clickhouse-client --query "INSERT INTO test.hits SELECT * FROM datasets.hits_v1 SETTINGS enable_filesystem_cache_on_write_operations=0"
@ -211,19 +220,29 @@ clickhouse-client --query "SYSTEM STOP THREAD FUZZER"
stop_server
# Let's enable S3 storage by default
export USE_S3_STORAGE_FOR_MERGE_TREE=1
export RANDOMIZE_OBJECT_KEY_TYPE=1
export ZOOKEEPER_FAULT_INJECTION=1
export THREAD_POOL_FAULT_INJECTION=1
configure
# But we still need default disk because some tables loaded only into it
sudo cat /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml \
| sed "s|<main><disk>s3</disk></main>|<main><disk>s3</disk></main><default><disk>default</disk></default>|" \
> /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml.tmp
mv /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml.tmp /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml
sudo chown clickhouse /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml
sudo chgrp clickhouse /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml
if [[ "$USE_S3_STORAGE_FOR_MERGE_TREE" == "1" ]]; then
# But we still need default disk because some tables loaded only into it
sudo cat /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml \
| sed "s|<main><disk>s3</disk></main>|<main><disk>s3</disk></main><default><disk>default</disk></default>|" \
> /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml.tmp
mv /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml.tmp /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml
sudo chown clickhouse /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml
sudo chgrp clickhouse /etc/clickhouse-server/config.d/s3_storage_policy_by_default.xml
elif [[ "$USE_AZURE_STORAGE_FOR_MERGE_TREE" == "1" ]]; then
# But we still need default disk because some tables loaded only into it
sudo cat /etc/clickhouse-server/config.d/azure_storage_policy_by_default.xml \
| sed "s|<main><disk>azure</disk></main>|<main><disk>azure</disk></main><default><disk>default</disk></default>|" \
> /etc/clickhouse-server/config.d/azure_storage_policy_by_default.xml.tmp
mv /etc/clickhouse-server/config.d/azure_storage_policy_by_default.xml.tmp /etc/clickhouse-server/config.d/azure_storage_policy_by_default.xml
sudo chown clickhouse /etc/clickhouse-server/config.d/azure_storage_policy_by_default.xml
sudo chgrp clickhouse /etc/clickhouse-server/config.d/azure_storage_policy_by_default.xml
fi
sudo cat /etc/clickhouse-server/config.d/logger_trace.xml \
| sed "s|<level>trace</level>|<level>test</level>|" \

20
programs/CMakeLists.txt
View File

@ -66,18 +66,18 @@ else()
message(STATUS "Library bridge mode: OFF")
endif()
if (ENABLE_CLICKHOUSE_KEEPER)
message(STATUS "ClickHouse keeper mode: ON")
else()
message(STATUS "ClickHouse keeper mode: OFF")
endif()
if (ENABLE_CLICKHOUSE_KEEPER_CONVERTER)
message(STATUS "ClickHouse keeper-converter mode: ON")
else()
message(STATUS "ClickHouse keeper-converter mode: OFF")
endif()
if (ENABLE_CLICKHOUSE_KEEPER)
message(STATUS "ClickHouse Keeper: ON")
else()
message(STATUS "ClickHouse Keeper: OFF")
endif()
if (ENABLE_CLICKHOUSE_KEEPER_CLIENT)
message(STATUS "ClickHouse keeper-client mode: ON")
else()
@ -131,10 +131,6 @@ add_subdirectory (static-files-disk-uploader)
add_subdirectory (su)
add_subdirectory (disks)
if (ENABLE_CLICKHOUSE_KEEPER)
add_subdirectory (keeper)
endif()
if (ENABLE_CLICKHOUSE_KEEPER_CONVERTER)
add_subdirectory (keeper-converter)
endif()
@ -143,6 +139,10 @@ if (ENABLE_CLICKHOUSE_KEEPER_CLIENT)
add_subdirectory (keeper-client)
endif()
if (ENABLE_CLICKHOUSE_KEEPER)
add_subdirectory (keeper)
endif()
if (ENABLE_CLICKHOUSE_ODBC_BRIDGE)
add_subdirectory (odbc-bridge)
endif ()

196
programs/keeper/CMakeLists.txt
View File

@ -1,4 +1,5 @@
set(CLICKHOUSE_KEEPER_SOURCES
keeper_main.cpp
Keeper.cpp
)
@ -8,6 +9,9 @@ set (CLICKHOUSE_KEEPER_LINK
clickhouse_common_io
clickhouse_common_zookeeper
daemon
clickhouse-keeper-converter-lib
clickhouse-keeper-client-lib
clickhouse_functions
dbms
)
@ -17,199 +21,11 @@ install(FILES keeper_config.xml DESTINATION "${CLICKHOUSE_ETC_DIR}/clickhouse-ke
if (BUILD_STANDALONE_KEEPER)
# Straight list of all required sources
set(CLICKHOUSE_KEEPER_STANDALONE_SOURCES
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperReconfiguration.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/RaftServerConfig.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/ACLMap.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/Changelog.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/CoordinationSettings.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/FourLetterCommand.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/InMemoryLogStore.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperConnectionStats.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperDispatcher.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperLogStore.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperServer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperContext.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperFeatureFlags.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperSnapshotManager.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperSnapshotManagerS3.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperStateMachine.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperContext.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperStateManager.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperStorage.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperConstants.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperAsynchronousMetrics.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/KeeperCommon.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/SessionExpiryQueue.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/SummingStateMachine.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/WriteBufferFromNuraftBuffer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/ZooKeeperDataReader.cpp
clickhouse_add_executable(clickhouse-keeper ${CLICKHOUSE_KEEPER_SOURCES})
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Core/SettingsFields.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Core/BaseSettings.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Core/ServerSettings.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Core/Field.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Core/SettingsEnums.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Core/ServerUUID.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Core/UUID.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Core/BackgroundSchedulePool.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/IO/ReadBuffer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/HTTPPathHints.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/KeeperTCPHandler.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/TCPServer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/NotFoundHandler.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/ProtocolServerAdapter.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/CertificateReloader.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/PrometheusRequestHandler.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/PrometheusMetricsWriter.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/waitServersToFinish.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/ServerType.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/HTTPRequestHandlerFactoryMain.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/KeeperReadinessHandler.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/CloudPlacementInfo.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/HTTP/HTTPServer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/HTTP/ReadHeaders.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/HTTP/HTTPServerConnection.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/HTTP/HTTPServerRequest.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/HTTP/HTTPServerResponse.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/HTTP/HTTPServerConnectionFactory.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Server/HTTP/WriteBufferFromHTTPServerResponse.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CachedCompressedReadBuffer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CheckingCompressedReadBuffer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressedReadBufferBase.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressedReadBuffer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressedReadBufferFromFile.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressedWriteBuffer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecEncrypted.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecLZ4.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecMultiple.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecNone.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionCodecZSTD.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/CompressionFactory.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/ICompressionCodec.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Compression/LZ4_decompress_faster.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/CurrentThread.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/NamedCollections/NamedCollections.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/NamedCollections/NamedCollectionConfiguration.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/Jemalloc.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/ZooKeeper/IKeeper.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/ZooKeeper/TestKeeper.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/ZooKeeper/ZooKeeperCommon.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/ZooKeeper/ZooKeeperConstants.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/ZooKeeper/ZooKeeper.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/ZooKeeper/ZooKeeperImpl.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/ZooKeeper/ZooKeeperIO.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/ZooKeeper/ZooKeeperLock.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Common/ZooKeeper/ZooKeeperNodeCache.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/registerDisks.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IDisk.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/DiskFactory.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/DiskSelector.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/DiskLocal.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/DiskLocalCheckThread.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/LocalDirectorySyncGuard.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/TemporaryFileOnDisk.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/loadLocalDiskConfig.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/DiskType.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/IObjectStorage.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/MetadataOperationsHolder.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/MetadataStorageFromPlainObjectStorageOperations.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/MetadataStorageFromPlainObjectStorage.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/MetadataStorageFromPlainRewritableObjectStorage.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/MetadataStorageFromDisk.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/MetadataStorageTransactionState.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/DiskObjectStorageMetadata.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/MetadataStorageFromDiskTransactionOperations.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/DiskObjectStorage.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/DiskObjectStorageTransaction.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/DiskObjectStorageRemoteMetadataRestoreHelper.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/ObjectStorageIteratorAsync.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/ObjectStorageIterator.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/StoredObject.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/S3/S3ObjectStorage.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/S3/S3Capabilities.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/S3/diskSettings.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/S3/DiskS3Utils.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/CommonPathPrefixKeyGenerator.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/ObjectStorageFactory.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/MetadataStorageFactory.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/ObjectStorages/RegisterDiskObjectStorage.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/createReadBufferFromFileBase.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/ReadBufferFromRemoteFSGather.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/IOUringReader.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/getIOUringReader.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/WriteBufferFromTemporaryFile.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/WriteBufferWithFinalizeCallback.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/AsynchronousBoundedReadBuffer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/getThreadPoolReader.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/ThreadPoolRemoteFSReader.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Disks/IO/ThreadPoolReader.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Daemon/BaseDaemon.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Daemon/SentryWriter.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Daemon/GraphiteWriter.cpp
${CMAKE_CURRENT_BINARY_DIR}/../../src/Daemon/GitHash.generated.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/Standalone/Context.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/Standalone/Settings.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Coordination/Standalone/ThreadStatusExt.cpp
Keeper.cpp
clickhouse-keeper.cpp
)
# List of resources for clickhouse-keeper client
if (ENABLE_CLICKHOUSE_KEEPER_CLIENT)
list(APPEND CLICKHOUSE_KEEPER_STANDALONE_SOURCES
${CMAKE_CURRENT_SOURCE_DIR}/../../programs/keeper-client/KeeperClient.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../programs/keeper-client/Commands.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../programs/keeper-client/Parser.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Client/LineReader.cpp
${CMAKE_CURRENT_SOURCE_DIR}/../../src/Client/ReplxxLineReader.cpp
)
endif()
clickhouse_add_executable(clickhouse-keeper ${CLICKHOUSE_KEEPER_STANDALONE_SOURCES})
# Remove some redundant dependencies
target_compile_definitions (clickhouse-keeper PRIVATE -DCLICKHOUSE_KEEPER_STANDALONE_BUILD)
target_compile_definitions (clickhouse-keeper PUBLIC -DWITHOUT_TEXT_LOG)
if (ENABLE_CLICKHOUSE_KEEPER_CLIENT AND TARGET ch_rust::skim)
target_link_libraries(clickhouse-keeper PRIVATE ch_rust::skim)
endif()
target_link_libraries(clickhouse-keeper
PRIVATE
ch_contrib::abseil_swiss_tables
ch_contrib::nuraft
ch_contrib::lz4
ch_contrib::zstd
ch_contrib::cityhash
ch_contrib::jemalloc
common ch_contrib::double_conversion
ch_contrib::dragonbox_to_chars
pcg_random
ch_contrib::pdqsort
ch_contrib::miniselect
clickhouse_common_config_no_zookeeper_log
loggers_no_text_log
clickhouse_common_io
clickhouse_parsers # Otherwise compression will not built. FIXME.
)
target_link_libraries(clickhouse-keeper PUBLIC ${CLICKHOUSE_KEEPER_LINK})
set_target_properties(clickhouse-keeper PROPERTIES RUNTIME_OUTPUT_DIRECTORY ../)
if (SPLIT_DEBUG_SYMBOLS)
clickhouse_split_debug_symbols(TARGET clickhouse-keeper DESTINATION_DIR ${CMAKE_CURRENT_BINARY_DIR}/../${SPLITTED_DEBUG_SYMBOLS_DIR} BINARY_PATH ../clickhouse-keeper)
else()

10
programs/keeper/Keeper.cpp
View File

@ -75,16 +75,6 @@ int mainEntryClickHouseKeeper(int argc, char ** argv)
}
}
#ifdef CLICKHOUSE_KEEPER_STANDALONE_BUILD
// Weak symbols don't work correctly on Darwin
// so we have a stub implementation to avoid linker errors
void collectCrashLog(
Int32, UInt64, const String &, const StackTrace &)
{}
#endif
namespace DB
{

30
programs/keeper/clickhouse-keeper.cpp
View File

@ -1,30 +0,0 @@
#include <Common/StringUtils.h>
#include "config_tools.h"
int mainEntryClickHouseKeeper(int argc, char ** argv);
#if ENABLE_CLICKHOUSE_KEEPER_CLIENT
int mainEntryClickHouseKeeperClient(int argc, char ** argv);
#endif
int main(int argc_, char ** argv_)
{
#if ENABLE_CLICKHOUSE_KEEPER_CLIENT
if (argc_ >= 2)
{
/// 'clickhouse-keeper --client ...' and 'clickhouse-keeper client ...' are OK
if (strcmp(argv_[1], "--client") == 0 || strcmp(argv_[1], "client") == 0)
{
argv_[1] = argv_[0];
return mainEntryClickHouseKeeperClient(--argc_, argv_ + 1);
}
}
if (argc_ > 0 && (strcmp(argv_[0], "clickhouse-keeper-client") == 0 || endsWith(argv_[0], "/clickhouse-keeper-client")))
return mainEntryClickHouseKeeperClient(argc_, argv_);
#endif
return mainEntryClickHouseKeeper(argc_, argv_);
}

189
programs/keeper/keeper_main.cpp Normal file
View File

@ -0,0 +1,189 @@
#include <unistd.h>
#include <fcntl.h>
#include <new>
#include <iostream>
#include <vector>
#include <string_view>
#include <utility> /// pair
#include <fmt/format.h>
#include "config.h"
#include "config_tools.h"
#include <Common/EnvironmentChecks.h>
#include <Common/Coverage.h>
#include <Common/StringUtils.h>
#include <Common/getHashOfLoadedBinary.h>
#include <Common/IO.h>
#include <base/phdr_cache.h>
#include <base/coverage.h>
int mainEntryClickHouseKeeper(int argc, char ** argv);
#if ENABLE_CLICKHOUSE_KEEPER_CONVERTER
int mainEntryClickHouseKeeperConverter(int argc, char ** argv);
#endif
#if ENABLE_CLICKHOUSE_KEEPER_CLIENT
int mainEntryClickHouseKeeperClient(int argc, char ** argv);
#endif
namespace
{
using MainFunc = int (*)(int, char**);
/// Add an item here to register new application
std::pair<std::string_view, MainFunc> clickhouse_applications[] =
{
// keeper
{"keeper", mainEntryClickHouseKeeper},
#if ENABLE_CLICKHOUSE_KEEPER_CONVERTER
{"converter", mainEntryClickHouseKeeperConverter},
{"keeper-converter", mainEntryClickHouseKeeperConverter},
#endif
#if ENABLE_CLICKHOUSE_KEEPER_CLIENT
{"client", mainEntryClickHouseKeeperClient},
{"keeper-client", mainEntryClickHouseKeeperClient},
#endif
};
int printHelp(int, char **)
{
std::cerr << "Use one of the following commands:" << std::endl;
for (auto & application : clickhouse_applications)
std::cerr << "clickhouse " << application.first << " [args] " << std::endl;
return -1;
}
}
bool isClickhouseApp(std::string_view app_suffix, std::vector<char *> & argv)
{
/// Use app if the first arg 'app' is passed (the arg should be quietly removed)
if (argv.size() >= 2)
{
auto first_arg = argv.begin() + 1;
/// 'clickhouse --client ...' and 'clickhouse client ...' are Ok
if (*first_arg == app_suffix
|| (std::string_view(*first_arg).starts_with("--") && std::string_view(*first_arg).substr(2) == app_suffix))
{
argv.erase(first_arg);
return true;
}
}
/// keeper suffix is default which will be used if no other app is detected
if (app_suffix == "keeper")
return false;
/// Use app if clickhouse binary is run through symbolic link with name clickhouse-app
std::string app_name = "clickhouse-" + std::string(app_suffix);
return !argv.empty() && (app_name == argv[0] || endsWith(argv[0], "/" + app_name));
}
/// Don't allow dlopen in the main ClickHouse binary, because it is harmful and insecure.
/// We don't use it. But it can be used by some libraries for implementation of "plugins".
/// We absolutely discourage the ancient technique of loading
/// 3rd-party uncontrolled dangerous libraries into the process address space,
/// because it is insane.
#if !defined(USE_MUSL)
extern "C"
{
void * dlopen(const char *, int)
{
return nullptr;
}
void * dlmopen(long, const char *, int) // NOLINT
{
return nullptr;
}
int dlclose(void *)
{
return 0;
}
const char * dlerror()
{
return "ClickHouse does not allow dynamic library loading";
}
}
#endif
/// Prevent messages from JeMalloc in the release build.
/// Some of these messages are non-actionable for the users, such as:
/// <jemalloc>: Number of CPUs detected is not deterministic. Per-CPU arena disabled.
#if USE_JEMALLOC && defined(NDEBUG) && !defined(SANITIZER)
extern "C" void (*malloc_message)(void *, const char *s);
__attribute__((constructor(0))) void init_je_malloc_message() { malloc_message = [](void *, const char *){}; }
#endif
/// This allows to implement assert to forbid initialization of a class in static constructors.
/// Usage:
///
/// extern bool inside_main;
/// class C { C() { assert(inside_main); } };
bool inside_main = false;
int main(int argc_, char ** argv_)
{
inside_main = true;
SCOPE_EXIT({ inside_main = false; });
/// PHDR cache is required for query profiler to work reliably
/// It also speed up exception handling, but exceptions from dynamically loaded libraries (dlopen)
/// will work only after additional call of this function.
/// Note: we forbid dlopen in our code.
updatePHDRCache();
#if !defined(USE_MUSL)
checkHarmfulEnvironmentVariables(argv_);
#endif
/// This is used for testing. For example,
/// clickhouse-local should be able to run a simple query without throw/catch.
if (getenv("CLICKHOUSE_TERMINATE_ON_ANY_EXCEPTION")) // NOLINT(concurrency-mt-unsafe)
DB::terminate_on_any_exception = true;
/// Reset new handler to default (that throws std::bad_alloc)
/// It is needed because LLVM library clobbers it.
std::set_new_handler(nullptr);
std::vector<char *> argv(argv_, argv_ + argc_);
/// Print a basic help if nothing was matched
MainFunc main_func = mainEntryClickHouseKeeper;
if (isClickhouseApp("help", argv))
{
main_func = printHelp;
}
else
{
for (auto & application : clickhouse_applications)
{
if (isClickhouseApp(application.first, argv))
{
main_func = application.second;
break;
}
}
}
int exit_code = main_func(static_cast<int>(argv.size()), argv.data());
#if defined(SANITIZE_COVERAGE)
dumpCoverage();
#endif
return exit_code;
}

267
programs/main.cpp
View File

@ -1,5 +1,3 @@
#include <csignal>
#include <csetjmp>
#include <unistd.h>
#include <fcntl.h>
@ -7,7 +5,6 @@
#include <iostream>
#include <vector>
#include <string>
#include <tuple>
#include <string_view>
#include <utility> /// pair
@ -16,6 +13,8 @@
#include "config.h"
#include "config_tools.h"
#include <Common/EnvironmentChecks.h>
#include <Common/Coverage.h>
#include <Common/StringUtils.h>
#include <Common/getHashOfLoadedBinary.h>
#include <Common/IO.h>
@ -119,268 +118,6 @@ std::pair<std::string_view, std::string_view> clickhouse_short_names[] =
{"chc", "client"},
};
enum class InstructionFail : uint8_t
{
NONE = 0,
SSE3 = 1,
SSSE3 = 2,
SSE4_1 = 3,
SSE4_2 = 4,
POPCNT = 5,
AVX = 6,
AVX2 = 7,
AVX512 = 8
};
auto instructionFailToString(InstructionFail fail)
{
switch (fail)
{
#define ret(x) return std::make_tuple(STDERR_FILENO, x, sizeof(x) - 1)
case InstructionFail::NONE:
ret("NONE");
case InstructionFail::SSE3:
ret("SSE3");
case InstructionFail::SSSE3:
ret("SSSE3");
case InstructionFail::SSE4_1:
ret("SSE4.1");
case InstructionFail::SSE4_2:
ret("SSE4.2");
case InstructionFail::POPCNT:
ret("POPCNT");
case InstructionFail::AVX:
ret("AVX");
case InstructionFail::AVX2:
ret("AVX2");
case InstructionFail::AVX512:
ret("AVX512");
#undef ret
}
}
sigjmp_buf jmpbuf;
[[noreturn]] void sigIllCheckHandler(int, siginfo_t *, void *)
{
siglongjmp(jmpbuf, 1);
}
/// Check if necessary SSE extensions are available by trying to execute some sse instructions.
/// If instruction is unavailable, SIGILL will be sent by kernel.
void checkRequiredInstructionsImpl(volatile InstructionFail & fail)
{
#if defined(__SSE3__)
fail = InstructionFail::SSE3;
__asm__ volatile ("addsubpd %%xmm0, %%xmm0" : : : "xmm0");
#endif
#if defined(__SSSE3__)
fail = InstructionFail::SSSE3;
__asm__ volatile ("pabsw %%xmm0, %%xmm0" : : : "xmm0");
#endif
#if defined(__SSE4_1__)
fail = InstructionFail::SSE4_1;
__asm__ volatile ("pmaxud %%xmm0, %%xmm0" : : : "xmm0");
#endif
#if defined(__SSE4_2__)
fail = InstructionFail::SSE4_2;
__asm__ volatile ("pcmpgtq %%xmm0, %%xmm0" : : : "xmm0");
#endif
/// Defined by -msse4.2
#if defined(__POPCNT__)
fail = InstructionFail::POPCNT;
{
uint64_t a = 0;
uint64_t b = 0;
__asm__ volatile ("popcnt %1, %0" : "=r"(a) :"r"(b) :);
}
#endif
#if defined(__AVX__)
fail = InstructionFail::AVX;
__asm__ volatile ("vaddpd %%ymm0, %%ymm0, %%ymm0" : : : "ymm0");
#endif
#if defined(__AVX2__)
fail = InstructionFail::AVX2;
__asm__ volatile ("vpabsw %%ymm0, %%ymm0" : : : "ymm0");
#endif
#if defined(__AVX512__)
fail = InstructionFail::AVX512;
__asm__ volatile ("vpabsw %%zmm0, %%zmm0" : : : "zmm0");
#endif
fail = InstructionFail::NONE;
}
/// Macros to avoid using strlen(), since it may fail if SSE is not supported.
#define writeError(data) do \
{ \
static_assert(__builtin_constant_p(data)); \
if (!writeRetry(STDERR_FILENO, data, sizeof(data) - 1)) \
_Exit(1); \
} while (false)
/// Check SSE and others instructions availability. Calls exit on fail.
/// This function must be called as early as possible, even before main, because static initializers may use unavailable instructions.
void checkRequiredInstructions()
{
struct sigaction sa{};
struct sigaction sa_old{};
sa.sa_sigaction = sigIllCheckHandler;
sa.sa_flags = SA_SIGINFO;
auto signal = SIGILL;
if (sigemptyset(&sa.sa_mask) != 0
|| sigaddset(&sa.sa_mask, signal) != 0
|| sigaction(signal, &sa, &sa_old) != 0)
{
/// You may wonder about strlen.
/// Typical implementation of strlen is using SSE4.2 or AVX2.
/// But this is not the case because it's compiler builtin and is executed at compile time.
writeError("Can not set signal handler\n");
_Exit(1);
}
volatile InstructionFail fail = InstructionFail::NONE;
if (sigsetjmp(jmpbuf, 1))
{
writeError("Instruction check fail. The CPU does not support ");
if (!std::apply(writeRetry, instructionFailToString(fail)))
_Exit(1);
writeError(" instruction set.\n");
_Exit(1);
}
checkRequiredInstructionsImpl(fail);
if (sigaction(signal, &sa_old, nullptr))
{
writeError("Can not set signal handler\n");
_Exit(1);
}
}
struct Checker
{
Checker()
{
checkRequiredInstructions();
}
} checker
#ifndef OS_DARWIN
__attribute__((init_priority(101))) /// Run before other static initializers.
#endif
;
#if !defined(USE_MUSL)
/// NOTE: We will migrate to full static linking or our own dynamic loader to make this code obsolete.
void checkHarmfulEnvironmentVariables(char ** argv)
{
std::initializer_list<const char *> harmful_env_variables = {
/// The list is a selection from "man ld-linux".
"LD_PRELOAD",
"LD_LIBRARY_PATH",
"LD_ORIGIN_PATH",
"LD_AUDIT",
"LD_DYNAMIC_WEAK",
/// The list is a selection from "man dyld" (osx).
"DYLD_LIBRARY_PATH",
"DYLD_FALLBACK_LIBRARY_PATH",
"DYLD_VERSIONED_LIBRARY_PATH",
"DYLD_INSERT_LIBRARIES",
};
bool require_reexec = false;
for (const auto * var : harmful_env_variables)
{
if (const char * value = getenv(var); value && value[0]) // NOLINT(concurrency-mt-unsafe)
{
/// NOTE: setenv() is used over unsetenv() since unsetenv() marked as harmful
if (setenv(var, "", true)) // NOLINT(concurrency-mt-unsafe) // this is safe if not called concurrently
{
fmt::print(stderr, "Cannot override {} environment variable", var);
_exit(1);
}
require_reexec = true;
}
}
if (require_reexec)
{
/// Use execvp() over execv() to search in PATH.
///
/// This should be safe, since:
/// - if argv[0] is relative path - it is OK
/// - if argv[0] has only basename, the it will search in PATH, like shell will do.
///
/// Also note, that this (search in PATH) because there is no easy and
/// portable way to get absolute path of argv[0].
/// - on linux there is /proc/self/exec and AT_EXECFN
/// - but on other OSes there is no such thing (especially on OSX).
///
/// And since static linking will be done someday anyway,
/// let's not pollute the code base with special cases.
int error = execvp(argv[0], argv);
_exit(error);
}
}
#endif
#if defined(SANITIZE_COVERAGE)
__attribute__((no_sanitize("coverage")))
void dumpCoverage()
{
/// A user can request to dump the coverage information into files at exit.
/// This is useful for non-server applications such as clickhouse-format or clickhouse-client,
/// that cannot introspect it with SQL functions at runtime.
/// The CLICKHOUSE_WRITE_COVERAGE environment variable defines a prefix for a filename 'prefix.pid'
/// containing the list of addresses of covered .
/// The format is even simpler than Clang's "sancov": an array of 64-bit addresses, native byte order, no header.
if (const char * coverage_filename_prefix = getenv("CLICKHOUSE_WRITE_COVERAGE")) // NOLINT(concurrency-mt-unsafe)
{
auto dump = [](const std::string & name, auto span)
{
/// Write only non-zeros.
std::vector<uintptr_t> data;
data.reserve(span.size());
for (auto addr : span)
if (addr)
data.push_back(addr);
int fd = ::open(name.c_str(), O_WRONLY | O_CREAT | O_TRUNC | O_CLOEXEC, 0400);
if (-1 == fd)
{
writeError("Cannot open a file to write the coverage data\n");
}
else
{
if (!writeRetry(fd, reinterpret_cast<const char *>(data.data()), data.size() * sizeof(data[0])))
writeError("Cannot write the coverage data to a file\n");
if (0 != ::close(fd))
writeError("Cannot close the file with coverage data\n");
}
};
dump(fmt::format("{}.{}", coverage_filename_prefix, getpid()), getCumulativeCoverage());
}
}
#endif
}
bool isClickhouseApp(std::string_view app_suffix, std::vector<char *> & argv)

