Merge branch 'master' into enable_constant_column_search

.clang-tidy

@@ -22,6 +22,8 @@ Checks: '*,
     -bugprone-implicit-widening-of-multiplication-result,
     -bugprone-narrowing-conversions,
     -bugprone-not-null-terminated-result,
+    -bugprone-unchecked-optional-access,
+    -bugprone-assignment-in-if-condition,
 
     -cert-dcl16-c,
     -cert-err58-cpp,
@@ -103,6 +105,7 @@ Checks: '*,
 
     -misc-no-recursion,
     -misc-non-private-member-variables-in-classes,
+    -misc-const-correctness,
 
     -modernize-avoid-c-arrays,
     -modernize-concat-nested-namespaces,
@@ -114,6 +117,7 @@ Checks: '*,
     -modernize-use-nodiscard,
     -modernize-use-override,
     -modernize-use-trailing-return-type,
+    -modernize-macro-to-enum,
 
     -performance-inefficient-string-concatenation,
     -performance-no-int-to-ptr,
@@ -135,6 +139,7 @@ Checks: '*,
     -readability-suspicious-call-argument,
     -readability-uppercase-literal-suffix,
     -readability-use-anyofallof,
+    -readability-simplify-boolean-expr,
 
     -zirkon-*,
 '

cmake/clang_tidy.cmake

@@ -3,7 +3,7 @@ option (ENABLE_CLANG_TIDY "Use clang-tidy static analyzer" OFF)
 
 if (ENABLE_CLANG_TIDY)
 
-    find_program (CLANG_TIDY_PATH NAMES "clang-tidy" "clang-tidy-14" "clang-tidy-13" "clang-tidy-12")
+    find_program (CLANG_TIDY_PATH NAMES "clang-tidy" "clang-tidy-15" "clang-tidy-14" "clang-tidy-13" "clang-tidy-12")
 
     if (CLANG_TIDY_PATH)
         message(STATUS

cmake/target.cmake

@@ -45,6 +45,7 @@ if (CMAKE_CROSSCOMPILING)
     endif ()
 
     if (USE_MUSL)
+        # use of undeclared identifier 'PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP'
         set (ENABLE_SENTRY OFF CACHE INTERNAL "")
         set (ENABLE_ODBC OFF CACHE INTERNAL "")
         set (ENABLE_GRPC OFF CACHE INTERNAL "")

contrib/NuRaft

@@ -1 +1 @@
-Subproject commit 1be805e7cb2494aa8170015493474379b0362dfc
+Subproject commit e15858f8ad0ce8aba85cf74e3763874c76bf927c

contrib/jemalloc-cmake/include_linux_x86_64_musl/jemalloc/internal/jemalloc_internal_defs.h.in

@@ -415,7 +415,7 @@
 /*
  * Defined if strerror_r returns char * if _GNU_SOURCE is defined.
  */
-#define JEMALLOC_STRERROR_R_RETURNS_CHAR_WITH_GNU_SOURCE
+/* #undef JEMALLOC_STRERROR_R_RETURNS_CHAR_WITH_GNU_SOURCE */
 
 /* Performs additional safety checks when defined. */
 /* #undef JEMALLOC_OPT_SAFETY_CHECKS */

contrib/krb5-cmake/autoconf_linux.h

@@ -440,7 +440,9 @@
 #define HAVE_STRERROR 1
 
 /* Define to 1 if you have the `strerror_r' function. */
+#ifndef USE_MUSL
 #define HAVE_STRERROR_R 1
+#endif
 
 /* Define to 1 if you have the <strings.h> header file. */
 #define HAVE_STRINGS_H 1

contrib/libcpuid

@@ -1 +1 @@
-Subproject commit 8db3b8d2d32d22437f063ce692a1b9bb15e42d18
+Subproject commit 503083acb77edf9fbce22a05826307dff2ce96e6

contrib/libpq-cmake/CMakeLists.txt

@@ -63,6 +63,13 @@ target_include_directories (_libpq SYSTEM PUBLIC ${LIBPQ_SOURCE_DIR})
 target_include_directories (_libpq SYSTEM PUBLIC "${LIBPQ_SOURCE_DIR}/include")
 target_include_directories (_libpq SYSTEM PRIVATE "${LIBPQ_SOURCE_DIR}/configs")
 
+# NOTE: this is a dirty hack to avoid and instead pg_config.h should be shipped
+# for different OS'es like for jemalloc, not one generic for all OS'es like
+# now.
+if (OS_DARWIN OR OS_FREEBSD OR USE_MUSL)
+    target_compile_definitions(_libpq PRIVATE -DSTRERROR_R_INT=1)
+endif()
+
 target_link_libraries (_libpq PRIVATE OpenSSL::SSL)
 
 add_library(ch_contrib::libpq ALIAS _libpq)

contrib/librdkafka

@@ -1 +1 @@
-Subproject commit ff32b4e9eeafd0b276f010ee969179e4e9e6d0b2
+Subproject commit 6f3b483426a8c8ec950e27e446bec175cf8b553f

contrib/llvm

@@ -1 +1 @@
-Subproject commit 20607e61728e97c969e536644c3c0c1bb1a50672
+Subproject commit 0db5bf5bd2452cd8f1283a1fcdc04845af705bfc

contrib/sentry-native

@@ -1 +1 @@
-Subproject commit f431047ac8da13179c488018dddf1c0d0771a997
+Subproject commit ae10fb8c224c3f41571446e1ed7fd57b9e5e366b

contrib/vectorscan

@@ -1 +1 @@
-Subproject commit 73695e419c27af7fe2a099c7aa57931cc02aea5d
+Subproject commit f6250ae3e5a3085000239313ad0689cc1e00cdc2

contrib/vectorscan-cmake/CMakeLists.txt

@@ -304,7 +304,7 @@ target_include_directories (_vectorscan SYSTEM PUBLIC "${LIBRARY_DIR}/src")
 # Please regenerate these files if you update vectorscan.
 
 if (ARCH_AMD64)
-    target_include_directories (_vectorscan PRIVATE x86_64)
+    target_include_directories (_vectorscan PRIVATE amd64)
 endif ()
 
 if (ARCH_AARCH64)

docker/packager/binary/Dockerfile

@@ -67,24 +67,5 @@ ENV GOCACHE=/workdir/
 RUN mkdir /workdir && chmod 777 /workdir
 WORKDIR /workdir
 
-# NOTE: thread sanitizer is broken in clang-14, we have to build it with clang-15
-# https://github.com/ClickHouse/ClickHouse/pull/39450
-# https://github.com/google/sanitizers/issues/1540
-# https://github.com/google/sanitizers/issues/1552
-
-RUN export CODENAME="$(lsb_release --codename --short | tr 'A-Z' 'a-z')" \
-    && echo "deb [trusted=yes] https://apt.llvm.org/${CODENAME}/ llvm-toolchain-${CODENAME}-15 main" >> \
-        /etc/apt/sources.list.d/clang.list \
-    && apt-get update \
-    && apt-get install \
-        clang-15 \
-        llvm-15 \
-        clang-tidy-15 \
-        --yes --no-install-recommends \
-    && apt-get clean
-
-# for external_symbolizer_path
-RUN ln -s /usr/bin/llvm-symbolizer-15 /usr/bin/llvm-symbolizer
-
 COPY build.sh /
 CMD ["bash", "-c", "/build.sh 2>&1"]

docker/packager/packager

@@ -339,17 +339,16 @@ if __name__ == "__main__":
     parser.add_argument(
         "--compiler",
         choices=(
-            "clang-15",  # For TSAN builds, see #39450
+            "clang-15",
-            "clang-14",
+            "clang-15-darwin",
-            "clang-14-darwin",
+            "clang-15-darwin-aarch64",
-            "clang-14-darwin-aarch64",
+            "clang-15-aarch64",
-            "clang-14-aarch64",
+            "clang-15-ppc64le",
-            "clang-14-ppc64le",
+            "clang-15-amd64sse2",
-            "clang-14-amd64sse2",
+            "clang-15-freebsd",
-            "clang-14-freebsd",
             "gcc-11",
         ),
-        default="clang-14",
+        default="clang-15",
         help="a compiler to use",
     )
     parser.add_argument(

docker/test/base/Dockerfile

@@ -16,11 +16,10 @@ RUN apt-get update \
 # and MEMORY_LIMIT_EXCEEDED exceptions in Functional tests (total memory limit in Functional tests is ~55.24 GiB).
 # TSAN will flush shadow memory when reaching this limit.
 # It may cause false-negatives, but it's better than OOM.
-RUN echo "TSAN_OPTIONS='verbosity=1000 halt_on_error=1 history_size=7 memory_limit_mb=46080'" >> /etc/environment; \
+RUN echo "TSAN_OPTIONS='verbosity=1000 halt_on_error=1 history_size=7 memory_limit_mb=46080'" >> /etc/environment
-  echo "UBSAN_OPTIONS='print_stacktrace=1'" >> /etc/environment; \
+RUN echo "UBSAN_OPTIONS='print_stacktrace=1'" >> /etc/environment
-  echo "MSAN_OPTIONS='abort_on_error=1 poison_in_dtor=1'" >> /etc/environment; \
+RUN echo "MSAN_OPTIONS='abort_on_error=1 poison_in_dtor=1'" >> /etc/environment
-  echo "LSAN_OPTIONS='suppressions=/usr/share/clickhouse-test/config/lsan_suppressions.txt'" >> /etc/environment; \
+RUN echo "LSAN_OPTIONS='suppressions=/usr/share/clickhouse-test/config/lsan_suppressions.txt'" >> /etc/environment
-  ln -s /usr/lib/llvm-${LLVM_VERSION}/bin/llvm-symbolizer /usr/bin/llvm-symbolizer;
 # Sanitizer options for current shell (not current, but the one that will be spawned on "docker run")
 # (but w/o verbosity for TSAN, otherwise test.reference will not match)
 ENV TSAN_OPTIONS='halt_on_error=1 history_size=7 memory_limit_mb=46080'

docker/test/codebrowser/Dockerfile

@@ -8,16 +8,41 @@ FROM clickhouse/binary-builder:$FROM_TAG
 ARG apt_archive="http://archive.ubuntu.com"
 RUN sed -i "s|http://archive.ubuntu.com|$apt_archive|g" /etc/apt/sources.list
 
-RUN apt-get update && apt-get --yes --allow-unauthenticated install clang-14 libllvm14 libclang-14-dev libmlir-14-dev
+RUN apt-get update && apt-get --yes --allow-unauthenticated install libclang-${LLVM_VERSION}-dev libmlir-${LLVM_VERSION}-dev
+
+# libclang-15-dev does not contain proper symlink:
+#
+# This is what cmake will search for:
+#
+#     # readlink -f /usr/lib/llvm-15/lib/libclang-15.so.1
+#     /usr/lib/x86_64-linux-gnu/libclang-15.so.1
+#
+# This is what exists:
+#
+#     # ls -l /usr/lib/x86_64-linux-gnu/libclang-15*
+#     lrwxrwxrwx 1 root root       16 Sep  5 13:31 /usr/lib/x86_64-linux-gnu/libclang-15.so -> libclang-15.so.1
+#     lrwxrwxrwx 1 root root       21 Sep  5 13:31 /usr/lib/x86_64-linux-gnu/libclang-15.so.15 -> libclang-15.so.15.0.0
+#     -rw-r--r-- 1 root root 31835760 Sep  5 13:31 /usr/lib/x86_64-linux-gnu/libclang-15.so.15.0.0
+#
+ARG TARGETARCH
+RUN arch=${TARGETARCH:-amd64} \
+    && case $arch in \
+        amd64) rarch=x86_64 ;; \
+        arm64) rarch=aarch64 ;; \
+        *) exit 1 ;; \
+    esac \
+    && ln -rsf /usr/lib/$rarch-linux-gnu/libclang-15.so.15 /usr/lib/$rarch-linux-gnu/libclang-15.so.1
 
 # repo versions doesn't work correctly with C++17
 # also we push reports to s3, so we add index.html to subfolder urls
 # https://github.com/ClickHouse-Extras/woboq_codebrowser/commit/37e15eaf377b920acb0b48dbe82471be9203f76b
 # TODO: remove branch in a few weeks after merge, e.g. in May or June 2022
-RUN git clone https://github.com/ClickHouse-Extras/woboq_codebrowser --branch llvm-14 \
+#
+# FIXME: update location of a repo
+RUN git clone https://github.com/azat/woboq_codebrowser --branch llvm-15 \
   && cd woboq_codebrowser \
-  && cmake . -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_COMPILER=clang\+\+-14 -DCMAKE_C_COMPILER=clang-14 \
+  && cmake . -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_COMPILER=clang\+\+-${LLVM_VERSION} -DCMAKE_C_COMPILER=clang-${LLVM_VERSION} \
-  && make -j \
+  && ninja \
   && cd .. \
   && rm -rf woboq_codebrowser
 
@@ -32,7 +57,7 @@ ENV SHA=nosha
 ENV DATA="https://s3.amazonaws.com/clickhouse-test-reports/codebrowser/data"
 
 CMD mkdir -p $BUILD_DIRECTORY && cd $BUILD_DIRECTORY && \
-    cmake $SOURCE_DIRECTORY -DCMAKE_CXX_COMPILER=/usr/bin/clang\+\+-14 -DCMAKE_C_COMPILER=/usr/bin/clang-14 -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DENABLE_EMBEDDED_COMPILER=0 -DENABLE_S3=0 && \
+    cmake $SOURCE_DIRECTORY -DCMAKE_CXX_COMPILER=/usr/bin/clang\+\+-${LLVM_VERSION} -DCMAKE_C_COMPILER=/usr/bin/clang-${LLVM_VERSION} -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DENABLE_EMBEDDED_COMPILER=0 -DENABLE_S3=0 && \
     mkdir -p $HTML_RESULT_DIRECTORY && \
     $CODEGEN -b $BUILD_DIRECTORY -a -o $HTML_RESULT_DIRECTORY -p ClickHouse:$SOURCE_DIRECTORY:$SHA -d $DATA | ts '%Y-%m-%d %H:%M:%S' && \
     cp -r $STATIC_DATA $HTML_RESULT_DIRECTORY/ &&\

docker/test/fuzzer/run-fuzzer.sh

@@ -19,7 +19,7 @@ stage=${stage:-}
 script_dir="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
 echo "$script_dir"
 repo_dir=ch
-BINARY_TO_DOWNLOAD=${BINARY_TO_DOWNLOAD:="clang-14_debug_none_unsplitted_disable_False_binary"}
+BINARY_TO_DOWNLOAD=${BINARY_TO_DOWNLOAD:="clang-15_debug_none_unsplitted_disable_False_binary"}
 BINARY_URL_TO_DOWNLOAD=${BINARY_URL_TO_DOWNLOAD:="https://clickhouse-builds.s3.amazonaws.com/$PR_TO_TEST/$SHA_TO_TEST/clickhouse_build_check/$BINARY_TO_DOWNLOAD/clickhouse"}
 
 function clone

docker/test/keeper-jepsen/run.sh

@@ -2,7 +2,7 @@
 set -euo pipefail
 
 
-CLICKHOUSE_PACKAGE=${CLICKHOUSE_PACKAGE:="https://clickhouse-builds.s3.amazonaws.com/$PR_TO_TEST/$SHA_TO_TEST/clickhouse_build_check/clang-14_relwithdebuginfo_none_unsplitted_disable_False_binary/clickhouse"}
+CLICKHOUSE_PACKAGE=${CLICKHOUSE_PACKAGE:="https://clickhouse-builds.s3.amazonaws.com/$PR_TO_TEST/$SHA_TO_TEST/clickhouse_build_check/clang-15_relwithdebuginfo_none_unsplitted_disable_False_binary/clickhouse"}
 CLICKHOUSE_REPO_PATH=${CLICKHOUSE_REPO_PATH:=""}
 
