build: support building from aarch64 (#3807)

* build: support building from aarch64 - include pre-compiled thirdparty and zetasql for aarch64 - fix python sdk naming under aarch64 * docs(compile): briefly introduce arm building
4paradigm · Mar 22, 2024 · 0a5fa74 · 0a5fa74
1 parent 524dba1
commit 0a5fa74
Show file tree

Hide file tree

Showing 8 changed files with 96 additions and 17 deletions.
diff --git a/docker/Dockerfile b/docker/Dockerfile
@@ -15,7 +15,7 @@
 
 FROM centos:7
 
-ARG ZETASQL_VERSION=0.3.4-b0
+ARG ZETASQL_VERSION=0.3.4
 ARG THIRDPARTY_VERSION=0.7.0
 ARG TARGETARCH
 

diff --git a/docker/setup_deps.sh b/docker/setup_deps.sh
@@ -88,11 +88,10 @@ if [[ "$ARCH" = "x86_64" ]]; then
     curl -Lo zetasql.tar.gz "https://github.com/4paradigm/zetasql/releases/download/v${ZETASQL_VERSION}/libzetasql-${ZETASQL_VERSION}-linux-gnu-x86_64-centos.tar.gz" && \
         echo "downloaed zetasql.tar.gz version $ZETASQL_VERSION for $ARCH"
 elif [[ "$ARCH" = "aarch64" ]]; then
-    # NOTE(aceforeverd): thirdparty for arm is out-of-date
-    curl -Lo thirdparty.tar.gz https://github.com/4paradigm/hybridsql-asserts/releases/download/v0.4.0/thirdparty-2021-08-03-linux-gnu-aarch64.tar.gz && \
-        echo "downloaded thirdparty.tar.gz version 0.4.0 for $ARCH"
-    curl -Lo zetasql.tar.gz "https://github.com/4paradigm/zetasql/releases/download/v0.2.6/libzetasql-0.2.6-linux-gnu-aarch64-centos.tar.gz" && \
-        echo "downloaed zetasql.tar.gz for version 0.2.6 $ARCH"
+    curl -Lo thirdparty.tar.gz "https://github.com/4paradigm/hybridsql-asserts/releases/download/v${THIRDPARTY_VERSION}/thirdparty-${THIRDPARTY_VERSION}-linux-gnu-${ARCH}.tar.gz" && \
+        echo "downloaded thirdparty.tar.gz version $THIRDPARTY_VERSION for $ARCH"
+    curl -Lo zetasql.tar.gz "https://github.com/4paradigm/zetasql/releases/download/v${ZETASQL_VERSION}/libzetasql-${ZETASQL_VERSION}-linux-gnu-${ARCH}.tar.gz" && \
+        echo "downloaed zetasql.tar.gz version $ZETASQL_VERSION for $ARCH"
 else
     echo "no pre-compiled deps for arch=$ARCH"
     exit 1

diff --git a/docs/en/deploy/compile.md b/docs/en/deploy/compile.md
@@ -216,6 +216,34 @@ After forking the OpenMLDB repository, you can trigger the `Other OS Build` work
 
 Please note that this compilation process involves building third-party dependencies from source code, and it may take a while to complete due to limited resources. The approximate time for this process is around 3 hours and 5 minutes (2 hours for third-party dependencies and 1 hour for OpenMLDB). However, the workflow caches the compilation output for third-party dependencies, so the second compilation will be much faster, taking approximately 1 hour and 15 minutes for OpenMLDB.
 
+### Linux for ARM64
+
+It's possible to run OpenMLDB on Linux for ARM64(AArch64), which you may compile the source from scratch. ARM version of compile image `ghcr.io/4paradigm/hybridsql` is recommended:
+
+```sh
+docker run -it ghcr.io/4paradigm/hybridsql:latest
+
+# inside docker container
+git clone https://github.com/4paradigm/OpenMLDB
+cd OpenMLDB
+
+#  necessary deps for all third-party
+yum install -y flex autoconf automake unzip bc expect libtool \
+    rh-python38-python-devel gettext byacc xz tcl cppunit-devel rh-python38-python-wheel patch java-1.8.0-openjdk-devel
+# bazel
+curl --create-dirs -SLo /usr/local/bin/bazel https://github.com/bazelbuild/bazelisk/releases/download/v1.19.0/bazelisk-linux-arm64
+chmod +x /usr/local/bin/bazel
+
+# third-party
+cmake -S third-party -B .deps -DBUILD_BUNDELD=ON -DMAKEOPTS=-j$(nproc)
+cmake --build .deps
+
+# OpenMLDB source
+cmake -S . -B build -DCMKAE_PREFIX_PATH=$(pwd)/.deps/usr -DSQL_JAVASDK_ENABLE=ON -DSQL_PYSDK_ENABLE=ON
+cmake --build build -- -j$(nproc)
+```
+
+
 ### Macos 10.15, 11
 
 MacOS doesn't require compiling third-party dependencies from source code, so compilation is relatively faster, taking about 1 hour and 15 minutes. Local compilation is similar to the steps outlined in the [Detailed Instructions for Build](#detailed-instructions-for-build) and does not require compiling third-party dependencies (`BUILD_BUNDLED=OFF`). For cloud compilation on macOS, trigger the `Other OS Build` workflow in `Actions` with the specified macOS version (`os name` as `macos10` or `macos11`). You can also disable Java or Python SDK compilation if they are not needed, by setting `java sdk enable` or `python sdk enable` to `OFF`.

diff --git a/docs/zh/deploy/compile.md b/docs/zh/deploy/compile.md
@@ -210,6 +210,34 @@ Fork OpenMLDB仓库后，可以使用在`Actions`中触发workflow `Other OS Bui
 