17
programs/self-extracting/CMakeLists.txt
View File

@ -10,9 +10,24 @@ else ()
set (COMPRESSOR "${PROJECT_BINARY_DIR}/utils/self-extracting-executable/compressor")
endif ()
add_custom_target (self-extracting ALL
add_custom_target (self-extracting-server ALL
${CMAKE_COMMAND} -E remove clickhouse clickhouse-stripped
COMMAND ${COMPRESSOR} ${DECOMPRESSOR} clickhouse ../clickhouse
COMMAND ${COMPRESSOR} ${DECOMPRESSOR} clickhouse-stripped ../clickhouse-stripped
DEPENDS clickhouse clickhouse-stripped compressor
)
set(self_extracting_deps "self-extracting-server")
if (BUILD_STANDALONE_KEEPER)
add_custom_target (self-extracting-keeper ALL
${CMAKE_COMMAND} -E remove clickhouse-keeper
COMMAND ${COMPRESSOR} ${DECOMPRESSOR} clickhouse-keeper ../clickhouse-keeper
DEPENDS compressor clickhouse-keeper
)
list(APPEND self_extracting_deps "self-extracting-keeper")
endif()
add_custom_target (self-extracting ALL
DEPENDS ${self_extracting_deps}
)

6
src/Columns/ColumnObject.cpp
View File

@ -1093,4 +1093,10 @@ void ColumnObject::finalize()
checkObjectHasNoAmbiguosPaths(getKeys());
}
void ColumnObject::updateHashFast(SipHash & hash) const
{
for (const auto & entry : subcolumns)
for (auto & part : entry->data.data)
part->updateHashFast(hash);
}
}

2
src/Columns/ColumnObject.h
View File

@ -242,7 +242,7 @@ public:
const char * skipSerializedInArena(const char *) const override { throwMustBeConcrete(); }
void updateHashWithValue(size_t, SipHash &) const override { throwMustBeConcrete(); }
void updateWeakHash32(WeakHash32 &) const override { throwMustBeConcrete(); }
void updateHashFast(SipHash &) const override { throwMustBeConcrete(); }
void updateHashFast(SipHash & hash) const override;
void expand(const Filter &, bool) override { throwMustBeConcrete(); }
bool hasEqualValues() const override { throwMustBeConcrete(); }
size_t byteSizeAt(size_t) const override { throwMustBeConcrete(); }

184
src/Common/CollectionOfDerived.h Normal file
View File

@ -0,0 +1,184 @@
#pragma once
#include <base/defines.h>
#include <Common/Exception.h>
#include <algorithm>
#include <memory>
#include <typeindex>
#include <vector>
#include <string>
namespace DB
{
namespace ErrorCodes
{
extern const int LOGICAL_ERROR;
}
/* This is a collections of objects derived from ItemBase.
* Collection contains no more than one instance for each derived type.
* The derived type is used to access the instance.
*/
template<class ItemBase>
class CollectionOfDerivedItems
{
public:
using Self = CollectionOfDerivedItems<ItemBase>;
using ItemPtr = std::shared_ptr<ItemBase>;
private:
struct Rec
{
std::type_index type_idx;
ItemPtr ptr;
bool operator<(const Rec & other) const
{
return type_idx < other.type_idx;
}
bool operator<(const std::type_index & value) const
{
return type_idx < value;
}
bool operator==(const Rec & other) const
{
return type_idx == other.type_idx;
}
};
using Records = std::vector<Rec>;
public:
void swap(Self & other) noexcept
{
records.swap(other.records);
}
void clear()
{
records.clear();
}
bool empty() const
{
return records.empty();
}
size_t size() const
{
return records.size();
}
Self clone() const
{
Self result;
result.records.reserve(records.size());
for (const auto & rec : records)
result.records.emplace_back(rec.type_idx, rec.ptr->clone());
return result;
}
void append(Self && other)
{
auto middle_idx = records.size();
std::move(other.records.begin(), other.records.end(), std::back_inserter(records));
std::inplace_merge(records.begin(), records.begin() + middle_idx, records.end());
chassert(isUniqTypes());
}
template <class T>
void add(std::shared_ptr<T> info)
{
static_assert(std::is_base_of_v<ItemBase, T>, "Template parameter must inherit items base class");
return addImpl(std::type_index(typeid(T)), std::move(info));
}
template <class T>
std::shared_ptr<T> get() const
{
static_assert(std::is_base_of_v<ItemBase, T>, "Template parameter must inherit items base class");
auto it = getImpl(std::type_index(typeid(T)));
if (it == records.cend())
return nullptr;
auto cast = std::dynamic_pointer_cast<T>(it->ptr);
chassert(cast);
return cast;
}
template <class T>
std::shared_ptr<T> extract()
{
static_assert(std::is_base_of_v<ItemBase, T>, "Template parameter must inherit items base class");
auto it = getImpl(std::type_index(typeid(T)));
if (it == records.cend())
return nullptr;
auto cast = std::dynamic_pointer_cast<T>(it->ptr);
chassert(cast);
records.erase(it);
return cast;
}
std::string debug() const
{
std::string result;
for (auto & rec : records)
{
result.append(rec.type_idx.name());
result.append(" ");
}
return result;
}
private:
bool isUniqTypes() const
{
auto uniq_it = std::adjacent_find(records.begin(), records.end());
return uniq_it == records.end();
}
void addImpl(std::type_index type_idx, ItemPtr item)
{
auto it = std::lower_bound(records.begin(), records.end(), type_idx);
if (it == records.end())
{
records.emplace_back(type_idx, item);
return;
}
if (it->type_idx == type_idx)
throw Exception(ErrorCodes::LOGICAL_ERROR, "inserted items must be unique by their type, type {} is inserted twice", type_idx.name());
records.emplace(it, type_idx, item);
chassert(isUniqTypes());
}
Records::const_iterator getImpl(std::type_index type_idx) const
{
auto it = std::lower_bound(records.cbegin(), records.cend(), type_idx);
if (it == records.cend())
return records.cend();
if (it->type_idx != type_idx)
return records.cend();
return it;
}
Records records;
};
}

65
src/Common/Coverage.cpp Normal file
View File

@ -0,0 +1,65 @@
#include <Common/Coverage.h>
#if defined(SANITIZE_COVERAGE)
#include <fcntl.h>
#include <unistd.h>
#include <string>
#include <vector>
#include <Common/IO.h>
#include <base/coverage.h>
#include <fmt/format.h>
/// Macros to avoid using strlen(), since it may fail if SSE is not supported.
#define writeError(data) do \
{ \
static_assert(__builtin_constant_p(data)); \
if (!writeRetry(STDERR_FILENO, data, sizeof(data) - 1)) \
_Exit(1); \
} while (false)
__attribute__((no_sanitize("coverage")))
void dumpCoverage()
{
/// A user can request to dump the coverage information into files at exit.
/// This is useful for non-server applications such as clickhouse-format or clickhouse-client,
/// that cannot introspect it with SQL functions at runtime.
/// The CLICKHOUSE_WRITE_COVERAGE environment variable defines a prefix for a filename 'prefix.pid'
/// containing the list of addresses of covered .
/// The format is even simpler than Clang's "sancov": an array of 64-bit addresses, native byte order, no header.
if (const char * coverage_filename_prefix = getenv("CLICKHOUSE_WRITE_COVERAGE")) // NOLINT(concurrency-mt-unsafe)
{
auto dump = [](const std::string & name, auto span)
{
/// Write only non-zeros.
std::vector<uintptr_t> data;
data.reserve(span.size());
for (auto addr : span)
if (addr)
data.push_back(addr);
int fd = ::open(name.c_str(), O_WRONLY | O_CREAT | O_TRUNC | O_CLOEXEC, 0400);
if (-1 == fd)
{
writeError("Cannot open a file to write the coverage data\n");
}
else
{
if (!writeRetry(fd, reinterpret_cast<const char *>(data.data()), data.size() * sizeof(data[0])))
writeError("Cannot write the coverage data to a file\n");
if (0 != ::close(fd))
writeError("Cannot close the file with coverage data\n");
}
};
dump(fmt::format("{}.{}", coverage_filename_prefix, getpid()), getCumulativeCoverage());
}
}
#endif

5
src/Common/Coverage.h Normal file
View File

@ -0,0 +1,5 @@
#pragma once
#if defined(SANITIZE_COVERAGE)
void dumpCoverage();
#endif

234
src/Common/EnvironmentChecks.cpp Normal file
View File

@ -0,0 +1,234 @@
#include <Common/EnvironmentChecks.h>
#include <Common/IO.h>
#include <fmt/format.h>
#include <csignal>
#include <csetjmp>
#include <cstdint>
#include <tuple>
#include <unistd.h>
namespace
{
enum class InstructionFail : uint8_t
{
NONE = 0,
SSE3 = 1,
SSSE3 = 2,
SSE4_1 = 3,
SSE4_2 = 4,
POPCNT = 5,
AVX = 6,
AVX2 = 7,
AVX512 = 8
};
auto instructionFailToString(InstructionFail fail)
{
switch (fail)
{
#define ret(x) return std::make_tuple(STDERR_FILENO, x, sizeof(x) - 1)
case InstructionFail::NONE:
ret("NONE");
case InstructionFail::SSE3:
ret("SSE3");
case InstructionFail::SSSE3:
ret("SSSE3");
case InstructionFail::SSE4_1:
ret("SSE4.1");
case InstructionFail::SSE4_2:
ret("SSE4.2");
case InstructionFail::POPCNT:
ret("POPCNT");
case InstructionFail::AVX:
ret("AVX");
case InstructionFail::AVX2:
ret("AVX2");
case InstructionFail::AVX512:
ret("AVX512");
#undef ret
}
}
sigjmp_buf jmpbuf;
[[noreturn]] void sigIllCheckHandler(int, siginfo_t *, void *)
{
siglongjmp(jmpbuf, 1);
}
/// Check if necessary SSE extensions are available by trying to execute some sse instructions.
/// If instruction is unavailable, SIGILL will be sent by kernel.
void checkRequiredInstructionsImpl(volatile InstructionFail & fail)
{
#if defined(__SSE3__)
fail = InstructionFail::SSE3;
__asm__ volatile ("addsubpd %%xmm0, %%xmm0" : : : "xmm0");
#endif
#if defined(__SSSE3__)
fail = InstructionFail::SSSE3;
__asm__ volatile ("pabsw %%xmm0, %%xmm0" : : : "xmm0");
#endif
#if defined(__SSE4_1__)
fail = InstructionFail::SSE4_1;
__asm__ volatile ("pmaxud %%xmm0, %%xmm0" : : : "xmm0");
#endif
#if defined(__SSE4_2__)
fail = InstructionFail::SSE4_2;
__asm__ volatile ("pcmpgtq %%xmm0, %%xmm0" : : : "xmm0");
#endif
/// Defined by -msse4.2
#if defined(__POPCNT__)
fail = InstructionFail::POPCNT;
{
uint64_t a = 0;
uint64_t b = 0;
__asm__ volatile ("popcnt %1, %0" : "=r"(a) :"r"(b) :);
}
#endif
#if defined(__AVX__)
fail = InstructionFail::AVX;
__asm__ volatile ("vaddpd %%ymm0, %%ymm0, %%ymm0" : : : "ymm0");
#endif
#if defined(__AVX2__)
fail = InstructionFail::AVX2;
__asm__ volatile ("vpabsw %%ymm0, %%ymm0" : : : "ymm0");
#endif
#if defined(__AVX512__)
fail = InstructionFail::AVX512;
__asm__ volatile ("vpabsw %%zmm0, %%zmm0" : : : "zmm0");
#endif
fail = InstructionFail::NONE;
}
/// Macros to avoid using strlen(), since it may fail if SSE is not supported.
#define writeError(data) do \
{ \
static_assert(__builtin_constant_p(data)); \
if (!writeRetry(STDERR_FILENO, data, sizeof(data) - 1)) \
_Exit(1); \
} while (false)
/// Check SSE and others instructions availability. Calls exit on fail.
/// This function must be called as early as possible, even before main, because static initializers may use unavailable instructions.
void checkRequiredInstructions()
{
struct sigaction sa{};
struct sigaction sa_old{};
sa.sa_sigaction = sigIllCheckHandler;
sa.sa_flags = SA_SIGINFO;
auto signal = SIGILL;
if (sigemptyset(&sa.sa_mask) != 0
|| sigaddset(&sa.sa_mask, signal) != 0
|| sigaction(signal, &sa, &sa_old) != 0)
{
/// You may wonder about strlen.
/// Typical implementation of strlen is using SSE4.2 or AVX2.
/// But this is not the case because it's compiler builtin and is executed at compile time.
writeError("Can not set signal handler\n");
_Exit(1);
}
volatile InstructionFail fail = InstructionFail::NONE;
if (sigsetjmp(jmpbuf, 1))
{
writeError("Instruction check fail. The CPU does not support ");
if (!std::apply(writeRetry, instructionFailToString(fail)))
_Exit(1);
writeError(" instruction set.\n");
_Exit(1);
}
checkRequiredInstructionsImpl(fail);
if (sigaction(signal, &sa_old, nullptr))
{
writeError("Can not set signal handler\n");
_Exit(1);
}
}
struct Checker
{
Checker()
{
checkRequiredInstructions();
}
} checker
#ifndef OS_DARWIN
__attribute__((init_priority(101))) /// Run before other static initializers.
#endif
;
}
#if !defined(USE_MUSL)
/// NOTE: We will migrate to full static linking or our own dynamic loader to make this code obsolete.
void checkHarmfulEnvironmentVariables(char ** argv)
{
std::initializer_list<const char *> harmful_env_variables = {
/// The list is a selection from "man ld-linux".
"LD_PRELOAD",
"LD_LIBRARY_PATH",
"LD_ORIGIN_PATH",
"LD_AUDIT",
"LD_DYNAMIC_WEAK",
/// The list is a selection from "man dyld" (osx).
"DYLD_LIBRARY_PATH",
"DYLD_FALLBACK_LIBRARY_PATH",
"DYLD_VERSIONED_LIBRARY_PATH",
"DYLD_INSERT_LIBRARIES",
};
bool require_reexec = false;
for (const auto * var : harmful_env_variables)
{
if (const char * value = getenv(var); value && value[0]) // NOLINT(concurrency-mt-unsafe)
{
/// NOTE: setenv() is used over unsetenv() since unsetenv() marked as harmful
if (setenv(var, "", true)) // NOLINT(concurrency-mt-unsafe) // this is safe if not called concurrently
{
fmt::print(stderr, "Cannot override {} environment variable", var);
_exit(1);
}
require_reexec = true;
}
}
if (require_reexec)
{
/// Use execvp() over execv() to search in PATH.
///
/// This should be safe, since:
/// - if argv[0] is relative path - it is OK
/// - if argv[0] has only basename, the it will search in PATH, like shell will do.
///
/// Also note, that this (search in PATH) because there is no easy and
/// portable way to get absolute path of argv[0].
/// - on linux there is /proc/self/exec and AT_EXECFN
/// - but on other OSes there is no such thing (especially on OSX).
///
/// And since static linking will be done someday anyway,
/// let's not pollute the code base with special cases.
int error = execvp(argv[0], argv);
_exit(error);
}
}
#endif

5
src/Common/EnvironmentChecks.h Normal file
View File

@ -0,0 +1,5 @@
#pragma once
#if !defined(USE_MUSL)
void checkHarmfulEnvironmentVariables(char ** argv);
#endif

4
src/Compression/CompressionFactory.cpp
View File

@ -185,7 +185,6 @@ void registerCodecDeflateQpl(CompressionCodecFactory & factory);
/// Keeper use only general-purpose codecs, so we don't need these special codecs
/// in standalone build
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
void registerCodecDelta(CompressionCodecFactory & factory);
void registerCodecT64(CompressionCodecFactory & factory);
void registerCodecDoubleDelta(CompressionCodecFactory & factory);
@ -193,7 +192,6 @@ void registerCodecGorilla(CompressionCodecFactory & factory);
void registerCodecEncrypted(CompressionCodecFactory & factory);
void registerCodecFPC(CompressionCodecFactory & factory);
void registerCodecGCD(CompressionCodecFactory & factory);
#endif
CompressionCodecFactory::CompressionCodecFactory()
{
@ -205,7 +203,6 @@ CompressionCodecFactory::CompressionCodecFactory()
#endif
registerCodecLZ4HC(*this);
registerCodecMultiple(*this);
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
registerCodecDelta(*this);
registerCodecT64(*this);
registerCodecDoubleDelta(*this);
@ -216,7 +213,6 @@ CompressionCodecFactory::CompressionCodecFactory()
registerCodecDeflateQpl(*this);
#endif
registerCodecGCD(*this);
#endif
default_codec = get("LZ4", {});
}

486
src/Coordination/Standalone/Context.cpp
View File

@ -1,486 +0,0 @@
#include <Interpreters/Context.h>
#include <Common/Config/ConfigProcessor.h>
#include <Common/Macros.h>
#include <Common/ThreadPool.h>
#include <Common/callOnce.h>
#include <Disks/IO/IOUringReader.h>
#include <IO/S3Settings.h>
#include <Disks/IO/getIOUringReader.h>
#include <Core/ServerSettings.h>
#include <boost/noncopyable.hpp>
#include <memory>
#include <cassert>
namespace ProfileEvents
{
extern const Event ContextLock;
extern const Event ContextLockWaitMicroseconds;
}
namespace CurrentMetrics
{
extern const Metric ContextLockWait;
extern const Metric BackgroundSchedulePoolTask;
extern const Metric BackgroundSchedulePoolSize;
extern const Metric IOWriterThreads;
extern const Metric IOWriterThreadsActive;
extern const Metric IOWriterThreadsScheduled;
}
namespace DB
{
namespace ErrorCodes
{
extern const int LOGICAL_ERROR;
extern const int UNSUPPORTED_METHOD;
}
struct ContextSharedPart : boost::noncopyable
{
ContextSharedPart()
: macros(std::make_unique<Macros>())
{}
~ContextSharedPart()
{
if (keeper_dispatcher)
{
try
{
keeper_dispatcher->shutdown();
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
/// Wait for thread pool for background reads and writes,
/// since it may use per-user MemoryTracker which will be destroyed here.
if (asynchronous_remote_fs_reader)
{
try
{
asynchronous_remote_fs_reader->wait();
asynchronous_remote_fs_reader.reset();
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
if (asynchronous_local_fs_reader)
{
try
{
asynchronous_local_fs_reader->wait();
asynchronous_local_fs_reader.reset();
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
if (synchronous_local_fs_reader)
{
try
{
synchronous_local_fs_reader->wait();
synchronous_local_fs_reader.reset();
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
if (threadpool_writer)
{
try
{
threadpool_writer->wait();
threadpool_writer.reset();
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
}
/// For access of most of shared objects.
mutable SharedMutex mutex;
ServerSettings server_settings;
String path; /// Path to the data directory, with a slash at the end.
ConfigurationPtr config; /// Global configuration settings.
MultiVersion<Macros> macros; /// Substitutions extracted from config.
OnceFlag schedule_pool_initialized;
mutable std::unique_ptr<BackgroundSchedulePool> schedule_pool; /// A thread pool that can run different jobs in background
RemoteHostFilter remote_host_filter; /// Allowed URL from config.xml
mutable OnceFlag readers_initialized;
mutable std::unique_ptr<IAsynchronousReader> asynchronous_remote_fs_reader;
mutable std::unique_ptr<IAsynchronousReader> asynchronous_local_fs_reader;
mutable std::unique_ptr<IAsynchronousReader> synchronous_local_fs_reader;
#if USE_LIBURING
mutable OnceFlag io_uring_reader_initialized;
mutable std::unique_ptr<IOUringReader> io_uring_reader;
#endif
mutable OnceFlag threadpool_writer_initialized;
mutable std::unique_ptr<ThreadPool> threadpool_writer;
mutable ThrottlerPtr remote_read_throttler; /// A server-wide throttler for remote IO reads
mutable ThrottlerPtr remote_write_throttler; /// A server-wide throttler for remote IO writes
mutable ThrottlerPtr local_read_throttler; /// A server-wide throttler for local IO reads
mutable ThrottlerPtr local_write_throttler; /// A server-wide throttler for local IO writes
std::optional<S3SettingsByEndpoint> storage_s3_settings TSA_GUARDED_BY(mutex); /// Settings of S3 storage
mutable std::mutex keeper_dispatcher_mutex;
mutable std::shared_ptr<KeeperDispatcher> keeper_dispatcher TSA_GUARDED_BY(keeper_dispatcher_mutex);
};
ContextData::ContextData() = default;
ContextData::ContextData(const ContextData &) = default;
Context::Context() = default;
Context::Context(const Context & rhs) : ContextData(rhs), std::enable_shared_from_this<Context>(rhs) {}
Context::~Context() = default;
SharedContextHolder::SharedContextHolder(SharedContextHolder &&) noexcept = default;
SharedContextHolder & SharedContextHolder::operator=(SharedContextHolder &&) noexcept = default;
SharedContextHolder::SharedContextHolder() = default;
SharedContextHolder::~SharedContextHolder() = default;
SharedContextHolder::SharedContextHolder(std::unique_ptr<ContextSharedPart> shared_context)
: shared(std::move(shared_context)) {}
void SharedContextHolder::reset() { shared.reset(); }
void Context::makeGlobalContext()
{
initGlobal();
global_context = shared_from_this();
}
ContextMutablePtr Context::createGlobal(ContextSharedPart * shared_part)
{
auto res = std::shared_ptr<Context>(new Context);
res->shared = shared_part;
return res;
}
void Context::initGlobal()
{
assert(!global_context_instance);
global_context_instance = shared_from_this();
}
SharedContextHolder Context::createShared()
{
return SharedContextHolder(std::make_unique<ContextSharedPart>());
}
ContextMutablePtr Context::getGlobalContext() const
{
auto ptr = global_context.lock();
if (!ptr) throw Exception(ErrorCodes::LOGICAL_ERROR, "There is no global context or global context has expired");
return ptr;
}
std::unique_lock<SharedMutex> Context::getGlobalLock() const
{
ProfileEvents::increment(ProfileEvents::ContextLock);
CurrentMetrics::Increment increment{CurrentMetrics::ContextLockWait};
Stopwatch watch;
auto lock = std::unique_lock(shared->mutex);
ProfileEvents::increment(ProfileEvents::ContextLockWaitMicroseconds, watch.elapsedMicroseconds());
return lock;
}
std::shared_lock<SharedMutex> Context::getGlobalSharedLock() const
{
ProfileEvents::increment(ProfileEvents::ContextLock);
CurrentMetrics::Increment increment{CurrentMetrics::ContextLockWait};
Stopwatch watch;
auto lock = std::shared_lock(shared->mutex);
ProfileEvents::increment(ProfileEvents::ContextLockWaitMicroseconds, watch.elapsedMicroseconds());
return lock;
}
std::unique_lock<SharedMutex> Context::getLocalLock() const
{
ProfileEvents::increment(ProfileEvents::ContextLock);
CurrentMetrics::Increment increment{CurrentMetrics::ContextLockWait};
Stopwatch watch;
auto lock = std::unique_lock(mutex);
ProfileEvents::increment(ProfileEvents::ContextLockWaitMicroseconds, watch.elapsedMicroseconds());
return lock;
}
std::shared_lock<SharedMutex> Context::getLocalSharedLock() const
{
ProfileEvents::increment(ProfileEvents::ContextLock);
CurrentMetrics::Increment increment{CurrentMetrics::ContextLockWait};
Stopwatch watch;
auto lock = std::shared_lock(mutex);
ProfileEvents::increment(ProfileEvents::ContextLockWaitMicroseconds, watch.elapsedMicroseconds());
return lock;
}
String Context::getPath() const
{
auto lock = getGlobalSharedLock();
return shared->path;
}
void Context::setPath(const String & path)
{
auto lock = getGlobalLock();
shared->path = path;
}
MultiVersion<Macros>::Version Context::getMacros() const
{
return shared->macros.get();
}
void Context::setMacros(std::unique_ptr<Macros> && macros)
{
shared->macros.set(std::move(macros));
}
BackgroundSchedulePool & Context::getSchedulePool() const
{
callOnce(shared->schedule_pool_initialized, [&] {
shared->schedule_pool = std::make_unique<BackgroundSchedulePool>(
shared->server_settings.background_schedule_pool_size,
CurrentMetrics::BackgroundSchedulePoolTask,
CurrentMetrics::BackgroundSchedulePoolSize,
"BgSchPool");
});
return *shared->schedule_pool;
}
void Context::setRemoteHostFilter(const Poco::Util::AbstractConfiguration & config)
{
shared->remote_host_filter.setValuesFromConfig(config);
}
const RemoteHostFilter & Context::getRemoteHostFilter() const
{
return shared->remote_host_filter;
}
IAsynchronousReader & Context::getThreadPoolReader(FilesystemReaderType type) const
{
callOnce(shared->readers_initialized, [&] {
const auto & config = getConfigRef();
shared->asynchronous_remote_fs_reader = createThreadPoolReader(FilesystemReaderType::ASYNCHRONOUS_REMOTE_FS_READER, config);
shared->asynchronous_local_fs_reader = createThreadPoolReader(FilesystemReaderType::ASYNCHRONOUS_LOCAL_FS_READER, config);
shared->synchronous_local_fs_reader = createThreadPoolReader(FilesystemReaderType::SYNCHRONOUS_LOCAL_FS_READER, config);
});
switch (type)
{
case FilesystemReaderType::ASYNCHRONOUS_REMOTE_FS_READER:
return *shared->asynchronous_remote_fs_reader;
case FilesystemReaderType::ASYNCHRONOUS_LOCAL_FS_READER:
return *shared->asynchronous_local_fs_reader;
case FilesystemReaderType::SYNCHRONOUS_LOCAL_FS_READER:
return *shared->synchronous_local_fs_reader;
}
}
#if USE_LIBURING
IOUringReader & Context::getIOUringReader() const
{
callOnce(shared->io_uring_reader_initialized, [&] {
shared->io_uring_reader = createIOUringReader();
});
return *shared->io_uring_reader;
}
#endif
std::shared_ptr<FilesystemCacheLog> Context::getFilesystemCacheLog() const
{
return nullptr;
}
std::shared_ptr<FilesystemReadPrefetchesLog> Context::getFilesystemReadPrefetchesLog() const
{
return nullptr;
}
std::shared_ptr<BlobStorageLog> Context::getBlobStorageLog() const
{
return nullptr;
}
void Context::setConfig(const ConfigurationPtr & config)
{
auto lock = getGlobalLock();
shared->config = config;
}
const Poco::Util::AbstractConfiguration & Context::getConfigRef() const
{
auto lock = getGlobalSharedLock();
return shared->config ? *shared->config : Poco::Util::Application::instance().config();
}
std::shared_ptr<AsyncReadCounters> Context::getAsyncReadCounters() const
{
auto lock = getLocalLock();
if (!async_read_counters)
async_read_counters = std::make_shared<AsyncReadCounters>();
return async_read_counters;
}
ThreadPool & Context::getThreadPoolWriter() const
{
callOnce(shared->threadpool_writer_initialized, [&] {
const auto & config = getConfigRef();
auto pool_size = config.getUInt(".threadpool_writer_pool_size", 100);
auto queue_size = config.getUInt(".threadpool_writer_queue_size", 1000000);
shared->threadpool_writer = std::make_unique<ThreadPool>(
CurrentMetrics::IOWriterThreads, CurrentMetrics::IOWriterThreadsActive, CurrentMetrics::IOWriterThreadsScheduled, pool_size, pool_size, queue_size);
});
return *shared->threadpool_writer;
}
ThrottlerPtr Context::getRemoteReadThrottler() const
{
return nullptr;
}
ThrottlerPtr Context::getRemoteWriteThrottler() const
{
return nullptr;
}
ThrottlerPtr Context::getLocalReadThrottler() const
{
return nullptr;
}
ThrottlerPtr Context::getLocalWriteThrottler() const
{
return nullptr;
}
ReadSettings Context::getReadSettings() const
{
return ReadSettings{};
}
ResourceManagerPtr Context::getResourceManager() const
{
return nullptr;
}
ClassifierPtr Context::getWorkloadClassifier() const
{
return nullptr;
}
void Context::initializeKeeperDispatcher([[maybe_unused]] bool start_async) const
{
const auto & config_ref = getConfigRef();
std::lock_guard lock(shared->keeper_dispatcher_mutex);
if (shared->keeper_dispatcher)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Trying to initialize Keeper multiple times");
if (config_ref.has("keeper_server"))
{
shared->keeper_dispatcher = std::make_shared<KeeperDispatcher>();
shared->keeper_dispatcher->initialize(config_ref, true, start_async, getMacros());
}
}
std::shared_ptr<KeeperDispatcher> Context::getKeeperDispatcher() const
{
std::lock_guard lock(shared->keeper_dispatcher_mutex);
if (!shared->keeper_dispatcher)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Keeper must be initialized before requests");
return shared->keeper_dispatcher;
}
std::shared_ptr<KeeperDispatcher> Context::tryGetKeeperDispatcher() const
{
std::lock_guard lock(shared->keeper_dispatcher_mutex);
return shared->keeper_dispatcher;
}
void Context::shutdownKeeperDispatcher() const
{
std::lock_guard lock(shared->keeper_dispatcher_mutex);
if (shared->keeper_dispatcher)
{
shared->keeper_dispatcher->shutdown();
shared->keeper_dispatcher.reset();
}
}
void Context::updateKeeperConfiguration([[maybe_unused]] const Poco::Util::AbstractConfiguration & config_)
{
std::lock_guard lock(shared->keeper_dispatcher_mutex);
if (!shared->keeper_dispatcher)
return;
shared->keeper_dispatcher->updateConfiguration(config_, getMacros());
}
std::shared_ptr<zkutil::ZooKeeper> Context::getZooKeeper() const
{
throw Exception(ErrorCodes::UNSUPPORTED_METHOD, "Cannot connect to ZooKeeper from Keeper");
}
const S3SettingsByEndpoint & Context::getStorageS3Settings() const
{
std::lock_guard lock(shared->mutex);
if (!shared->storage_s3_settings)
{
const auto & config = shared->config ? *shared->config : Poco::Util::Application::instance().config();
shared->storage_s3_settings.emplace().loadFromConfig(config, "s3", getSettingsRef());
}
return *shared->storage_s3_settings;
}
const ServerSettings & Context::getServerSettings() const
{
return shared->server_settings;
}
bool Context::hasTraceCollector() const
{
return false;
}
bool Context::isBackgroundOperationContext() const
{
return false;
}
}