 

docker/test/performance-comparison/compare.sh

@@ -61,7 +61,7 @@ function configure
     cp -rv right/config left ||:
 
     # Start a temporary server to rename the tables
-    while pkill clickhouse-serv; do echo . ; sleep 1 ; done
+    while pkill -f clickhouse-serv ; do echo . ; sleep 1 ; done
     echo all killed
 
     set -m # Spawn temporary in its own process groups
@@ -88,7 +88,7 @@ function configure
     clickhouse-client --port $LEFT_SERVER_PORT --query "create database test" ||:
     clickhouse-client --port $LEFT_SERVER_PORT --query "rename table datasets.hits_v1 to test.hits" ||:
 
-    while pkill clickhouse-serv; do echo . ; sleep 1 ; done
+    while pkill -f clickhouse-serv ; do echo . ; sleep 1 ; done
     echo all killed
 
     # Make copies of the original db for both servers. Use hardlinks instead
@@ -106,7 +106,7 @@ function configure
 
 function restart
 {
-    while pkill clickhouse-serv; do echo . ; sleep 1 ; done
+    while pkill -f clickhouse-serv ; do echo . ; sleep 1 ; done
     echo all killed
 
     # Change the jemalloc settings here.
@@ -1400,7 +1400,7 @@ case "$stage" in
     while env kill -- -$watchdog_pid ; do sleep 1; done
 
     # Stop the servers to free memory for the subsequent query analysis.
-    while pkill clickhouse-serv; do echo . ; sleep 1 ; done
+    while pkill -f clickhouse-serv ; do echo . ; sleep 1 ; done
     echo Servers stopped.
     ;&
 "analyze_queries")

docker/test/util/Dockerfile

@@ -5,7 +5,7 @@ FROM ubuntu:20.04
 ARG apt_archive="http://archive.ubuntu.com"
 RUN sed -i "s|http://archive.ubuntu.com|$apt_archive|g" /etc/apt/sources.list
 
-ENV DEBIAN_FRONTEND=noninteractive LLVM_VERSION=14
+ENV DEBIAN_FRONTEND=noninteractive LLVM_VERSION=15
 
 RUN apt-get update \
     && apt-get install \
@@ -56,6 +56,8 @@ RUN apt-get update \
 
 # This symlink required by gcc to find lld compiler
 RUN ln -s /usr/bin/lld-${LLVM_VERSION} /usr/bin/ld.lld
+# for external_symbolizer_path
+RUN ln -s /usr/bin/llvm-symbolizer-${LLVM_VERSION} /usr/bin/llvm-symbolizer
 
 ARG CCACHE_VERSION=4.6.1
 RUN mkdir /tmp/ccache \

docs/en/operations/storing-data.md

@@ -134,6 +134,13 @@ Example of configuration for versions later or equal to 22.8:
                 <max_size>10000000</max_size>
             </cache>
         </disks>
+        <policies>
+            <volumes>
+                <main>
+                    <disk>cache</disk>
+                </main>
+            </volumes>
+        <policies>
     </storage_configuration>
 ```
 
@@ -151,6 +158,13 @@ Example of configuration for versions earlier than 22.8:
                 <data_cache_size>10000000</data_cache_size>
             </s3>
         </disks>
+        <policies>
+            <volumes>
+                <main>
+                    <disk>s3</disk>
+                </main>
+            </volumes>
+        <policies>
     </storage_configuration>
 ```
 

docs/en/sql-reference/table-functions/file.md

@@ -13,7 +13,7 @@ Creates a table from a file. This table function is similar to [url](../../sql-r
 **Syntax**
 
 ``` sql
-file(path, format, structure)
+file(path [,format] [,structure])
 ```
 
 **Parameters**

docs/en/sql-reference/table-functions/s3.md

@@ -11,7 +11,7 @@ Provides table-like interface to select/insert files in [Amazon S3](https://aws.
 **Syntax**
 
 ``` sql
-s3(path, [aws_access_key_id, aws_secret_access_key,] format, structure, [compression])
+s3(path [,aws_access_key_id, aws_secret_access_key] [,format] [,structure] [,compression])
 ```
 
 **Arguments**

docs/en/sql-reference/table-functions/s3Cluster.md

@@ -10,7 +10,7 @@ Allows processing files from [Amazon S3](https://aws.amazon.com/s3/) in parallel
 **Syntax**
 
 ``` sql
-s3Cluster(cluster_name, source, [access_key_id, secret_access_key,] format, structure)
+s3Cluster(cluster_name, source, [,access_key_id, secret_access_key] [,format] [,structure])
 ```
 
 **Arguments**

docs/en/sql-reference/table-functions/url.md

@@ -13,7 +13,7 @@ sidebar_label: url
 **Syntax**
 
 ``` sql
-url(URL, format, structure)
+url(URL [,format] [,structure])
 ```
 
 **Parameters**

docs/ru/sql-reference/table-functions/file.md

@@ -13,7 +13,7 @@ sidebar_label: file
 **Синтаксис**
 
 ``` sql
-file(path, format, structure)
+file(path [,format] [,structure])
 ```
 
 **Параметры**

docs/ru/sql-reference/table-functions/s3.md

@@ -11,7 +11,7 @@ sidebar_label: s3
 **Синтаксис**
 
 ``` sql
-s3(path, [aws_access_key_id, aws_secret_access_key,] format, structure, [compression])
+s3(path [,aws_access_key_id, aws_secret_access_key] [,format] [,structure] [,compression])
 ```
 
 **Aргументы**

docs/ru/sql-reference/table-functions/s3Cluster.md

@@ -11,7 +11,7 @@ sidebar_label: s3Cluster
 **Синтаксис**
 
 ``` sql
-s3Cluster(cluster_name, source, [access_key_id, secret_access_key,] format, structure)
+s3Cluster(cluster_name, source, [,access_key_id, secret_access_key] [,format] [,structure])
 ```
 
 **Аргументы**

docs/ru/sql-reference/table-functions/url.md

@@ -13,7 +13,7 @@ sidebar_label: url
 **Синтаксис**
 
 ``` sql
-url(URL, format, structure)
+url(URL [,format] [,structure])
 ```
 
 **Параметры**

src/AggregateFunctions/AggregateFunctionIf.cpp

@@ -278,6 +278,71 @@ public:
         }
     }
 
+    void addBatchSinglePlace(
+        size_t row_begin, size_t row_end, AggregateDataPtr __restrict place, const IColumn ** columns, Arena * arena, ssize_t) const final
+    {
+        std::unique_ptr<UInt8[]> final_null_flags = std::make_unique<UInt8[]>(row_end);
+        const size_t filter_column_num = number_of_arguments - 1;
+
+        if (is_nullable[filter_column_num])
+        {
+            const ColumnNullable * nullable_column = assert_cast<const ColumnNullable *>(columns[filter_column_num]);
+            const IColumn & filter_column = nullable_column->getNestedColumn();
+            const UInt8 * filter_null_map = nullable_column->getNullMapColumn().getData().data();
+            const UInt8 * filter_values = assert_cast<const ColumnUInt8 &>(filter_column).getData().data();
+
+            for (size_t i = row_begin; i < row_end; i++)
+            {
+                final_null_flags[i] = (null_is_skipped && filter_null_map[i]) || !filter_values[i];
+            }
+        }
+        else
+        {
+            const IColumn * filter_column = columns[filter_column_num];
+            const UInt8 * filter_values = assert_cast<const ColumnUInt8 *>(filter_column)->getData().data();
+            for (size_t i = row_begin; i < row_end; i++)
+                final_null_flags[i] = !filter_values[i];
+        }
+
+        const IColumn * nested_columns[number_of_arguments];
+        for (size_t arg = 0; arg < number_of_arguments; arg++)
+        {
+            if (is_nullable[arg])
+            {
+                const ColumnNullable & nullable_col = assert_cast<const ColumnNullable &>(*columns[arg]);
+                if (null_is_skipped && (arg != filter_column_num))
+                {
+                    const ColumnUInt8 & nullmap_column = nullable_col.getNullMapColumn();
+                    const UInt8 * col_null_map = nullmap_column.getData().data();
+                    for (size_t r = row_begin; r < row_end; r++)
+                    {
+                        final_null_flags[r] |= col_null_map[r];
+                    }
+                }
+                nested_columns[arg] = &nullable_col.getNestedColumn();
+            }
+            else
+                nested_columns[arg] = columns[arg];
+        }
+
+        bool at_least_one = false;
+        for (size_t i = row_begin; i < row_end; i++)
+        {
+            if (!final_null_flags[i])
+            {
+                at_least_one = true;
+                break;
+            }
+        }
+
+        if (at_least_one)
+        {
+            this->setFlag(place);
+            this->nested_function->addBatchSinglePlaceNotNull(
+                row_begin, row_end, this->nestedPlace(place), nested_columns, final_null_flags.get(), arena, -1);
+        }
+    }
+
 #if USE_EMBEDDED_COMPILER
 
     void compileAdd(llvm::IRBuilderBase & builder, llvm::Value * aggregate_data_ptr, const DataTypes & arguments_types, const std::vector<llvm::Value *> & argument_values) const override

src/AggregateFunctions/AggregateFunctionNull.h

@@ -414,6 +414,109 @@ public:
         this->nested_function->add(this->nestedPlace(place), nested_columns, row_num, arena);
     }
 
+    void addBatchSinglePlace(
+        size_t row_begin,
+        size_t row_end,
+        AggregateDataPtr __restrict place,
+        const IColumn ** columns,
+        Arena * arena,
+        ssize_t if_argument_pos) const final
+    {
+        /// We are going to merge all the flags into a single one to be able to call the nested batching functions
+        std::vector<const UInt8 *> nullable_filters;
+        const IColumn * nested_columns[number_of_arguments];
+
+        std::unique_ptr<UInt8[]> final_flags = nullptr;
+        const UInt8 * final_flags_ptr = nullptr;
+
+        if (if_argument_pos >= 0)
+        {
+            final_flags = std::make_unique<UInt8[]>(row_end);
+            final_flags_ptr = final_flags.get();
+
+            bool included_elements = 0;
+            const auto & flags = assert_cast<const ColumnUInt8 &>(*columns[if_argument_pos]).getData();
+            for (size_t i = row_begin; i < row_end; i++)
+            {
+                final_flags[i] = !flags.data()[i];
+                included_elements += !!flags.data()[i];
+            }
+
+            if (included_elements == 0)
+                return;
+            if (included_elements != (row_end - row_begin))
+            {
+                nullable_filters.push_back(final_flags_ptr);
+            }
+        }
+
+        for (size_t i = 0; i < number_of_arguments; ++i)
+        {
+            if (is_nullable[i])
+            {
+                const ColumnNullable & nullable_col = assert_cast<const ColumnNullable &>(*columns[i]);
+                nested_columns[i] = &nullable_col.getNestedColumn();
+                if constexpr (null_is_skipped)
+                {
+                    const ColumnUInt8 & nullmap_column = nullable_col.getNullMapColumn();
+                    nullable_filters.push_back(nullmap_column.getData().data());
+                }
+            }
+            else
+            {
+                nested_columns[i] = columns[i];
+            }
+        }
+
+        bool found_one = false;
+
+        chassert(nullable_filters.size() > 0); /// We work under the assumption that we reach this because one argument was NULL
+        if (nullable_filters.size() == 1)
+        {
+            /// We can avoid making copies of the only filter but we still need to check that there is data to be added
+            final_flags_ptr = nullable_filters[0];
+            for (size_t i = row_begin; i < row_end; i++)
+            {
+                if (!final_flags_ptr[i])
+                {
+                    found_one = true;
+                    break;
+                }
+            }
+        }
+        else
+        {
+            if (!final_flags)
+            {
+                final_flags = std::make_unique<UInt8[]>(row_end);
+                final_flags_ptr = final_flags.get();
+            }
+
+            const size_t filter_start = nullable_filters[0] == final_flags_ptr ? 1 : 0;
+            for (size_t filter = filter_start; filter < nullable_filters.size(); filter++)
+            {
+                for (size_t i = row_begin; i < row_end; i++)
+                    final_flags[i] |= nullable_filters[filter][i];
+            }
+
+            for (size_t i = row_begin; i < row_end; i++)
+            {
+                if (!final_flags_ptr[i])
+                {
+                    found_one = true;
+                    break;
+                }
+            }
+        }
+
+        if (!found_one)
+            return; // Nothing to do and nothing to mark
+
+        this->setFlag(place);
+        this->nested_function->addBatchSinglePlaceNotNull(
+            row_begin, row_end, this->nestedPlace(place), nested_columns, final_flags_ptr, arena, -1);
+    }
+
 
 #if USE_EMBEDDED_COMPILER
 

src/AggregateFunctions/UniquesHashSet.h

@@ -424,6 +424,8 @@ public:
 
         alloc(new_size_degree);
 
+        if (m_size <= 1)
+        {
             for (size_t i = 0; i < m_size; ++i)
             {
                 HashValue x = 0;
@@ -434,6 +436,20 @@ public:
                     reinsertImpl(x);
             }
         }
+        else
+        {
+            auto hs = std::make_unique<HashValue[]>(m_size);
+            rb.readStrict(reinterpret_cast<char *>(hs.get()), m_size * sizeof(HashValue));
+
+            for (size_t i = 0; i < m_size; ++i)
+            {
+                if (hs[i] == 0)
+                    has_zero = true;
+                else
+                    reinsertImpl(hs[i]);
+            }
+        }
+    }
 
     void readAndMerge(DB::ReadBuffer & rb)
     {
@@ -458,6 +474,8 @@ public:
             resize(new_size_degree);
         }
 
+        if (rhs_size <= 1)
+        {
             for (size_t i = 0; i < rhs_size; ++i)
             {
                 HashValue x = 0;
@@ -465,6 +483,17 @@ public:
                 insertHash(x);
             }
         }
+        else
+        {
+            auto hs = std::make_unique<HashValue[]>(rhs_size);
+            rb.readStrict(reinterpret_cast<char *>(hs.get()), rhs_size * sizeof(HashValue));
+
+            for (size_t i = 0; i < rhs_size; ++i)
+            {
+                insertHash(hs[i]);
+            }
+        }
+    }
 
     static void skip(DB::ReadBuffer & rb)
     {

src/Common/ArrayCache.h

@@ -722,5 +722,3 @@ public:
         return res;
     }
 };
-
-template <typename Key, typename Payload> constexpr size_t ArrayCache<Key, Payload>::min_chunk_size;

src/Common/MemoryStatisticsOS.cpp

@@ -135,7 +135,7 @@ MemoryStatisticsOS::Data MemoryStatisticsOS::get() const
     struct kinfo_proc kp;
     size_t len = sizeof(struct kinfo_proc);
 
-    if (-1 == ::sysctl(mib, 4, &kp, &len, NULL, 0))
+    if (-1 == ::sysctl(mib, 4, &kp, &len, nullptr, 0))
         throwFromErrno("Cannot sysctl(kern.proc.pid." + std::to_string(self) + ")", ErrorCodes::SYSTEM_ERROR);
 
     if (sizeof(struct kinfo_proc) != len)

src/Coordination/CoordinationSettings.cpp

@@ -189,6 +189,9 @@ KeeperConfigurationAndSettings::loadFromConfig(const Poco::Util::AbstractConfigu
 
     ret->coordination_settings->loadFromConfig("keeper_server.coordination_settings", config);
 
