Changes to facilitate Linux build for KVService

CMakeLists.txt

@@ -17,6 +17,7 @@ if(UNIX)
     message(STATUS "Building for Linux: KVClient + KVPlayground")
     add_subdirectory(KVClient)
     add_subdirectory(KVPlayground)
+    add_subdirectory(KVService)
 elseif(WIN32)
     message(WARNING "For Windows builds, use KVService/build_with_local_sdk.ps1")
     message(WARNING "This root CMakeLists.txt is for Linux builds only")

KVPlayground/CMakeLists.txt

@@ -38,8 +38,13 @@ set_target_properties(KVPlayground PROPERTIES
     CXX_EXTENSIONS NO
 )
 
+find_package(Python3 REQUIRED)
+
 # Copy chunk.bin and conversation_tokens.json to output directory after build
 add_custom_command(TARGET KVPlayground POST_BUILD
+    COMMAND ${Python3_EXECUTABLE} ${CMAKE_CURRENT_LIST_DIR}/precompute_tokens.py
+            ${CMAKE_CURRENT_LIST_DIR}/conversation_template.txt
+            ${CMAKE_CURRENT_LIST_DIR}/conversation_tokens.json
     COMMAND ${CMAKE_COMMAND} -E copy_if_different
         ${CMAKE_CURRENT_SOURCE_DIR}/chunk.bin
         $<TARGET_FILE_DIR:KVPlayground>/chunk.bin

KVService/CMakeLists.txt

@@ -3,7 +3,57 @@ cmake_minimum_required(VERSION 3.15)
 # Option to use local Azure SDK build (for multi-NIC support)
 # NOTE: Local SDK must be built first with custom multi-NIC features
 # Set to ON to use local Azure SDK with multi-NIC support
-option(USE_LOCAL_AZURE_SDK "Use local Azure SDK build instead of vcpkg" ON)
+if(WIN32)
+	option(USE_LOCAL_AZURE_SDK "Use local Azure SDK build instead of vcpkg" ON)
+else()
+	option(USE_LOCAL_AZURE_SDK "Use local Azure SDK build instead of vcpkg" OFF)
+	# These are currently only tested for Linux builds.
+	# Set only one of tcmalloc or jemalloc.
+	option(USE_TCMALLOC "Use tcmalloc for malloc/free/new/delete" ON)
+	option(USE_JEMALLOC "Use jemalloc for malloc/free/new/delete" OFF)
+endif()
+
+if(USE_JEMALLOC)
+    find_package(PkgConfig REQUIRED)
+    pkg_search_module(JEMALLOC QUIET jemalloc)
+
+    if(NOT JEMALLOC_FOUND)
+        message(STATUS "jemalloc not found, trying to install libjemalloc-dev!")
+        execute_process(COMMAND sudo apt install -y libjemalloc-dev
+                RESULT_VARIABLE JEMALLOC_INSTALL_RESULT)
+        if(NOT JEMALLOC_INSTALL_RESULT EQUAL "0")
+            message(FATAL_ERROR "apt install libjemalloc-dev failed with ${JEMALLOC_INSTALL_RESULT}, try installing libjemalloc-dev manually and then run cmake again")
+        else()
+            # Call once more to ensure above install completed fine and also
+            # will set JEMALLOC_LIBRARY_DIRS variable.
+            pkg_search_module(JEMALLOC REQUIRED jemalloc)
+        endif()
+    endif()
+
+    # We want to link against the static jemalloc lib.
+    message(STATUS "Using jemalloc static lib ${JEMALLOC_LIBRARY_DIRS}/libjemalloc.a")
+endif()
+
+if(USE_TCMALLOC)
+    find_package(PkgConfig REQUIRED)
+    pkg_check_modules(TCMALLOC QUIET libtcmalloc)
+
+    if(NOT TCMALLOC_FOUND)
+        message(STATUS "tcmalloc not found, trying to install libgoogle-perftools-dev!")
+        execute_process(COMMAND sudo apt install -y libgoogle-perftools-dev
+                RESULT_VARIABLE TCMALLOC_INSTALL_RESULT)
+        if(NOT TCMALLOC_INSTALL_RESULT EQUAL "0")
+            message(FATAL_ERROR "apt install libgoogle-perftools-dev failed with ${TCMALLOC_INSTALL_RESULT}, try installing libgoogle-perftools-dev manually and then run cmake again")
+        else()
+            # Call once more to ensure above install completed fine and also
+            # will set tcmalloc_INCLUDE_DIR and TCMALLOC_LIBRARIES variables.
+            pkg_check_modules(TCMALLOC REQUIRED libtcmalloc)
+        endif()
+    else()
+        message(STATUS "Using tcmalloc lib ${TCMALLOC_LIBRARIES}")
+        message(STATUS "Using tcmalloc include dir ${TCMALLOC_INCLUDE_DIRS}")
+    endif()
+endif()
 
 # Configure local Azure SDK if enabled (BEFORE vcpkg toolchain)
 if(USE_LOCAL_AZURE_SDK)
@@ -147,11 +197,26 @@ else()
         Azure::azure-storage-blobs
         Azure::azure-identity
         Azure::azure-core
-        # Windows system libraries
-        winhttp
-        bcrypt
-        crypt32
     )
+
+    if(USE_JEMALLOC)
+        target_link_libraries(AzureStorageKVStoreLibV2 PUBLIC
+                          ${JEMALLOC_LIBRARY_DIRS}/libjemalloc.a)
+    endif()
+
+    if(USE_TCMALLOC)
+        target_link_libraries(AzureStorageKVStoreLibV2 PUBLIC
+                          ${TCMALLOC_LIBRARIES})
+    endif()
+
+    if(WIN32)
+        target_link_libraries(AzureStorageKVStoreLibV2 PUBLIC
+            # Windows system libraries
+            winhttp
+            bcrypt
+            crypt32
+    )
+    endif()
 endif()
 