178
src/Coordination/Standalone/Context.h
View File

@ -1,178 +0,0 @@
#pragma once
#include <Interpreters/Context_fwd.h>
#include <Coordination/KeeperDispatcher.h>
#include <Common/MultiVersion.h>
#include <Common/RemoteHostFilter.h>
#include <Common/SharedMutex.h>
#include <Disks/IO/getThreadPoolReader.h>
#include <Core/Settings.h>
#include <Core/ServerSettings.h>
#include <Core/BackgroundSchedulePool.h>
#include <IO/AsyncReadCounters.h>
#include <Common/Scheduler/IResourceManager.h>
#include <Poco/Util/Application.h>
#include <memory>
#include "config.h"
namespace zkutil
{
class ZooKeeper;
using ZooKeeperPtr = std::shared_ptr<ZooKeeper>;
}
namespace DB
{
struct ContextSharedPart;
class Macros;
class FilesystemCacheLog;
class FilesystemReadPrefetchesLog;
class BlobStorageLog;
class IOUringReader;
class S3SettingsByEndpoint;
/// A small class which owns ContextShared.
/// We don't use something like unique_ptr directly to allow ContextShared type to be incomplete.
struct SharedContextHolder
{
~SharedContextHolder();
SharedContextHolder();
explicit SharedContextHolder(std::unique_ptr<ContextSharedPart> shared_context);
SharedContextHolder(SharedContextHolder &&) noexcept;
SharedContextHolder & operator=(SharedContextHolder &&) noexcept;
ContextSharedPart * get() const { return shared.get(); }
void reset();
private:
std::unique_ptr<ContextSharedPart> shared;
};
class ContextData
{
protected:
ContextWeakMutablePtr global_context;
inline static ContextPtr global_context_instance;
ContextSharedPart * shared;
/// Query metrics for reading data asynchronously with IAsynchronousReader.
mutable std::shared_ptr<AsyncReadCounters> async_read_counters;
Settings settings; /// Setting for query execution.
public:
/// Use copy constructor or createGlobal() instead
ContextData();
ContextData(const ContextData &);
};
class Context : public ContextData, public std::enable_shared_from_this<Context>
{
private:
/// ContextData mutex
mutable SharedMutex mutex;
Context();
Context(const Context &);
std::unique_lock<SharedMutex> getGlobalLock() const;
std::shared_lock<SharedMutex> getGlobalSharedLock() const;
std::unique_lock<SharedMutex> getLocalLock() const;
std::shared_lock<SharedMutex> getLocalSharedLock() const;
public:
/// Create initial Context with ContextShared and etc.
static ContextMutablePtr createGlobal(ContextSharedPart * shared_part);
static SharedContextHolder createShared();
ContextMutablePtr getGlobalContext() const;
static ContextPtr getGlobalContextInstance() { return global_context_instance; }
void makeGlobalContext();
void initGlobal();
~Context();
using ConfigurationPtr = Poco::AutoPtr<Poco::Util::AbstractConfiguration>;
/// Global application configuration settings.
void setConfig(const ConfigurationPtr & config);
const Poco::Util::AbstractConfiguration & getConfigRef() const;
const Settings & getSettingsRef() const { return settings; }
String getPath() const;
void setPath(const String & path);
MultiVersion<Macros>::Version getMacros() const;
void setMacros(std::unique_ptr<Macros> && macros);
BackgroundSchedulePool & getSchedulePool() const;
/// Storage of allowed hosts from config.xml
void setRemoteHostFilter(const Poco::Util::AbstractConfiguration & config);
const RemoteHostFilter & getRemoteHostFilter() const;
std::shared_ptr<FilesystemCacheLog> getFilesystemCacheLog() const;
std::shared_ptr<FilesystemReadPrefetchesLog> getFilesystemReadPrefetchesLog() const;
std::shared_ptr<BlobStorageLog> getBlobStorageLog() const;
enum class ApplicationType : uint8_t
{
KEEPER,
SERVER,
};
void setApplicationType(ApplicationType) {}
ApplicationType getApplicationType() const { return ApplicationType::KEEPER; }
IAsynchronousReader & getThreadPoolReader(FilesystemReaderType type) const;
#if USE_LIBURING
IOUringReader & getIOUringReader() const;
#endif
std::shared_ptr<AsyncReadCounters> getAsyncReadCounters() const;
ThreadPool & getThreadPoolWriter() const;
ThrottlerPtr getRemoteReadThrottler() const;
ThrottlerPtr getRemoteWriteThrottler() const;
ThrottlerPtr getLocalReadThrottler() const;
ThrottlerPtr getLocalWriteThrottler() const;
ReadSettings getReadSettings() const;
/// Resource management related
ResourceManagerPtr getResourceManager() const;
ClassifierPtr getWorkloadClassifier() const;
std::shared_ptr<KeeperDispatcher> getKeeperDispatcher() const;
std::shared_ptr<KeeperDispatcher> tryGetKeeperDispatcher() const;
void initializeKeeperDispatcher(bool start_async) const;
void shutdownKeeperDispatcher() const;
void updateKeeperConfiguration(const Poco::Util::AbstractConfiguration & config);
zkutil::ZooKeeperPtr getZooKeeper() const;
const S3SettingsByEndpoint & getStorageS3Settings() const;
const String & getUserName() const { static std::string user; return user; }
const ServerSettings & getServerSettings() const;
bool hasTraceCollector() const;
bool isBackgroundOperationContext() const;
};
}

24
src/Coordination/Standalone/Settings.cpp
View File

@ -1,24 +0,0 @@
#include <Core/Settings.h>
namespace DB
{
IMPLEMENT_SETTINGS_TRAITS(SettingsTraits, LIST_OF_SETTINGS)
std::vector<String> Settings::getAllRegisteredNames() const
{
std::vector<String> all_settings;
for (const auto & setting_field : all())
{
all_settings.push_back(setting_field.getName());
}
return all_settings;
}
void Settings::set(std::string_view name, const Field & value)
{
BaseSettings::set(name, value);
}
}

19
src/Coordination/Standalone/ThreadStatusExt.cpp
View File

@ -1,19 +0,0 @@
#include <Common/CurrentThread.h>
#include <Common/ThreadStatus.h>
namespace DB
{
void CurrentThread::detachFromGroupIfNotDetached()
{
}
void CurrentThread::attachToGroup(const ThreadGroupPtr &)
{
}
void ThreadStatus::initGlobalProfiler(UInt64 /*global_profiler_real_time_period*/, UInt64 /*global_profiler_cpu_time_period*/)
{
}
}

7
src/Core/Settings.h
View File

@ -36,7 +36,7 @@ class IColumn;
M(Dialect, dialect, Dialect::clickhouse, "Which dialect will be used to parse query", 0)\
M(UInt64, min_compress_block_size, 65536, "The actual size of the block to compress, if the uncompressed data less than max_compress_block_size is no less than this value and no less than the volume of data for one mark.", 0) \
M(UInt64, max_compress_block_size, 1048576, "The maximum size of blocks of uncompressed data before compressing for writing to a table.", 0) \
M(UInt64, max_block_size, DEFAULT_BLOCK_SIZE, "Maximum block size for reading", 0) \
M(UInt64, max_block_size, DEFAULT_BLOCK_SIZE, "Maximum block size in rows for reading", 0) \
M(UInt64, max_insert_block_size, DEFAULT_INSERT_BLOCK_SIZE, "The maximum block size for insertion, if we control the creation of blocks for insertion.", 0) \
M(UInt64, min_insert_block_size_rows, DEFAULT_INSERT_BLOCK_SIZE, "Squash blocks passed to INSERT query to specified size in rows, if blocks are not big enough.", 0) \
M(UInt64, min_insert_block_size_bytes, (DEFAULT_INSERT_BLOCK_SIZE * 256), "Squash blocks passed to INSERT query to specified size in bytes, if blocks are not big enough.", 0) \
@ -612,6 +612,7 @@ class IColumn;
M(UInt64, mutations_sync, 0, "Wait for synchronous execution of ALTER TABLE UPDATE/DELETE queries (mutations). 0 - execute asynchronously. 1 - wait current server. 2 - wait all replicas if they exist.", 0) \
M(Bool, enable_lightweight_delete, true, "Enable lightweight DELETE mutations for mergetree tables.", 0) ALIAS(allow_experimental_lightweight_delete) \
M(UInt64, lightweight_deletes_sync, 2, "The same as 'mutation_sync', but controls only execution of lightweight deletes", 0) \
M(LightweightMutationProjectionMode, lightweight_mutation_projection_mode, LightweightMutationProjectionMode::THROW, "When lightweight delete happens on a table with projection(s), the possible operations include throw the exception as projection exists, or drop all projection related to this table then do lightweight delete.", 0) \
M(Bool, apply_deleted_mask, true, "Enables filtering out rows deleted with lightweight DELETE. If disabled, a query will be able to read those rows. This is useful for debugging and \"undelete\" scenarios", 0) \
M(Bool, optimize_normalize_count_variants, true, "Rewrite aggregate functions that semantically equals to count() as count().", 0) \
M(Bool, optimize_injective_functions_inside_uniq, true, "Delete injective functions of one argument inside uniq*() functions.", 0) \
@ -630,9 +631,8 @@ class IColumn;
M(Bool, optimize_time_filter_with_preimage, true, "Optimize Date and DateTime predicates by converting functions into equivalent comparisons without conversions (e.g. toYear(col) = 2023 -> col >= '2023-01-01' AND col <= '2023-12-31')", 0) \
M(Bool, normalize_function_names, true, "Normalize function names to their canonical names", 0) \
M(Bool, enable_early_constant_folding, true, "Enable query optimization where we analyze function and subqueries results and rewrite query if there are constants there", 0) \
M(Bool, deduplicate_blocks_in_dependent_materialized_views, false, "Should deduplicate blocks for materialized views if the block is not a duplicate for the table. Use true to always deduplicate in dependent tables.", 0) \
M(Bool, deduplicate_blocks_in_dependent_materialized_views, false, "Should deduplicate blocks for materialized views. Use true to always deduplicate in dependent tables.", 0) \
M(Bool, throw_if_deduplication_in_dependent_materialized_views_enabled_with_async_insert, true, "Throw exception on INSERT query when the setting `deduplicate_blocks_in_dependent_materialized_views` is enabled along with `async_insert`. It guarantees correctness, because these features can't work together.", 0) \
M(Bool, update_insert_deduplication_token_in_dependent_materialized_views, false, "Should update insert deduplication token with table identifier during insert in dependent materialized views.", 0) \
M(Bool, materialized_views_ignore_errors, false, "Allows to ignore errors for MATERIALIZED VIEW, and deliver original block to the table regardless of MVs", 0) \
M(Bool, ignore_materialized_views_with_dropped_target_table, false, "Ignore MVs with dropped target table during pushing to views", 0) \
M(Bool, allow_experimental_refreshable_materialized_view, false, "Allow refreshable materialized views (CREATE MATERIALIZED VIEW <name> REFRESH ...).", 0) \
@ -949,6 +949,7 @@ class IColumn;
#define OBSOLETE_SETTINGS(M, ALIAS) \
/** Obsolete settings that do nothing but left for compatibility reasons. Remove each one after half a year of obsolescence. */ \
MAKE_OBSOLETE(M, Bool, update_insert_deduplication_token_in_dependent_materialized_views, 1) \
MAKE_OBSOLETE(M, UInt64, max_memory_usage_for_all_queries, 0) \
MAKE_OBSOLETE(M, UInt64, multiple_joins_rewriter_version, 0) \
MAKE_OBSOLETE(M, Bool, enable_debug_queries, false) \

1
src/Core/SettingsChangesHistory.cpp
View File

@ -61,6 +61,7 @@ static std::initializer_list<std::pair<ClickHouseVersion, SettingsChangesHistory
{"optimize_functions_to_subcolumns", false, true, "Enable optimization by default"},
{"input_format_json_ignore_key_case", false, false, "Ignore json key case while read json field from string."},
{"optimize_trivial_insert_select", true, false, "The optimization does not make sense in many cases."},
{"lightweight_mutation_projection_mode", "throw", "throw", "When lightweight delete happens on a table with projection(s), the possible operations include throw the exception as projection exists, or drop all projection related to this table then do lightweight delete."},
{"database_replicated_allow_heavy_create", true, false, "Long-running DDL queries (CREATE AS SELECT and POPULATE) for Replicated database engine was forbidden"},
}},
{"24.6", {{"materialize_skip_indexes_on_insert", true, true, "Added new setting to allow to disable materialization of skip indexes on insert"},

4
src/Core/SettingsEnums.cpp
View File

@ -173,6 +173,10 @@ IMPLEMENT_SETTING_ENUM(ParallelReplicasCustomKeyFilterType, ErrorCodes::BAD_ARGU
{{"default", ParallelReplicasCustomKeyFilterType::DEFAULT},
{"range", ParallelReplicasCustomKeyFilterType::RANGE}})
IMPLEMENT_SETTING_ENUM(LightweightMutationProjectionMode, ErrorCodes::BAD_ARGUMENTS,
{{"throw", LightweightMutationProjectionMode::THROW},
{"drop", LightweightMutationProjectionMode::DROP}})
IMPLEMENT_SETTING_AUTO_ENUM(LocalFSReadMethod, ErrorCodes::BAD_ARGUMENTS)
IMPLEMENT_SETTING_ENUM(ParquetVersion, ErrorCodes::BAD_ARGUMENTS,

8
src/Core/SettingsEnums.h
View File

@ -339,6 +339,14 @@ enum class ParallelReplicasCustomKeyFilterType : uint8_t
DECLARE_SETTING_ENUM(ParallelReplicasCustomKeyFilterType)
enum class LightweightMutationProjectionMode : uint8_t
{
THROW,
DROP,
};
DECLARE_SETTING_ENUM(LightweightMutationProjectionMode)
DECLARE_SETTING_ENUM(LocalFSReadMethod)
enum class ObjectStorageQueueMode : uint8_t

45
src/Core/SettingsFields.cpp
View File

@ -380,15 +380,6 @@ void SettingFieldString::readBinary(ReadBuffer & in)
*this = std::move(str);
}
/// Unbeautiful workaround for clickhouse-keeper standalone build ("-DBUILD_STANDALONE_KEEPER=1").
/// In this build, we don't build and link library dbms (to which SettingsField.cpp belongs) but
/// only build SettingsField.cpp. Further dependencies, e.g. DataTypeString and DataTypeMap below,
/// require building of further files for clickhouse-keeper. To keep dependencies slim, we don't do
/// that. The linker does not complain only because clickhouse-keeper does not call any of below
/// functions. A cleaner alternative would be more modular libraries, e.g. one for data types, which
/// could then be linked by the server and the linker.
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
SettingFieldMap::SettingFieldMap(const Field & f) : value(fieldToMap(f)) {}
String SettingFieldMap::toString() const
@ -428,42 +419,6 @@ void SettingFieldMap::readBinary(ReadBuffer & in)
*this = map;
}
#else
namespace ErrorCodes
{
extern const int LOGICAL_ERROR;
}
SettingFieldMap::SettingFieldMap(const Field &) : value(Map()) {}
String SettingFieldMap::toString() const
{
throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Setting of type Map not supported");
}
SettingFieldMap & SettingFieldMap::operator =(const Field &)
{
throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Setting of type Map not supported");
}
void SettingFieldMap::parseFromString(const String &)
{
throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Setting of type Map not supported");
}
void SettingFieldMap::writeBinary(WriteBuffer &) const
{
throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Setting of type Map not supported");
}
void SettingFieldMap::readBinary(ReadBuffer &)
{
throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Setting of type Map not supported");
}
#endif
namespace
{
char stringToChar(const String & str)

14
src/Core/SettingsFields.h
View File

@ -247,12 +247,6 @@ struct SettingFieldString
void readBinary(ReadBuffer & in);
};
#ifdef CLICKHOUSE_KEEPER_STANDALONE_BUILD
#define NORETURN [[noreturn]]
#else
#define NORETURN
#endif
struct SettingFieldMap
{
public:
@ -269,11 +263,11 @@ public:
operator const Map &() const { return value; } /// NOLINT
explicit operator Field() const { return value; }
NORETURN String toString() const;
NORETURN void parseFromString(const String & str);
String toString() const;
void parseFromString(const String & str);
NORETURN void writeBinary(WriteBuffer & out) const;
NORETURN void readBinary(ReadBuffer & in);
void writeBinary(WriteBuffer & out) const;
void readBinary(ReadBuffer & in);
};
#undef NORETURN

5
src/Daemon/BaseDaemon.cpp
View File

@ -502,9 +502,7 @@ private:
if (collectCrashLog)
collectCrashLog(sig, thread_num, query_id, stack_trace);
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
Context::getGlobalContextInstance()->handleCrash();
#endif
/// Send crash report to developers (if configured)
if (sig != SanitizerTrap)
@ -533,8 +531,6 @@ private:
}
}
/// ClickHouse Keeper does not link to some parts of Settings.
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
/// List changed settings.
if (!query_id.empty())
{
@ -549,7 +545,6 @@ private:
LOG_FATAL(log, "Changed settings: {}", changed_settings);
}
}
#endif
/// When everything is done, we will try to send these error messages to the client.
if (thread_ptr)

2
src/Daemon/SentryWriter.cpp
View File

@ -19,7 +19,7 @@
#include "config.h"
#include <Common/config_version.h>
#if USE_SENTRY && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_SENTRY
# include <sentry.h>
# include <cstdio>

2
src/Disks/IO/ReadBufferFromRemoteFSGather.cpp
View File

@ -78,7 +78,6 @@ SeekableReadBufferPtr ReadBufferFromRemoteFSGather::createImplementationBuffer(c
std::unique_ptr<ReadBufferFromFileBase> buf;
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
if (with_file_cache)
{
auto cache_key = settings.remote_fs_cache->createKeyForPath(object_path);
@ -96,7 +95,6 @@ SeekableReadBufferPtr ReadBufferFromRemoteFSGather::createImplementationBuffer(c
/* read_until_position */std::nullopt,
cache_log);
}
#endif
/// Can't wrap CachedOnDiskReadBufferFromFile in CachedInMemoryReadBufferFromFile because the
/// former doesn't support seeks.