+    if (ret->coordination_settings->quorum_reads)
+        LOG_WARNING(&Poco::Logger::get("KeeperConfigurationAndSettings"), "Setting 'quorum_reads' is deprecated. Please use 'read_mode'");
+
     return ret;
 }
 

src/Coordination/CoordinationSettings.h

@@ -26,6 +26,7 @@ struct Settings;
     M(Milliseconds, heart_beat_interval_ms, 500, "Heartbeat interval between quorum nodes", 0) \
     M(Milliseconds, election_timeout_lower_bound_ms, 1000, "Lower bound of election timer (avoid too often leader elections)", 0) \
     M(Milliseconds, election_timeout_upper_bound_ms, 2000, "Upper bound of election timer (avoid too often leader elections)", 0) \
+    M(Milliseconds, leadership_expiry, 0, "How often will leader node check if it still has majority. Set it lower or equal to election_timeout_lower_bound_ms to have linearizable reads.", 0) \
     M(UInt64, reserved_log_items, 100000, "How many log items to store (don't remove during compaction)", 0) \
     M(UInt64, snapshot_distance, 100000, "How many log items we have to collect to write new snapshot", 0) \
     M(Bool, auto_forwarding, true, "Allow to forward write requests from followers to leader", 0) \
@@ -38,11 +39,12 @@ struct Settings;
     M(UInt64, stale_log_gap, 10000, "When node became stale and should receive snapshots from leader", 0) \
     M(UInt64, fresh_log_gap, 200, "When node became fresh", 0) \
     M(UInt64, max_requests_batch_size, 100, "Max size of batch in requests count before it will be sent to RAFT", 0) \
-    M(Bool, quorum_reads, false, "Execute read requests as writes through whole RAFT consesus with similar speed", 0) \
+    M(Bool, quorum_reads, false, "Deprecated - use read_mode. Execute read requests as writes through whole RAFT consesus with similar speed", 0) \
     M(Bool, force_sync, true, "Call fsync on each change in RAFT changelog", 0) \
     M(Bool, compress_logs, true, "Write compressed coordination logs in ZSTD format", 0) \
     M(Bool, compress_snapshots_with_zstd_format, true, "Write compressed snapshots in ZSTD format (instead of custom LZ4)", 0) \
-    M(UInt64, configuration_change_tries_count, 20, "How many times we will try to apply configuration change (add/remove server) to the cluster", 0)
+    M(UInt64, configuration_change_tries_count, 20, "How many times we will try to apply configuration change (add/remove server) to the cluster", 0) \
+    M(String, read_mode, "nonlinear", "How should reads be processed. Valid values: 'nonlinear', 'fastlinear', 'quorum'. 'nonlinear' is the fastest option because there are no consistency requirements", 0)
 
 DECLARE_SETTINGS_TRAITS(CoordinationSettingsTraits, LIST_OF_COORDINATION_SETTINGS)
 

src/Coordination/KeeperDispatcher.cpp

@@ -1,4 +1,5 @@
 #include <Coordination/KeeperDispatcher.h>
+#include <libnuraft/async.hxx>
 #include <Common/setThreadName.h>
 #include <Common/ZooKeeper/KeeperException.h>
 #include <future>
@@ -6,6 +7,8 @@
 #include <Poco/Path.h>
 #include <Common/hex.h>
 #include <filesystem>
+#include <iterator>
+#include <limits>
 #include <Common/checkStackSize.h>
 #include <Common/CurrentMetrics.h>
 
@@ -30,22 +33,83 @@ namespace ErrorCodes
 
 KeeperDispatcher::KeeperDispatcher()
     : responses_queue(std::numeric_limits<size_t>::max())
+    , read_requests_queue(std::numeric_limits<size_t>::max())
+    , finalize_requests_queue(std::numeric_limits<size_t>::max())
     , configuration_and_settings(std::make_shared<KeeperConfigurationAndSettings>())
     , log(&Poco::Logger::get("KeeperDispatcher"))
 {
 }
 
 
+/// ZooKeepers has 2 requirements:
+/// - writes need to be linearizable
+/// - all requests from single session need to be processed in the order of their arrival
+///
+/// Because of that, we cannot process read and write requests from SAME session at the same time.
+/// To be able to process read and write requests in parallel we need to make sure that only 1 type
+/// of request is being processed from a single session.
+/// Multiple types from different sessions can be processed at the same time.
+///
+/// We do some in-session housekeeping to make sure that the multithreaded request processing is correct.
+/// When a request is received from a client, we check if there are requests being processed from that same
+/// session, and if yes, of what type. If the types are the same, and there are no requests of different
+/// type inbetetween, we can instanly add it to active request queue. Otherwise, we need to wait until
+/// all requests of the other type are processed.
+///
+/// There are multiple threads used for processing the request, each of them communicating with a queue.
+/// Assumption: only one type of request is being processed from a same session at any point in time (read or write).
+///
+/// requestThread -> requests currently being processed
+/// readRequestThread -> thread for processing read requests
+/// finalizeRequestThread -> thread for finalizing requests:
+///                          - in-session housekeeping, add requests to the active request queue if there are any
+///
+/// If reads are linearizable without quorum, a request can possibly wait for a certain log to be committed.
+/// In that case we add it to the waiting queue for that log.
+/// When that log is committed, the committing thread will send that read request to readRequestThread so it can be processed.
+///
 void KeeperDispatcher::requestThread()
 {
     setThreadName("KeeperReqT");
 
     /// Result of requests batch from previous iteration
-    RaftAppendResult prev_result = nullptr;
+    RaftResult prev_result = nullptr;
-    /// Requests from previous iteration. We store them to be able
+    const auto previous_quorum_done = [&] { return !prev_result || prev_result->has_result() || prev_result->get_result_code() != nuraft::cmd_result_code::OK; };
-    /// to send errors to the client.
-    KeeperStorage::RequestsForSessions prev_batch;
 
+    const auto needs_quorum = [](const auto & coordination_settings, const auto & request)
+    {
+        return coordination_settings->quorum_reads || coordination_settings->read_mode.toString() == "quorum" || !request.request->isReadRequest();
+    };
+
+    KeeperStorage::RequestsForSessions quorum_requests;
+    KeeperStorage::RequestsForSessions read_requests;
+
+    auto process_quorum_requests = [&, this]() mutable
+    {
+        /// Forcefully process all previous pending requests
+        if (prev_result)
+            forceWaitAndProcessResult(prev_result);
+
+        prev_result = server->putRequestBatch(quorum_requests);
+
+        if (prev_result)
+        {
+            prev_result->when_ready([&, requests_for_sessions = std::move(quorum_requests)](nuraft::cmd_result<nuraft::ptr<nuraft::buffer>> & result, nuraft::ptr<std::exception> &) mutable
+            {
+                if (!result.get_accepted() || result.get_result_code() == nuraft::cmd_result_code::TIMEOUT)
+                    addErrorResponses(requests_for_sessions, Coordination::Error::ZOPERATIONTIMEOUT);
+                else if (result.get_result_code() != nuraft::cmd_result_code::OK)
+                    addErrorResponses(requests_for_sessions, Coordination::Error::ZCONNECTIONLOSS);
+            });
+        }
+
+        quorum_requests.clear();
+    };
+
+    /// ZooKeeper requires that the requests inside a single session are processed in a strict order
+    /// (we cannot process later requests before all the previous once are processed)
+    /// By making sure that at this point we can either have just read requests or just write requests
+    /// from a single session, we can process them independently
     while (!shutdown_called)
     {
         KeeperStorage::RequestForSession request;
@@ -54,94 +118,67 @@ void KeeperDispatcher::requestThread()
         uint64_t max_wait = coordination_settings->operation_timeout_ms.totalMilliseconds();
         uint64_t max_batch_size = coordination_settings->max_requests_batch_size;
 
-        /// The code below do a very simple thing: batch all write (quorum) requests into vector until
-        /// previous write batch is not finished or max_batch size achieved. The main complexity goes from
-        /// the ability to process read requests without quorum (from local state). So when we are collecting
-        /// requests into a batch we must check that the new request is not read request. Otherwise we have to
-        /// process all already accumulated write requests, wait them synchronously and only after that process
-        /// read request. So reads are some kind of "separator" for writes.
         try
         {
-            if (requests_queue->tryPop(request, max_wait))
+            if (active_requests_queue->tryPop(request, max_wait))
             {
                 CurrentMetrics::sub(CurrentMetrics::KeeperOutstandingRequets);
                 if (shutdown_called)
                     break;
 
-                KeeperStorage::RequestsForSessions current_batch;
+                if (needs_quorum(coordination_settings, request))
-
+                    quorum_requests.emplace_back(request);
-                bool has_read_request = false;
+                else
-
+                    read_requests.emplace_back(request);
-                /// If new request is not read request or we must to process it through quorum.
-                /// Otherwise we will process it locally.
-                if (coordination_settings->quorum_reads || !request.request->isReadRequest())
-                {
-                    current_batch.emplace_back(request);
 
                 /// Waiting until previous append will be successful, or batch is big enough
                 /// has_result == false && get_result_code == OK means that our request still not processed.
                 /// Sometimes NuRaft set errorcode without setting result, so we check both here.
-                    while (prev_result && (!prev_result->has_result() && prev_result->get_result_code() == nuraft::cmd_result_code::OK) && current_batch.size() <= max_batch_size)
+                while (true)
                 {
+                    if (quorum_requests.size() > max_batch_size)
+                        break;
+
+                    if (read_requests.size() > max_batch_size)
+                    {
+                        processReadRequests(coordination_settings, read_requests);
+
+                        if (previous_quorum_done())
+                            break;
+                    }
+
                     /// Trying to get batch requests as fast as possible
-                        if (requests_queue->tryPop(request, 1))
+                    if (active_requests_queue->tryPop(request, 1))
                     {
                         CurrentMetrics::sub(CurrentMetrics::KeeperOutstandingRequets);
-                            /// Don't append read request into batch, we have to process them separately
+                        if (needs_quorum(coordination_settings, request))
-                            if (!coordination_settings->quorum_reads && request.request->isReadRequest())
+                            quorum_requests.emplace_back(request);
-                            {
+                        else
-                                has_read_request = true;
+                            read_requests.emplace_back(request);
-                                break;
                     }
                     else
                     {
+                        /// batch of read requests can send at most one request
+                        /// so we don't care if the previous batch hasn't received response
+                        if (!read_requests.empty())
+                            processReadRequests(coordination_settings, read_requests);
 
-                                current_batch.emplace_back(request);
+                        /// if we still didn't process previous batch we can
-                            }
+                        /// increase are current batch even more
+                        if (previous_quorum_done())
+                            break;
                     }
 
                     if (shutdown_called)
                         break;
                 }
-                }
-                else
-                    has_read_request = true;
 
                 if (shutdown_called)
                     break;
 
-                /// Forcefully process all previous pending requests
+                if (!quorum_requests.empty())
-                if (prev_result)
+                    process_quorum_requests();
-                    forceWaitAndProcessResult(prev_result, prev_batch);
 
-                /// Process collected write requests batch
-                if (!current_batch.empty())
-                {
-                    auto result = server->putRequestBatch(current_batch);
-
-                    if (result)
-                    {
-                        if (has_read_request) /// If we will execute read request next, than we have to process result now
-                            forceWaitAndProcessResult(result, current_batch);
-                    }
-                    else
-                    {
-                        addErrorResponses(current_batch, Coordination::Error::ZCONNECTIONLOSS);
-                        current_batch.clear();
-                    }
-
-                    prev_batch = std::move(current_batch);
-                    prev_result = result;
-                }
-
-                /// Read request always goes after write batch (last request)
-                if (has_read_request)
-                {
-                    if (server->isLeaderAlive())
-                        server->putLocalReadRequest(request);
-                    else
-                        addErrorResponses({request}, Coordination::Error::ZCONNECTIONLOSS);
-                }
             }
         }
         catch (...)
@@ -151,6 +188,72 @@ void KeeperDispatcher::requestThread()
     }
 }
 
+void KeeperDispatcher::processReadRequests(const CoordinationSettingsPtr & coordination_settings, KeeperStorage::RequestsForSessions & read_requests)
+{
+    if (coordination_settings->read_mode.toString() == "fastlinear")
+    {
+        // we just want to know what's the current latest committed log on Leader node
+        auto leader_info_result = server->getLeaderInfo();
+        if (leader_info_result)
+        {
+            leader_info_result->when_ready([&, requests_for_sessions = std::move(read_requests)](nuraft::cmd_result<nuraft::ptr<nuraft::buffer>> & result, nuraft::ptr<std::exception> & exception) mutable
+            {
+                if (!result.get_accepted() || result.get_result_code() == nuraft::cmd_result_code::TIMEOUT)
+                {
+                    addErrorResponses(requests_for_sessions, Coordination::Error::ZOPERATIONTIMEOUT);
+                    return;
+                }
+
+                if (result.get_result_code() != nuraft::cmd_result_code::OK)
+                {
+                    addErrorResponses(requests_for_sessions, Coordination::Error::ZCONNECTIONLOSS);
+                    return;
+                }
+
+                if (exception)
+                {
+                    LOG_INFO(log, "Got exception while waiting for read results {}", exception->what());
+                    addErrorResponses(requests_for_sessions, Coordination::Error::ZCONNECTIONLOSS);
+                    return;
+                }
+
+                auto & leader_info_ctx = result.get();
+
+                if (!leader_info_ctx)
+                {
+                    addErrorResponses(requests_for_sessions, Coordination::Error::ZCONNECTIONLOSS);
+                    return;
+                }
+
+                KeeperServer::NodeInfo leader_info;
+                leader_info.term = leader_info_ctx->get_ulong();
+                leader_info.last_committed_index = leader_info_ctx->get_ulong();
+                std::lock_guard lock(leader_waiter_mutex);
+                auto node_info = server->getNodeInfo();
+
+                /// we're behind, we need to wait
+                if (node_info.term < leader_info.term || node_info.last_committed_index < leader_info.last_committed_index)
+                {
+                    auto & leader_waiter = leader_waiters[leader_info];
+                    leader_waiter.insert(leader_waiter.end(), requests_for_sessions.begin(), requests_for_sessions.end());
+                    LOG_TRACE(log, "waiting for term {}, idx {}", leader_info.term, leader_info.last_committed_index);
+                }
+                /// process it in background thread
+                else if (!read_requests_queue.push(std::move(requests_for_sessions)))
+                    throw Exception(ErrorCodes::SYSTEM_ERROR, "Cannot push read requests to queue");
+            });
+        }
+    }
+    else
+    {
+        assert(coordination_settings->read_mode.toString() == "nonlinear");
+        if (!read_requests_queue.push(std::move(read_requests)))
+            throw Exception(ErrorCodes::SYSTEM_ERROR, "Cannot push read requests to queue");
+    }
+
+    read_requests.clear();
+}
+
 void KeeperDispatcher::responseThread()
 {
     setThreadName("KeeperRspT");
@@ -200,6 +303,65 @@ void KeeperDispatcher::snapshotThread()
     }
 }
 