 # KVStoreService library
@@ -182,6 +247,16 @@ target_link_libraries(KVStoreServiceLib PUBLIC
     protobuf::libprotobuf
 )
 
+if(USE_JEMALLOC)
+target_link_libraries(KVStoreServiceLib PUBLIC
+                      ${JEMALLOC_LIBRARY_DIRS}/libjemalloc.a)
+endif()
+
+if(USE_TCMALLOC)
+target_link_libraries(KVStoreServiceLib PUBLIC
+                      ${TCMALLOC_LIBRARIES})
+endif()
+
 # Force static runtime to match vcpkg libraries
 target_compile_options(KVStoreServiceLib PRIVATE
     $<$<CONFIG:Release>:/MT>
@@ -197,6 +272,15 @@ target_link_libraries(KVStoreServer PRIVATE
     KVStoreServiceLib
 )
 
+if(USE_JEMALLOC)
+	target_link_libraries(KVStoreServer PRIVATE
+                      ${JEMALLOC_LIBRARY_DIRS}/libjemalloc.a)
+endif()
+
+if(USE_TCMALLOC)
+	target_link_libraries(KVStoreServer PRIVATE
+                      ${TCMALLOC_LIBRARIES})
+endif()
 # Force static runtime to match vcpkg libraries
 target_compile_options(KVStoreServer PRIVATE
     $<$<CONFIG:Release>:/MT>

KVService/QUICKSTART.md

@@ -42,7 +42,7 @@ cmake --build build --config Release
 
 ```bash
 # Navigate to your kvStore directory
-cd ~/kvStore
+cd ~/kvStoreV2
 
 # Configure CMake
 cmake -B build -S . -DCMAKE_TOOLCHAIN_FILE=/path/to/vcpkg/scripts/buildsystems/vcpkg.cmake

KVService/README.md

@@ -267,7 +267,7 @@ Managed via vcpkg.json:
 ## Related Documentation
 
 - [ARCHITECTURE.md](../docs/ARCHITECTURE.md) - System architecture and deployment
-- [QUICKSTART.md](../docs/QUICKSTART.md) - Detailed setup guide
+- [QUICKSTART.md](QUICKSTART.md) - Detailed setup guide
 - [KVClient README](../KVClient/README.md) - Linux client library
 
 ## License

KVService/src/AzureStorageKVStoreLibV2.cpp

@@ -661,8 +661,8 @@ std::future<std::pair<bool, PromptChunk>> AzureStorageKVStoreLibV2::ReadAsync(co
             chunk.partitionKey = partitionKey;
             chunk.parentHash = parentHash;
             chunk.hash = hash;
-            chunk.buffer = buffer;
             chunk.bufferSize = buffer.size();
+            chunk.buffer = std::move(buffer);
 
             auto endTime = std::chrono::high_resolution_clock::now();
             auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(endTime - startTime).count();

README.md

@@ -141,7 +141,7 @@ export KVSTORE_GRPC_SERVER="your-windows-server:8085"
 ```bash
 sudo apt-get update
 sudo apt-get install -y build-essential cmake git pkg-config \
-    libssl-dev autoconf libtool curl unzip
+    libssl-dev autoconf libtool curl zip unzip python3-tiktoken
 ```
 
 2. **Install vcpkg**:
@@ -152,15 +152,15 @@ cd vcpkg
 ./vcpkg integrate install
 ```
 
-3. **Install gRPC and Protobuf**:
+3. **Install gRPC, Protobuf, Azure SDK and other important packages**:
 ```bash
-./vcpkg install grpc protobuf nlohmann-json
+./vcpkg install grpc protobuf nlohmann-json zlib azure-storage-blobs-cpp azure-identity-cpp
 ```
 
 ### Build Steps
 
 ```bash
-cd KVStoreV2
+cd kvStoreV2
 mkdir build && cd build
 cmake .. -DCMAKE_TOOLCHAIN_FILE=/path/to/vcpkg/scripts/buildsystems/vcpkg.cmake
 make -j$(nproc)

build-linux.sh

@@ -0,0 +1,10 @@
+#!/bin/bash
+
+# Navigate to your kvStore directory
+cd ~/kvStoreV2/
+
+# Configure CMake
+cmake -B build -S . -DCMAKE_TOOLCHAIN_FILE=/home/azureuser/vcpkg/scripts/buildsystems/vcpkg.cmake
+
+# Build
+cmake --build build --config Release

run-kvserver.sh

@@ -0,0 +1,14 @@
+#!/bin/bash
+
+#
+# On my VM the nic is bound to numa node 1, also one numa node is sufficient
+# to process the current load.
+# If one numa node is not sufficient then run one server process on each numa
+# node.
+#
+while :; do
+	numactl --cpunodebind=1 --membind=1 ./build/KVService/KVStoreServer --port 8085 --log-level error --disable-metrics --disable-multi-nic &
+	numactl --cpunodebind=1 --membind=1 ./build/KVService/KVStoreServer --port 8086 --log-level error --disable-metrics --disable-multi-nic &
+
+	wait
+done