2
src/Disks/ObjectStorages/DiskObjectStorage.h
View File

@ -195,7 +195,6 @@ public:
/// DiskObjectStorage(CachedObjectStorage(CachedObjectStorage(S3ObjectStorage)))
String getStructure() const { return fmt::format("DiskObjectStorage-{}({})", getName(), object_storage->getName()); }
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
/// Add a cache layer.
/// Example: DiskObjectStorage(S3ObjectStorage) -> DiskObjectStorage(CachedObjectStorage(S3ObjectStorage))
/// There can be any number of cache layers:
@ -204,7 +203,6 @@ public:
/// Get names of all cache layers. Name is how cache is defined in configuration file.
NameSet getCacheLayersNames() const override;
#endif
bool supportsStat() const override { return metadata_storage->supportsStat(); }
struct stat stat(const String & path) const override;

4
src/Disks/ObjectStorages/DiskObjectStorageMetadata.cpp
View File

@ -222,11 +222,7 @@ ObjectKeyWithMetadata DiskObjectStorageMetadata::popLastObject()
bool DiskObjectStorageMetadata::getWriteFullObjectKeySetting()
{
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
return Context::getGlobalContextInstance()->getServerSettings().storage_metadata_write_full_object_key;
#else
return false;
#endif
}
}

6
src/Disks/ObjectStorages/MetadataStorageFactory.cpp
View File

@ -2,9 +2,7 @@
#include <Disks/ObjectStorages/MetadataStorageFromDisk.h>
#include <Disks/ObjectStorages/MetadataStorageFromPlainObjectStorage.h>
#include <Disks/ObjectStorages/MetadataStorageFromPlainRewritableObjectStorage.h>
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
#include <Disks/ObjectStorages/Web/MetadataStorageFromStaticFilesWebServer.h>
#endif
#include <Disks/DiskLocal.h>
#include <Interpreters/Context.h>
@ -135,7 +133,6 @@ void registerPlainRewritableMetadataStorage(MetadataStorageFactory & factory)
});
}
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
void registerMetadataStorageFromStaticFilesWebServer(MetadataStorageFactory & factory)
{
factory.registerMetadataStorageType("web", [](
@ -147,7 +144,6 @@ void registerMetadataStorageFromStaticFilesWebServer(MetadataStorageFactory & fa
return std::make_shared<MetadataStorageFromStaticFilesWebServer>(assert_cast<const WebObjectStorage &>(*object_storage));
});
}
#endif
void registerMetadataStorages()
{
@ -155,9 +151,7 @@ void registerMetadataStorages()
registerMetadataStorageFromDisk(factory);
registerPlainMetadataStorage(factory);
registerPlainRewritableMetadataStorage(factory);
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
registerMetadataStorageFromStaticFilesWebServer(factory);
#endif
}
}

18
src/Disks/ObjectStorages/ObjectStorageFactory.cpp
View File

@ -7,19 +7,17 @@
#include <Disks/ObjectStorages/S3/S3ObjectStorage.h>
#include <Disks/ObjectStorages/S3/diskSettings.h>
#endif
#if USE_HDFS && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_HDFS
#include <Disks/ObjectStorages/HDFS/HDFSObjectStorage.h>
#include <Storages/ObjectStorage/HDFS/HDFSCommon.h>
#endif
#if USE_AZURE_BLOB_STORAGE && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_AZURE_BLOB_STORAGE
#include <Disks/ObjectStorages/AzureBlobStorage/AzureObjectStorage.h>
#include <Disks/ObjectStorages/AzureBlobStorage/AzureBlobStorageAuth.h>
#endif
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
#include <Disks/ObjectStorages/Web/WebObjectStorage.h>
#include <Disks/ObjectStorages/Local/LocalObjectStorage.h>
#include <Disks/loadLocalDiskConfig.h>
#endif
#include <Disks/ObjectStorages/MetadataStorageFactory.h>
#include <Disks/ObjectStorages/PlainObjectStorage.h>
#include <Disks/ObjectStorages/PlainRewritableObjectStorage.h>
@ -284,7 +282,7 @@ void registerS3PlainRewritableObjectStorage(ObjectStorageFactory & factory)
#endif
#if USE_HDFS && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_HDFS
void registerHDFSObjectStorage(ObjectStorageFactory & factory)
{
factory.registerObjectStorageType(
@ -309,7 +307,7 @@ void registerHDFSObjectStorage(ObjectStorageFactory & factory)
}
#endif
#if USE_AZURE_BLOB_STORAGE && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_AZURE_BLOB_STORAGE
void registerAzureObjectStorage(ObjectStorageFactory & factory)
{
auto creator = [](
@ -333,7 +331,6 @@ void registerAzureObjectStorage(ObjectStorageFactory & factory)
}
#endif
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
void registerWebObjectStorage(ObjectStorageFactory & factory)
{
factory.registerObjectStorageType("web", [](
@ -381,7 +378,6 @@ void registerLocalObjectStorage(ObjectStorageFactory & factory)
factory.registerObjectStorageType("local_blob_storage", creator);
factory.registerObjectStorageType("local", creator);
}
#endif
void registerObjectStorages()
{
@ -393,18 +389,16 @@ void registerObjectStorages()
registerS3PlainRewritableObjectStorage(factory);
#endif
#if USE_HDFS && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_HDFS
registerHDFSObjectStorage(factory);
#endif
#if USE_AZURE_BLOB_STORAGE && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_AZURE_BLOB_STORAGE
registerAzureObjectStorage(factory);
#endif
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
registerWebObjectStorage(factory);
registerLocalObjectStorage(factory);
#endif
}
}

12
src/Disks/ObjectStorages/createMetadataStorageMetrics.h
View File

@ -1,14 +1,14 @@
#pragma once
#include "config.h"
#if USE_AWS_S3
# include <Disks/ObjectStorages/S3/S3ObjectStorage.h>
#endif
#if USE_AZURE_BLOB_STORAGE && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_AZURE_BLOB_STORAGE
# include <Disks/ObjectStorages/AzureBlobStorage/AzureObjectStorage.h>
#endif
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
# include <Disks/ObjectStorages/Local/LocalObjectStorage.h>
#endif
#include <Disks/ObjectStorages/Local/LocalObjectStorage.h>
#include <Disks/ObjectStorages/MetadataStorageMetrics.h>
namespace ProfileEvents
@ -42,7 +42,7 @@ inline MetadataStorageMetrics MetadataStorageMetrics::create<S3ObjectStorage, Me
}
#endif
#if USE_AZURE_BLOB_STORAGE && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_AZURE_BLOB_STORAGE
template <>
inline MetadataStorageMetrics MetadataStorageMetrics::create<AzureObjectStorage, MetadataStorageType::PlainRewritable>()
{
@ -53,7 +53,6 @@ inline MetadataStorageMetrics MetadataStorageMetrics::create<AzureObjectStorage,
}
#endif
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
template <>
inline MetadataStorageMetrics MetadataStorageMetrics::create<LocalObjectStorage, MetadataStorageType::PlainRewritable>()
{
@ -62,6 +61,5 @@ inline MetadataStorageMetrics MetadataStorageMetrics::create<LocalObjectStorage,
.directory_removed = ProfileEvents::DiskPlainRewritableLocalDirectoryRemoved,
.directory_map_size = CurrentMetrics::DiskPlainRewritableLocalDirectoryMapSize};
}
#endif
}

15
src/Disks/registerDisks.cpp
View File

@ -17,8 +17,6 @@ void registerDiskCache(DiskFactory & factory, bool global_skip_access_check);
void registerDiskObjectStorage(DiskFactory & factory, bool global_skip_access_check);
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
void registerDisks(bool global_skip_access_check)
{
auto & factory = DiskFactory::instance();
@ -34,17 +32,4 @@ void registerDisks(bool global_skip_access_check)
registerDiskObjectStorage(factory, global_skip_access_check);
}
#else
void registerDisks(bool global_skip_access_check)
{
auto & factory = DiskFactory::instance();
registerDiskLocal(factory, global_skip_access_check);
registerDiskObjectStorage(factory, global_skip_access_check);
}
#endif
}

2
src/IO/S3/BlobStorageLogWriter.cpp
View File

@ -56,7 +56,6 @@ void BlobStorageLogWriter::addEvent(
BlobStorageLogWriterPtr BlobStorageLogWriter::create(const String & disk_name)
{
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD /// Keeper standalone build doesn't have a context
if (auto blob_storage_log = Context::getGlobalContextInstance()->getBlobStorageLog())
{
auto log_writer = std::make_shared<BlobStorageLogWriter>(std::move(blob_storage_log));
@ -67,7 +66,6 @@ BlobStorageLogWriterPtr BlobStorageLogWriter::create(const String & disk_name)
return log_writer;
}
#endif
return {};
}

19
src/Interpreters/AsynchronousInsertQueue.cpp
View File

@ -301,7 +301,13 @@ void AsynchronousInsertQueue::preprocessInsertQuery(const ASTPtr & query, const
auto & insert_query = query->as<ASTInsertQuery &>();
insert_query.async_insert_flush = true;
InterpreterInsertQuery interpreter(query, query_context, query_context->getSettingsRef().insert_allow_materialized_columns);
InterpreterInsertQuery interpreter(
query,
query_context,
query_context->getSettingsRef().insert_allow_materialized_columns,
/* no_squash */ false,
/* no_destination */ false,
/* async_insert */ false);
auto table = interpreter.getTable(insert_query);
auto sample_block = InterpreterInsertQuery::getSampleBlock(insert_query, table, table->getInMemoryMetadataPtr(), query_context);
@ -781,7 +787,12 @@ try
try
{
interpreter = std::make_unique<InterpreterInsertQuery>(
key.query, insert_context, key.settings.insert_allow_materialized_columns, false, false, true);
key.query,
insert_context,
key.settings.insert_allow_materialized_columns,
false,
false,
true);
pipeline = interpreter->execute().pipeline;
chassert(pipeline.pushing());
@ -1000,7 +1011,7 @@ Chunk AsynchronousInsertQueue::processEntriesWithParsing(
}
Chunk chunk(executor.getResultColumns(), total_rows);
chunk.setChunkInfo(std::move(chunk_info));
chunk.getChunkInfos().add(std::move(chunk_info));
return chunk;
}
@ -1052,7 +1063,7 @@ Chunk AsynchronousInsertQueue::processPreprocessedEntries(
}
Chunk chunk(std::move(result_columns), total_rows);
chunk.setChunkInfo(std::move(chunk_info));
chunk.getChunkInfos().add(std::move(chunk_info));
return chunk;
}

8
src/Interpreters/Context.h
View File

@ -1,7 +1,5 @@
#pragma once
#ifndef CLICKHOUSE_KEEPER_STANDALONE_BUILD
#include <base/types.h>
#include <Common/isLocalAddress.h>
#include <Common/MultiVersion.h>
@ -1451,9 +1449,3 @@ struct HTTPContext : public IHTTPContext
};
}
#else
#include <Coordination/Standalone/Context.h>
#endif

18
src/Interpreters/InterpreterCheckQuery.cpp
View File

@ -2,6 +2,7 @@
#include <Interpreters/InterpreterFactory.h>
#include <algorithm>
#include <memory>
#include <Access/Common/AccessFlags.h>
@ -22,6 +23,7 @@
#include <Parsers/ASTCheckQuery.h>
#include <Parsers/ASTSetQuery.h>
#include <Processors/Chunk.h>
#include <Processors/IAccumulatingTransform.h>
#include <Processors/IInflatingTransform.h>
#include <Processors/ISimpleTransform.h>
@ -91,7 +93,7 @@ Chunk getChunkFromCheckResult(const String & database, const String & table, con
return Chunk(std::move(columns), 1);
}
class TableCheckTask : public ChunkInfo
class TableCheckTask : public ChunkInfoCloneable<TableCheckTask>
{
public:
TableCheckTask(StorageID table_id, const std::variant<std::monostate, ASTPtr, String> & partition_or_part, ContextPtr context)
@ -110,6 +112,12 @@ public:
context->checkAccess(AccessType::SHOW_TABLES, table_->getStorageID());
}
TableCheckTask(const TableCheckTask & other)
: table(other.table)
, check_data_tasks(other.check_data_tasks)
, is_finished(other.is_finished.load())
{}
std::optional<CheckResult> checkNext() const
{
if (isFinished())
@ -121,8 +129,8 @@ public:
std::this_thread::sleep_for(sleep_time);
});
IStorage::DataValidationTasksPtr check_data_tasks_ = check_data_tasks;
auto result = table->checkDataNext(check_data_tasks_);
IStorage::DataValidationTasksPtr tmp = check_data_tasks;
auto result = table->checkDataNext(tmp);
is_finished = !result.has_value();
return result;
}
@ -180,7 +188,7 @@ protected:
/// source should return at least one row to start pipeline
result.addColumn(ColumnUInt8::create(1, 1));
/// actual data stored in chunk info
result.setChunkInfo(std::move(current_check_task));
result.getChunkInfos().add(std::move(current_check_task));
return result;
}
@ -280,7 +288,7 @@ public:
protected:
void transform(Chunk & chunk) override
{
auto table_check_task = std::dynamic_pointer_cast<const TableCheckTask>(chunk.getChunkInfo());
auto table_check_task = chunk.getChunkInfos().get<TableCheckTask>();
auto check_result = table_check_task->checkNext();
if (!check_result)
{

9
src/Interpreters/InterpreterCreateQuery.cpp
View File

@ -1776,8 +1776,13 @@ BlockIO InterpreterCreateQuery::fillTableIfNeeded(const ASTCreateQuery & create)
else
insert->select = create.select->clone();
return InterpreterInsertQuery(insert, getContext(),
getContext()->getSettingsRef().insert_allow_materialized_columns).execute();
return InterpreterInsertQuery(
insert,
getContext(),
getContext()->getSettingsRef().insert_allow_materialized_columns,
/* no_squash */ false,
/* no_destination */ false,
/* async_isnert */ false).execute();
}
return {};

91
src/Interpreters/InterpreterDeleteQuery.cpp
View File

@ -61,24 +61,7 @@ BlockIO InterpreterDeleteQuery::execute()
auto table_lock = table->lockForShare(getContext()->getCurrentQueryId(), getContext()->getSettingsRef().lock_acquire_timeout);
auto metadata_snapshot = table->getInMemoryMetadataPtr();
if (table->supportsDelete())
{
/// Convert to MutationCommand
MutationCommands mutation_commands;
MutationCommand mut_command;
mut_command.type = MutationCommand::Type::DELETE;
mut_command.predicate = delete_query.predicate;
mutation_commands.emplace_back(mut_command);
table->checkMutationIsPossible(mutation_commands, getContext()->getSettingsRef());
MutationsInterpreter::Settings settings(false);
MutationsInterpreter(table, metadata_snapshot, mutation_commands, getContext(), settings).validate();
table->mutate(mutation_commands, getContext());
return {};
}
else if (table->supportsLightweightDelete())
auto lightweightDelete = [&]()
{
if (!getContext()->getSettingsRef().enable_lightweight_delete)
throw Exception(ErrorCodes::SUPPORT_IS_DISABLED,
@ -105,17 +88,77 @@ BlockIO InterpreterDeleteQuery::execute()
context->setSetting("mutations_sync", Field(context->getSettingsRef().lightweight_deletes_sync));
InterpreterAlterQuery alter_interpreter(alter_ast, context);
return alter_interpreter.execute();
};
if (table->supportsDelete())
{
/// Convert to MutationCommand
MutationCommands mutation_commands;
MutationCommand mut_command;
mut_command.type = MutationCommand::Type::DELETE;
mut_command.predicate = delete_query.predicate;
mutation_commands.emplace_back(mut_command);
table->checkMutationIsPossible(mutation_commands, getContext()->getSettingsRef());
MutationsInterpreter::Settings settings(false);
MutationsInterpreter(table, metadata_snapshot, mutation_commands, getContext(), settings).validate();
table->mutate(mutation_commands, getContext());
return {};
}
else if (table->supportsLightweightDelete())
{
return lightweightDelete();
}
else
{
/// Currently just better exception for the case of a table with projection,
/// can act differently according to the setting.
if (table->hasProjection())
{
throw Exception(ErrorCodes::NOT_IMPLEMENTED,
"DELETE query is not supported for table {} as it has projections. "
"User should drop all the projections manually before running the query",
table->getStorageID().getFullTableName());
auto context = Context::createCopy(getContext());
auto mode = context->getSettingsRef().lightweight_mutation_projection_mode;
if (mode == LightweightMutationProjectionMode::THROW)
{
throw Exception(ErrorCodes::NOT_IMPLEMENTED,
"DELETE query is not supported for table {} as it has projections. "
"User should drop all the projections manually before running the query",
table->getStorageID().getFullTableName());
}
else if (mode == LightweightMutationProjectionMode::DROP)
{
std::vector<String> all_projections = metadata_snapshot->projections.getAllRegisteredNames();
context->setSetting("mutations_sync", Field(context->getSettingsRef().lightweight_deletes_sync));
/// Drop projections first so that lightweight delete can be performed.
for (const auto & projection : all_projections)
{
String alter_query =
"ALTER TABLE " + table->getStorageID().getFullTableName()
+ (delete_query.cluster.empty() ? "" : " ON CLUSTER " + backQuoteIfNeed(delete_query.cluster))
+ " DROP PROJECTION IF EXISTS " + projection;
ParserAlterQuery parser;
ASTPtr alter_ast = parseQuery(
parser,
alter_query.data(),
alter_query.data() + alter_query.size(),
"ALTER query",
0,
DBMS_DEFAULT_MAX_PARSER_DEPTH,
DBMS_DEFAULT_MAX_PARSER_BACKTRACKS);
InterpreterAlterQuery alter_interpreter(alter_ast, context);
alter_interpreter.execute();
}
}
else
{
throw Exception(ErrorCodes::BAD_ARGUMENTS,
"Unrecognized lightweight_mutation_projection_mode, only throw and drop are allowed.");
}
return lightweightDelete();
}
throw Exception(ErrorCodes::BAD_ARGUMENTS,

8
src/Interpreters/InterpreterExplainQuery.cpp
View File

@ -534,7 +534,13 @@ QueryPipeline InterpreterExplainQuery::executeImpl()
}
else if (dynamic_cast<const ASTInsertQuery *>(ast.getExplainedQuery().get()))
{
InterpreterInsertQuery insert(ast.getExplainedQuery(), getContext());
InterpreterInsertQuery insert(
ast.getExplainedQuery(),
getContext(),
/* allow_materialized */ false,
/* no_squash */ false,
/* no_destination */ false,
/* async_isnert */ false);
auto io = insert.execute();
printPipeline(io.pipeline.getProcessors(), buf);
}