+/// Background thread for processing read requests
+void KeeperDispatcher::readRequestThread()
+{
+    setThreadName("KeeperReadT");
+    while (!shutdown_called)
+    {
+        KeeperStorage::RequestsForSessions requests;
+        if (!read_requests_queue.pop(requests))
+            break;
+
+        if (shutdown_called)
+            break;
+
+        try
+        {
+            for (const auto & request_info : requests)
+            {
+                if (server->isLeaderAlive())
+                    server->putLocalReadRequest(request_info);
+                else
+                    addErrorResponses({request_info}, Coordination::Error::ZCONNECTIONLOSS);
+            }
+
+            if (!finalize_requests_queue.push(std::move(requests)))
+                throw Exception(ErrorCodes::SYSTEM_ERROR, "Cannot push read requests to queue");
+        }
+        catch (...)
+        {
+            tryLogCurrentException(__PRETTY_FUNCTION__);
+        }
+    }
+}
+
+/// We finalize requests every time we commit a single log with request
+/// or process a batch of read requests.
+/// Because it can get heavy, we do it in background thread.
+void KeeperDispatcher::finalizeRequestsThread()
+{
+    setThreadName("KeeperFinalT");
+    while (!shutdown_called)
+    {
+        KeeperStorage::RequestsForSessions requests;
+        if (!finalize_requests_queue.pop(requests))
+            break;
+
+        if (shutdown_called)
+            break;
+
+        try
+        {
+            finalizeRequests(requests);
+        }
+        catch (...)
+        {
+            tryLogCurrentException(__PRETTY_FUNCTION__);
+        }
+    }
+}
+
 void KeeperDispatcher::setResponse(int64_t session_id, const Coordination::ZooKeeperResponsePtr & response)
 {
     std::lock_guard lock(session_to_response_callback_mutex);
@@ -255,6 +417,30 @@ bool KeeperDispatcher::putRequest(const Coordination::ZooKeeperRequestPtr & requ
     request_info.time = duration_cast<milliseconds>(system_clock::now().time_since_epoch()).count();
     request_info.session_id = session_id;
 
+    {
+        std::lock_guard lock{unprocessed_request_mutex};
+        auto unprocessed_requests_it = unprocessed_requests_for_session.find(session_id);
+        if (unprocessed_requests_it == unprocessed_requests_for_session.end())
+        {
+            auto & unprocessed_requests = unprocessed_requests_for_session[session_id];
+            unprocessed_requests.unprocessed_num = 1;
+            unprocessed_requests.is_read = request->isReadRequest();
+        }
+        else
+        {
+            auto & unprocessed_requests = unprocessed_requests_it->second;
+
+            /// queue is not empty, or the request types don't match, put it in the waiting queue
+            if (!unprocessed_requests.request_queue.empty() || unprocessed_requests.is_read != request->isReadRequest())
+            {
+                unprocessed_requests.request_queue.push_back(std::move(request_info));
+                return true;
+            }
+
+            ++unprocessed_requests.unprocessed_num;
+        }
+    }
+
     std::lock_guard lock(push_request_mutex);
 
     if (shutdown_called)
@@ -263,10 +449,10 @@ bool KeeperDispatcher::putRequest(const Coordination::ZooKeeperRequestPtr & requ
     /// Put close requests without timeouts
     if (request->getOpNum() == Coordination::OpNum::Close)
     {
-        if (!requests_queue->push(std::move(request_info)))
+        if (!active_requests_queue->push(std::move(request_info)))
             throw Exception("Cannot push request to queue", ErrorCodes::SYSTEM_ERROR);
     }
-    else if (!requests_queue->tryPush(std::move(request_info), configuration_and_settings->coordination_settings->operation_timeout_ms.totalMilliseconds()))
+    else if (!active_requests_queue->tryPush(std::move(request_info), configuration_and_settings->coordination_settings->operation_timeout_ms.totalMilliseconds()))
     {
         throw Exception("Cannot push request to queue within operation timeout", ErrorCodes::TIMEOUT_EXCEEDED);
     }
@@ -279,13 +465,23 @@ void KeeperDispatcher::initialize(const Poco::Util::AbstractConfiguration & conf
     LOG_DEBUG(log, "Initializing storage dispatcher");
 
     configuration_and_settings = KeeperConfigurationAndSettings::loadFromConfig(config, standalone_keeper);
-    requests_queue = std::make_unique<RequestsQueue>(configuration_and_settings->coordination_settings->max_requests_batch_size);
+    active_requests_queue = std::make_unique<RequestsQueue>(configuration_and_settings->coordination_settings->max_requests_batch_size);
 
     request_thread = ThreadFromGlobalPool([this] { requestThread(); });
     responses_thread = ThreadFromGlobalPool([this] { responseThread(); });
     snapshot_thread = ThreadFromGlobalPool([this] { snapshotThread(); });
+    read_request_thread = ThreadFromGlobalPool([this] { readRequestThread(); });
+    finalize_requests_thread = ThreadFromGlobalPool([this] { finalizeRequestsThread(); });
 
-    server = std::make_unique<KeeperServer>(configuration_and_settings, config, responses_queue, snapshots_queue);
+    server = std::make_unique<KeeperServer>(
+        configuration_and_settings,
+        config,
+        responses_queue,
+        snapshots_queue,
+        [this](const KeeperStorage::RequestForSession & request_for_session, uint64_t log_term, uint64_t log_idx)
+        { onRequestCommit(request_for_session, log_term, log_idx); },
+        [this](uint64_t term, uint64_t last_idx)
+        { onApplySnapshot(term, last_idx); });
 
     try
     {
@@ -333,9 +529,9 @@ void KeeperDispatcher::shutdown()
             if (session_cleaner_thread.joinable())
                 session_cleaner_thread.join();
 
-            if (requests_queue)
+            if (active_requests_queue)
             {
-                requests_queue->finish();
+                active_requests_queue->finish();
 
                 if (request_thread.joinable())
                     request_thread.join();
@@ -349,6 +545,14 @@ void KeeperDispatcher::shutdown()
             if (snapshot_thread.joinable())
                 snapshot_thread.join();
 
+            read_requests_queue.finish();
+            if (read_request_thread.joinable())
+                read_request_thread.join();
+
+            finalize_requests_queue.finish();
+            if (finalize_requests_thread.joinable())
+                finalize_requests_thread.join();
+
             update_configuration_queue.finish();
             if (update_configuration_thread.joinable())
                 update_configuration_thread.join();
@@ -357,7 +561,7 @@ void KeeperDispatcher::shutdown()
         KeeperStorage::RequestForSession request_for_session;
 
         /// Set session expired for all pending requests
-        while (requests_queue && requests_queue->tryPop(request_for_session))
+        while (active_requests_queue && active_requests_queue->tryPop(request_for_session))
         {
             CurrentMetrics::sub(CurrentMetrics::KeeperOutstandingRequets);
             auto response = request_for_session.request->makeResponse();
@@ -474,7 +678,7 @@ void KeeperDispatcher::sessionCleanerTask()
                     };
                     {
                         std::lock_guard lock(push_request_mutex);
-                        if (!requests_queue->push(std::move(request_info)))
+                        if (!active_requests_queue->push(std::move(request_info)))
                             LOG_INFO(log, "Cannot push close request to queue while cleaning outdated sessions");
                         CurrentMetrics::add(CurrentMetrics::KeeperOutstandingRequets);
                     }
@@ -524,19 +728,12 @@ void KeeperDispatcher::addErrorResponses(const KeeperStorage::RequestsForSession
     }
 }
 
-void KeeperDispatcher::forceWaitAndProcessResult(RaftAppendResult & result, KeeperStorage::RequestsForSessions & requests_for_sessions)
+void KeeperDispatcher::forceWaitAndProcessResult(RaftResult & result)
 {
     if (!result->has_result())
         result->get();
 
-    /// If we get some errors, than send them to clients
-    if (!result->get_accepted() || result->get_result_code() == nuraft::cmd_result_code::TIMEOUT)
-        addErrorResponses(requests_for_sessions, Coordination::Error::ZOPERATIONTIMEOUT);
-    else if (result->get_result_code() != nuraft::cmd_result_code::OK)
-        addErrorResponses(requests_for_sessions, Coordination::Error::ZCONNECTIONLOSS);
-
     result = nullptr;
-    requests_for_sessions.clear();
 }
 
 int64_t KeeperDispatcher::getSessionID(int64_t session_timeout_ms)
@@ -584,7 +781,7 @@ int64_t KeeperDispatcher::getSessionID(int64_t session_timeout_ms)
     /// Push new session request to queue
     {
         std::lock_guard lock(push_request_mutex);
-        if (!requests_queue->tryPush(std::move(request_info), session_timeout_ms))
+        if (!active_requests_queue->tryPush(std::move(request_info), session_timeout_ms))
             throw Exception("Cannot push session id request to queue within session timeout", ErrorCodes::TIMEOUT_EXCEEDED);
         CurrentMetrics::add(CurrentMetrics::KeeperOutstandingRequets);
     }
@@ -657,6 +854,122 @@ void KeeperDispatcher::updateConfigurationThread()
     }
 }
 
+// Used to update the state for a session based on the requests
+// - update the number of current unprocessed requests for the session
+// - if the number of unprocessed requests is 0, we can start adding next type of requests
+//   from unprocessed requests queue to the active queue
+void KeeperDispatcher::finalizeRequests(const KeeperStorage::RequestsForSessions & requests_for_sessions)
+{
+    std::unordered_map<int64_t, size_t> counts_for_session;
+
+    for (const auto & request_for_session : requests_for_sessions)
+    {
+        ++counts_for_session[request_for_session.session_id];
+    }
+
+    std::lock_guard lock{unprocessed_request_mutex};
+    for (const auto [session_id, count] : counts_for_session)
+    {
+        auto unprocessed_requests_it = unprocessed_requests_for_session.find(session_id);
+        if (unprocessed_requests_it == unprocessed_requests_for_session.end())
+            continue;
+
+        auto & unprocessed_requests = unprocessed_requests_it->second;
+        unprocessed_requests.unprocessed_num -= count;
+
+        if (unprocessed_requests.unprocessed_num == 0)
+        {
+            if (!unprocessed_requests.request_queue.empty())
+            {
+                auto & unprocessed_requests_queue = unprocessed_requests.request_queue;
+                unprocessed_requests.is_read = !unprocessed_requests.is_read;
+                // start adding next type of requests
+                while (!unprocessed_requests_queue.empty() && unprocessed_requests_queue.front().request->isReadRequest() == unprocessed_requests.is_read)
+                {
+                    auto & front_request = unprocessed_requests_queue.front();
+
+                    /// Put close requests without timeouts
+                    if (front_request.request->getOpNum() == Coordination::OpNum::Close)
+                    {
+                        if (!active_requests_queue->push(std::move(front_request)))
+                            throw Exception("Cannot push request to queue", ErrorCodes::SYSTEM_ERROR);
+                    }
+                    else if (!active_requests_queue->tryPush(std::move(front_request), configuration_and_settings->coordination_settings->operation_timeout_ms.totalMilliseconds()))
+                    {
+                        throw Exception("Cannot push request to queue within operation timeout", ErrorCodes::TIMEOUT_EXCEEDED);
+                    }
+
+                    ++unprocessed_requests.unprocessed_num;
+                    unprocessed_requests_queue.pop_front();
+                }
+            }
+            else
+            {
+                unprocessed_requests_for_session.erase(unprocessed_requests_it);
+            }
+        }
+    }
+}
+
+// Finalize request
+// Process read requests that were waiting for this commit
+void KeeperDispatcher::onRequestCommit(const KeeperStorage::RequestForSession & request_for_session, uint64_t log_term, uint64_t log_idx)
+{
+    if (!finalize_requests_queue.push({request_for_session}))
+        throw Exception(ErrorCodes::SYSTEM_ERROR, "Cannot push read requests to queue");
+
+    KeeperStorage::RequestsForSessions requests;
+    {
+        std::lock_guard lock(leader_waiter_mutex);
+        auto request_queue_it = leader_waiters.find(KeeperServer::NodeInfo{.term = log_term, .last_committed_index = log_idx});
+        if (request_queue_it != leader_waiters.end())
+        {
+            requests = std::move(request_queue_it->second);
+            leader_waiters.erase(request_queue_it);
+        }
+    }
+
+    if (requests.empty())
+        return;
+
+    if (!read_requests_queue.push(std::move(requests)))
+        throw Exception(ErrorCodes::SYSTEM_ERROR, "Cannot push read requests to queue");
+}
+
+/// Process all read request that are waiting for lower or currently last processed log index
+void KeeperDispatcher::onApplySnapshot(uint64_t term, uint64_t last_idx)
+{
+    KeeperServer::NodeInfo current_node_info{term, last_idx};
+    KeeperStorage::RequestsForSessions requests;
+    {
+        std::lock_guard lock(leader_waiter_mutex);
+        for (auto leader_waiter_it = leader_waiters.begin(); leader_waiter_it != leader_waiters.end();)
+        {
+            auto waiting_node_info = leader_waiter_it->first;
+            if (waiting_node_info.term <= current_node_info.term
+                && waiting_node_info.last_committed_index <= current_node_info.last_committed_index)
+            {
+                for (auto & request : leader_waiter_it->second)
+                {
+                    requests.push_back(std::move(request));
+                }
+
+                leader_waiter_it = leader_waiters.erase(leader_waiter_it);
+            }
+            else
+            {
+                ++leader_waiter_it;
+            }
+        }
+    }
+
+    if (requests.empty())
+        return;
+
+    if (!read_requests_queue.push(std::move(requests)))
+        throw Exception(ErrorCodes::SYSTEM_ERROR, "Cannot push read requests to queue");
+}
+
 bool KeeperDispatcher::isServerActive() const
 {
     return checkInit() && hasLeader() && !server->isRecovering();
@@ -721,7 +1034,7 @@ Keeper4LWInfo KeeperDispatcher::getKeeper4LWInfo() const
     Keeper4LWInfo result = server->getPartiallyFilled4LWInfo();
     {
         std::lock_guard lock(push_request_mutex);
-        result.outstanding_requests_count = requests_queue->size();
+        result.outstanding_requests_count = active_requests_queue->size();
     }
     {
         std::lock_guard lock(session_to_response_callback_mutex);

src/Coordination/KeeperDispatcher.h

@@ -32,9 +32,12 @@ private:
     using UpdateConfigurationQueue = ConcurrentBoundedQueue<ConfigUpdateAction>;
 
     /// Size depends on coordination settings
-    std::unique_ptr<RequestsQueue> requests_queue;
+    /// Request currently being processed
+    std::unique_ptr<RequestsQueue> active_requests_queue;
     ResponsesQueue responses_queue;
     SnapshotsQueue snapshots_queue{1};
+    ConcurrentBoundedQueue<KeeperStorage::RequestsForSessions> read_requests_queue;
+    ConcurrentBoundedQueue<KeeperStorage::RequestsForSessions> finalize_requests_queue;
 
     /// More than 1k updates is definitely misconfiguration.
     UpdateConfigurationQueue update_configuration_queue{1000};
@@ -64,6 +67,8 @@ private:
     ThreadFromGlobalPool snapshot_thread;
     /// Apply or wait for configuration changes
     ThreadFromGlobalPool update_configuration_thread;
+    ThreadFromGlobalPool read_request_thread;
+    ThreadFromGlobalPool finalize_requests_thread;
 
     /// RAFT wrapper.
     std::unique_ptr<KeeperServer> server;
@@ -77,6 +82,34 @@ private:
     /// Counter for new session_id requests.
     std::atomic<int64_t> internal_session_id_counter{0};
 
