Implement jemalloc extent hooks memory tracker (#1250)

Should improve/fix memory usage exceeds --memory-limit issues

libs/CMakeLists.txt

@@ -15,7 +15,6 @@ set(GFLAGS_NOTHREADS OFF)
 # NOTE: config/generate.py depends on the gflags help XML format.
 find_package(gflags REQUIRED)
 find_package(fmt 8.0.1)
-find_package(Jemalloc REQUIRED)
 find_package(ZLIB 1.2.11 REQUIRED)
 
 set(LIB_DIR ${CMAKE_CURRENT_SOURCE_DIR})
@@ -99,6 +98,17 @@ macro(import_external_library name type library_location include_dir)
   import_library(${name} ${type} ${${_upper_name}_LIBRARY} ${${_upper_name}_INCLUDE_DIR})
 endmacro(import_external_library)
 
+
+macro(set_path_external_library name type library_location include_dir)
+  string(TOUPPER ${name} _upper_name)
+  set(${_upper_name}_LIBRARY ${library_location} CACHE FILEPATH
+      "Path to ${name} library" FORCE)
+  set(${_upper_name}_INCLUDE_DIR ${include_dir} CACHE FILEPATH
+      "Path to ${name} include directory" FORCE)
+  mark_as_advanced(${name}_LIBRARY ${name}_INCLUDE_DIR)
+endmacro(set_path_external_library)
+
+
 # setup antlr
 import_external_library(antlr4 STATIC
   ${CMAKE_CURRENT_SOURCE_DIR}/antlr4/runtime/Cpp/lib/libantlr4-runtime.a
@@ -265,3 +275,8 @@ import_header_library(ctre ${CMAKE_CURRENT_SOURCE_DIR})
 # setup absl (cmake sub_directory tolerant)
 set(ABSL_PROPAGATE_CXX_STD ON)
 add_subdirectory(absl EXCLUDE_FROM_ALL)
+
+# set Jemalloc
+set_path_external_library(jemalloc STATIC
+  ${CMAKE_CURRENT_SOURCE_DIR}/jemalloc/lib/libjemalloc.a
+  ${CMAKE_CURRENT_SOURCE_DIR}/jemalloc/include/)

libs/setup.sh

@@ -124,6 +124,7 @@ declare -A primary_urls=(
   ["librdtsc"]="http://$local_cache_host/git/librdtsc.git"
   ["ctre"]="http://$local_cache_host/file/hanickadot/compile-time-regular-expressions/v3.7.2/single-header/ctre.hpp"
   ["absl"]="https://$local_cache_host/git/abseil-cpp.git"
+  ["jemalloc"]="https://$local_cache_host/git/jemalloc.git"
 )
 
 # The goal of secondary urls is to have links to the "source of truth" of
@@ -151,6 +152,7 @@ declare -A secondary_urls=(
   ["librdtsc"]="https://github.com/gabrieleara/librdtsc.git"
   ["ctre"]="https://raw.githubusercontent.com/hanickadot/compile-time-regular-expressions/v3.7.2/single-header/ctre.hpp"
   ["absl"]="https://github.com/abseil/abseil-cpp.git"
+  ["jemalloc"]="https://github.com/jemalloc/jemalloc.git"
 )
 
 # antlr
@@ -252,3 +254,21 @@ cd ..
 # abseil 20230125.3
 absl_ref="20230125.3"
 repo_clone_try_double "${primary_urls[absl]}" "${secondary_urls[absl]}" "absl" "$absl_ref"
+
+
+# jemalloc ea6b3e973b477b8061e0076bb257dbd7f3faa756
+JEMALLOC_COMMIT_VERSION="5.2.1"
+repo_clone_try_double "${secondary_urls[jemalloc]}" "${secondary_urls[jemalloc]}" "jemalloc" "$JEMALLOC_COMMIT_VERSION"
+
+# this is hack for cmake in libs to set path, and for FindJemalloc to use Jemalloc_INCLUDE_DIR
+pushd jemalloc
+
+./autogen.sh
+MALLOC_CONF="retain:false,percpu_arena:percpu,oversize_threshold:0,muzzy_decay_ms:5000,dirty_decay_ms:5000" \
+./configure \
+  --disable-cxx \
+  --enable-shared=no --prefix=$working_dir \
+  --with-malloc-conf="retain:false,percpu_arena:percpu,oversize_threshold:0,muzzy_decay_ms:5000,dirty_decay_ms:5000"
+
+make -j$CPUS install
+popd