678
src/Interpreters/InterpreterInsertQuery.cpp
View File

@ -16,6 +16,7 @@
#include <Interpreters/getTableExpressions.h>
#include <Interpreters/processColumnTransformers.h>
#include <Interpreters/InterpreterSelectQueryAnalyzer.h>
#include <Interpreters/Context_fwd.h>
#include <Parsers/ASTFunction.h>
#include <Parsers/ASTInsertQuery.h>
#include <Parsers/ASTSelectQuery.h>
@ -26,6 +27,7 @@
#include <Processors/Transforms/CountingTransform.h>
#include <Processors/Transforms/ExpressionTransform.h>
#include <Processors/Transforms/MaterializingTransform.h>
#include <Processors/Transforms/DeduplicationTokenTransforms.h>
#include <Processors/Transforms/SquashingTransform.h>
#include <Processors/Transforms/PlanSquashingTransform.h>
#include <Processors/Transforms/getSourceFromASTInsertQuery.h>
@ -38,6 +40,7 @@
#include <Common/ThreadStatus.h>
#include <Common/checkStackSize.h>
#include <Common/ProfileEvents.h>
#include "base/defines.h"
namespace ProfileEvents
@ -394,28 +397,358 @@ Chain InterpreterInsertQuery::buildPreSinkChain(
return out;
}
std::pair<std::vector<Chain>, std::vector<Chain>> InterpreterInsertQuery::buildPreAndSinkChains(size_t presink_streams, size_t sink_streams, StoragePtr table, const StorageMetadataPtr & metadata_snapshot, const Block & query_sample_block)
{
chassert(presink_streams > 0);
chassert(sink_streams > 0);
ThreadGroupPtr running_group;
if (current_thread)
running_group = current_thread->getThreadGroup();
if (!running_group)
running_group = std::make_shared<ThreadGroup>(getContext());
std::vector<Chain> sink_chains;
std::vector<Chain> presink_chains;
for (size_t i = 0; i < sink_streams; ++i)
{
auto out = buildSink(table, metadata_snapshot, /* thread_status_holder= */ nullptr,
running_group, /* elapsed_counter_ms= */ nullptr);
sink_chains.emplace_back(std::move(out));
}
for (size_t i = 0; i < presink_streams; ++i)
{
auto out = buildPreSinkChain(sink_chains[0].getInputHeader(), table, metadata_snapshot, query_sample_block);
presink_chains.emplace_back(std::move(out));
}
return {std::move(presink_chains), std::move(sink_chains)};
}
QueryPipeline InterpreterInsertQuery::buildInsertSelectPipeline(ASTInsertQuery & query, StoragePtr table)
{
const Settings & settings = getContext()->getSettingsRef();
auto metadata_snapshot = table->getInMemoryMetadataPtr();
auto query_sample_block = getSampleBlock(query, table, metadata_snapshot, getContext(), no_destination, allow_materialized);
bool is_trivial_insert_select = false;
if (settings.optimize_trivial_insert_select)
{
const auto & select_query = query.select->as<ASTSelectWithUnionQuery &>();
const auto & selects = select_query.list_of_selects->children;
const auto & union_modes = select_query.list_of_modes;
/// ASTSelectWithUnionQuery is not normalized now, so it may pass some queries which can be Trivial select queries
const auto mode_is_all = [](const auto & mode) { return mode == SelectUnionMode::UNION_ALL; };
is_trivial_insert_select =
std::all_of(union_modes.begin(), union_modes.end(), std::move(mode_is_all))
&& std::all_of(selects.begin(), selects.end(), isTrivialSelect);
}
ContextPtr select_context = getContext();
if (is_trivial_insert_select)
{
/** When doing trivial INSERT INTO ... SELECT ... FROM table,
* don't need to process SELECT with more than max_insert_threads
* and it's reasonable to set block size for SELECT to the desired block size for INSERT
* to avoid unnecessary squashing.
*/
Settings new_settings = select_context->getSettings();
new_settings.max_threads = std::max<UInt64>(1, settings.max_insert_threads);
if (table->prefersLargeBlocks())
{
if (settings.min_insert_block_size_rows)
new_settings.max_block_size = settings.min_insert_block_size_rows;
if (settings.min_insert_block_size_bytes)
new_settings.preferred_block_size_bytes = settings.min_insert_block_size_bytes;
}
auto context_for_trivial_select = Context::createCopy(context);
context_for_trivial_select->setSettings(new_settings);
context_for_trivial_select->setInsertionTable(getContext()->getInsertionTable(), getContext()->getInsertionTableColumnNames());
select_context = context_for_trivial_select;
}
QueryPipelineBuilder pipeline;
{
auto select_query_options = SelectQueryOptions(QueryProcessingStage::Complete, 1);
if (settings.allow_experimental_analyzer)
{
InterpreterSelectQueryAnalyzer interpreter_select_analyzer(query.select, select_context, select_query_options);
pipeline = interpreter_select_analyzer.buildQueryPipeline();
}
else
{
InterpreterSelectWithUnionQuery interpreter_select(query.select, select_context, select_query_options);
pipeline = interpreter_select.buildQueryPipeline();
}
}
pipeline.dropTotalsAndExtremes();
/// Allow to insert Nullable into non-Nullable columns, NULL values will be added as defaults values.
if (getContext()->getSettingsRef().insert_null_as_default)
{
const auto & input_columns = pipeline.getHeader().getColumnsWithTypeAndName();
const auto & query_columns = query_sample_block.getColumnsWithTypeAndName();
const auto & output_columns = metadata_snapshot->getColumns();
if (input_columns.size() == query_columns.size())
{
for (size_t col_idx = 0; col_idx < query_columns.size(); ++col_idx)
{
/// Change query sample block columns to Nullable to allow inserting nullable columns, where NULL values will be substituted with
/// default column values (in AddingDefaultsTransform), so all values will be cast correctly.
if (isNullableOrLowCardinalityNullable(input_columns[col_idx].type)
&& !isNullableOrLowCardinalityNullable(query_columns[col_idx].type)
&& !isVariant(query_columns[col_idx].type)
&& !isDynamic(query_columns[col_idx].type)
&& output_columns.has(query_columns[col_idx].name))
{
query_sample_block.setColumn(
col_idx,
ColumnWithTypeAndName(
makeNullableOrLowCardinalityNullable(query_columns[col_idx].column),
makeNullableOrLowCardinalityNullable(query_columns[col_idx].type),
query_columns[col_idx].name));
}
}
}
}
auto actions_dag = ActionsDAG::makeConvertingActions(
pipeline.getHeader().getColumnsWithTypeAndName(),
query_sample_block.getColumnsWithTypeAndName(),
ActionsDAG::MatchColumnsMode::Position);
auto actions = std::make_shared<ExpressionActions>(actions_dag, ExpressionActionsSettings::fromContext(getContext(), CompileExpressions::yes));
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
{
return std::make_shared<ExpressionTransform>(in_header, actions);
});
/// We need to convert Sparse columns to full, because it's destination storage
/// may not support it or may have different settings for applying Sparse serialization.
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
{
return std::make_shared<MaterializingTransform>(in_header);
});
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
{
auto context_ptr = getContext();
auto counting = std::make_shared<CountingTransform>(in_header, nullptr, context_ptr->getQuota());
counting->setProcessListElement(context_ptr->getProcessListElement());
counting->setProgressCallback(context_ptr->getProgressCallback());
return counting;
});
size_t num_select_threads = pipeline.getNumThreads();
pipeline.resize(1);
if (shouldAddSquashingFroStorage(table))
{
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
{
return std::make_shared<PlanSquashingTransform>(
in_header,
table->prefersLargeBlocks() ? settings.min_insert_block_size_rows : settings.max_block_size,
table->prefersLargeBlocks() ? settings.min_insert_block_size_bytes : 0ULL);
});
}
pipeline.addSimpleTransform([&](const Block &in_header) -> ProcessorPtr
{
return std::make_shared<DeduplicationToken::AddTokenInfoTransform>(in_header);
});
if (!settings.insert_deduplication_token.value.empty())
{
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
{
return std::make_shared<DeduplicationToken::SetUserTokenTransform>(settings.insert_deduplication_token.value, in_header);
});
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
{
return std::make_shared<DeduplicationToken::SetSourceBlockNumberTransform>(in_header);
});
}
/// Number of streams works like this:
/// * For the SELECT, use `max_threads`, or `max_insert_threads`, or whatever
/// InterpreterSelectQuery ends up with.
/// * Use `max_insert_threads` streams for various insert-preparation steps, e.g.
/// materializing and squashing (too slow to do in one thread). That's `presink_chains`.
/// * If the table supports parallel inserts, use max_insert_threads for writing to IStorage.
/// Otherwise ResizeProcessor them down to 1 stream.
size_t presink_streams_size = std::max<size_t>(settings.max_insert_threads, pipeline.getNumStreams());
size_t sink_streams_size = table->supportsParallelInsert() ? std::max<size_t>(1, settings.max_insert_threads) : 1;
if (!settings.parallel_view_processing)
{
auto table_id = table->getStorageID();
auto views = DatabaseCatalog::instance().getDependentViews(table_id);
if (table->isView() || !views.empty())
sink_streams_size = 1;
}
auto [presink_chains, sink_chains] = buildPreAndSinkChains(
presink_streams_size, sink_streams_size,
table, metadata_snapshot, query_sample_block);
pipeline.resize(presink_chains.size());
if (shouldAddSquashingFroStorage(table))
{
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
{
return std::make_shared<ApplySquashingTransform>(
in_header,
table->prefersLargeBlocks() ? settings.min_insert_block_size_rows : settings.max_block_size,
table->prefersLargeBlocks() ? settings.min_insert_block_size_bytes : 0ULL);
});
}
for (auto & chain : presink_chains)
pipeline.addResources(chain.detachResources());
pipeline.addChains(std::move(presink_chains));
pipeline.resize(sink_streams_size);
for (auto & chain : sink_chains)
pipeline.addResources(chain.detachResources());
pipeline.addChains(std::move(sink_chains));
if (!settings.parallel_view_processing)
{
/// Don't use more threads for INSERT than for SELECT to reduce memory consumption.
if (pipeline.getNumThreads() > num_select_threads)
pipeline.setMaxThreads(num_select_threads);
}
else if (pipeline.getNumThreads() < settings.max_threads)
{
/// It is possible for query to have max_threads=1, due to optimize_trivial_insert_select,
/// however in case of parallel_view_processing and multiple views, views can still be processed in parallel.
///
/// Note, number of threads will be limited by buildPushingToViewsChain() to max_threads.
pipeline.setMaxThreads(settings.max_threads);
}
pipeline.setSinks([&](const Block & cur_header, QueryPipelineBuilder::StreamType) -> ProcessorPtr
{
return std::make_shared<EmptySink>(cur_header);
});
return QueryPipelineBuilder::getPipeline(std::move(pipeline));
}
QueryPipeline InterpreterInsertQuery::buildInsertPipeline(ASTInsertQuery & query, StoragePtr table)
{
const Settings & settings = getContext()->getSettingsRef();
auto metadata_snapshot = table->getInMemoryMetadataPtr();
auto query_sample_block = getSampleBlock(query, table, metadata_snapshot, getContext(), no_destination, allow_materialized);
Chain chain;
{
auto [presink_chains, sink_chains] = buildPreAndSinkChains(
/* presink_streams */1, /* sink_streams */1,
table, metadata_snapshot, query_sample_block);
chain = std::move(presink_chains.front());
chain.appendChain(std::move(sink_chains.front()));
}
if (!settings.insert_deduplication_token.value.empty())
{
chain.addSource(std::make_shared<DeduplicationToken::SetSourceBlockNumberTransform>(chain.getInputHeader()));
chain.addSource(std::make_shared<DeduplicationToken::SetUserTokenTransform>(settings.insert_deduplication_token.value, chain.getInputHeader()));
}
chain.addSource(std::make_shared<DeduplicationToken::AddTokenInfoTransform>(chain.getInputHeader()));
if (shouldAddSquashingFroStorage(table))
{
bool table_prefers_large_blocks = table->prefersLargeBlocks();
auto squashing = std::make_shared<ApplySquashingTransform>(
chain.getInputHeader(),
table_prefers_large_blocks ? settings.min_insert_block_size_rows : settings.max_block_size,
table_prefers_large_blocks ? settings.min_insert_block_size_bytes : 0ULL);
chain.addSource(std::move(squashing));
auto balancing = std::make_shared<PlanSquashingTransform>(
chain.getInputHeader(),
table_prefers_large_blocks ? settings.min_insert_block_size_rows : settings.max_block_size,
table_prefers_large_blocks ? settings.min_insert_block_size_bytes : 0ULL);
chain.addSource(std::move(balancing));
}
auto context_ptr = getContext();
auto counting = std::make_shared<CountingTransform>(chain.getInputHeader(), nullptr, context_ptr->getQuota());
counting->setProcessListElement(context_ptr->getProcessListElement());
counting->setProgressCallback(context_ptr->getProgressCallback());
chain.addSource(std::move(counting));
QueryPipeline pipeline = QueryPipeline(std::move(chain));
pipeline.setNumThreads(std::min<size_t>(pipeline.getNumThreads(), settings.max_threads));
pipeline.setConcurrencyControl(settings.use_concurrency_control);
if (query.hasInlinedData() && !async_insert)
{
/// can execute without additional data
auto format = getInputFormatFromASTInsertQuery(query_ptr, true, query_sample_block, getContext(), nullptr);
for (auto && buffer : owned_buffers)
format->addBuffer(std::move(buffer));
auto pipe = getSourceFromInputFormat(query_ptr, std::move(format), getContext(), nullptr);
pipeline.complete(std::move(pipe));
}
return pipeline;
}
BlockIO InterpreterInsertQuery::execute()
{
const Settings & settings = getContext()->getSettingsRef();
auto & query = query_ptr->as<ASTInsertQuery &>();
QueryPipelineBuilder pipeline;
std::optional<QueryPipeline> distributed_pipeline;
QueryPlanResourceHolder resources;
StoragePtr table = getTable(query);
checkStorageSupportsTransactionsIfNeeded(table, getContext());
StoragePtr inner_table;
if (const auto * mv = dynamic_cast<const StorageMaterializedView *>(table.get()))
inner_table = mv->getTargetTable();
if (query.partition_by && !table->supportsPartitionBy())
throw Exception(ErrorCodes::NOT_IMPLEMENTED, "PARTITION BY clause is not supported by storage");
auto table_lock = table->lockForShare(getContext()->getInitialQueryId(), settings.lock_acquire_timeout);
auto metadata_snapshot = table->getInMemoryMetadataPtr();
auto metadata_snapshot = table->getInMemoryMetadataPtr();
auto query_sample_block = getSampleBlock(query, table, metadata_snapshot, getContext(), no_destination, allow_materialized);
/// For table functions we check access while executing
@ -423,320 +756,43 @@ BlockIO InterpreterInsertQuery::execute()
if (!query.table_function)
getContext()->checkAccess(AccessType::INSERT, query.table_id, query_sample_block.getNames());
if (query.select && settings.parallel_distributed_insert_select)
// Distributed INSERT SELECT
distributed_pipeline = table->distributedWrite(query, getContext());
std::vector<Chain> presink_chains;
std::vector<Chain> sink_chains;
if (!distributed_pipeline)
if (!allow_materialized)
{
/// Number of streams works like this:
/// * For the SELECT, use `max_threads`, or `max_insert_threads`, or whatever
/// InterpreterSelectQuery ends up with.
/// * Use `max_insert_threads` streams for various insert-preparation steps, e.g.
/// materializing and squashing (too slow to do in one thread). That's `presink_chains`.
/// * If the table supports parallel inserts, use the same streams for writing to IStorage.
/// Otherwise ResizeProcessor them down to 1 stream.
/// * If it's not an INSERT SELECT, forget all that and use one stream.
size_t pre_streams_size = 1;
size_t sink_streams_size = 1;
if (query.select)
{
bool is_trivial_insert_select = false;
if (settings.optimize_trivial_insert_select)
{
const auto & select_query = query.select->as<ASTSelectWithUnionQuery &>();
const auto & selects = select_query.list_of_selects->children;
const auto & union_modes = select_query.list_of_modes;
/// ASTSelectWithUnionQuery is not normalized now, so it may pass some queries which can be Trivial select queries
const auto mode_is_all = [](const auto & mode) { return mode == SelectUnionMode::UNION_ALL; };
is_trivial_insert_select =
std::all_of(union_modes.begin(), union_modes.end(), std::move(mode_is_all))
&& std::all_of(selects.begin(), selects.end(), isTrivialSelect);
}
if (is_trivial_insert_select)
{
/** When doing trivial INSERT INTO ... SELECT ... FROM table,
* don't need to process SELECT with more than max_insert_threads
* and it's reasonable to set block size for SELECT to the desired block size for INSERT
* to avoid unnecessary squashing.
*/
Settings new_settings = getContext()->getSettings();
new_settings.max_threads = std::max<UInt64>(1, settings.max_insert_threads);
if (table->prefersLargeBlocks())
{
if (settings.min_insert_block_size_rows)
new_settings.max_block_size = settings.min_insert_block_size_rows;
if (settings.min_insert_block_size_bytes)
new_settings.preferred_block_size_bytes = settings.min_insert_block_size_bytes;
}
auto new_context = Context::createCopy(context);
new_context->setSettings(new_settings);
new_context->setInsertionTable(getContext()->getInsertionTable(), getContext()->getInsertionTableColumnNames());
auto select_query_options = SelectQueryOptions(QueryProcessingStage::Complete, 1);
if (settings.allow_experimental_analyzer)
{
InterpreterSelectQueryAnalyzer interpreter_select_analyzer(query.select, new_context, select_query_options);
pipeline = interpreter_select_analyzer.buildQueryPipeline();
}
else
{
InterpreterSelectWithUnionQuery interpreter_select(query.select, new_context, select_query_options);
pipeline = interpreter_select.buildQueryPipeline();
}
}
else
{
/// Passing 1 as subquery_depth will disable limiting size of intermediate result.
auto select_query_options = SelectQueryOptions(QueryProcessingStage::Complete, 1);
if (settings.allow_experimental_analyzer)
{
InterpreterSelectQueryAnalyzer interpreter_select_analyzer(query.select, getContext(), select_query_options);
pipeline = interpreter_select_analyzer.buildQueryPipeline();
}
else
{
InterpreterSelectWithUnionQuery interpreter_select(query.select, getContext(), select_query_options);
pipeline = interpreter_select.buildQueryPipeline();
}
}
pipeline.dropTotalsAndExtremes();
if (settings.max_insert_threads > 1)
{
auto table_id = table->getStorageID();
auto views = DatabaseCatalog::instance().getDependentViews(table_id);
/// It breaks some views-related tests and we have dedicated `parallel_view_processing` for views, so let's just skip them.
/// Also it doesn't make sense to reshuffle data if storage doesn't support parallel inserts.
const bool resize_to_max_insert_threads = !table->isView() && views.empty() && table->supportsParallelInsert();
pre_streams_size = resize_to_max_insert_threads ? settings.max_insert_threads
: std::min<size_t>(settings.max_insert_threads, pipeline.getNumStreams());
/// Deduplication when passing insert_deduplication_token breaks if using more than one thread
if (!settings.insert_deduplication_token.toString().empty())
{
LOG_DEBUG(
getLogger("InsertQuery"),
"Insert-select query using insert_deduplication_token, setting streams to 1 to avoid deduplication issues");
pre_streams_size = 1;
}
if (table->supportsParallelInsert())
sink_streams_size = pre_streams_size;
}
pipeline.resize(pre_streams_size);
/// Allow to insert Nullable into non-Nullable columns, NULL values will be added as defaults values.
if (getContext()->getSettingsRef().insert_null_as_default)
{
const auto & input_columns = pipeline.getHeader().getColumnsWithTypeAndName();
const auto & query_columns = query_sample_block.getColumnsWithTypeAndName();
const auto & output_columns = metadata_snapshot->getColumns();
if (input_columns.size() == query_columns.size())
{
for (size_t col_idx = 0; col_idx < query_columns.size(); ++col_idx)
{
/// Change query sample block columns to Nullable to allow inserting nullable columns, where NULL values will be substituted with
/// default column values (in AddingDefaultsTransform), so all values will be cast correctly.
if (isNullableOrLowCardinalityNullable(input_columns[col_idx].type)
&& !isNullableOrLowCardinalityNullable(query_columns[col_idx].type)
&& !isVariant(query_columns[col_idx].type)
&& !isDynamic(query_columns[col_idx].type)
&& output_columns.has(query_columns[col_idx].name))
query_sample_block.setColumn(col_idx, ColumnWithTypeAndName(makeNullableOrLowCardinalityNullable(query_columns[col_idx].column), makeNullableOrLowCardinalityNullable(query_columns[col_idx].type), query_columns[col_idx].name));
}
}
}
}
ThreadGroupPtr running_group;
if (current_thread)
running_group = current_thread->getThreadGroup();
if (!running_group)
running_group = std::make_shared<ThreadGroup>(getContext());
for (size_t i = 0; i < sink_streams_size; ++i)
{
auto out = buildSink(table, metadata_snapshot, /* thread_status_holder= */ nullptr,
running_group, /* elapsed_counter_ms= */ nullptr);
sink_chains.emplace_back(std::move(out));
}
for (size_t i = 0; i < pre_streams_size; ++i)
{
auto out = buildPreSinkChain(sink_chains[0].getInputHeader(), table, metadata_snapshot, query_sample_block);
presink_chains.emplace_back(std::move(out));
}
for (const auto & column : metadata_snapshot->getColumns())
if (column.default_desc.kind == ColumnDefaultKind::Materialized && query_sample_block.has(column.name))
throw Exception(ErrorCodes::ILLEGAL_COLUMN, "Cannot insert column {}, because it is MATERIALIZED column.", column.name);
}
BlockIO res;
/// What type of query: INSERT or INSERT SELECT or INSERT WATCH?
if (distributed_pipeline)
if (query.select)
{
res.pipeline = std::move(*distributed_pipeline);
}
else if (query.select)
{
const auto & header = presink_chains.at(0).getInputHeader();
auto actions_dag = ActionsDAG::makeConvertingActions(
pipeline.getHeader().getColumnsWithTypeAndName(),
header.getColumnsWithTypeAndName(),
ActionsDAG::MatchColumnsMode::Position);
auto actions = std::make_shared<ExpressionActions>(actions_dag, ExpressionActionsSettings::fromContext(getContext(), CompileExpressions::yes));
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
if (settings.parallel_distributed_insert_select)
{
return std::make_shared<ExpressionTransform>(in_header, actions);
});
/// We need to convert Sparse columns to full, because it's destination storage
/// may not support it or may have different settings for applying Sparse serialization.
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
{
return std::make_shared<MaterializingTransform>(in_header);
});
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
{
auto context_ptr = getContext();
auto counting = std::make_shared<CountingTransform>(in_header, nullptr, context_ptr->getQuota());
counting->setProcessListElement(context_ptr->getProcessListElement());
counting->setProgressCallback(context_ptr->getProgressCallback());
return counting;
});
if (shouldAddSquashingFroStorage(table))
{
bool table_prefers_large_blocks = table->prefersLargeBlocks();
size_t threads = presink_chains.size();
pipeline.resize(1);
pipeline.addTransform(std::make_shared<PlanSquashingTransform>(
header,
table_prefers_large_blocks ? settings.min_insert_block_size_rows : settings.max_block_size,
table_prefers_large_blocks ? settings.min_insert_block_size_bytes : 0ULL));
pipeline.resize(threads);
pipeline.addSimpleTransform([&](const Block & in_header) -> ProcessorPtr
auto distributed = table->distributedWrite(query, getContext());
if (distributed)
{
return std::make_shared<ApplySquashingTransform>(
in_header,
table_prefers_large_blocks ? settings.min_insert_block_size_rows : settings.max_block_size,
table_prefers_large_blocks ? settings.min_insert_block_size_bytes : 0ULL);
});
res.pipeline = std::move(*distributed);
}
else
{
res.pipeline = buildInsertSelectPipeline(query, table);
}
}
size_t num_select_threads = pipeline.getNumThreads();
for (auto & chain : presink_chains)
resources = chain.detachResources();
for (auto & chain : sink_chains)
resources = chain.detachResources();
pipeline.addChains(std::move(presink_chains));
pipeline.resize(sink_chains.size());
pipeline.addChains(std::move(sink_chains));
if (!settings.parallel_view_processing)
else
{
/// Don't use more threads for INSERT than for SELECT to reduce memory consumption.
if (pipeline.getNumThreads() > num_select_threads)
pipeline.setMaxThreads(num_select_threads);
res.pipeline = buildInsertSelectPipeline(query, table);
}
else if (pipeline.getNumThreads() < settings.max_threads)
{
/// It is possible for query to have max_threads=1, due to optimize_trivial_insert_select,
/// however in case of parallel_view_processing and multiple views, views can still be processed in parallel.
///
/// Note, number of threads will be limited by buildPushingToViewsChain() to max_threads.
pipeline.setMaxThreads(settings.max_threads);
}
pipeline.setSinks([&](const Block & cur_header, QueryPipelineBuilder::StreamType) -> ProcessorPtr
{
return std::make_shared<EmptySink>(cur_header);
});
if (!allow_materialized)
{
for (const auto & column : metadata_snapshot->getColumns())
if (column.default_desc.kind == ColumnDefaultKind::Materialized && header.has(column.name))
throw Exception(ErrorCodes::ILLEGAL_COLUMN, "Cannot insert column {}, because it is MATERIALIZED column.", column.name);
}
res.pipeline = QueryPipelineBuilder::getPipeline(std::move(pipeline));
}
else
{
auto & chain = presink_chains.at(0);
chain.appendChain(std::move(sink_chains.at(0)));
if (shouldAddSquashingFroStorage(table))
{
bool table_prefers_large_blocks = table->prefersLargeBlocks();
auto squashing = std::make_shared<ApplySquashingTransform>(
chain.getInputHeader(),
table_prefers_large_blocks ? settings.min_insert_block_size_rows : settings.max_block_size,
table_prefers_large_blocks ? settings.min_insert_block_size_bytes : 0ULL);
chain.addSource(std::move(squashing));
auto balancing = std::make_shared<PlanSquashingTransform>(
chain.getInputHeader(),
table_prefers_large_blocks ? settings.min_insert_block_size_rows : settings.max_block_size,
table_prefers_large_blocks ? settings.min_insert_block_size_bytes : 0ULL);
chain.addSource(std::move(balancing));
}
auto context_ptr = getContext();
auto counting = std::make_shared<CountingTransform>(chain.getInputHeader(), nullptr, context_ptr->getQuota());
counting->setProcessListElement(context_ptr->getProcessListElement());
counting->setProgressCallback(context_ptr->getProgressCallback());
chain.addSource(std::move(counting));
res.pipeline = QueryPipeline(std::move(presink_chains[0]));
res.pipeline.setNumThreads(std::min<size_t>(res.pipeline.getNumThreads(), settings.max_threads));
res.pipeline.setConcurrencyControl(settings.use_concurrency_control);
if (query.hasInlinedData() && !async_insert)
{
/// can execute without additional data
auto format = getInputFormatFromASTInsertQuery(query_ptr, true, query_sample_block, getContext(), nullptr);
for (auto && buffer : owned_buffers)
format->addBuffer(std::move(buffer));
auto pipe = getSourceFromInputFormat(query_ptr, std::move(format), getContext(), nullptr);
res.pipeline.complete(std::move(pipe));
}
res.pipeline = buildInsertPipeline(query, table);
}
res.pipeline.addResources(std::move(resources));
res.pipeline.addStorageHolder(table);
if (inner_table)
res.pipeline.addStorageHolder(inner_table);
if (const auto * mv = dynamic_cast<const StorageMaterializedView *>(table.get()))
res.pipeline.addStorageHolder(mv->getTargetTable());
return res;
}
@ -757,17 +813,27 @@ void InterpreterInsertQuery::extendQueryLogElemImpl(QueryLogElement & elem, Cont
}
}
void InterpreterInsertQuery::extendQueryLogElemImpl(QueryLogElement & elem, const ASTPtr &, ContextPtr context_) const
{
extendQueryLogElemImpl(elem, context_);
}
void registerInterpreterInsertQuery(InterpreterFactory & factory)
{
auto create_fn = [] (const InterpreterFactory::Arguments & args)
{
return std::make_unique<InterpreterInsertQuery>(args.query, args.context, args.allow_materialized);
return std::make_unique<InterpreterInsertQuery>(
args.query,
args.context,
args.allow_materialized,
/* no_squash */false,
/* no_destination */false,
/* async_insert */false);
};
factory.registerInterpreter("InterpreterInsertQuery", create_fn);
}
}

17
src/Interpreters/InterpreterInsertQuery.h
View File

@ -23,10 +23,10 @@ public:
InterpreterInsertQuery(
const ASTPtr & query_ptr_,
ContextPtr context_,
bool allow_materialized_ = false,
bool no_squash_ = false,
bool no_destination_ = false,
bool async_insert_ = false);
bool allow_materialized_,
bool no_squash_,
bool no_destination,
bool async_insert_);
/** Prepare a request for execution. Return block streams
* - the stream into which you can write data to execute the query, if INSERT;
@ -73,12 +73,17 @@ private:
ASTPtr query_ptr;
const bool allow_materialized;
const bool no_squash;
const bool no_destination;
bool no_squash = false;
bool no_destination = false;
const bool async_insert;
std::vector<std::unique_ptr<ReadBuffer>> owned_buffers;
std::pair<std::vector<Chain>, std::vector<Chain>> buildPreAndSinkChains(size_t presink_streams, size_t sink_streams, StoragePtr table, const StorageMetadataPtr & metadata_snapshot, const Block & query_sample_block);
QueryPipeline buildInsertSelectPipeline(ASTInsertQuery & query, StoragePtr table);
QueryPipeline buildInsertPipeline(ASTInsertQuery & query, StoragePtr table);
Chain buildSink(
const StoragePtr & table,
const StorageMetadataPtr & metadata_snapshot,

124
src/Interpreters/Squashing.cpp
View File

@ -1,6 +1,7 @@
#include <vector>
#include <Interpreters/Squashing.h>
#include <Common/CurrentThread.h>
#include <base/defines.h>
namespace DB
@ -11,24 +12,33 @@ namespace ErrorCodes
}
Squashing::Squashing(Block header_, size_t min_block_size_rows_, size_t min_block_size_bytes_)
: header(header_)
, min_block_size_rows(min_block_size_rows_)
: min_block_size_rows(min_block_size_rows_)
, min_block_size_bytes(min_block_size_bytes_)
, header(header_)
{
}
Chunk Squashing::flush()
{
return convertToChunk(std::move(chunks_to_merge_vec));
if (!accumulated)
return {};
auto result = convertToChunk(accumulated.extract());
chassert(result);
return result;
}
Chunk Squashing::squash(Chunk && input_chunk)
{
if (!input_chunk.hasChunkInfo())
if (!input_chunk)
return Chunk();
const auto *info = getInfoFromChunk(input_chunk);
return squash(info->chunks);
auto squash_info = input_chunk.getChunkInfos().extract<ChunksToSquash>();
if (!squash_info)
throw Exception(ErrorCodes::LOGICAL_ERROR, "There is no ChunksToSquash in ChunkInfoPtr");
return squash(std::move(squash_info->chunks), std::move(input_chunk.getChunkInfos()));
}
Chunk Squashing::add(Chunk && input_chunk)
@ -37,48 +47,37 @@ Chunk Squashing::add(Chunk && input_chunk)
return {};
/// Just read block is already enough.
if (isEnoughSize(input_chunk.getNumRows(), input_chunk.bytes()))
if (isEnoughSize(input_chunk))
{
/// If no accumulated data, return just read block.
if (chunks_to_merge_vec.empty())
if (!accumulated)
{
chunks_to_merge_vec.push_back(std::move(input_chunk));
Chunk res_chunk = convertToChunk(std::move(chunks_to_merge_vec));
chunks_to_merge_vec.clear();
return res_chunk;
accumulated.add(std::move(input_chunk));
return convertToChunk(accumulated.extract());
}
/// Return accumulated data (maybe it has small size) and place new block to accumulated data.
Chunk res_chunk = convertToChunk(std::move(chunks_to_merge_vec));
chunks_to_merge_vec.clear();
changeCurrentSize(input_chunk.getNumRows(), input_chunk.bytes());
chunks_to_merge_vec.push_back(std::move(input_chunk));
Chunk res_chunk = convertToChunk(accumulated.extract());
accumulated.add(std::move(input_chunk));
return res_chunk;
}
/// Accumulated block is already enough.
if (isEnoughSize(accumulated_size.rows, accumulated_size.bytes))
if (isEnoughSize())
{
/// Return accumulated data and place new block to accumulated data.
Chunk res_chunk = convertToChunk(std::move(chunks_to_merge_vec));
chunks_to_merge_vec.clear();
changeCurrentSize(input_chunk.getNumRows(), input_chunk.bytes());
chunks_to_merge_vec.push_back(std::move(input_chunk));
Chunk res_chunk = convertToChunk(accumulated.extract());
accumulated.add(std::move(input_chunk));
return res_chunk;
}
/// Pushing data into accumulating vector
expandCurrentSize(input_chunk.getNumRows(), input_chunk.bytes());
chunks_to_merge_vec.push_back(std::move(input_chunk));
accumulated.add(std::move(input_chunk));
/// If accumulated data is big enough, we send it
if (isEnoughSize(accumulated_size.rows, accumulated_size.bytes))
{
Chunk res_chunk = convertToChunk(std::move(chunks_to_merge_vec));
changeCurrentSize(0, 0);
chunks_to_merge_vec.clear();
return res_chunk;
}
if (isEnoughSize())
return convertToChunk(accumulated.extract());
return {};
}
@ -90,14 +89,15 @@ Chunk Squashing::convertToChunk(std::vector<Chunk> && chunks) const
auto info = std::make_shared<ChunksToSquash>();
info->chunks = std::move(chunks);
chunks.clear();
return Chunk(header.cloneEmptyColumns(), 0, info);
// It is imortant that chunk is not empty, it has to have columns even if they are empty
auto aggr_chunk = Chunk(header.getColumns(), 0);
aggr_chunk.getChunkInfos().add(std::move(info));
chassert(aggr_chunk);
return aggr_chunk;
}
Chunk Squashing::squash(std::vector<Chunk> & input_chunks)
Chunk Squashing::squash(std::vector<Chunk> && input_chunks, Chunk::ChunkInfoCollection && infos)
{
Chunk accumulated_chunk;
std::vector<IColumn::MutablePtr> mutable_columns = {};
size_t rows = 0;
for (const Chunk & chunk : input_chunks)
@ -119,35 +119,17 @@ Chunk Squashing::squash(std::vector<Chunk> & input_chunks)
for (size_t j = 0, size = mutable_columns.size(); j < size; ++j)
{
const auto source_column = columns[j];
mutable_columns[j]->insertRangeFrom(*source_column, 0, source_column->size());
}
}
accumulated_chunk.setColumns(std::move(mutable_columns), rows);
return accumulated_chunk;
}
const ChunksToSquash* Squashing::getInfoFromChunk(const Chunk & chunk)
{
const auto& info = chunk.getChunkInfo();
const auto * agg_info = typeid_cast<const ChunksToSquash *>(info.get());
Chunk result;
result.setColumns(std::move(mutable_columns), rows);
result.setChunkInfos(infos);
result.getChunkInfos().append(std::move(input_chunks.back().getChunkInfos()));
if (!agg_info)
throw Exception(ErrorCodes::LOGICAL_ERROR, "There is no ChunksToSquash in ChunkInfoPtr");
return agg_info;
}
void Squashing::expandCurrentSize(size_t rows, size_t bytes)
{
accumulated_size.rows += rows;
accumulated_size.bytes += bytes;
}
void Squashing::changeCurrentSize(size_t rows, size_t bytes)
{
accumulated_size.rows = rows;
accumulated_size.bytes = bytes;
chassert(result);
return result;
}
bool Squashing::isEnoughSize(size_t rows, size_t bytes) const
@ -156,4 +138,28 @@ bool Squashing::isEnoughSize(size_t rows, size_t bytes) const
|| (min_block_size_rows && rows >= min_block_size_rows)
|| (min_block_size_bytes && bytes >= min_block_size_bytes);
}
bool Squashing::isEnoughSize() const
{
return isEnoughSize(accumulated.getRows(), accumulated.getBytes());
};
bool Squashing::isEnoughSize(const Chunk & chunk) const
{
return isEnoughSize(chunk.getNumRows(), chunk.bytes());
}
void Squashing::CurrentSize::add(Chunk && chunk)
{
rows += chunk.getNumRows();
bytes += chunk.bytes();
chunks.push_back(std::move(chunk));
}
std::vector<Chunk> Squashing::CurrentSize::extract()
{
auto result = std::move(chunks);
*this = {};
return result;
}
}