+    /// A read request needs to have at least the log it was the last committed log on the leader
+    /// at the time the request was being made.
+    /// If the node is stale, we need to wait to commit that log before doing local read requests to achieve
+    /// linearizability.
+    std::unordered_map<KeeperServer::NodeInfo, KeeperStorage::RequestsForSessions> leader_waiters;
+    std::mutex leader_waiter_mutex;
+
+    /// We can be actively processing one type of requests (either read or write) from a single session.
+    /// If we receive a request of a type that is not currently being processed, we put it in the waiting queue.
+    /// Also, we want to process them in ariving order, so if we have a different type in the queue, we cannot process that request
+    /// but wait for all the previous requests to finish.
+    /// E.g. READ -> WRITE -> READ, the last READ will go to the waiting queue even though we are currently processing the first READ
+    /// because we have WRITE request before it that needs to be processed.
+    struct UnprocessedRequests
+    {
+        /// how many requests are currently in the active request queue
+        size_t unprocessed_num{0};
+        /// is_read currently being processed
+        bool is_read{false};
+        std::list<KeeperStorage::RequestForSession> request_queue;
+    };
+
+    // Called every time a batch of requests are processed.
+    void finalizeRequests(const KeeperStorage::RequestsForSessions & requests_for_sessions);
+
+    std::unordered_map<int64_t, UnprocessedRequests> unprocessed_requests_for_session;
+    std::mutex unprocessed_request_mutex;
+
     /// Thread put requests to raft
     void requestThread();
     /// Thread put responses for subscribed sessions
@@ -88,6 +121,12 @@ private:
     /// Thread apply or wait configuration changes from leader
     void updateConfigurationThread();
 
+    void readRequestThread();
+
+    void finalizeRequestsThread();
+
+    void processReadRequests(const CoordinationSettingsPtr & coordination_settings, KeeperStorage::RequestsForSessions & read_requests);
+
     void setResponse(int64_t session_id, const Coordination::ZooKeeperResponsePtr & response);
 
     /// Add error responses for requests to responses queue.
@@ -96,7 +135,7 @@ private:
 
     /// Forcefully wait for result and sets errors if something when wrong.
     /// Clears both arguments
-    void forceWaitAndProcessResult(RaftAppendResult & result, KeeperStorage::RequestsForSessions & requests_for_sessions);
+    static void forceWaitAndProcessResult(RaftResult & result);
 
 public:
     /// Just allocate some objects, real initialization is done by `intialize method`
@@ -116,6 +155,12 @@ public:
         return server && server->checkInit();
     }
 
+    /// Called when a single log with request is committed.
+    void onRequestCommit(const KeeperStorage::RequestForSession & request_for_session, uint64_t log_term, uint64_t log_idx);
+
+    /// Called when a snapshot is applied
+    void onApplySnapshot(uint64_t term, uint64_t last_idx);
+
     /// Is server accepting requests, i.e. connected to the cluster
     /// and achieved quorum
     bool isServerActive() const;

src/Coordination/KeeperServer.cpp

@@ -105,7 +105,9 @@ KeeperServer::KeeperServer(
     const KeeperConfigurationAndSettingsPtr & configuration_and_settings_,
     const Poco::Util::AbstractConfiguration & config,
     ResponsesQueue & responses_queue_,
-    SnapshotsQueue & snapshots_queue_)
+    SnapshotsQueue & snapshots_queue_,
+    KeeperStateMachine::CommitCallback commit_callback,
+    KeeperStateMachine::ApplySnapshotCallback apply_snapshot_callback)
     : server_id(configuration_and_settings_->server_id)
     , coordination_settings(configuration_and_settings_->coordination_settings)
     , log(&Poco::Logger::get("KeeperServer"))
@@ -113,7 +115,7 @@ KeeperServer::KeeperServer(
     , keeper_context{std::make_shared<KeeperContext>()}
     , create_snapshot_on_exit(config.getBool("keeper_server.create_snapshot_on_exit", true))
 {
-    if (coordination_settings->quorum_reads)
+    if (coordination_settings->quorum_reads || coordination_settings->read_mode.toString() == "quorum")
         LOG_WARNING(log, "Quorum reads enabled, Keeper will work slower.");
 
     keeper_context->digest_enabled = config.getBool("keeper_server.digest_enabled", false);
@@ -125,7 +127,9 @@ KeeperServer::KeeperServer(
         configuration_and_settings_->snapshot_storage_path,
         coordination_settings,
         keeper_context,
-        checkAndGetSuperdigest(configuration_and_settings_->super_digest));
+        checkAndGetSuperdigest(configuration_and_settings_->super_digest),
+        std::move(commit_callback),
+        std::move(apply_snapshot_callback));
 
     state_manager = nuraft::cs_new<KeeperStateManager>(
         server_id,
@@ -176,6 +180,13 @@ struct KeeperServer::KeeperRaftServer : public nuraft::raft_server
         reconfigure(new_config);
     }
 
+    RaftResult getLeaderInfo()
+    {
+        nuraft::ptr<nuraft::req_msg> req
+            = nuraft::cs_new<nuraft::req_msg>(0ull, nuraft::msg_type::leader_status_request, 0, 0, 0ull, 0ull, 0ull);
+        return send_msg_to_leader(req);
+    }
+
     void commit_in_bg() override
     {
         // For NuRaft, if any commit fails (uncaught exception) the whole server aborts as a safety
@@ -269,6 +280,20 @@ void KeeperServer::launchRaftServer(const Poco::Util::AbstractConfiguration & co
         coordination_settings->election_timeout_lower_bound_ms.totalMilliseconds(), "election_timeout_lower_bound_ms", log);
     params.election_timeout_upper_bound_ = getValueOrMaxInt32AndLogWarning(
         coordination_settings->election_timeout_upper_bound_ms.totalMilliseconds(), "election_timeout_upper_bound_ms", log);
+
+    params.leadership_expiry_ = getValueOrMaxInt32AndLogWarning(coordination_settings->leadership_expiry.totalMilliseconds(), "leadership_expiry", log);
+
+    if (coordination_settings->read_mode.toString() == "fastlinear")
+    {
+        if (params.leadership_expiry_ == 0)
+            params.leadership_expiry_ = params.election_timeout_lower_bound_;
+        else if (params.leadership_expiry_ > params.election_timeout_lower_bound_)
+        {
+            LOG_WARNING(log, "To use fast linearizable reads, leadership_expiry should be set to a value that is less or equal to the election_timeout_upper_bound_ms. "
+                    "Based on current settings, there are no guarantees for linearizability of reads.");
+        }
+    }
+
     params.reserved_log_items_ = getValueOrMaxInt32AndLogWarning(coordination_settings->reserved_log_items, "reserved_log_items", log);
     params.snapshot_distance_ = getValueOrMaxInt32AndLogWarning(coordination_settings->snapshot_distance, "snapshot_distance", log);
 
@@ -487,7 +512,7 @@ void KeeperServer::putLocalReadRequest(const KeeperStorage::RequestForSession &
     state_machine->processReadRequest(request_for_session);
 }
 
-RaftAppendResult KeeperServer::putRequestBatch(const KeeperStorage::RequestsForSessions & requests_for_sessions)
+RaftResult KeeperServer::putRequestBatch(const KeeperStorage::RequestsForSessions & requests_for_sessions)
 {
     std::vector<nuraft::ptr<nuraft::buffer>> entries;
     for (const auto & request_for_session : requests_for_sessions)
@@ -713,6 +738,20 @@ std::vector<int64_t> KeeperServer::getDeadSessions()
     return state_machine->getDeadSessions();
 }
 
+RaftResult KeeperServer::getLeaderInfo()
+{
+    std::lock_guard lock{server_write_mutex};
+    if (is_recovering)
+        return nullptr;
+
+    return raft_instance->getLeaderInfo();
+}
+
+KeeperServer::NodeInfo KeeperServer::getNodeInfo()
+{
+    return { .term = raft_instance->get_term(), .last_committed_index = state_machine->last_commit_index() };
+}
+
 ConfigUpdateActions KeeperServer::getConfigurationDiff(const Poco::Util::AbstractConfiguration & config)
 {
     auto diff = state_manager->getConfigurationDiff(config);

src/Coordination/KeeperServer.h

@@ -14,7 +14,7 @@
 namespace DB
 {
 
-using RaftAppendResult = nuraft::ptr<nuraft::cmd_result<nuraft::ptr<nuraft::buffer>>>;
+using RaftResult = nuraft::ptr<nuraft::cmd_result<nuraft::ptr<nuraft::buffer>>>;
 
 class KeeperServer
 {
@@ -71,7 +71,9 @@ public:
         const KeeperConfigurationAndSettingsPtr & settings_,
         const Poco::Util::AbstractConfiguration & config_,
         ResponsesQueue & responses_queue_,
-        SnapshotsQueue & snapshots_queue_);
+        SnapshotsQueue & snapshots_queue_,
+        KeeperStateMachine::CommitCallback commit_callback,
+        KeeperStateMachine::ApplySnapshotCallback apply_snapshot_callback);
 
     /// Load state machine from the latest snapshot and load log storage. Start NuRaft with required settings.
     void startup(const Poco::Util::AbstractConfiguration & config, bool enable_ipv6 = true);
@@ -84,7 +86,7 @@ public:
 
     /// Put batch of requests into Raft and get result of put. Responses will be set separately into
     /// responses_queue.
-    RaftAppendResult putRequestBatch(const KeeperStorage::RequestsForSessions & requests);
+    RaftResult putRequestBatch(const KeeperStorage::RequestsForSessions & requests);
 
     /// Return set of the non-active sessions
     std::vector<int64_t> getDeadSessions();
@@ -119,6 +121,17 @@ public:
 
     int getServerID() const { return server_id; }
 
+    struct NodeInfo
+    {
+        uint64_t term;
+        uint64_t last_committed_index;
+
+        bool operator==(const NodeInfo &) const = default;
+    };
+
+    RaftResult getLeaderInfo();
+    NodeInfo getNodeInfo();
+
     /// Get configuration diff between current configuration in RAFT and in XML file
     ConfigUpdateActions getConfigurationDiff(const Poco::Util::AbstractConfiguration & config);
 
@@ -126,10 +139,23 @@ public:
     /// Synchronously check for update results with retries.
     void applyConfigurationUpdate(const ConfigUpdateAction & task);
 
-
     /// Wait configuration update for action. Used by followers.
     /// Return true if update was successfully received.
     bool waitConfigurationUpdate(const ConfigUpdateAction & task);
 };
 
 }
+namespace std
+{
+    template <>
+    struct hash<DB::KeeperServer::NodeInfo>
+    {
+        size_t operator()(const DB::KeeperServer::NodeInfo & info) const
+        {
+            SipHash hash_state;
+            hash_state.update(info.term);
+            hash_state.update(info.last_committed_index);
+            return hash_state.get64();
+        }
+    };
+}

src/Coordination/KeeperStateMachine.cpp

@@ -44,7 +44,9 @@ KeeperStateMachine::KeeperStateMachine(
     const std::string & snapshots_path_,
     const CoordinationSettingsPtr & coordination_settings_,
     const KeeperContextPtr & keeper_context_,
-    const std::string & superdigest_)
+    const std::string & superdigest_,
+    CommitCallback commit_callback_,
+    ApplySnapshotCallback apply_snapshot_callback_)
     : coordination_settings(coordination_settings_)
     , snapshot_manager(
           snapshots_path_,
@@ -58,6 +60,8 @@ KeeperStateMachine::KeeperStateMachine(
     , last_committed_idx(0)
     , log(&Poco::Logger::get("KeeperStateMachine"))
     , superdigest(superdigest_)
+    , commit_callback(std::move(commit_callback_))
+    , apply_snapshot_callback(std::move(apply_snapshot_callback_))
     , keeper_context(keeper_context_)
 {
 }
@@ -223,11 +227,11 @@ bool KeeperStateMachine::preprocess(const KeeperStorage::RequestForSession & req
     return true;
 }
 
-nuraft::ptr<nuraft::buffer> KeeperStateMachine::commit(const uint64_t log_idx, nuraft::buffer & data)
+nuraft::ptr<nuraft::buffer> KeeperStateMachine::commit_ext(const ext_op_params & params)
 {
-    auto request_for_session = parseRequest(data);
+    auto request_for_session = parseRequest(*params.data);
     if (!request_for_session.zxid)
-        request_for_session.zxid = log_idx;
+        request_for_session.zxid = params.log_idx;
 
     /// Special processing of session_id request
     if (request_for_session.request->getOpNum() == Coordination::OpNum::SessionID)
@@ -272,8 +276,9 @@ nuraft::ptr<nuraft::buffer> KeeperStateMachine::commit(const uint64_t log_idx, n
             assertDigest(*request_for_session.digest, storage->getNodesDigest(true), *request_for_session.request, true);
     }
 
+    last_committed_idx = params.log_idx;
+    commit_callback(request_for_session, params.log_term, params.log_idx);
     ProfileEvents::increment(ProfileEvents::KeeperCommits);
-    last_committed_idx = log_idx;
     return nullptr;
 }
 
@@ -306,6 +311,7 @@ bool KeeperStateMachine::apply_snapshot(nuraft::snapshot & s)
 
     ProfileEvents::increment(ProfileEvents::KeeperSnapshotApplys);
     last_committed_idx = s.get_last_log_idx();
+    apply_snapshot_callback(s.get_last_log_term(), s.get_last_log_idx());
     return true;
 }
 
@@ -320,6 +326,10 @@ void KeeperStateMachine::commit_config(const uint64_t /* log_idx */, nuraft::ptr
 void KeeperStateMachine::rollback(uint64_t log_idx, nuraft::buffer & data)
 {
     auto request_for_session = parseRequest(data);
+
+    if (request_for_session.request->getOpNum() == Coordination::OpNum::SessionID)
+        return;
+
     // If we received a log from an older node, use the log_idx as the zxid
     // log_idx will always be larger or equal to the zxid so we can safely do this
     // (log_idx is increased for all logs, while zxid is only increased for requests)

src/Coordination/KeeperStateMachine.h

@@ -20,13 +20,18 @@ using SnapshotsQueue = ConcurrentBoundedQueue<CreateSnapshotTask>;
 class KeeperStateMachine : public nuraft::state_machine
 {
 public:
+    using CommitCallback = std::function<void(const KeeperStorage::RequestForSession &, uint64_t, uint64_t)>;
+    using ApplySnapshotCallback = std::function<void(uint64_t, uint64_t)>;
+
     KeeperStateMachine(
         ResponsesQueue & responses_queue_,
         SnapshotsQueue & snapshots_queue_,
         const std::string & snapshots_path_,
         const CoordinationSettingsPtr & coordination_settings_,
         const KeeperContextPtr & keeper_context_,
-        const std::string & superdigest_ = "");
+        const std::string & superdigest_ = "",
+        CommitCallback commit_callback_ = [](const KeeperStorage::RequestForSession &, uint64_t, uint64_t){},
+        ApplySnapshotCallback apply_snapshot_callback_ = [](uint64_t, uint64_t){});
 
     /// Read state from the latest snapshot
     void init();
@@ -37,7 +42,7 @@ public:
 
     nuraft::ptr<nuraft::buffer> pre_commit(uint64_t log_idx, nuraft::buffer & data) override;
 
-    nuraft::ptr<nuraft::buffer> commit(const uint64_t log_idx, nuraft::buffer & data) override; /// NOLINT
+    nuraft::ptr<nuraft::buffer> commit_ext(const ext_op_params & params) override; /// NOLINT
 
     /// Save new cluster config to our snapshot (copy of the config stored in StateManager)
     void commit_config(const uint64_t log_idx, nuraft::ptr<nuraft::cluster_config> & new_conf) override; /// NOLINT
@@ -145,6 +150,11 @@ private:
     /// Special part of ACL system -- superdigest specified in server config.
     const std::string superdigest;
 
+    /// call when a request is committed
+    const CommitCallback commit_callback;
+    /// call when snapshot is applied
+    const ApplySnapshotCallback apply_snapshot_callback;
+
     KeeperContextPtr keeper_context;
 };
 

src/Coordination/tests/gtest_coordination.cpp

@@ -1330,8 +1330,9 @@ void testLogAndStateMachine(Coordination::CoordinationSettingsPtr settings, uint
         changelog.append(entry);
         changelog.end_of_append_batch(0, 0);
 
-        state_machine->pre_commit(i, changelog.entry_at(i)->get_buf());
+        auto entry_buf = changelog.entry_at(i)->get_buf_ptr();
-        state_machine->commit(i, changelog.entry_at(i)->get_buf());
+        state_machine->pre_commit(i, *entry_buf);
+        state_machine->commit_ext(nuraft::state_machine::ext_op_params{i, entry_buf});
         bool snapshot_created = false;
         if (i % settings->snapshot_distance == 0)
         {
@@ -1375,8 +1376,9 @@ void testLogAndStateMachine(Coordination::CoordinationSettingsPtr settings, uint
 
     for (size_t i = restore_machine->last_commit_index() + 1; i < restore_changelog.next_slot(); ++i)
     {
-        restore_machine->pre_commit(i, changelog.entry_at(i)->get_buf());
+        auto entry = changelog.entry_at(i)->get_buf_ptr();
-        restore_machine->commit(i, changelog.entry_at(i)->get_buf());
+        restore_machine->pre_commit(i, *entry);
+        restore_machine->commit_ext(nuraft::state_machine::ext_op_params{i, entry});
     }
 
     auto & source_storage = state_machine->getStorage();
@@ -1477,18 +1479,18 @@ TEST_P(CoordinationTest, TestEphemeralNodeRemove)
     std::shared_ptr<ZooKeeperCreateRequest> request_c = std::make_shared<ZooKeeperCreateRequest>();
     request_c->path = "/hello";
     request_c->is_ephemeral = true;
-    auto entry_c = getLogEntryFromZKRequest(0, 1, state_machine->getNextZxid(), request_c);
+    auto entry_c = getLogEntryFromZKRequest(0, 1, state_machine->getNextZxid(), request_c)->get_buf_ptr();
-    state_machine->pre_commit(1, entry_c->get_buf());
+    state_machine->pre_commit(1, *entry_c);
-    state_machine->commit(1, entry_c->get_buf());
+    state_machine->commit_ext(nuraft::state_machine::ext_op_params{1, entry_c});
     const auto & storage = state_machine->getStorage();
 
     EXPECT_EQ(storage.ephemerals.size(), 1);
     std::shared_ptr<ZooKeeperRemoveRequest> request_d = std::make_shared<ZooKeeperRemoveRequest>();
     request_d->path = "/hello";
     /// Delete from other session
-    auto entry_d = getLogEntryFromZKRequest(0, 2, state_machine->getNextZxid(), request_d);
+    auto entry_d = getLogEntryFromZKRequest(0, 2, state_machine->getNextZxid(), request_d)->get_buf_ptr();
-    state_machine->pre_commit(2, entry_d->get_buf());
+    state_machine->pre_commit(2, *entry_d);
-    state_machine->commit(2, entry_d->get_buf());
+    state_machine->commit_ext(nuraft::state_machine::ext_op_params{2, entry_d});
 
     EXPECT_EQ(storage.ephemerals.size(), 0);
 }

src/Core/Settings.h

@@ -481,7 +481,7 @@ static constexpr UInt64 operator""_GiB(unsigned long long value)
     M(Bool, optimize_if_chain_to_multiif, false, "Replace if(cond1, then1, if(cond2, ...)) chains to multiIf. Currently it's not beneficial for numeric types.", 0) \
     M(Bool, optimize_multiif_to_if, true, "Replace 'multiIf' with only one condition to 'if'.", 0) \
     M(Bool, optimize_if_transform_strings_to_enum, false, "Replaces string-type arguments in If and Transform to enum. Disabled by default cause it could make inconsistent change in distributed query that would lead to its fail.", 0) \
-    M(Bool, optimize_monotonous_functions_in_order_by, true, "Replace monotonous function with its argument in ORDER BY", 0) \
+    M(Bool, optimize_monotonous_functions_in_order_by, false, "Replace monotonous function with its argument in ORDER BY", 0) \
     M(Bool, optimize_functions_to_subcolumns, false, "Transform functions to subcolumns, if possible, to reduce amount of read data. E.g. 'length(arr)' -> 'arr.size0', 'col IS NULL' -> 'col.null' ", 0) \
     M(Bool, optimize_using_constraints, false, "Use constraints for query optimization", 0)                                                                                                                                           \
     M(Bool, optimize_substitute_columns, false, "Use constraints for column substitution", 0)                                                                                                                                         \

src/Core/examples/coro.cpp

@@ -14,7 +14,7 @@
 
 namespace std // NOLINT(cert-dcl58-cpp)
 {
-    using namespace experimental::coroutines_v1;
+    using namespace experimental::coroutines_v1; // NOLINT(cert-dcl58-cpp)
 }
 
 #if __has_warning("-Wdeprecated-experimental-coroutine")

src/Disks/IO/CachedOnDiskReadBufferFromFile.cpp

@@ -143,9 +143,11 @@ void CachedOnDiskReadBufferFromFile::initialize(size_t offset, size_t size)
 }
 
 CachedOnDiskReadBufferFromFile::ImplementationBufferPtr
-CachedOnDiskReadBufferFromFile::getCacheReadBuffer(size_t offset) const
+CachedOnDiskReadBufferFromFile::getCacheReadBuffer(const FileSegment & file_segment) const
 {
-    auto path = cache->getPathInLocalCache(cache_key, offset, is_persistent);
+    /// Use is_persistent flag from in-memory state of the filesegment,
+    /// because it is consistent with what is written on disk.
+    auto path = file_segment.getPathInLocalCache();
 
     ReadSettings local_read_settings{settings};
     /// Do not allow to use asynchronous version of LocalFSReadMethod.
@@ -237,8 +239,6 @@ bool CachedOnDiskReadBufferFromFile::canStartFromCache(size_t current_offset, co
 CachedOnDiskReadBufferFromFile::ImplementationBufferPtr
 CachedOnDiskReadBufferFromFile::getReadBufferForFileSegment(FileSegmentPtr & file_segment)
 {
-    auto range = file_segment->range();
-
     auto download_state = file_segment->state();
     LOG_TEST(log, "getReadBufferForFileSegment: {}", file_segment->getInfoForLog());
 
@@ -247,7 +247,7 @@ CachedOnDiskReadBufferFromFile::getReadBufferForFileSegment(FileSegmentPtr & fil
         if (download_state == FileSegment::State::DOWNLOADED)
         {
             read_type = ReadType::CACHED;
-            return getCacheReadBuffer(range.left);
+            return getCacheReadBuffer(*file_segment);
         }
         else
         {
@@ -280,7 +280,7 @@ CachedOnDiskReadBufferFromFile::getReadBufferForFileSegment(FileSegmentPtr & fil
                     ///                     file_offset_of_buffer_end
 
                     read_type = ReadType::CACHED;
-                    return getCacheReadBuffer(range.left);
+                    return getCacheReadBuffer(*file_segment);
                 }
 
                 download_state = file_segment->wait();
@@ -289,7 +289,7 @@ CachedOnDiskReadBufferFromFile::getReadBufferForFileSegment(FileSegmentPtr & fil
             case FileSegment::State::DOWNLOADED:
             {
                 read_type = ReadType::CACHED;
-                return getCacheReadBuffer(range.left);
+                return getCacheReadBuffer(*file_segment);
             }
             case FileSegment::State::EMPTY:
             case FileSegment::State::PARTIALLY_DOWNLOADED:
@@ -305,7 +305,7 @@ CachedOnDiskReadBufferFromFile::getReadBufferForFileSegment(FileSegmentPtr & fil
                     ///                     file_offset_of_buffer_end
 
                     read_type = ReadType::CACHED;
-                    return getCacheReadBuffer(range.left);
+                    return getCacheReadBuffer(*file_segment);
                 }
 
                 auto downloader_id = file_segment->getOrSetDownloader();
@@ -323,7 +323,7 @@ CachedOnDiskReadBufferFromFile::getReadBufferForFileSegment(FileSegmentPtr & fil
 
                         read_type = ReadType::CACHED;
                         file_segment->resetDownloader();
-                        return getCacheReadBuffer(range.left);
+                        return getCacheReadBuffer(*file_segment);
                     }
 
                     if (file_segment->getCurrentWriteOffset() < file_offset_of_buffer_end)
@@ -339,7 +339,7 @@ CachedOnDiskReadBufferFromFile::getReadBufferForFileSegment(FileSegmentPtr & fil
                         LOG_TEST(log, "Predownload. File segment info: {}", file_segment->getInfoForLog());
                         chassert(file_offset_of_buffer_end > file_segment->getCurrentWriteOffset());
                         bytes_to_predownload = file_offset_of_buffer_end - file_segment->getCurrentWriteOffset();
-                        chassert(bytes_to_predownload < range.size());
+                        chassert(bytes_to_predownload < file_segment->range().size());
                     }
 
                     read_type = ReadType::REMOTE_FS_READ_AND_PUT_IN_CACHE;
@@ -354,7 +354,7 @@ CachedOnDiskReadBufferFromFile::getReadBufferForFileSegment(FileSegmentPtr & fil
                 if (canStartFromCache(file_offset_of_buffer_end, *file_segment))
                 {
                     read_type = ReadType::CACHED;
-                    return getCacheReadBuffer(range.left);
+                    return getCacheReadBuffer(*file_segment);
                 }
                 else
                 {

src/Disks/IO/CachedOnDiskReadBufferFromFile.h

@@ -68,7 +68,7 @@ private:
 
     ImplementationBufferPtr getReadBufferForFileSegment(FileSegmentPtr & file_segment);
 
-    ImplementationBufferPtr getCacheReadBuffer(size_t offset) const;
+    ImplementationBufferPtr getCacheReadBuffer(const FileSegment & file_segment) const;
 
     std::optional<size_t> getLastNonDownloadedOffset() const;
 

src/Disks/ObjectStorages/DiskObjectStorageMetadata.cpp

@@ -13,7 +13,6 @@ namespace DB
 namespace ErrorCodes
 {
     extern const int UNKNOWN_FORMAT;
-    extern const int LOGICAL_ERROR;
 }
 
 void DiskObjectStorageMetadata::deserialize(ReadBuffer & buf)
@@ -131,9 +130,6 @@ DiskObjectStorageMetadata::DiskObjectStorageMetadata(
 
 void DiskObjectStorageMetadata::addObject(const String & path, size_t size)
 {
-    if (!object_storage_root_path.empty() && path.starts_with(object_storage_root_path))
-        throw Exception(ErrorCodes::LOGICAL_ERROR, "Expected relative path");
-
     total_size += size;
     storage_objects.emplace_back(path, size);
 }

src/Formats/FormatFactory.cpp

@@ -530,6 +530,7 @@ String FormatFactory::getFormatFromFileDescriptor(int fd)
         return getFormatFromFileName(file_path, false);
     return "";
 #else
+    (void)fd;
     return "";
 #endif
 }

src/IO/ReadBufferFromFileDescriptor.cpp

@@ -233,7 +233,7 @@ void ReadBufferFromFileDescriptor::rewind()
 
 
 /// Assuming file descriptor supports 'select', check that we have data to read or wait until timeout.
-bool ReadBufferFromFileDescriptor::poll(size_t timeout_microseconds)
+bool ReadBufferFromFileDescriptor::poll(size_t timeout_microseconds) const
 {
     fd_set fds;
     FD_ZERO(&fds);

src/IO/ReadBufferFromFileDescriptor.h

@@ -66,7 +66,7 @@ public:
 
 private:
     /// Assuming file descriptor supports 'select', check that we have data to read or wait until timeout.
-    bool poll(size_t timeout_microseconds);
+    bool poll(size_t timeout_microseconds) const;
 };
 
 

src/Interpreters/DDLWorker.h

@@ -61,18 +61,23 @@ public:
         return host_fqdn_id;
     }
 
+    std::string getQueueDir() const
+    {
+        return queue_dir;
+    }
+
     void startup();
     virtual void shutdown();
 
     bool isCurrentlyActive() const { return initialized && !stop_flag; }
 
-protected:
 
     /// Returns cached ZooKeeper session (possibly expired).
     ZooKeeperPtr tryGetZooKeeper() const;
     /// If necessary, creates a new session and caches it.
     ZooKeeperPtr getAndSetZooKeeper();
 
+protected:
     /// Iterates through queue tasks in ZooKeeper, runs execution of new tasks
     void scheduleTasks(bool reinitialized);
 

src/Interpreters/JIT/compileFunction.cpp

@@ -739,7 +739,10 @@ CompiledAggregateFunctions compileAggregateFunctions(CHJIT & jit, const std::vec
     {
         compileCreateAggregateStatesFunctions(module, functions, create_aggregate_states_functions_name);
         compileAddIntoAggregateStatesFunctions(module, functions, add_aggregate_states_functions_name);
-        compileAddIntoAggregateStatesFunctionsSinglePlace(module, functions, add_aggregate_states_functions_name_single_place);
+        /// FIXME: this leads to use-of-uninitialized-value in llvm
+        /// But for now, it is safe, since it is not used by Aggregator anyway
+        (void)compileAddIntoAggregateStatesFunctionsSinglePlace;
+        /// compileAddIntoAggregateStatesFunctionsSinglePlace(module, functions, add_aggregate_states_functions_name_single_place);
         compileMergeAggregatesStates(module, functions, merge_aggregate_states_functions_name);
         compileInsertAggregatesIntoResultColumns(module, functions, insert_aggregate_states_functions_name);
     });
@@ -752,7 +755,7 @@ CompiledAggregateFunctions compileAggregateFunctions(CHJIT & jit, const std::vec
 
     assert(create_aggregate_states_function);
     assert(add_into_aggregate_states_function);
-    assert(add_into_aggregate_states_function_single_place);
+    /// assert(add_into_aggregate_states_function_single_place); /// FIXME
     assert(merge_aggregate_states_function);
     assert(insert_aggregate_states_function);
 

src/Processors/Formats/Impl/MsgPackRowInputFormat.cpp

@@ -2,6 +2,15 @@
 
 #if USE_MSGPACK
 
+/// FIXME: there is some issue with clang-15, that incorrectly detect a
+/// "Attempt to free released memory" in msgpack::unpack(), because of delete
+/// operator for zone (from msgpack/v1/detail/cpp11_zone.hpp), hence NOLINT
+///
+/// NOTE: that I was not able to suppress it locally, only with
+/// NOLINTBEGIN/NOLINTEND
+//
+// NOLINTBEGIN(clang-analyzer-cplusplus.NewDelete)
+
 #include <cstdlib>
 #include <Common/assert_cast.h>
 #include <IO/ReadHelpers.h>
@@ -235,8 +244,10 @@ static void insertNull(IColumn & column, DataTypePtr type)
     assert_cast<ColumnNullable &>(column).insertDefault();
 }
 