46
src/Interpreters/Squashing.h
View File

@ -8,9 +8,18 @@
namespace DB
{
struct ChunksToSquash : public ChunkInfo
class ChunksToSquash : public ChunkInfoCloneable<ChunksToSquash>
{
mutable std::vector<Chunk> chunks = {};
public:
ChunksToSquash() = default;
ChunksToSquash(const ChunksToSquash & other)
{
chunks.reserve(other.chunks.size());
for (const auto & chunk: other.chunks)
chunks.push_back(chunk.clone());
}
std::vector<Chunk> chunks = {};
};
/** Merging consecutive passed blocks to specified minimum size.
@ -36,32 +45,35 @@ public:
static Chunk squash(Chunk && input_chunk);
Chunk flush();
bool isDataLeft()
{
return !chunks_to_merge_vec.empty();
}
void setHeader(Block header_) { header = std::move(header_); }
const Block & getHeader() const { return header; }
Block header;
private:
struct CurrentSize
class CurrentSize
{
std::vector<Chunk> chunks = {};
size_t rows = 0;
size_t bytes = 0;
public:
explicit operator bool () const { return !chunks.empty(); }
size_t getRows() const { return rows; }
size_t getBytes() const { return bytes; }
void add(Chunk && chunk);
std::vector<Chunk> extract();
};
std::vector<Chunk> chunks_to_merge_vec = {};
size_t min_block_size_rows;
size_t min_block_size_bytes;
const size_t min_block_size_rows;
const size_t min_block_size_bytes;
Block header;
CurrentSize accumulated_size;
CurrentSize accumulated;
static const ChunksToSquash * getInfoFromChunk(const Chunk & chunk);
static Chunk squash(std::vector<Chunk> && input_chunks, Chunk::ChunkInfoCollection && infos);
static Chunk squash(std::vector<Chunk> & input_chunks);
void expandCurrentSize(size_t rows, size_t bytes);
void changeCurrentSize(size_t rows, size_t bytes);
bool isEnoughSize() const;
bool isEnoughSize(size_t rows, size_t bytes) const;
bool isEnoughSize(const Chunk & chunk) const;
Chunk convertToChunk(std::vector<Chunk> && chunks) const;
};

8
src/Interpreters/SystemLog.cpp
View File

@ -538,7 +538,13 @@ void SystemLog<LogElement>::flushImpl(const std::vector<LogElement> & to_flush,
insert_context->makeQueryContext();
addSettingsForQuery(insert_context, IAST::QueryKind::Insert);
InterpreterInsertQuery interpreter(query_ptr, insert_context);
InterpreterInsertQuery interpreter(
query_ptr,
insert_context,
/* allow_materialized */ false,
/* no_squash */ false,
/* no_destination */ false,
/* async_isnert */ false);
BlockIO io = interpreter.execute();
PushingPipelineExecutor executor(io.pipeline);

2
src/Interpreters/TreeRewriter.cpp
View File

@ -1188,7 +1188,7 @@ bool TreeRewriterResult::collectUsedColumns(const ASTPtr & query, bool is_select
}
}
/// Check for dynamic subcolums in unknown required columns.
/// Check for dynamic subcolumns in unknown required columns.
if (!unknown_required_source_columns.empty())
{
for (const NameAndTypePair & pair : source_columns_ordinary)

20
src/Processors/Chunk.cpp
View File

@ -19,14 +19,6 @@ Chunk::Chunk(DB::Columns columns_, UInt64 num_rows_) : columns(std::move(columns
checkNumRowsIsConsistent();
}
Chunk::Chunk(Columns columns_, UInt64 num_rows_, ChunkInfoPtr chunk_info_)
: columns(std::move(columns_))
, num_rows(num_rows_)
, chunk_info(std::move(chunk_info_))
{
checkNumRowsIsConsistent();
}
static Columns unmuteColumns(MutableColumns && mutable_columns)
{
Columns columns;
@ -43,17 +35,11 @@ Chunk::Chunk(MutableColumns columns_, UInt64 num_rows_)
checkNumRowsIsConsistent();
}
Chunk::Chunk(MutableColumns columns_, UInt64 num_rows_, ChunkInfoPtr chunk_info_)
: columns(unmuteColumns(std::move(columns_)))
, num_rows(num_rows_)
, chunk_info(std::move(chunk_info_))
{
checkNumRowsIsConsistent();
}
Chunk Chunk::clone() const
{
return Chunk(getColumns(), getNumRows(), chunk_info);
auto tmp = Chunk(getColumns(), getNumRows());
tmp.setChunkInfos(chunk_infos.clone());
return tmp;
}
void Chunk::setColumns(Columns columns_, UInt64 num_rows_)

58
src/Processors/Chunk.h
View File

@ -1,7 +1,9 @@
#pragma once
#include <Common/CollectionOfDerived.h>
#include <Columns/IColumn.h>
#include <unordered_map>
#include <memory>
namespace DB
{
@ -9,11 +11,29 @@ namespace DB
class ChunkInfo
{
public:
virtual ~ChunkInfo() = default;
using Ptr = std::shared_ptr<ChunkInfo>;
ChunkInfo() = default;
ChunkInfo(const ChunkInfo&) = default;
ChunkInfo(ChunkInfo&&) = default;
virtual Ptr clone() const = 0;
virtual ~ChunkInfo() = default;
};
using ChunkInfoPtr = std::shared_ptr<const ChunkInfo>;
template<class Derived>
class ChunkInfoCloneable : public ChunkInfo
{
public:
ChunkInfoCloneable() = default;
ChunkInfoCloneable(const ChunkInfoCloneable & other) = default;
Ptr clone() const override
{
return std::static_pointer_cast<ChunkInfo>(std::make_shared<Derived>(*static_cast<const Derived*>(this)));
}
};
/**
* Chunk is a list of columns with the same length.
@ -32,26 +52,26 @@ using ChunkInfoPtr = std::shared_ptr<const ChunkInfo>;
class Chunk
{
public:
using ChunkInfoCollection = CollectionOfDerivedItems<ChunkInfo>;
Chunk() = default;
Chunk(const Chunk & other) = delete;
Chunk(Chunk && other) noexcept
: columns(std::move(other.columns))
, num_rows(other.num_rows)
, chunk_info(std::move(other.chunk_info))
, chunk_infos(std::move(other.chunk_infos))
{
other.num_rows = 0;
}
Chunk(Columns columns_, UInt64 num_rows_);
Chunk(Columns columns_, UInt64 num_rows_, ChunkInfoPtr chunk_info_);
Chunk(MutableColumns columns_, UInt64 num_rows_);
Chunk(MutableColumns columns_, UInt64 num_rows_, ChunkInfoPtr chunk_info_);
Chunk & operator=(const Chunk & other) = delete;
Chunk & operator=(Chunk && other) noexcept
{
columns = std::move(other.columns);
chunk_info = std::move(other.chunk_info);
chunk_infos = std::move(other.chunk_infos);
num_rows = other.num_rows;
other.num_rows = 0;
return *this;
@ -62,15 +82,15 @@ public:
void swap(Chunk & other) noexcept
{
columns.swap(other.columns);
chunk_info.swap(other.chunk_info);
std::swap(num_rows, other.num_rows);
chunk_infos.swap(other.chunk_infos);
}
void clear()
{
num_rows = 0;
columns.clear();
chunk_info.reset();
chunk_infos.clear();
}
const Columns & getColumns() const { return columns; }
@ -81,9 +101,9 @@ public:
/** Get empty columns with the same types as in block. */
MutableColumns cloneEmptyColumns() const;
const ChunkInfoPtr & getChunkInfo() const { return chunk_info; }
bool hasChunkInfo() const { return chunk_info != nullptr; }
void setChunkInfo(ChunkInfoPtr chunk_info_) { chunk_info = std::move(chunk_info_); }
ChunkInfoCollection & getChunkInfos() { return chunk_infos; }
const ChunkInfoCollection & getChunkInfos() const { return chunk_infos; }
void setChunkInfos(ChunkInfoCollection chunk_infos_) { chunk_infos = std::move(chunk_infos_); }
UInt64 getNumRows() const { return num_rows; }
UInt64 getNumColumns() const { return columns.size(); }
@ -107,7 +127,7 @@ public:
private:
Columns columns;
UInt64 num_rows = 0;
ChunkInfoPtr chunk_info;
ChunkInfoCollection chunk_infos;
void checkNumRowsIsConsistent();
};
@ -117,11 +137,15 @@ using Chunks = std::vector<Chunk>;
/// AsyncInsert needs two kinds of information:
/// - offsets of different sub-chunks
/// - tokens of different sub-chunks, which are assigned by setting `insert_deduplication_token`.
class AsyncInsertInfo : public ChunkInfo
class AsyncInsertInfo : public ChunkInfoCloneable<AsyncInsertInfo>
{
public:
AsyncInsertInfo() = default;
explicit AsyncInsertInfo(const std::vector<size_t> & offsets_, const std::vector<String> & tokens_) : offsets(offsets_), tokens(tokens_) {}
AsyncInsertInfo(const AsyncInsertInfo & other) = default;
AsyncInsertInfo(const std::vector<size_t> & offsets_, const std::vector<String> & tokens_)
: offsets(offsets_)
, tokens(tokens_)
{}
std::vector<size_t> offsets;
std::vector<String> tokens;
@ -130,9 +154,11 @@ public:
using AsyncInsertInfoPtr = std::shared_ptr<AsyncInsertInfo>;
/// Extension to support delayed defaults. AddingDefaultsProcessor uses it to replace missing values with column defaults.
class ChunkMissingValues : public ChunkInfo
class ChunkMissingValues : public ChunkInfoCloneable<ChunkMissingValues>
{
public:
ChunkMissingValues(const ChunkMissingValues & other) = default;
using RowsBitMask = std::vector<bool>; /// a bit per row for a column
const RowsBitMask & getDefaultsBitmask(size_t column_idx) const;

9
src/Processors/Executors/PullingAsyncPipelineExecutor.cpp
View File

@ -147,13 +147,10 @@ bool PullingAsyncPipelineExecutor::pull(Block & block, uint64_t milliseconds)
block = lazy_format->getPort(IOutputFormat::PortKind::Main).getHeader().cloneWithColumns(chunk.detachColumns());
if (auto chunk_info = chunk.getChunkInfo())
if (auto agg_info = chunk.getChunkInfos().get<AggregatedChunkInfo>())
{
if (const auto * agg_info = typeid_cast<const AggregatedChunkInfo *>(chunk_info.get()))
{
block.info.bucket_num = agg_info->bucket_num;
block.info.is_overflows = agg_info->is_overflows;
}
block.info.bucket_num = agg_info->bucket_num;
block.info.is_overflows = agg_info->is_overflows;
}
return true;

9
src/Processors/Executors/PullingPipelineExecutor.cpp
View File

@ -73,13 +73,10 @@ bool PullingPipelineExecutor::pull(Block & block)
}
block = pulling_format->getPort(IOutputFormat::PortKind::Main).getHeader().cloneWithColumns(chunk.detachColumns());
if (auto chunk_info = chunk.getChunkInfo())
if (auto agg_info = chunk.getChunkInfos().get<AggregatedChunkInfo>())
{
if (const auto * agg_info = typeid_cast<const AggregatedChunkInfo *>(chunk_info.get()))
{
block.info.bucket_num = agg_info->bucket_num;
block.info.is_overflows = agg_info->is_overflows;
}
block.info.bucket_num = agg_info->bucket_num;
block.info.is_overflows = agg_info->is_overflows;
}
return true;

4
src/Processors/Formats/Impl/ParquetBlockOutputFormat.cpp
View File

@ -179,7 +179,9 @@ void ParquetBlockOutputFormat::consume(Chunk chunk)
columns[i]->insertRangeFrom(*concatenated.getColumns()[i], offset, count);
Chunks piece;
piece.emplace_back(std::move(columns), count, concatenated.getChunkInfo());
piece.emplace_back(std::move(columns), count);
piece.back().setChunkInfos(concatenated.getChunkInfos());
writeRowGroup(std::move(piece));
}
}

5
src/Processors/IAccumulatingTransform.cpp
View File

@ -8,8 +8,9 @@ namespace ErrorCodes
}
IAccumulatingTransform::IAccumulatingTransform(Block input_header, Block output_header)
: IProcessor({std::move(input_header)}, {std::move(output_header)}),
input(inputs.front()), output(outputs.front())
: IProcessor({std::move(input_header)}, {std::move(output_header)})
, input(inputs.front())
, output(outputs.front())
{
}

10
src/Processors/Merges/Algorithms/FinishAggregatingInOrderAlgorithm.cpp
View File

@ -53,13 +53,11 @@ void FinishAggregatingInOrderAlgorithm::consume(Input & input, size_t source_num
if (!input.chunk.hasRows())
return;
const auto & info = input.chunk.getChunkInfo();
if (!info)
if (input.chunk.getChunkInfos().empty())
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk info was not set for chunk in FinishAggregatingInOrderAlgorithm");
Int64 allocated_bytes = 0;
/// Will be set by AggregatingInOrderTransform during local aggregation; will be nullptr during merging on initiator.
if (const auto * arenas_info = typeid_cast<const ChunkInfoWithAllocatedBytes *>(info.get()))
if (auto arenas_info = input.chunk.getChunkInfos().get<ChunkInfoWithAllocatedBytes>())
allocated_bytes = arenas_info->allocated_bytes;
states[source_num] = State{input.chunk, description, allocated_bytes};
@ -136,7 +134,7 @@ Chunk FinishAggregatingInOrderAlgorithm::prepareToMerge()
info->chunk_num = chunk_num++;
Chunk chunk;
chunk.setChunkInfo(std::move(info));
chunk.getChunkInfos().add(std::move(info));
return chunk;
}
@ -163,7 +161,7 @@ void FinishAggregatingInOrderAlgorithm::addToAggregation()
chunks.emplace_back(std::move(new_columns), current_rows);
}
chunks.back().setChunkInfo(std::make_shared<AggregatedChunkInfo>());
chunks.back().getChunkInfos().add(std::make_shared<AggregatedChunkInfo>());
states[i].current_row = states[i].to_row;
/// We assume that sizes in bytes of rows are almost the same.

12
src/Processors/Merges/Algorithms/MergeTreePartLevelInfo.h
View File

@ -6,18 +6,22 @@ namespace DB
{
/// To carry part level if chunk is produced by a merge tree source
class MergeTreePartLevelInfo : public ChunkInfo
class MergeTreePartLevelInfo : public ChunkInfoCloneable<MergeTreePartLevelInfo>
{
public:
MergeTreePartLevelInfo() = delete;
explicit MergeTreePartLevelInfo(ssize_t part_level) : origin_merge_tree_part_level(part_level) { }
explicit MergeTreePartLevelInfo(ssize_t part_level)
: origin_merge_tree_part_level(part_level)
{ }
MergeTreePartLevelInfo(const MergeTreePartLevelInfo & other) = default;
size_t origin_merge_tree_part_level = 0;
};
inline size_t getPartLevelFromChunk(const Chunk & chunk)
{
const auto & info = chunk.getChunkInfo();
if (const auto * part_level_info = typeid_cast<const MergeTreePartLevelInfo *>(info.get()))
const auto part_level_info = chunk.getChunkInfos().get<MergeTreePartLevelInfo>();
if (part_level_info)
return part_level_info->origin_merge_tree_part_level;
return 0;
}

2
src/Processors/Merges/Algorithms/ReplacingSortedAlgorithm.cpp
View File

@ -17,7 +17,7 @@ namespace ErrorCodes
static IMergingAlgorithm::Status emitChunk(detail::SharedChunkPtr & chunk, bool finished = false)
{
chunk->setChunkInfo(std::make_shared<ChunkSelectFinalIndices>(std::move(chunk->replace_final_selection)));
chunk->getChunkInfos().add(std::make_shared<ChunkSelectFinalIndices>(std::move(chunk->replace_final_selection)));
return IMergingAlgorithm::Status(std::move(*chunk), finished);
}

7
src/Processors/Merges/Algorithms/ReplacingSortedAlgorithm.h
View File

@ -3,6 +3,7 @@
#include <Processors/Merges/Algorithms/MergedData.h>
#include <Processors/Transforms/ColumnGathererTransform.h>
#include <Processors/Merges/Algorithms/RowRef.h>
#include <Processors/Chunk.h>
namespace Poco
{
@ -14,11 +15,13 @@ namespace DB
/** Use in skipping final to keep list of indices of selected row after merging final
*/
struct ChunkSelectFinalIndices : public ChunkInfo
struct ChunkSelectFinalIndices : public ChunkInfoCloneable<ChunkSelectFinalIndices>
{
explicit ChunkSelectFinalIndices(MutableColumnPtr select_final_indices_);
ChunkSelectFinalIndices(const ChunkSelectFinalIndices & other) = default;
const ColumnPtr column_holder;
const ColumnUInt64 * select_final_indices = nullptr;
explicit ChunkSelectFinalIndices(MutableColumnPtr select_final_indices_);
};
/** Merges several sorted inputs into one.

2
src/Processors/Merges/IMergingTransform.cpp
View File

@ -157,7 +157,7 @@ IProcessor::Status IMergingTransformBase::prepare()
bool is_port_full = !output.canPush();
/// Push if has data.
if ((state.output_chunk || state.output_chunk.hasChunkInfo()) && !is_port_full)
if ((state.output_chunk || !state.output_chunk.getChunkInfos().empty()) && !is_port_full)
output.push(std::move(state.output_chunk));
if (!is_initialized)

2
src/Processors/Merges/IMergingTransform.h
View File

@ -129,7 +129,7 @@ public:
IMergingAlgorithm::Status status = algorithm.merge();
if ((status.chunk && status.chunk.hasRows()) || status.chunk.hasChunkInfo())
if ((status.chunk && status.chunk.hasRows()) || !status.chunk.getChunkInfos().empty())
{
// std::cerr << "Got chunk with " << status.chunk.getNumRows() << " rows" << std::endl;
state.output_chunk = std::move(status.chunk);

2
src/Processors/Sinks/RemoteSink.h
View File

@ -20,7 +20,7 @@ public:
}
String getName() const override { return "RemoteSink"; }
void consume (Chunk chunk) override { write(RemoteInserter::getHeader().cloneWithColumns(chunk.detachColumns())); }
void consume (Chunk & chunk) override { write(RemoteInserter::getHeader().cloneWithColumns(chunk.getColumns())); }
void onFinish() override { RemoteInserter::onFinish(); }
};

5
src/Processors/Sinks/SinkToStorage.cpp
View File

@ -15,9 +15,8 @@ void SinkToStorage::onConsume(Chunk chunk)
*/
Nested::validateArraySizes(getHeader().cloneWithColumns(chunk.getColumns()));
consume(chunk.clone());
if (!lastBlockIsDuplicate())
cur_chunk = std::move(chunk);
consume(chunk);
cur_chunk = std::move(chunk);
}
SinkToStorage::GenerateResult SinkToStorage::onGenerate()

5
src/Processors/Sinks/SinkToStorage.h
View File

@ -18,8 +18,7 @@ public:
void addTableLock(const TableLockHolder & lock) { table_locks.push_back(lock); }
protected:
virtual void consume(Chunk chunk) = 0;
virtual bool lastBlockIsDuplicate() const { return false; }
virtual void consume(Chunk & chunk) = 0;
private:
std::vector<TableLockHolder> table_locks;
@ -38,7 +37,7 @@ class NullSinkToStorage : public SinkToStorage
public:
using SinkToStorage::SinkToStorage;
std::string getName() const override { return "NullSinkToStorage"; }
void consume(Chunk) override {}
void consume(Chunk &) override {}
};
using SinkPtr = std::shared_ptr<SinkToStorage>;

5
src/Processors/Sources/BlocksSource.h
View File

@ -43,7 +43,10 @@ protected:
info->bucket_num = res.info.bucket_num;
info->is_overflows = res.info.is_overflows;
return Chunk(res.getColumns(), res.rows(), std::move(info));
auto chunk = Chunk(res.getColumns(), res.rows());
chunk.getChunkInfos().add(std::move(info));
return chunk;
}
private:

2
src/Processors/Sources/RemoteSource.cpp
View File

@ -176,7 +176,7 @@ std::optional<Chunk> RemoteSource::tryGenerate()
auto info = std::make_shared<AggregatedChunkInfo>();
info->bucket_num = block.info.bucket_num;
info->is_overflows = block.info.is_overflows;
chunk.setChunkInfo(std::move(info));
chunk.getChunkInfos().add(std::move(info));
}
return chunk;

6
src/Processors/Sources/SourceFromSingleChunk.cpp
View File

@ -5,7 +5,9 @@
namespace DB
{
SourceFromSingleChunk::SourceFromSingleChunk(Block header, Chunk chunk_) : ISource(std::move(header)), chunk(std::move(chunk_)) {}
SourceFromSingleChunk::SourceFromSingleChunk(Block header, Chunk chunk_) : ISource(std::move(header)), chunk(std::move(chunk_))
{
}
SourceFromSingleChunk::SourceFromSingleChunk(Block data) : ISource(data.cloneEmpty()), chunk(data.getColumns(), data.rows())
{
@ -20,7 +22,7 @@ SourceFromSingleChunk::SourceFromSingleChunk(Block data) : ISource(data.cloneEmp
auto info = std::make_shared<AggregatedChunkInfo>();
info->bucket_num = data.info.bucket_num;
info->is_overflows = data.info.is_overflows;
chunk.setChunkInfo(std::move(info));
chunk.getChunkInfos().add(std::move(info));
}
}

11
src/Processors/Transforms/AggregatingInOrderTransform.cpp
View File

@ -332,7 +332,7 @@ void AggregatingInOrderTransform::generate()
variants.aggregates_pool = variants.aggregates_pools.at(0).get();
/// Pass info about used memory by aggregate functions further.
to_push_chunk.setChunkInfo(std::make_shared<ChunkInfoWithAllocatedBytes>(cur_block_bytes));
to_push_chunk.getChunkInfos().add(std::make_shared<ChunkInfoWithAllocatedBytes>(cur_block_bytes));
cur_block_bytes = 0;
cur_block_size = 0;
@ -351,11 +351,12 @@ FinalizeAggregatedTransform::FinalizeAggregatedTransform(Block header, Aggregati
void FinalizeAggregatedTransform::transform(Chunk & chunk)
{
if (params->final)
finalizeChunk(chunk, aggregates_mask);
else if (!chunk.getChunkInfo())
{
auto info = std::make_shared<AggregatedChunkInfo>();
chunk.setChunkInfo(std::move(info));
finalizeChunk(chunk, aggregates_mask);
}
else if (!chunk.getChunkInfos().get<AggregatedChunkInfo>())
{
chunk.getChunkInfos().add(std::make_shared<AggregatedChunkInfo>());
}
}

5
src/Processors/Transforms/AggregatingInOrderTransform.h
View File

@ -5,6 +5,7 @@
#include <Processors/ISimpleTransform.h>
#include <Processors/Transforms/AggregatingTransform.h>
#include <Processors/Transforms/finalizeChunk.h>
#include <Processors/Chunk.h>
namespace DB
{
@ -12,10 +13,12 @@ namespace DB
struct InputOrderInfo;
using InputOrderInfoPtr = std::shared_ptr<const InputOrderInfo>;
struct ChunkInfoWithAllocatedBytes : public ChunkInfo
struct ChunkInfoWithAllocatedBytes : public ChunkInfoCloneable<ChunkInfoWithAllocatedBytes>
{
ChunkInfoWithAllocatedBytes(const ChunkInfoWithAllocatedBytes & other) = default;
explicit ChunkInfoWithAllocatedBytes(Int64 allocated_bytes_)
: allocated_bytes(allocated_bytes_) {}
Int64 allocated_bytes;
};

16
src/Processors/Transforms/AggregatingTransform.cpp
View File

@ -35,7 +35,7 @@ Chunk convertToChunk(const Block & block)
UInt64 num_rows = block.rows();
Chunk chunk(block.getColumns(), num_rows);
chunk.setChunkInfo(std::move(info));
chunk.getChunkInfos().add(std::move(info));
return chunk;
}
@ -44,15 +44,11 @@ namespace
{
const AggregatedChunkInfo * getInfoFromChunk(const Chunk & chunk)
{
const auto & info = chunk.getChunkInfo();
if (!info)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk info was not set for chunk.");
const auto * agg_info = typeid_cast<const AggregatedChunkInfo *>(info.get());
auto agg_info = chunk.getChunkInfos().get<AggregatedChunkInfo>();
if (!agg_info)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk should have AggregatedChunkInfo.");
return agg_info;
return agg_info.get();
}
/// Reads chunks from file in native format. Provide chunks with aggregation info.
@ -210,11 +206,7 @@ private:
void process(Chunk && chunk)
{
if (!chunk.hasChunkInfo())
throw Exception(ErrorCodes::LOGICAL_ERROR, "Expected chunk with chunk info in {}", getName());
const auto & info = chunk.getChunkInfo();
const auto * chunks_to_merge = typeid_cast<const ChunksToMerge *>(info.get());
auto chunks_to_merge = chunk.getChunkInfos().get<ChunksToMerge>();
if (!chunks_to_merge)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Expected chunk with ChunksToMerge info in {}", getName());

3
src/Processors/Transforms/AggregatingTransform.h
View File

@ -2,6 +2,7 @@
#include <Compression/CompressedReadBuffer.h>
#include <IO/ReadBufferFromFile.h>
#include <Interpreters/Aggregator.h>
#include <Processors/Chunk.h>
#include <Processors/IAccumulatingTransform.h>
#include <Common/Stopwatch.h>
#include <Common/setThreadName.h>
@ -19,7 +20,7 @@ namespace CurrentMetrics
namespace DB
{
class AggregatedChunkInfo : public ChunkInfo
class AggregatedChunkInfo : public ChunkInfoCloneable<AggregatedChunkInfo>
{
public:
bool is_overflows = false;

14
src/Processors/Transforms/ApplySquashingTransform.h
View File

@ -27,18 +27,12 @@ public:
}
ExceptionKeepingTransform::work();
if (finish_chunk)
{
data.chunk = std::move(finish_chunk);
ready_output = true;
}
}
protected:
void onConsume(Chunk chunk) override
{
if (auto res_chunk = DB::Squashing::squash(std::move(chunk)))
cur_chunk.setColumns(res_chunk.getColumns(), res_chunk.getNumRows());
cur_chunk = Squashing::squash(std::move(chunk));
}
GenerateResult onGenerate() override
@ -48,16 +42,10 @@ protected:
res.is_done = true;
return res;
}
void onFinish() override
{
auto chunk = DB::Squashing::squash({});
finish_chunk.setColumns(chunk.getColumns(), chunk.getNumRows());
}
private:
Squashing squashing;
Chunk cur_chunk;
Chunk finish_chunk;
};
}

5
src/Processors/Transforms/CountingTransform.cpp
View File

@ -1,6 +1,7 @@
#include <Interpreters/ProcessList.h>
#include <Processors/Transforms/CountingTransform.h>
#include <IO/Progress.h>
#include <Interpreters/ProcessList.h>
#include <Common/ProfileEvents.h>
#include <Common/ThreadStatus.h>

236
src/Processors/Transforms/DeduplicationTokenTransforms.cpp Normal file
View File

@ -0,0 +1,236 @@
#include <Processors/Transforms/DeduplicationTokenTransforms.h>
#include <IO/WriteHelpers.h>
#include <Common/logger_useful.h>
#include <Common/Exception.h>
#include <Common/SipHash.h>
#include <fmt/core.h>
namespace DB
{
namespace ErrorCodes
{
extern const int LOGICAL_ERROR;
}
void RestoreChunkInfosTransform::transform(Chunk & chunk)
{
chunk.getChunkInfos().append(chunk_infos.clone());
}
namespace DeduplicationToken
{
String TokenInfo::getToken() const
{
if (!isDefined())
throw Exception(ErrorCodes::LOGICAL_ERROR, "token is not defined, stage {}, token {}", stage, debugToken());
return getTokenImpl();
}
String TokenInfo::getTokenImpl() const
{
String result;
result.reserve(getTotalSize());
for (const auto & part : parts)
{
if (!result.empty())
result.append(":");
result.append(part);
}
return result;
}
String TokenInfo::debugToken() const
{
return getTokenImpl();
}
void TokenInfo::addChunkHash(String part)
{
if (stage == UNDEFINED && empty())
stage = DEFINE_SOURCE_WITH_HASHES;
if (stage != DEFINE_SOURCE_WITH_HASHES)
throw Exception(ErrorCodes::LOGICAL_ERROR, "token is in wrong stage {}, token {}", stage, debugToken());
addTokenPart(std::move(part));
}
void TokenInfo::finishChunkHashes()
{
if (stage == UNDEFINED && empty())
stage = DEFINE_SOURCE_WITH_HASHES;
if (stage != DEFINE_SOURCE_WITH_HASHES)
throw Exception(ErrorCodes::LOGICAL_ERROR, "token is in wrong stage {}, token {}", stage, debugToken());
stage = DEFINED;
}
void TokenInfo::setUserToken(const String & token)
{
if (stage == UNDEFINED && empty())
stage = DEFINE_SOURCE_USER_TOKEN;
if (stage != DEFINE_SOURCE_USER_TOKEN)
throw Exception(ErrorCodes::LOGICAL_ERROR, "token is in wrong stage {}, token {}", stage, debugToken());
addTokenPart(fmt::format("user-token-{}", token));
}
void TokenInfo::setSourceWithUserToken(size_t block_number)
{
if (stage != DEFINE_SOURCE_USER_TOKEN)
throw Exception(ErrorCodes::LOGICAL_ERROR, "token is in wrong stage {}, token {}", stage, debugToken());
addTokenPart(fmt::format("source-number-{}", block_number));
stage = DEFINED;
}
void TokenInfo::setViewID(const String & id)
{
if (stage == DEFINED)
stage = DEFINE_VIEW;
if (stage != DEFINE_VIEW)
throw Exception(ErrorCodes::LOGICAL_ERROR, "token is in wrong stage {}, token {}", stage, debugToken());
addTokenPart(fmt::format("view-id-{}", id));
}
void TokenInfo::setViewBlockNumber(size_t block_number)
{
if (stage != DEFINE_VIEW)
throw Exception(ErrorCodes::LOGICAL_ERROR, "token is in wrong stage {}, token {}", stage, debugToken());
addTokenPart(fmt::format("view-block-{}", block_number));
stage = DEFINED;
}
void TokenInfo::reset()
{
stage = UNDEFINED;
parts.clear();
}
void TokenInfo::addTokenPart(String part)
{
parts.push_back(std::move(part));
}
size_t TokenInfo::getTotalSize() const
{
if (parts.empty())
return 0;
size_t size = 0;
for (const auto & part : parts)
size += part.size();
// we reserve more size here to be able to add delimenter between parts.
return size + parts.size() - 1;
}
#ifdef ABORT_ON_LOGICAL_ERROR
void CheckTokenTransform::transform(Chunk & chunk)
{
auto token_info = chunk.getChunkInfos().get<TokenInfo>();
if (!token_info)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk has to have DedupTokenInfo as ChunkInfo, {}", debug);
LOG_DEBUG(log, "debug: {}, token: {}", debug, token_info->debugToken());
}
#endif
String DefineSourceWithChunkHashTransform::getChunkHash(const Chunk & chunk)
{
SipHash hash;
for (const auto & colunm : chunk.getColumns())
colunm->updateHashFast(hash);
const auto hash_value = hash.get128();
return toString(hash_value.items[0]) + "_" + toString(hash_value.items[1]);
}
void DefineSourceWithChunkHashTransform::transform(Chunk & chunk)
{
auto token_info = chunk.getChunkInfos().get<TokenInfo>();
if (!token_info)
throw Exception(
ErrorCodes::LOGICAL_ERROR,
"TokenInfo is expected for consumed chunk in DefineSourceWithChunkHashesTransform");
if (token_info->isDefined())
return;
token_info->addChunkHash(getChunkHash(chunk));
token_info->finishChunkHashes();
}
void SetUserTokenTransform::transform(Chunk & chunk)
{
auto token_info = chunk.getChunkInfos().get<TokenInfo>();
if (!token_info)
throw Exception(
ErrorCodes::LOGICAL_ERROR,
"TokenInfo is expected for consumed chunk in SetUserTokenTransform");
token_info->setUserToken(user_token);
}
void SetSourceBlockNumberTransform::transform(Chunk & chunk)
{
auto token_info = chunk.getChunkInfos().get<TokenInfo>();
if (!token_info)
throw Exception(
ErrorCodes::LOGICAL_ERROR,
"TokenInfo is expected for consumed chunk in SetSourceBlockNumberTransform");
token_info->setSourceWithUserToken(block_number++);
}
void SetViewIDTransform::transform(Chunk & chunk)
{
auto token_info = chunk.getChunkInfos().get<TokenInfo>();
if (!token_info)
throw Exception(
ErrorCodes::LOGICAL_ERROR,
"TokenInfo is expected for consumed chunk in SetViewIDTransform");
token_info->setViewID(view_id);
}
void SetViewBlockNumberTransform::transform(Chunk & chunk)
{
auto token_info = chunk.getChunkInfos().get<TokenInfo>();
if (!token_info)
throw Exception(
ErrorCodes::LOGICAL_ERROR,
"TokenInfo is expected for consumed chunk in SetViewBlockNumberTransform");
token_info->setViewBlockNumber(block_number++);
}
void ResetTokenTransform::transform(Chunk & chunk)
{
auto token_info = chunk.getChunkInfos().get<TokenInfo>();
if (!token_info)
throw Exception(
ErrorCodes::LOGICAL_ERROR,
"TokenInfo is expected for consumed chunk in ResetTokenTransform");
token_info->reset();
}
}
}

237
src/Processors/Transforms/DeduplicationTokenTransforms.h Normal file
View File

@ -0,0 +1,237 @@
#pragma once
#include <Processors/Chunk.h>
#include <Processors/ISimpleTransform.h>
#include <base/defines.h>
#include "Common/Logger.h"
namespace DB
{
class RestoreChunkInfosTransform : public ISimpleTransform
{
public:
RestoreChunkInfosTransform(Chunk::ChunkInfoCollection chunk_infos_, const Block & header_)
: ISimpleTransform(header_, header_, true)
, chunk_infos(std::move(chunk_infos_))
{}
String getName() const override { return "RestoreChunkInfosTransform"; }
void transform(Chunk & chunk) override;
private:
Chunk::ChunkInfoCollection chunk_infos;
};
namespace DeduplicationToken
{
class TokenInfo : public ChunkInfoCloneable<TokenInfo>
{
public:
TokenInfo() = default;
TokenInfo(const TokenInfo & other) = default;
String getToken() const;
String debugToken() const;
bool empty() const { return parts.empty(); }
bool isDefined() const { return stage == DEFINED; }
void addChunkHash(String part);
void finishChunkHashes();
void setUserToken(const String & token);
void setSourceWithUserToken(size_t block_number);
void setViewID(const String & id);
void setViewBlockNumber(size_t block_number);
void reset();
private:
String getTokenImpl() const;
void addTokenPart(String part);
size_t getTotalSize() const;
/* Token has to be prepared in a particular order.
* BuildingStage ensures that token is expanded according the following order.
* Firstly token is expanded with information about the source.
* It could be done with two ways: add several hash sums from the source chunks or provide user defined deduplication token and its sequentional block number.
*
* transition // method
* UNDEFINED -> DEFINE_SOURCE_WITH_HASHES // addChunkHash
* DEFINE_SOURCE_WITH_HASHES -> DEFINE_SOURCE_WITH_HASHES // addChunkHash
* DEFINE_SOURCE_WITH_HASHES -> DEFINED // defineSourceWithChankHashes
*
* transition // method
* UNDEFINED -> DEFINE_SOURCE_USER_TOKEN // setUserToken
* DEFINE_SOURCE_USER_TOKEN -> DEFINED // defineSourceWithUserToken
*
* After token is defined, it could be extended with view id and view block number. Actually it has to be expanded with view details if there is one or several views.
*
* transition // method
* DEFINED -> DEFINE_VIEW // setViewID
* DEFINE_VIEW -> DEFINED // defineViewID
*/
enum BuildingStage
{
UNDEFINED,
DEFINE_SOURCE_WITH_HASHES,
DEFINE_SOURCE_USER_TOKEN,
DEFINE_VIEW,
DEFINED,
};
BuildingStage stage = UNDEFINED;
std::vector<String> parts;
};
#ifdef ABORT_ON_LOGICAL_ERROR
/// use that class only with debug builds in CI for introspection
class CheckTokenTransform : public ISimpleTransform
{
public:
CheckTokenTransform(String debug_, const Block & header_)
: ISimpleTransform(header_, header_, true)
, debug(std::move(debug_))
{
}
String getName() const override { return "DeduplicationToken::CheckTokenTransform"; }
void transform(Chunk & chunk) override;
private:
String debug;
LoggerPtr log = getLogger("CheckInsertDeduplicationTokenTransform");
};
#endif
class AddTokenInfoTransform : public ISimpleTransform
{
public:
explicit AddTokenInfoTransform(const Block & header_)
: ISimpleTransform(header_, header_, true)
{
}
String getName() const override { return "DeduplicationToken::AddTokenInfoTransform"; }
void transform(Chunk & chunk) override
{
chunk.getChunkInfos().add(std::make_shared<TokenInfo>());
}
};
class DefineSourceWithChunkHashTransform : public ISimpleTransform
{
public:
explicit DefineSourceWithChunkHashTransform(const Block & header_)
: ISimpleTransform(header_, header_, true)
{
}
String getName() const override { return "DeduplicationToken::DefineSourceWithChunkHashesTransform"; }
// Usually MergeTreeSink/ReplicatedMergeTreeSink calls addChunkHash for the deduplication token with hashes from the parts.
// But if there is some table with different engine, we still need to define the source of the data in deduplication token
// We use that transform to define the source as a hash of entire block in deduplication token
void transform(Chunk & chunk) override;
static String getChunkHash(const Chunk & chunk);
};
class ResetTokenTransform : public ISimpleTransform
{
public:
explicit ResetTokenTransform(const Block & header_)
: ISimpleTransform(header_, header_, true)
{
}
String getName() const override { return "DeduplicationToken::ResetTokenTransform"; }
void transform(Chunk & chunk) override;
};
class SetUserTokenTransform : public ISimpleTransform
{
public:
SetUserTokenTransform(String user_token_, const Block & header_)
: ISimpleTransform(header_, header_, true)
, user_token(std::move(user_token_))
{
}
String getName() const override { return "DeduplicationToken::SetUserTokenTransform"; }
void transform(Chunk & chunk) override;
private:
String user_token;
};
class SetSourceBlockNumberTransform : public ISimpleTransform
{
public:
explicit SetSourceBlockNumberTransform(const Block & header_)
: ISimpleTransform(header_, header_, true)
{
}
String getName() const override { return "DeduplicationToken::SetSourceBlockNumberTransform"; }
void transform(Chunk & chunk) override;
private:
size_t block_number = 0;
};
class SetViewIDTransform : public ISimpleTransform
{
public:
SetViewIDTransform(String view_id_, const Block & header_)
: ISimpleTransform(header_, header_, true)
, view_id(std::move(view_id_))
{
}
String getName() const override { return "DeduplicationToken::SetViewIDTransform"; }
void transform(Chunk & chunk) override;
private:
String view_id;
};
class SetViewBlockNumberTransform : public ISimpleTransform
{
public:
explicit SetViewBlockNumberTransform(const Block & header_)
: ISimpleTransform(header_, header_, true)
{
}
String getName() const override { return "DeduplicationToken::SetViewBlockNumberTransform"; }
void transform(Chunk & chunk) override;
private:
size_t block_number = 0;
};
}
}

2
src/Processors/Transforms/ExpressionTransform.cpp
View File

@ -1,5 +1,7 @@
#include <Processors/Transforms/ExpressionTransform.h>
#include <Interpreters/ExpressionActions.h>
namespace DB
{

9
src/Processors/Transforms/JoiningTransform.cpp
View File

@ -365,10 +365,9 @@ IProcessor::Status DelayedJoinedBlocksWorkerTransform::prepare()
return Status::Finished;
}
if (!data.chunk.hasChunkInfo())
task = data.chunk.getChunkInfos().get<DelayedBlocksTask>();
if (!task)
throw Exception(ErrorCodes::LOGICAL_ERROR, "DelayedJoinedBlocksWorkerTransform must have chunk info");
task = std::dynamic_pointer_cast<const DelayedBlocksTask>(data.chunk.getChunkInfo());
}
else
{
@ -479,7 +478,7 @@ IProcessor::Status DelayedJoinedBlocksTransform::prepare()
if (output.isFinished())
continue;
Chunk chunk;
chunk.setChunkInfo(std::make_shared<DelayedBlocksTask>());
chunk.getChunkInfos().add(std::make_shared<DelayedBlocksTask>());
output.push(std::move(chunk));
output.finish();
}
@ -496,7 +495,7 @@ IProcessor::Status DelayedJoinedBlocksTransform::prepare()
{
Chunk chunk;
auto task = std::make_shared<DelayedBlocksTask>(delayed_blocks, left_delayed_stream_finished_counter);
chunk.setChunkInfo(task);
chunk.getChunkInfos().add(std::move(task));
output.push(std::move(chunk));
}
delayed_blocks = nullptr;

6
src/Processors/Transforms/JoiningTransform.h
View File

@ -1,6 +1,7 @@
#pragma once
#include <Processors/IProcessor.h>
#include <Processors/Chunk.h>
#include <memory>
namespace DB
{
@ -111,11 +112,12 @@ private:
};
class DelayedBlocksTask : public ChunkInfo
class DelayedBlocksTask : public ChunkInfoCloneable<DelayedBlocksTask>
{
public:
DelayedBlocksTask() = default;
DelayedBlocksTask(const DelayedBlocksTask & other) = default;
explicit DelayedBlocksTask(IBlocksStreamPtr delayed_blocks_, JoiningTransform::FinishCounterPtr left_delayed_stream_finish_counter_)
: delayed_blocks(std::move(delayed_blocks_))
, left_delayed_stream_finish_counter(left_delayed_stream_finish_counter_)

1
src/Processors/Transforms/MaterializingTransform.cpp
View File

@ -1,6 +1,7 @@
#include <Processors/Transforms/MaterializingTransform.h>
#include <Columns/ColumnSparse.h>
namespace DB
{

6
src/Processors/Transforms/MemoryBoundMerging.h
View File

@ -150,11 +150,7 @@ private:
if (!chunk.hasRows())
return;
const auto & info = chunk.getChunkInfo();
if (!info)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk info was not set for chunk in SortingAggregatedForMemoryBoundMergingTransform.");
const auto * agg_info = typeid_cast<const AggregatedChunkInfo *>(info.get());
const auto & agg_info = chunk.getChunkInfos().get<AggregatedChunkInfo>();
if (!agg_info)
throw Exception(
ErrorCodes::LOGICAL_ERROR, "Chunk should have AggregatedChunkInfo in SortingAggregatedForMemoryBoundMergingTransform.");

36
src/Processors/Transforms/MergingAggregatedMemoryEfficientTransform.cpp
View File

@ -30,10 +30,10 @@ void GroupingAggregatedTransform::pushData(Chunks chunks, Int32 bucket, bool is_
auto info = std::make_shared<ChunksToMerge>();
info->bucket_num = bucket;
info->is_overflows = is_overflows;
info->chunks = std::make_unique<Chunks>(std::move(chunks));
info->chunks = std::make_shared<Chunks>(std::move(chunks));
Chunk chunk;
chunk.setChunkInfo(std::move(info));
chunk.getChunkInfos().add(std::move(info));
output.push(std::move(chunk));
}
@ -255,11 +255,10 @@ void GroupingAggregatedTransform::addChunk(Chunk chunk, size_t input)
if (!chunk.hasRows())
return;
const auto & info = chunk.getChunkInfo();
if (!info)
if (chunk.getChunkInfos().empty())
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk info was not set for chunk in GroupingAggregatedTransform.");
if (const auto * agg_info = typeid_cast<const AggregatedChunkInfo *>(info.get()))
if (auto agg_info = chunk.getChunkInfos().get<AggregatedChunkInfo>())
{
Int32 bucket = agg_info->bucket_num;
bool is_overflows = agg_info->is_overflows;
@ -275,7 +274,7 @@ void GroupingAggregatedTransform::addChunk(Chunk chunk, size_t input)
last_bucket_number[input] = bucket;
}
}
else if (typeid_cast<const ChunkInfoWithAllocatedBytes *>(info.get()))
else if (chunk.getChunkInfos().get<ChunkInfoWithAllocatedBytes>())
{
single_level_chunks.emplace_back(std::move(chunk));
}
@ -304,7 +303,11 @@ void GroupingAggregatedTransform::work()
Int32 bucket = cur_block.info.bucket_num;
auto chunk_info = std::make_shared<AggregatedChunkInfo>();
chunk_info->bucket_num = bucket;
chunks_map[bucket].emplace_back(Chunk(cur_block.getColumns(), cur_block.rows(), std::move(chunk_info)));
auto chunk = Chunk(cur_block.getColumns(), cur_block.rows());
chunk.getChunkInfos().add(std::move(chunk_info));
chunks_map[bucket].emplace_back(std::move(chunk));
}
}
}
@ -319,9 +322,7 @@ MergingAggregatedBucketTransform::MergingAggregatedBucketTransform(
void MergingAggregatedBucketTransform::transform(Chunk & chunk)
{
const auto & info = chunk.getChunkInfo();
const auto * chunks_to_merge = typeid_cast<const ChunksToMerge *>(info.get());
auto chunks_to_merge = chunk.getChunkInfos().get<ChunksToMerge>();
if (!chunks_to_merge)
throw Exception(ErrorCodes::LOGICAL_ERROR, "MergingAggregatedSimpleTransform chunk must have ChunkInfo with type ChunksToMerge.");
@ -330,11 +331,10 @@ void MergingAggregatedBucketTransform::transform(Chunk & chunk)
BlocksList blocks_list;
for (auto & cur_chunk : *chunks_to_merge->chunks)
{
const auto & cur_info = cur_chunk.getChunkInfo();
if (!cur_info)
if (cur_chunk.getChunkInfos().empty())
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk info was not set for chunk in MergingAggregatedBucketTransform.");
if (const auto * agg_info = typeid_cast<const AggregatedChunkInfo *>(cur_info.get()))
if (auto agg_info = cur_chunk.getChunkInfos().get<AggregatedChunkInfo>())
{
Block block = header.cloneWithColumns(cur_chunk.detachColumns());
block.info.is_overflows = agg_info->is_overflows;
@ -342,7 +342,7 @@ void MergingAggregatedBucketTransform::transform(Chunk & chunk)
blocks_list.emplace_back(std::move(block));
}
else if (typeid_cast<const ChunkInfoWithAllocatedBytes *>(cur_info.get()))
else if (cur_chunk.getChunkInfos().get<ChunkInfoWithAllocatedBytes>())
{
Block block = header.cloneWithColumns(cur_chunk.detachColumns());
block.info.is_overflows = false;
@ -361,7 +361,7 @@ void MergingAggregatedBucketTransform::transform(Chunk & chunk)
res_info->is_overflows = chunks_to_merge->is_overflows;
res_info->bucket_num = chunks_to_merge->bucket_num;
res_info->chunk_num = chunks_to_merge->chunk_num;
chunk.setChunkInfo(std::move(res_info));
chunk.getChunkInfos().add(std::move(res_info));
auto block = params->aggregator.mergeBlocks(blocks_list, params->final, is_cancelled);
@ -405,11 +405,7 @@ bool SortingAggregatedTransform::tryPushChunk()
void SortingAggregatedTransform::addChunk(Chunk chunk, size_t from_input)
{
const auto & info = chunk.getChunkInfo();
if (!info)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk info was not set for chunk in SortingAggregatedTransform.");
const auto * agg_info = typeid_cast<const AggregatedChunkInfo *>(info.get());
auto agg_info = chunk.getChunkInfos().get<AggregatedChunkInfo>();
if (!agg_info)
throw Exception(ErrorCodes::LOGICAL_ERROR,
"Chunk should have AggregatedChunkInfo in SortingAggregatedTransform.");

5
src/Processors/Transforms/MergingAggregatedMemoryEfficientTransform.h
View File

@ -3,6 +3,7 @@
#include <Core/SortDescription.h>
#include <Common/HashTable/HashSet.h>
#include <Interpreters/Aggregator.h>
#include <Processors/Chunk.h>
#include <Processors/IProcessor.h>
#include <Processors/ISimpleTransform.h>
#include <Processors/ResizeProcessor.h>
@ -142,9 +143,9 @@ private:
void addChunk(Chunk chunk, size_t from_input);
};
struct ChunksToMerge : public ChunkInfo
struct ChunksToMerge : public ChunkInfoCloneable<ChunksToMerge>
{
std::unique_ptr<Chunks> chunks;
std::shared_ptr<Chunks> chunks;
Int32 bucket_num = -1;
bool is_overflows = false;
UInt64 chunk_num = 0; // chunk number in order of generation, used during memory bound merging to restore chunks order

10
src/Processors/Transforms/MergingAggregatedTransform.cpp
View File

@ -32,11 +32,10 @@ void MergingAggregatedTransform::consume(Chunk chunk)
total_input_rows += input_rows;
++total_input_blocks;
const auto & info = chunk.getChunkInfo();
if (!info)
if (chunk.getChunkInfos().empty())
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk info was not set for chunk in MergingAggregatedTransform.");
if (const auto * agg_info = typeid_cast<const AggregatedChunkInfo *>(info.get()))
if (auto agg_info = chunk.getChunkInfos().get<AggregatedChunkInfo>())
{
/** If the remote servers used a two-level aggregation method,
* then blocks will contain information about the number of the bucket.
@ -49,7 +48,7 @@ void MergingAggregatedTransform::consume(Chunk chunk)
bucket_to_blocks[agg_info->bucket_num].emplace_back(std::move(block));
}
else if (typeid_cast<const ChunkInfoWithAllocatedBytes *>(info.get()))
else if (chunk.getChunkInfos().get<ChunkInfoWithAllocatedBytes>())
{
auto block = getInputPort().getHeader().cloneWithColumns(chunk.getColumns());
block.info.is_overflows = false;
@ -89,7 +88,8 @@ Chunk MergingAggregatedTransform::generate()
UInt64 num_rows = block.rows();
Chunk chunk(block.getColumns(), num_rows);
chunk.setChunkInfo(std::move(info));
chunk.getChunkInfos().add(std::move(info));
return chunk;
}