-static void insertUUID(IColumn & column, DataTypePtr /*type*/, const char * value, size_t size)
+static void insertUUID(IColumn & column, DataTypePtr type, const char * value, size_t size)
 {
+    if (!isUUID(type))
+        throw Exception(ErrorCodes::ILLEGAL_COLUMN, "Cannot insert MessagePack UUID into column with type {}.", type->getName());
     ReadBufferFromMemory buf(value, size);
     UUID uuid;
     readBinaryBigEndian(uuid.toUnderType().items[0], buf);
@@ -551,6 +562,8 @@ void registerMsgPackSchemaReader(FormatFactory & factory)
 
 }
 
+// NOLINTEND(clang-analyzer-cplusplus.NewDelete)
+
 #else
 
 namespace DB

src/Storages/HDFS/StorageHDFS.cpp

@@ -255,7 +255,7 @@ private:
 class HDFSSource::URISIterator::Impl
 {
 public:
-    explicit Impl(const std::vector<const String> & uris_, ContextPtr context)
+    explicit Impl(const std::vector<String> & uris_, ContextPtr context)
     {
         auto path_and_uri = getPathFromUriAndUriWithoutPath(uris_[0]);
         HDFSBuilderWrapper builder = createHDFSBuilder(path_and_uri.second + "/", context->getGlobalContext()->getConfigRef());
@@ -293,7 +293,7 @@ String HDFSSource::DisclosedGlobIterator::next()
     return pimpl->next();
 }
 
-HDFSSource::URISIterator::URISIterator(const std::vector<const String> & uris_, ContextPtr context)
+HDFSSource::URISIterator::URISIterator(const std::vector<String> & uris_, ContextPtr context)
     : pimpl(std::make_shared<HDFSSource::URISIterator::Impl>(uris_, context))
 {
 }

src/Storages/HDFS/StorageHDFS.h

@@ -86,7 +86,7 @@ private:
         const String & format_name,
         const ContextPtr & ctx);
 
-    std::vector<const String> uris;
+    std::vector<String> uris;
     String format_name;
     String compression_method;
     const bool distributed_processing;
@@ -116,7 +116,7 @@ public:
     class URISIterator
     {
         public:
-            URISIterator(const std::vector<const String> & uris_, ContextPtr context);
+            URISIterator(const std::vector<String> & uris_, ContextPtr context);
             String next();
         private:
             class Impl;

src/Storages/System/StorageSystemDDLWorkerQueue.cpp

@@ -205,9 +205,9 @@ static void fillStatusColumns(MutableColumns & res_columns, size_t & col,
 
 void StorageSystemDDLWorkerQueue::fillData(MutableColumns & res_columns, ContextPtr context, const SelectQueryInfo &) const
 {
-    zkutil::ZooKeeperPtr zookeeper = context->getZooKeeper();
+    auto& ddl_worker = context->getDDLWorker();
-    fs::path ddl_zookeeper_path = context->getConfigRef().getString("distributed_ddl.path", "/clickhouse/task_queue/ddl/");
+    fs::path ddl_zookeeper_path = ddl_worker.getQueueDir();
-
+    zkutil::ZooKeeperPtr zookeeper = ddl_worker.getAndSetZooKeeper();
     Strings ddl_task_paths = zookeeper->getChildren(ddl_zookeeper_path);
 
     GetResponseFutures ddl_task_futures;

tests/ci/build_check.py

@@ -291,7 +291,9 @@ def main():
 
     logging.info("Will try to fetch cache for our build")
     try:
-        get_ccache_if_not_exists(ccache_path, s3_helper, pr_info.number, TEMP_PATH)
+        get_ccache_if_not_exists(
+            ccache_path, s3_helper, pr_info.number, TEMP_PATH, pr_info.release_pr
+        )
     except Exception as e:
         # In case there are issues with ccache, remove the path and do not fail a build
         logging.info("Failed to get ccache, building without it. Error: %s", e)

tests/ci/ccache_utils.py

@@ -11,6 +11,7 @@ import requests  # type: ignore
 
 from compress_files import decompress_fast, compress_fast
 from env_helper import S3_DOWNLOAD, S3_BUILDS_BUCKET
+from s3_helper import S3Helper
 
 DOWNLOAD_RETRIES_COUNT = 5
 
@@ -57,12 +58,19 @@ def dowload_file_with_progress(url, path):
 
 
 def get_ccache_if_not_exists(
-    path_to_ccache_dir, s3_helper, current_pr_number, temp_path
+    path_to_ccache_dir: str,
+    s3_helper: S3Helper,
+    current_pr_number: int,
+    temp_path: str,
+    release_pr: int,
 ) -> int:
     """returns: number of PR for downloaded PR. -1 if ccache not found"""
     ccache_name = os.path.basename(path_to_ccache_dir)
     cache_found = False
     prs_to_check = [current_pr_number]
+    # Release PR is either 0 or defined
+    if release_pr:
+        prs_to_check.append(release_pr)
     ccache_pr = -1
     if current_pr_number != 0:
         prs_to_check.append(0)

tests/ci/ci_config.py

@@ -8,7 +8,7 @@ BuildConfig = Dict[str, ConfValue]
 CI_CONFIG = {
     "build_config": {
         "package_release": {
-            "compiler": "clang-14",
+            "compiler": "clang-15",
             "build_type": "",
             "sanitizer": "",
             "package_type": "deb",
@@ -19,7 +19,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "coverity": {
-            "compiler": "clang-14",
+            "compiler": "clang-15",
             "build_type": "",
             "sanitizer": "",
             "package_type": "coverity",
@@ -29,7 +29,7 @@ CI_CONFIG = {
             "official": False,
         },
         "package_aarch64": {
-            "compiler": "clang-14-aarch64",
+            "compiler": "clang-15-aarch64",
             "build_type": "",
             "sanitizer": "",
             "package_type": "deb",
@@ -40,7 +40,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "package_asan": {
-            "compiler": "clang-14",
+            "compiler": "clang-15",
             "build_type": "",
             "sanitizer": "address",
             "package_type": "deb",
@@ -49,7 +49,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "package_ubsan": {
-            "compiler": "clang-14",
+            "compiler": "clang-15",
             "build_type": "",
             "sanitizer": "undefined",
             "package_type": "deb",
@@ -67,7 +67,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "package_msan": {
-            "compiler": "clang-14",
+            "compiler": "clang-15",
             "build_type": "",
             "sanitizer": "memory",
             "package_type": "deb",
@@ -76,7 +76,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "package_debug": {
-            "compiler": "clang-14",
+            "compiler": "clang-15",
             "build_type": "debug",
             "sanitizer": "",
             "package_type": "deb",
@@ -85,7 +85,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "binary_release": {
-            "compiler": "clang-14",
+            "compiler": "clang-15",
             "build_type": "",
             "sanitizer": "",
             "package_type": "binary",
@@ -94,7 +94,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "binary_tidy": {
-            "compiler": "clang-14",
+            "compiler": "clang-15",
             "build_type": "debug",
             "sanitizer": "",
             "package_type": "binary",
@@ -104,7 +104,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "binary_shared": {
-            "compiler": "clang-14",
+            "compiler": "clang-15",
             "build_type": "",
             "sanitizer": "",
             "package_type": "binary",
@@ -113,7 +113,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "binary_darwin": {
-            "compiler": "clang-14-darwin",
+            "compiler": "clang-15-darwin",
             "build_type": "",
             "sanitizer": "",
             "package_type": "binary",
@@ -123,7 +123,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "binary_aarch64": {
-            "compiler": "clang-14-aarch64",
+            "compiler": "clang-15-aarch64",
             "build_type": "",
             "sanitizer": "",
             "package_type": "binary",
@@ -132,7 +132,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "binary_freebsd": {
-            "compiler": "clang-14-freebsd",
+            "compiler": "clang-15-freebsd",
             "build_type": "",
             "sanitizer": "",
             "package_type": "binary",
@@ -142,7 +142,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "binary_darwin_aarch64": {
-            "compiler": "clang-14-darwin-aarch64",
+            "compiler": "clang-15-darwin-aarch64",
             "build_type": "",
             "sanitizer": "",
             "package_type": "binary",
@@ -152,7 +152,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "binary_ppc64le": {
-            "compiler": "clang-14-ppc64le",
+            "compiler": "clang-15-ppc64le",
             "build_type": "",
             "sanitizer": "",
             "package_type": "binary",
@@ -162,7 +162,7 @@ CI_CONFIG = {
             "with_coverage": False,
         },
         "binary_amd64sse2": {
-            "compiler": "clang-14-amd64sse2",
+            "compiler": "clang-15-amd64sse2",
             "build_type": "",
             "sanitizer": "",
             "package_type": "binary",

tests/ci/fast_test_check.py

@@ -125,7 +125,7 @@ if __name__ == "__main__":
 
     logging.info("Will try to fetch cache for our build")
     ccache_for_pr = get_ccache_if_not_exists(
-        cache_path, s3_helper, pr_info.number, temp_path
+        cache_path, s3_helper, pr_info.number, temp_path, pr_info.release_pr
     )
     upload_master_ccache = ccache_for_pr in (-1, 0)
 

tests/ci/pr_info.py

@@ -86,7 +86,7 @@ class PRInfo:
         self.changed_files = set()  # type: Set[str]
         self.body = ""
         self.diff_urls = []
-        self.release_pr = ""
+        self.release_pr = 0
         ref = github_event.get("ref", "refs/head/master")
         if ref and ref.startswith("refs/heads/"):
             ref = ref[11:]

tests/integration/test_merge_tree_s3/configs/config.d/storage_conf.xml

@@ -38,6 +38,20 @@
                 <path>/jbod1/</path>
                 <max_size>1000000000</max_size>
             </s3_with_cache_and_jbod>
+            <s3_r>
+                <type>s3</type>
+                <endpoint>http://minio1:9001/root/data/</endpoint>
+                <access_key_id>minio</access_key_id>
+                <secret_access_key>minio123</secret_access_key>
+                <s3_max_single_part_upload_size>33554432</s3_max_single_part_upload_size>
+            </s3_r>
+            <s3_cache_r>
+                <type>cache</type>
+                <disk>s3_r</disk>
+                <path>/s3_cache_r/</path>
+                <max_size>1000000000</max_size>
+                <do_not_evict_index_and_mark_files>1</do_not_evict_index_and_mark_files>
+            </s3_cache_r>
         </disks>
         <policies>
             <s3>
@@ -78,6 +92,13 @@
                     </main>
                 </volumes>
             </s3_with_cache_and_jbod>
+            <s3_cache_r>
+                <volumes>
+                    <main>
+                        <disk>s3_cache_r</disk>
+                    </main>
+                </volumes>
+            </s3_cache_r>
         </policies>
     </storage_configuration>
 

tests/integration/test_merge_tree_s3/test.py

@@ -6,7 +6,6 @@ import pytest
 from helpers.cluster import ClickHouseCluster
 from helpers.utility import generate_values, replace_config, SafeThread
 
-
 SCRIPT_DIR = os.path.dirname(os.path.realpath(__file__))
 
 
@@ -36,6 +35,7 @@ def cluster():
                 "/jbod1:size=2M",
             ],
         )
+
         logging.info("Starting cluster...")
         cluster.start()
         logging.info("Cluster started")
@@ -742,3 +742,79 @@ def test_store_cleanup_disk_s3(cluster, node_name):
         "CREATE TABLE s3_test UUID '00000000-1000-4000-8000-000000000001' (n UInt64) Engine=MergeTree() ORDER BY n SETTINGS storage_policy='s3';"
     )
     node.query("INSERT INTO s3_test SELECT 1")
+
+
+@pytest.mark.parametrize("node_name", ["node"])
+def test_cache_setting_compatibility(cluster, node_name):
+    node = cluster.instances[node_name]
+
+    node.query("DROP TABLE IF EXISTS s3_test NO DELAY")
+
+    node.query(
+        "CREATE TABLE s3_test (key UInt32, value String) Engine=MergeTree() ORDER BY key SETTINGS storage_policy='s3_cache_r';"
+    )
+    node.query(
+        "INSERT INTO s3_test SELECT * FROM generateRandom('key UInt32, value String') LIMIT 500"
+    )
+
+    result = node.query("SYSTEM DROP FILESYSTEM CACHE")
+
+    result = node.query(
+        "SELECT count() FROM system.filesystem_cache WHERE cache_path LIKE '%persistent'"
+    )
+    assert int(result) == 0
+
+    node.query("SELECT * FROM s3_test")
+
+    result = node.query(
+        "SELECT count() FROM system.filesystem_cache WHERE cache_path LIKE '%persistent'"
+    )
+    assert int(result) > 0
+
+    config_path = os.path.join(
+        SCRIPT_DIR,
+        f"./{cluster.instances_dir_name}/node/configs/config.d/storage_conf.xml",
+    )
+
+    replace_config(
+        config_path,
+        "<do_not_evict_index_and_mark_files>1</do_not_evict_index_and_mark_files>",
+        "<do_not_evict_index_and_mark_files>0</do_not_evict_index_and_mark_files>",
+    )
+
+    result = node.query("DESCRIBE CACHE 's3_cache_r'")
+    assert result.strip().endswith("1")
+
+    node.restart_clickhouse()
+
+    result = node.query("DESCRIBE CACHE 's3_cache_r'")
+    assert result.strip().endswith("0")
+
+    result = node.query(
+        "SELECT count() FROM system.filesystem_cache WHERE cache_path LIKE '%persistent'"
+    )
+    assert int(result) > 0
+
+    node.query("SELECT * FROM s3_test FORMAT Null")
+
+    assert not node.contains_in_log("No such file or directory: Cache info:")
+
+    replace_config(
+        config_path,
+        "<do_not_evict_index_and_mark_files>0</do_not_evict_index_and_mark_files>",
+        "<do_not_evict_index_and_mark_files>1</do_not_evict_index_and_mark_files>",
+    )
+
+    result = node.query(
+        "SELECT count() FROM system.filesystem_cache WHERE cache_path LIKE '%persistent'"
+    )
+    assert int(result) > 0
+
+    node.restart_clickhouse()
+
+    result = node.query("DESCRIBE CACHE 's3_cache_r'")
+    assert result.strip().endswith("1")
+
+    node.query("SELECT * FROM s3_test FORMAT Null")
+
+    assert not node.contains_in_log("No such file or directory: Cache info:")

tests/integration/test_send_crash_reports/test.py

@@ -36,8 +36,10 @@ def started_node():
 
 
 def test_send_segfault(started_node):
+    # NOTE: another option is to increase waiting time.
     if (
         started_node.is_built_with_thread_sanitizer()
+        or started_node.is_built_with_address_sanitizer()
         or started_node.is_built_with_memory_sanitizer()
     ):
         pytest.skip("doesn't fit in timeouts for stacktrace generation")

tests/jepsen.clickhouse-keeper/resources/keeper_config.xml

@@ -21,7 +21,8 @@
             <heart_beat_interval_ms>1000</heart_beat_interval_ms>
             <election_timeout_lower_bound_ms>2000</election_timeout_lower_bound_ms>
             <election_timeout_upper_bound_ms>4000</election_timeout_upper_bound_ms>
-            <quorum_reads>{quorum_reads}</quorum_reads>
+            <quorum_reads>0</quorum_reads>
+            <read_mode>fastlinear</read_mode>
             <snapshot_distance>{snapshot_distance}</snapshot_distance>
             <stale_log_gap>{stale_log_gap}</stale_log_gap>
             <reserved_log_items>{reserved_log_items}</reserved_log_items>

tests/jepsen.clickhouse-keeper/src/jepsen/clickhouse_keeper/counter.clj

@@ -27,7 +27,12 @@
 
   (invoke! [this test op]
     (case (:f op)
-      :read (exec-with-retries 30 (fn []
+      :read (try
+             (assoc op
+               :type :ok
+               :value (count (zk-list conn root-path)))
+             (catch Exception _ (assoc op :type :info, :error :connect-error)))
+      :final-read (exec-with-retries 30 (fn []
                                     (assoc op
                                            :type :ok
                                            :value (count (zk-list conn root-path)))))
@@ -49,7 +54,5 @@
    :checker   (checker/compose
                 {:counter (checker/counter)
                  :perf    (checker/perf)})
-   :generator (->> (range)
+   :generator (gen/mix [r add])
-                   (map (fn [x]
+   :final-generator (gen/once {:type :invoke, :f :final-read, :value nil})})
-                          (->> (gen/mix [r add])))))
-   :final-generator (gen/once {:type :invoke, :f :read, :value nil})})

tests/jepsen.clickhouse-keeper/src/jepsen/clickhouse_keeper/db.clj

@@ -98,7 +98,6 @@
                          #"\{srv2\}" (get nodes 1)
                          #"\{srv3\}" (get nodes 2)
                          #"\{id\}" (str (inc (.indexOf nodes node)))
-                         #"\{quorum_reads\}" (str (boolean (:quorum test)))
                          #"\{snapshot_distance\}" (str (:snapshot-distance test))
                          #"\{stale_log_gap\}" (str (:stale-log-gap test))
                          #"\{reserved_log_items\}" (str (:reserved-log-items test))}]

tests/jepsen.clickhouse-keeper/src/jepsen/clickhouse_keeper/main.clj

@@ -103,7 +103,7 @@
         current-nemesis (get custom-nemesis/custom-nemesises (:nemesis opts))]
     (merge tests/noop-test
            opts
-           {:name (str "clickhouse-keeper-quorum=" quorum "-"  (name (:workload opts)) "-" (name (:nemesis opts)))
+           {:name (str "clickhouse-keeper-" (name (:workload opts)) "-" (name (:nemesis opts)))
             :os ubuntu/os
             :db (get-db opts)
             :pure-generators true

tests/jepsen.clickhouse-keeper/src/jepsen/clickhouse_keeper/register.clj

@@ -20,7 +20,8 @@
     (assoc this :conn (zk-connect node 9181 30000)))
 