15
src/Processors/Transforms/PlanSquashingTransform.cpp
View File

@ -10,20 +10,20 @@ namespace ErrorCodes
}
PlanSquashingTransform::PlanSquashingTransform(
const Block & header, size_t min_block_size_rows, size_t min_block_size_bytes)
: IInflatingTransform(header, header), squashing(header, min_block_size_rows, min_block_size_bytes)
Block header_, size_t min_block_size_rows, size_t min_block_size_bytes)
: IInflatingTransform(header_, header_)
, squashing(header_, min_block_size_rows, min_block_size_bytes)
{
}
void PlanSquashingTransform::consume(Chunk chunk)
{
if (Chunk current_chunk = squashing.add(std::move(chunk)); current_chunk.hasChunkInfo())
squashed_chunk.swap(current_chunk);
squashed_chunk = squashing.add(std::move(chunk));
}
Chunk PlanSquashingTransform::generate()
{
if (!squashed_chunk.hasChunkInfo())
if (!squashed_chunk)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Can't generate chunk in SimpleSquashingChunksTransform");
Chunk result_chunk;
@ -33,12 +33,11 @@ Chunk PlanSquashingTransform::generate()
bool PlanSquashingTransform::canGenerate()
{
return squashed_chunk.hasChunkInfo();
return bool(squashed_chunk);
}
Chunk PlanSquashingTransform::getRemaining()
{
Chunk current_chunk = squashing.flush();
return current_chunk;
return squashing.flush();
}
}

3
src/Processors/Transforms/PlanSquashingTransform.h
View File

@ -10,7 +10,7 @@ class PlanSquashingTransform : public IInflatingTransform
{
public:
PlanSquashingTransform(
const Block & header, size_t min_block_size_rows, size_t min_block_size_bytes);
Block header_, size_t min_block_size_rows, size_t min_block_size_bytes);
String getName() const override { return "PlanSquashingTransform"; }
@ -23,7 +23,6 @@ protected:
private:
Squashing squashing;
Chunk squashed_chunk;
Chunk finish_chunk;
};
}

3
src/Processors/Transforms/SelectByIndicesTransform.h
View File

@ -26,7 +26,7 @@ public:
void transform(Chunk & chunk) override
{
size_t num_rows = chunk.getNumRows();
const auto * select_final_indices_info = typeid_cast<const ChunkSelectFinalIndices *>(chunk.getChunkInfo().get());
auto select_final_indices_info = chunk.getChunkInfos().extract<ChunkSelectFinalIndices>();
if (!select_final_indices_info || !select_final_indices_info->select_final_indices)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk passed to SelectByIndicesTransform without indices column");
@ -41,7 +41,6 @@ public:
chunk.setColumns(std::move(columns), index_column->size());
}
chunk.setChunkInfo(nullptr);
}
};

18
src/Processors/Transforms/SquashingTransform.cpp
View File

@ -18,9 +18,7 @@ SquashingTransform::SquashingTransform(
void SquashingTransform::onConsume(Chunk chunk)
{
Chunk planned_chunk = squashing.add(std::move(chunk));
if (planned_chunk.hasChunkInfo())
cur_chunk = DB::Squashing::squash(std::move(planned_chunk));
cur_chunk = Squashing::squash(squashing.add(std::move(chunk)));
}
SquashingTransform::GenerateResult SquashingTransform::onGenerate()
@ -33,10 +31,7 @@ SquashingTransform::GenerateResult SquashingTransform::onGenerate()
void SquashingTransform::onFinish()
{
Chunk chunk = squashing.flush();
if (chunk.hasChunkInfo())
chunk = DB::Squashing::squash(std::move(chunk));
finish_chunk.setColumns(chunk.getColumns(), chunk.getNumRows());
finish_chunk = Squashing::squash(squashing.flush());
}
void SquashingTransform::work()
@ -49,6 +44,7 @@ void SquashingTransform::work()
}
ExceptionKeepingTransform::work();
if (finish_chunk)
{
data.chunk = std::move(finish_chunk);
@ -67,18 +63,14 @@ void SimpleSquashingTransform::transform(Chunk & chunk)
{
if (!finished)
{
Chunk planned_chunk = squashing.add(std::move(chunk));
if (planned_chunk.hasChunkInfo())
chunk = DB::Squashing::squash(std::move(planned_chunk));
chunk = Squashing::squash(squashing.add(std::move(chunk)));
}
else
{
if (chunk.hasRows())
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk expected to be empty, otherwise it will be lost");
chunk = squashing.flush();
if (chunk.hasChunkInfo())
chunk = DB::Squashing::squash(std::move(chunk));
chunk = Squashing::squash(squashing.flush());
}
}

6
src/Processors/Transforms/TotalsHavingTransform.cpp
View File

@ -150,11 +150,7 @@ void TotalsHavingTransform::transform(Chunk & chunk)
/// Block with values not included in `max_rows_to_group_by`. We'll postpone it.
if (overflow_row)
{
const auto & info = chunk.getChunkInfo();
if (!info)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk info was not set for chunk in TotalsHavingTransform.");
const auto * agg_info = typeid_cast<const AggregatedChunkInfo *>(info.get());
const auto & agg_info = chunk.getChunkInfos().get<AggregatedChunkInfo>();
if (!agg_info)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Chunk should have AggregatedChunkInfo in TotalsHavingTransform.");

127
src/Processors/Transforms/buildPushingToViewsChain.cpp
View File

@ -5,7 +5,9 @@
#include <Interpreters/InterpreterSelectQuery.h>
#include <Interpreters/InterpreterSelectQueryAnalyzer.h>
#include <Parsers/ASTInsertQuery.h>
#include <Processors/Chunk.h>
#include <Processors/Transforms/CountingTransform.h>
#include <Processors/Transforms/DeduplicationTokenTransforms.h>
#include <Processors/Transforms/PlanSquashingTransform.h>
#include <Processors/Transforms/SquashingTransform.h>
#include <Processors/Transforms/ExpressionTransform.h>
@ -16,6 +18,7 @@
#include <Storages/StorageMaterializedView.h>
#include <Storages/StorageValues.h>
#include <QueryPipeline/QueryPipelineBuilder.h>
#include <Common/Logger.h>
#include <Common/Exception.h>
#include <Common/CurrentThread.h>
#include <Common/MemoryTracker.h>
@ -24,9 +27,12 @@
#include <Common/ThreadStatus.h>
#include <Common/checkStackSize.h>
#include <Common/logger_useful.h>
#include "base/defines.h"
#include <Core/Field.h>
#include <atomic>
#include <chrono>
#include <memory>
namespace ProfileEvents
@ -105,7 +111,7 @@ private:
class ExecutingInnerQueryFromViewTransform final : public ExceptionKeepingTransform
{
public:
ExecutingInnerQueryFromViewTransform(const Block & header, ViewRuntimeData & view_, ViewsDataPtr views_data_);
ExecutingInnerQueryFromViewTransform(const Block & header, ViewRuntimeData & view_, ViewsDataPtr views_data_, bool disable_deduplication_for_children_);
String getName() const override { return "ExecutingInnerQueryFromView"; }
@ -116,6 +122,7 @@ protected:
private:
ViewsDataPtr views_data;
ViewRuntimeData & view;
bool disable_deduplication_for_children;
struct State
{
@ -138,7 +145,7 @@ class PushingToLiveViewSink final : public SinkToStorage
public:
PushingToLiveViewSink(const Block & header, StorageLiveView & live_view_, StoragePtr storage_holder_, ContextPtr context_);
String getName() const override { return "PushingToLiveViewSink"; }
void consume(Chunk chunk) override;
void consume(Chunk & chunk) override;
private:
StorageLiveView & live_view;
@ -152,7 +159,7 @@ class PushingToWindowViewSink final : public SinkToStorage
public:
PushingToWindowViewSink(const Block & header, StorageWindowView & window_view_, StoragePtr storage_holder_, ContextPtr context_);
String getName() const override { return "PushingToWindowViewSink"; }
void consume(Chunk chunk) override;
void consume(Chunk & chunk) override;
private:
StorageWindowView & window_view;
@ -216,45 +223,10 @@ std::optional<Chain> generateViewChain(
const auto & insert_settings = insert_context->getSettingsRef();
// Do not deduplicate insertions into MV if the main insertion is Ok
if (disable_deduplication_for_children)
{
insert_context->setSetting("insert_deduplicate", Field{false});
}
else if (insert_settings.update_insert_deduplication_token_in_dependent_materialized_views &&
!insert_settings.insert_deduplication_token.value.empty())
{
/** Update deduplication token passed to dependent MV with current view id. So it is possible to properly handle
* deduplication in complex INSERT flows.
*
* Example:
*
* landing -┬--> mv_1_1 ---> ds_1_1 ---> mv_2_1 --┬-> ds_2_1 ---> mv_3_1 ---> ds_3_1
* | |
* └--> mv_1_2 ---> ds_1_2 ---> mv_2_2 --┘
*
* Here we want to avoid deduplication for two different blocks generated from `mv_2_1` and `mv_2_2` that will
* be inserted into `ds_2_1`.
*
* We are forced to use view id instead of table id because there are some possible INSERT flows where no tables
* are involved.
*
* Example:
*
* landing -┬--> mv_1_1 --┬-> ds_1_1
* | |
* └--> mv_1_2 --┘
*
*/
auto insert_deduplication_token = insert_settings.insert_deduplication_token.value;
if (view_id.hasUUID())
insert_deduplication_token += "_" + toString(view_id.uuid);
else
insert_deduplication_token += "_" + view_id.getFullNameNotQuoted();
insert_context->setSetting("insert_deduplication_token", insert_deduplication_token);
}
// Processing of blocks for MVs is done block by block, and there will
// be no parallel reading after (plus it is not a costless operation)
@ -361,7 +333,13 @@ std::optional<Chain> generateViewChain(
insert_columns.emplace_back(column.name);
}
InterpreterInsertQuery interpreter(nullptr, insert_context, false, false, false);
InterpreterInsertQuery interpreter(
nullptr,
insert_context,
/* allow_materialized */ false,
/* no_squash */ false,
/* no_destination */ false,
/* async_isnert */ false);
/// TODO: remove sql_security_type check after we turn `ignore_empty_sql_security_in_create_view_query=false`
bool check_access = !materialized_view->hasInnerTable() && materialized_view->getInMemoryMetadataPtr()->sql_security_type;
@ -378,6 +356,10 @@ std::optional<Chain> generateViewChain(
table_prefers_large_blocks ? settings.min_insert_block_size_bytes : 0ULL));
}
#ifdef ABORT_ON_LOGICAL_ERROR
out.addSource(std::make_shared<DeduplicationToken::CheckTokenTransform>("Before squashing", out.getInputHeader()));
#endif
auto counting = std::make_shared<CountingTransform>(out.getInputHeader(), current_thread, insert_context->getQuota());
counting->setProcessListElement(insert_context->getProcessListElement());
counting->setProgressCallback(insert_context->getProgressCallback());
@ -420,11 +402,19 @@ std::optional<Chain> generateViewChain(
if (type == QueryViewsLogElement::ViewType::MATERIALIZED)
{
#ifdef ABORT_ON_LOGICAL_ERROR
out.addSource(std::make_shared<DeduplicationToken::CheckTokenTransform>("Right after Inner query", out.getInputHeader()));
#endif
auto executing_inner_query = std::make_shared<ExecutingInnerQueryFromViewTransform>(
storage_header, views_data->views.back(), views_data);
storage_header, views_data->views.back(), views_data, disable_deduplication_for_children);
executing_inner_query->setRuntimeData(view_thread_status, view_counter_ms);
out.addSource(std::move(executing_inner_query));
#ifdef ABORT_ON_LOGICAL_ERROR
out.addSource(std::make_shared<DeduplicationToken::CheckTokenTransform>("Right before Inner query", out.getInputHeader()));
#endif
}
return out;
@ -465,11 +455,7 @@ Chain buildPushingToViewsChain(
*/
result_chain.addTableLock(storage->lockForShare(context->getInitialQueryId(), context->getSettingsRef().lock_acquire_timeout));
/// If the "root" table deduplicates blocks, there are no need to make deduplication for children
/// Moreover, deduplication for AggregatingMergeTree children could produce false positives due to low size of inserting blocks
bool disable_deduplication_for_children = false;
if (!context->getSettingsRef().deduplicate_blocks_in_dependent_materialized_views)
disable_deduplication_for_children = !no_destination && storage->supportsDeduplication();
bool disable_deduplication_for_children = !context->getSettingsRef().deduplicate_blocks_in_dependent_materialized_views;
auto table_id = storage->getStorageID();
auto views = DatabaseCatalog::instance().getDependentViews(table_id);
@ -560,12 +546,25 @@ Chain buildPushingToViewsChain(
auto sink = std::make_shared<PushingToLiveViewSink>(live_view_header, *live_view, storage, context);
sink->setRuntimeData(thread_status, elapsed_counter_ms);
result_chain.addSource(std::move(sink));
result_chain.addSource(std::make_shared<DeduplicationToken::DefineSourceWithChunkHashTransform>(result_chain.getInputHeader()));
}
else if (auto * window_view = dynamic_cast<StorageWindowView *>(storage.get()))
{
auto sink = std::make_shared<PushingToWindowViewSink>(window_view->getInputHeader(), *window_view, storage, context);
sink->setRuntimeData(thread_status, elapsed_counter_ms);
result_chain.addSource(std::move(sink));
result_chain.addSource(std::make_shared<DeduplicationToken::DefineSourceWithChunkHashTransform>(result_chain.getInputHeader()));
}
else if (dynamic_cast<StorageMaterializedView *>(storage.get()))
{
auto sink = storage->write(query_ptr, metadata_snapshot, context, async_insert);
metadata_snapshot->check(sink->getHeader().getColumnsWithTypeAndName());
sink->setRuntimeData(thread_status, elapsed_counter_ms);
result_chain.addSource(std::move(sink));
result_chain.addSource(std::make_shared<DeduplicationToken::DefineSourceWithChunkHashTransform>(result_chain.getInputHeader()));
}
/// Do not push to destination table if the flag is set
else if (!no_destination)
@ -573,8 +572,15 @@ Chain buildPushingToViewsChain(
auto sink = storage->write(query_ptr, metadata_snapshot, context, async_insert);
metadata_snapshot->check(sink->getHeader().getColumnsWithTypeAndName());
sink->setRuntimeData(thread_status, elapsed_counter_ms);
result_chain.addSource(std::make_shared<DeduplicationToken::DefineSourceWithChunkHashTransform>(sink->getHeader()));
result_chain.addSource(std::move(sink));
}
else
{
result_chain.addSource(std::make_shared<DeduplicationToken::DefineSourceWithChunkHashTransform>(storage_header));
}
if (result_chain.empty())
result_chain.addSink(std::make_shared<NullSinkToStorage>(storage_header));
@ -590,7 +596,7 @@ Chain buildPushingToViewsChain(
return result_chain;
}
static QueryPipeline process(Block block, ViewRuntimeData & view, const ViewsData & views_data)
static QueryPipeline process(Block block, ViewRuntimeData & view, const ViewsData & views_data, Chunk::ChunkInfoCollection && chunk_infos, bool disable_deduplication_for_children)
{
const auto & context = view.context;
@ -637,6 +643,19 @@ static QueryPipeline process(Block block, ViewRuntimeData & view, const ViewsDat
pipeline.getHeader(),
std::make_shared<ExpressionActions>(std::move(converting))));
pipeline.addTransform(std::make_shared<RestoreChunkInfosTransform>(std::move(chunk_infos), pipeline.getHeader()));
if (!disable_deduplication_for_children)
{
String materialize_view_id = view.table_id.hasUUID() ? toString(view.table_id.uuid) : view.table_id.getFullNameNotQuoted();
pipeline.addTransform(std::make_shared<DeduplicationToken::SetViewIDTransform>(std::move(materialize_view_id), pipeline.getHeader()));
pipeline.addTransform(std::make_shared<DeduplicationToken::SetViewBlockNumberTransform>(pipeline.getHeader()));
}
else
{
pipeline.addTransform(std::make_shared<DeduplicationToken::ResetTokenTransform>(pipeline.getHeader()));
}
return QueryPipelineBuilder::getPipeline(std::move(pipeline));
}
@ -728,17 +747,19 @@ IProcessor::Status CopyingDataToViewsTransform::prepare()
ExecutingInnerQueryFromViewTransform::ExecutingInnerQueryFromViewTransform(
const Block & header,
ViewRuntimeData & view_,
std::shared_ptr<ViewsData> views_data_)
std::shared_ptr<ViewsData> views_data_,
bool disable_deduplication_for_children_)
: ExceptionKeepingTransform(header, view_.sample_block)
, views_data(std::move(views_data_))
, view(view_)
, disable_deduplication_for_children(disable_deduplication_for_children_)
{
}
void ExecutingInnerQueryFromViewTransform::onConsume(Chunk chunk)
{
auto block = getInputPort().getHeader().cloneWithColumns(chunk.getColumns());
state.emplace(process(block, view, *views_data));
auto block = getInputPort().getHeader().cloneWithColumns(chunk.detachColumns());
state.emplace(process(std::move(block), view, *views_data, std::move(chunk.getChunkInfos()), disable_deduplication_for_children));
}
@ -770,10 +791,10 @@ PushingToLiveViewSink::PushingToLiveViewSink(const Block & header, StorageLiveVi
{
}
void PushingToLiveViewSink::consume(Chunk chunk)
void PushingToLiveViewSink::consume(Chunk & chunk)
{
Progress local_progress(chunk.getNumRows(), chunk.bytes(), 0);
live_view.writeBlock(getHeader().cloneWithColumns(chunk.detachColumns()), context);
live_view.writeBlock(live_view, getHeader().cloneWithColumns(chunk.detachColumns()), std::move(chunk.getChunkInfos()), context);
if (auto process = context->getProcessListElement())
process->updateProgressIn(local_progress);
@ -793,11 +814,11 @@ PushingToWindowViewSink::PushingToWindowViewSink(
{
}
void PushingToWindowViewSink::consume(Chunk chunk)
void PushingToWindowViewSink::consume(Chunk & chunk)
{
Progress local_progress(chunk.getNumRows(), chunk.bytes(), 0);
StorageWindowView::writeIntoWindowView(
window_view, getHeader().cloneWithColumns(chunk.detachColumns()), context);
window_view, getHeader().cloneWithColumns(chunk.detachColumns()), std::move(chunk.getChunkInfos()), context);
if (auto process = context->getProcessListElement())
process->updateProgressIn(local_progress);

2
src/QueryPipeline/QueryPipelineBuilder.h
View File

@ -193,7 +193,7 @@ public:
return concurrency_control;
}
void addResources(QueryPlanResourceHolder resources_) { resources = std::move(resources_); }
void addResources(QueryPlanResourceHolder resources_) { resources.append(std::move(resources_)); }
void setQueryIdHolder(std::shared_ptr<QueryIdHolder> query_id_holder) { resources.query_id_holders.emplace_back(std::move(query_id_holder)); }
void addContext(ContextPtr context) { resources.interpreter_context.emplace_back(std::move(context)); }

8
src/QueryPipeline/QueryPlanResourceHolder.cpp
View File

@ -5,7 +5,7 @@
namespace DB
{
QueryPlanResourceHolder & QueryPlanResourceHolder::operator=(QueryPlanResourceHolder && rhs) noexcept
QueryPlanResourceHolder & QueryPlanResourceHolder::append(QueryPlanResourceHolder && rhs) noexcept
{
table_locks.insert(table_locks.end(), rhs.table_locks.begin(), rhs.table_locks.end());
storage_holders.insert(storage_holders.end(), rhs.storage_holders.begin(), rhs.storage_holders.end());
@ -16,6 +16,12 @@ QueryPlanResourceHolder & QueryPlanResourceHolder::operator=(QueryPlanResourceHo
return *this;
}
QueryPlanResourceHolder & QueryPlanResourceHolder::operator=(QueryPlanResourceHolder && rhs) noexcept
{
append(std::move(rhs));
return *this;
}
QueryPlanResourceHolder::QueryPlanResourceHolder() = default;
QueryPlanResourceHolder::QueryPlanResourceHolder(QueryPlanResourceHolder &&) noexcept = default;
QueryPlanResourceHolder::~QueryPlanResourceHolder() = default;

3
src/QueryPipeline/QueryPlanResourceHolder.h
View File

@ -20,8 +20,11 @@ struct QueryPlanResourceHolder
QueryPlanResourceHolder(QueryPlanResourceHolder &&) noexcept;
~QueryPlanResourceHolder();
QueryPlanResourceHolder & operator=(QueryPlanResourceHolder &) = delete;
/// Custom move assignment does not destroy data from lhs. It appends data from rhs to lhs.
QueryPlanResourceHolder & operator=(QueryPlanResourceHolder &&) noexcept;
QueryPlanResourceHolder & append(QueryPlanResourceHolder &&) noexcept;
/// Some processors may implicitly use Context or temporary Storage created by Interpreter.
/// But lifetime of Streams is not nested in lifetime of Interpreters, so we have to store it here,

102
src/Server/GRPCServer.cpp
View File

@ -1735,10 +1735,19 @@ namespace
class GRPCServer::Runner
{
public:
explicit Runner(GRPCServer & owner_) : owner(owner_) {}
explicit Runner(GRPCServer & owner_) : owner(owner_), log(owner.log) {}
~Runner()
{
try
{
stop();
}
catch (...)
{
tryLogCurrentException(log, "~Runner");
}
if (queue_thread.joinable())
queue_thread.join();
}
@ -1756,13 +1765,27 @@ public:
}
catch (...)
{
tryLogCurrentException("GRPCServer");
tryLogCurrentException(log, "run");
}
};
queue_thread = ThreadFromGlobalPool{runner_function};
}
void stop() { stopReceivingNewCalls(); }
void stop()
{
std::lock_guard lock{mutex};
should_stop = true;
if (current_calls.empty())
{
/// If there are no current calls then we call shutdownQueue() to signal the queue to stop waiting for next events.
/// The following line will make CompletionQueue::Next() stop waiting if the queue is empty and return false instead.
shutdownQueue();
/// If there are some current calls then we can't call shutdownQueue() right now because we want to let the current calls finish.
/// In this case function shutdownQueue() will be called later in run().
}
}
size_t getNumCurrentCalls() const
{
@ -1789,12 +1812,6 @@ private:
[this, call_type](bool ok) { onNewCall(call_type, ok); });
}
void stopReceivingNewCalls()
{
std::lock_guard lock{mutex};
should_stop = true;
}
void onNewCall(CallType call_type, bool responder_started_ok)
{
std::lock_guard lock{mutex};
@ -1827,38 +1844,47 @@ private:
void run()
{
setThreadName("GRPCServerQueue");
while (true)
bool ok = false;
void * tag = nullptr;
while (owner.queue->Next(&tag, &ok))
{
{
std::lock_guard lock{mutex};
finished_calls.clear(); /// Destroy finished calls.
/// If (should_stop == true) we continue processing until there is no active calls.
if (should_stop && current_calls.empty())
{
bool all_responders_gone = std::all_of(
responders_for_new_calls.begin(), responders_for_new_calls.end(),
[](std::unique_ptr<BaseResponder> & responder) { return !responder; });
if (all_responders_gone)
break;
}
}
bool ok = false;
void * tag = nullptr;
if (!owner.queue->Next(&tag, &ok))
{
/// Queue shutted down.
break;
}
auto & callback = *static_cast<CompletionCallback *>(tag);
callback(ok);
std::lock_guard lock{mutex};
finished_calls.clear(); /// Destroy finished calls.
/// If (should_stop == true) we continue processing while there are current calls.
if (should_stop && current_calls.empty())
shutdownQueue();
}
/// CompletionQueue::Next() returns false if the queue is fully drained and shut down.
}
/// Shutdown the queue if that isn't done yet.
void shutdownQueue()
{
chassert(should_stop);
if (queue_is_shut_down)
return;
queue_is_shut_down = true;
/// Server should be shut down before CompletionQueue.
if (owner.grpc_server)
owner.grpc_server->Shutdown();
if (owner.queue)
owner.queue->Shutdown();
}
GRPCServer & owner;
LoggerRawPtr log;
ThreadFromGlobalPool queue_thread;
bool queue_is_shut_down = false;
std::vector<std::unique_ptr<BaseResponder>> responders_for_new_calls;
std::map<Call *, std::unique_ptr<Call>> current_calls;
std::vector<std::unique_ptr<Call>> finished_calls;
@ -1876,16 +1902,6 @@ GRPCServer::GRPCServer(IServer & iserver_, const Poco::Net::SocketAddress & addr
GRPCServer::~GRPCServer()
{
/// Server should be shutdown before CompletionQueue.
if (grpc_server)
grpc_server->Shutdown();
/// Completion Queue should be shutdown before destroying the runner,
/// because the runner is now probably executing CompletionQueue::Next() on queue_thread
/// which is blocked until an event is available or the queue is shutting down.
if (queue)
queue->Shutdown();
runner.reset();
}

3
src/Server/PrometheusRequestHandler.cpp
View File

@ -18,9 +18,6 @@ void PrometheusRequestHandler::handleRequest(HTTPServerRequest & request, HTTPSe
{
try
{
/// Raw config reference is used here to avoid dependency on Context and ServerSettings.
/// This is painful, because this class is also used in a build with CLICKHOUSE_KEEPER_STANDALONE_BUILD=1
/// And there ordinary Context is replaced with a tiny clone.
const auto & config = server.config();
unsigned keep_alive_timeout = config.getUInt("keep_alive_timeout", DEFAULT_HTTP_KEEP_ALIVE_TIMEOUT);

4
src/Server/ProtocolServerAdapter.cpp
View File

@ -1,7 +1,7 @@
#include <Server/ProtocolServerAdapter.h>
#include <Server/TCPServer.h>
#if USE_GRPC && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_GRPC
#include <Server/GRPCServer.h>
#endif
@ -37,7 +37,7 @@ ProtocolServerAdapter::ProtocolServerAdapter(
{
}
#if USE_GRPC && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_GRPC
class ProtocolServerAdapter::GRPCServerAdapterImpl : public Impl
{
public:

2
src/Server/ProtocolServerAdapter.h
View File

@ -23,7 +23,7 @@ public:
ProtocolServerAdapter & operator =(ProtocolServerAdapter && src) = default;
ProtocolServerAdapter(const std::string & listen_host_, const char * port_name_, const std::string & description_, std::unique_ptr<TCPServer> tcp_server_);
#if USE_GRPC && !defined(CLICKHOUSE_KEEPER_STANDALONE_BUILD)
#if USE_GRPC
ProtocolServerAdapter(const std::string & listen_host_, const char * port_name_, const std::string & description_, std::unique_ptr<GRPCServer> grpc_server_);
#endif

Some files were not shown because too many files have changed in this diff Show More