   (setup! [this test]
-    (zk-create-range conn 300)) ; 300 nodes to be sure
+    (exec-with-retries 30 (fn []
+                            (zk-create-range conn 300))))
 
   (invoke! [_ test op]
     (let [[k v] (:value op)

tests/jepsen.clickhouse-keeper/src/jepsen/clickhouse_keeper/utils.clj

@@ -45,7 +45,7 @@
 
 (defn zk-connect
   [host port timeout]
-  (exec-with-retries 30 (fn [] (zk/connect (str host ":" port) :timeout-msec timeout))))
+  (zk/connect (str host ":" port) :timeout-msec timeout))
 
 (defn zk-create-range
   [conn n]

tests/performance/avg_weighted.xml

@@ -32,5 +32,21 @@
     <query>SELECT avgWeighted(num_u, num)   FROM perf_avg FORMAT Null</query>
     <query>SELECT avgWeighted(num_u, num_u) FROM perf_avg FORMAT Null</query>
 
+    <query>SELECT avgWeighted(num_f,                num_f)             FROM perf_avg FORMAT Null</query>
+    <query>SELECT avgWeighted(toNullable(num_f),    num_f)             FROM perf_avg FORMAT Null</query>
+    <query>SELECT avgWeighted(num_f,                toNullable(num_f)) FROM perf_avg FORMAT Null</query>
+    <query>SELECT avgWeighted(toNullable(num_f),    toNullable(num_f)) FROM perf_avg FORMAT Null</query>
+
+    <query>SELECT avgWeightedIf(num_f,                num_f,             num % 10) FROM perf_avg FORMAT Null</query>
+    <query>SELECT avgWeightedIf(toNullable(num_f),    num_f,             num % 10) FROM perf_avg FORMAT Null</query>
+    <query>SELECT avgWeightedIf(num_f,                toNullable(num_f), num % 10) FROM perf_avg FORMAT Null</query>
+    <query>SELECT avgWeightedIf(toNullable(num_f),    toNullable(num_f), num % 10) FROM perf_avg FORMAT Null</query>
+
+    <query>SELECT avgWeightedIf(num_f,                num_f,             toNullable(num) % 10) FROM perf_avg FORMAT Null</query>
+    <query>SELECT avgWeightedIf(toNullable(num_f),    num_f,             toNullable(num) % 10) FROM perf_avg FORMAT Null</query>
+    <query>SELECT avgWeightedIf(num_f,                toNullable(num_f), toNullable(num) % 10) FROM perf_avg FORMAT Null</query>
+    <query>SELECT avgWeightedIf(toNullable(num_f),    toNullable(num_f), toNullable(num) % 10) FROM perf_avg FORMAT Null</query>
+
+
     <drop_query>DROP TABLE IF EXISTS perf_avg</drop_query>
 </test>

tests/performance/uniq_stored.xml

@@ -0,0 +1,58 @@
+<test>
+    <create_query>
+        create table matview_1
+        (
+            a String,
+            b_count AggregateFunction(uniq, UInt64)
+        ) Engine=MergeTree partition by tuple()
+        ORDER by tuple()
+        SETTINGS index_granularity = 1024;
+    </create_query>
+
+    <create_query>
+        create table matview_10000
+        (
+            a String,
+            b_count AggregateFunction(uniq, String)
+        ) Engine=MergeTree partition by tuple()
+        ORDER by tuple()
+        SETTINGS index_granularity = 1024;
+    </create_query>
+
+
+    <drop_query>DROP TABLE IF EXISTS matview_1</drop_query>
+    <drop_query>DROP TABLE IF EXISTS matview_10000</drop_query>
+
+    <fill_query>
+        INSERT INTO matview_10000
+        SELECT a, uniqState(b) b_count
+        FROM
+        (
+            SELECT toString(rand() % 1000) a, toString(number % 10000) b
+            FROM numbers_mt(20000000)
+        )
+        GROUP BY a
+        SETTINGS max_insert_threads=8;
+    </fill_query>
+    <fill_query>OPTIMIZE TABLE matview_10000 FINAL</fill_query>
+
+    <fill_query>
+        INSERT INTO matview_1
+        SELECT '1', uniqState(number) b_count
+        FROM
+        (
+            SELECT *
+            FROM numbers_mt(2000000)
+        )
+        GROUP BY number
+        SETTINGS max_insert_threads=8;
+    </fill_query>
+    <fill_query>OPTIMIZE TABLE matview_1 FINAL</fill_query>
+
+    <!-- Test with ~10000 elements per state -->
+    <query>select a, uniqMerge(b_count) as b_count from matview_10000 prewhere a='55' group by a FORMAT Null SETTINGS max_threads=1;</query>
+    <query>select uniqMerge(b_count) as b_count from matview_10000 FORMAT Null SETTINGS max_threads=1;</query>
+
+    <!-- Test with ~1 elements per state -->
+    <query>select uniqMerge(b_count) as b_count FROM matview_1 FORMAT Null SETTINGS max_threads=1;</query>
+</test>

tests/queries/0_stateless/01321_monotonous_functions_in_order_by_bug.reference

@@ -0,0 +1,2 @@
+2020-01-01 01:00:00	1
+2020-01-01 01:00:00	999

tests/queries/0_stateless/01321_monotonous_functions_in_order_by_bug.sql

@@ -0,0 +1,7 @@
+SELECT
+    toStartOfHour(c1) AS _c1,
+    c2
+FROM values((toDateTime('2020-01-01 01:01:01'), 999), (toDateTime('2020-01-01 01:01:59'), 1))
+ORDER BY
+    _c1 ASC,
+    c2 ASC

tests/queries/0_stateless/01576_alias_column_rewrite.sql

@@ -17,7 +17,7 @@ INSERT INTO test_table(timestamp, value) SELECT toDateTime('2020-01-01 12:00:00'
 INSERT INTO test_table(timestamp, value) SELECT toDateTime('2020-01-02 12:00:00'), 1 FROM numbers(10);
 INSERT INTO test_table(timestamp, value) SELECT toDateTime('2020-01-03 12:00:00'), 1 FROM numbers(10);
 
-set optimize_respect_aliases = 1;
+set optimize_respect_aliases = 1, optimize_monotonous_functions_in_order_by = 1;
 SELECT 'test-partition-prune';
 
 SELECT COUNT() = 10 FROM test_table WHERE day = '2020-01-01' SETTINGS max_rows_to_read = 10;

tests/queries/0_stateless/02149_read_in_order_fixed_prefix.sql

@@ -56,7 +56,13 @@ ENGINE = MergeTree ORDER BY (toStartOfDay(dt), d);
 INSERT INTO t_read_in_order SELECT toDateTime('2020-10-10 00:00:00') + number, 1 / (number % 100 + 1), number FROM numbers(1000);
 
 EXPLAIN PIPELINE SELECT toStartOfDay(dt) as date, d FROM t_read_in_order ORDER BY date, round(d) LIMIT 5;
-SELECT toStartOfDay(dt) as date, d FROM t_read_in_order ORDER BY date, round(d) LIMIT 5;
+SELECT * from (
+  SELECT toStartOfDay(dt) as date, d FROM t_read_in_order ORDER BY date, round(d) LIMIT 50000000000
+  -- subquery with limit 50000000 to stabilize a test result and prevent order by d pushdown
+) order by d limit 5;
 
 EXPLAIN PIPELINE SELECT toStartOfDay(dt) as date, d FROM t_read_in_order ORDER BY date, round(d) LIMIT 5;
-SELECT toStartOfDay(dt) as date, d FROM t_read_in_order WHERE date = '2020-10-10' ORDER BY round(d) LIMIT 5;
+SELECT * from (
+  SELECT toStartOfDay(dt) as date, d FROM t_read_in_order WHERE date = '2020-10-10' ORDER BY round(d) LIMIT 50000000000
+  -- subquery with limit 50000000 to stabilize a test result and prevent order by d pushdown
+) order by d limit 5;

tests/queries/0_stateless/02417_null_variadic_behaviour.reference

@@ -0,0 +1,65 @@
+-- { echo }
+SELECT avgWeighted(number, number) t, toTypeName(t) FROM numbers(1);
+nan	Float64
+SELECT avgWeighted(number, number + 1) t, toTypeName(t) FROM numbers(0);
+nan	Float64
+SELECT avgWeighted(toNullable(number), number) t, toTypeName(t) FROM numbers(1);
+nan	Nullable(Float64)
+SELECT avgWeighted(if(number < 10000, NULL, number), number) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeighted(if(number < 50, NULL, number), number) t, toTypeName(t) FROM numbers(100);
+77.29530201342281	Nullable(Float64)
+SELECT avgWeighted(number, if(number < 10000, NULL, number)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeighted(number, if(number < 50, NULL, number)) t, toTypeName(t) FROM numbers(100);
+77.29530201342281	Nullable(Float64)
+SELECT avgWeighted(toNullable(number), if(number < 10000, NULL, number)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeighted(toNullable(number), if(number < 50, NULL, number)) t, toTypeName(t) FROM numbers(100);
+77.29530201342281	Nullable(Float64)
+SELECT avgWeighted(if(number < 10000, NULL, number), toNullable(number)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeighted(if(number < 50, NULL, number), toNullable(number)) t, toTypeName(t) FROM numbers(100);
+77.29530201342281	Nullable(Float64)
+SELECT avgWeighted(if(number < 10000, NULL, number), if(number < 10000, NULL, number)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeighted(if(number < 50, NULL, number), if(number < 10000, NULL, number)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeighted(if(number < 10000, NULL, number), if(number < 50, NULL, number)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeighted(if(number < 50, NULL, number), if(number < 50, NULL, number)) t, toTypeName(t) FROM numbers(100);
+77.29530201342281	Nullable(Float64)
+SELECT avgWeightedIf(number, number, number % 10) t, toTypeName(t) FROM numbers(100);
+66.63333333333334	Float64
+SELECT avgWeightedIf(number, number, toNullable(number % 10)) t, toTypeName(t) FROM numbers(100);
+66.63333333333334	Float64
+SELECT avgWeightedIf(number, number, if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+nan	Float64
+SELECT avgWeightedIf(number, number, if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+77.75555555555556	Float64
+SELECT avgWeightedIf(number, number, if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+66.63333333333334	Float64
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 10000, NULL, number), if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 10000, NULL, number), if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 50, NULL, number), if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 50, NULL, number), if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 10000, NULL, number), if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 10000, NULL, number), if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 50, NULL, number), if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 50, NULL, number), if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+77.75555555555556	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 10000, NULL, number), if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 10000, NULL, number), if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 50, NULL, number), if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+\N	Nullable(Float64)
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 50, NULL, number), if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+77.75555555555556	Nullable(Float64)

tests/queries/0_stateless/02417_null_variadic_behaviour.sql

@@ -0,0 +1,41 @@
+-- { echo }
+SELECT avgWeighted(number, number) t, toTypeName(t) FROM numbers(1);
+SELECT avgWeighted(number, number + 1) t, toTypeName(t) FROM numbers(0);
+
+SELECT avgWeighted(toNullable(number), number) t, toTypeName(t) FROM numbers(1);
+SELECT avgWeighted(if(number < 10000, NULL, number), number) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeighted(if(number < 50, NULL, number), number) t, toTypeName(t) FROM numbers(100);
+
+SELECT avgWeighted(number, if(number < 10000, NULL, number)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeighted(number, if(number < 50, NULL, number)) t, toTypeName(t) FROM numbers(100);
+
+SELECT avgWeighted(toNullable(number), if(number < 10000, NULL, number)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeighted(toNullable(number), if(number < 50, NULL, number)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeighted(if(number < 10000, NULL, number), toNullable(number)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeighted(if(number < 50, NULL, number), toNullable(number)) t, toTypeName(t) FROM numbers(100);
+
+SELECT avgWeighted(if(number < 10000, NULL, number), if(number < 10000, NULL, number)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeighted(if(number < 50, NULL, number), if(number < 10000, NULL, number)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeighted(if(number < 10000, NULL, number), if(number < 50, NULL, number)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeighted(if(number < 50, NULL, number), if(number < 50, NULL, number)) t, toTypeName(t) FROM numbers(100);
+
+SELECT avgWeightedIf(number, number, number % 10) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(number, number, toNullable(number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(number, number, if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(number, number, if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(number, number, if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 10000, NULL, number), if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 10000, NULL, number), if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 50, NULL, number), if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 50, NULL, number), if(number < 10000, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 10000, NULL, number), if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 10000, NULL, number), if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 50, NULL, number), if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 50, NULL, number), if(number < 50, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 10000, NULL, number), if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 10000, NULL, number), if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(if(number < 10000, NULL, number), if(number < 50, NULL, number), if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);
+SELECT avgWeightedIf(if(number < 50, NULL, number), if(number < 50, NULL, number), if(number < 0, NULL, number % 10)) t, toTypeName(t) FROM numbers(100);

tests/queries/0_stateless/02422_msgpack_uuid_wrong_column.sql

@@ -0,0 +1,4 @@
+-- Tags: no-parallel, no-fasttest
+
+insert into function file(02422_data.msgpack) select toUUID('f4cdd80d-5d15-4bdc-9527-adcca635ec1f') as uuid settings output_format_msgpack_uuid_representation='ext';
+select * from file(02422_data.msgpack, auto, 'x Int32'); -- {serverError ILLEGAL_COLUMN}