 此编译流程需要从源码编译thirdparty，且资源较少，无法开启较高的并发编译。因此编译时间较长，大约需要3h5m（2h thirdparty+1h OpenMLDB）。workflow会缓存thirdparty的编译产出，因此第二次编译会快很多（1h15m OpenMLDB）。
 
+### Linux for ARM64
+
+支持在 AArch64 架构的 Linux 系统上编译, 建议选择编译镜像 `ghcr.io/4paradigm/hybridsql` 在容器内编译. 在容器内:
+
+```sh
+# 建议在编译镜像内进行
+docker run -it ghcr.io/4paradigm/hybridsql:latest
+ 
+# inside docker container
+git clone https://github.com/4paradigm/OpenMLDB
+cd OpenMLDB
+ 
+# 安装 third-party 编译 deps
+yum install -y flex autoconf automake unzip bc expect libtool \
+    rh-python38-python-devel gettext byacc xz tcl cppunit-devel rh-python38-python-wheel patch java-1.8.0-openjdk-devel
+# bazel
+curl --create-dirs -SLo /usr/local/bin/bazel https://github.com/bazelbuild/bazelisk/releases/download/v1.19.0/bazelisk-linux-arm64
+chmod +x /usr/local/bin/bazel
+ 
+# third-party
+cmake -S third-party -B .deps -DBUILD_BUNDELD=ON -DMAKEOPTS=-j$(nproc)
+cmake --build .deps
+ 
+# OpenMLDB source
+cmake -S . -B build -DCMKAE_PREFIX_PATH=$(pwd)/.deps/usr -DSQL_JAVASDK_ENABLE=ON -DSQL_PYSDK_ENABLE=ON
+cmake --build build -- -j$(nproc)
+```
+
 ### Macos 10.15, 11
 
 Macos适配不需要从源码编译thirdparty，所以云编译耗时不会太长，大约1h15m。本地编译与[从源码全量编译](#从源码全量编译)章节相同，无需编译thirdparty（`BUILD_BUNDLED=OFF`）。云编译需要在`Actions`中触发workflow `Other OS Build`，编译产出在`Actions`的`Artifacts`中。workflow 配置 `os name`为`macos10`/`macos11`，同样可配置`java sdk enable`或`python sdk enable`为`OFF`。
diff --git a/src/sdk/CMakeLists.txt b/src/sdk/CMakeLists.txt
@@ -130,7 +130,18 @@ if(SQL_PYSDK_ENABLE)
                 COMMAND ${CMAKE_COMMAND} -E copy $<TARGET_FILE:sql_router_sdk> ${PROJECT_SOURCE_DIR}/python/openmldb_sdk/openmldb/native/
                 COMMAND echo "Do not strip library for MacOS, refer to https://github.com/4paradigm/OpenMLDB/issues/905")
     else()
-        set(PYTHON_PLATFORM manylinux1_x86_64)
+        # ref https://peps.python.org/pep-0600/ 
+        # centos7 is glibc 2.17, so 'manylinux2014' should preferred, 
+        # 'manylinux2014' is just legacy alias 'manylinux_2_17', we keep this legacy name 
+        # since there still installation issue on python 3.8 for new the name.
+        if (CMAKE_SYSTEM_PROCESSOR MATCHES "(amd64)|(x86_64)")
+            set(PYTHON_PLATFORM manylinux1_x86_64)
+        elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "(arm64)|(ARM64)|(aarch64)|(AARCH64)")
+            set(PYTHON_PLATFORM manylinux2014_aarch64)
+        else()
+            message(FATAL_ERROR "unsupported architecture ${CMAKE_SYSTEM_PROCESSOR}")
+        endif()
+
         add_custom_target(strip_python_so ALL DEPENDS sql_router_sdk
             COMMAND ${CMAKE_COMMAND} -E copy $<TARGET_FILE:sql_router_sdk> ${PROJECT_SOURCE_DIR}/python/openmldb_sdk/openmldb/native/
             COMMAND strip ${PROJECT_SOURCE_DIR}/python/openmldb_sdk/openmldb/native/_sql_router_sdk.so)

diff --git a/third-party/CMakeLists.txt b/third-party/CMakeLists.txt
@@ -88,9 +88,8 @@ function(init_hybridsql_thirdparty_urls)
     get_linux_lsb_release_information()
 
     if(CMAKE_SYSTEM_PROCESSOR MATCHES "(arm64)|(ARM64)|(aarch64)|(AARCH64)")
-      # NOTE(#1138): pre-compiled thirdparty is out-of-date
-      set(HYBRIDSQL_ASSERTS_URL "https://github.com/4paradigm/hybridsql-asserts/releases/download/v0.4.0/thirdparty-2021-08-03-linux-gnu-aarch64.tar.gz" PARENT_SCOPE)
-      set(HYBRIDSQL_ASSERTS_HASH 128e2147903c6b4f5649284326c60ba048ff369f2925f9957f2973a515c65d75 PARENT_SCOPE)
+      set(HYBRIDSQL_ASSERTS_URL "${HYBRIDSQL_ASSERTS_HOME}/releases/download/v${HYBRIDSQL_ASSERTS_VERSION}/thirdparty-${HYBRIDSQL_ASSERTS_VERSION}-linux-gnu-aarch64.tar.gz" PARENT_SCOPE)
+      set(HYBRIDSQL_ASSERTS_HASH 323ca51573b46bb71b781a1a10a91a4112559fa54c18ae2d7bacc7e342f49f3d PARENT_SCOPE)
     else()
       if (LSB_RELEASE_ID_SHORT STREQUAL "centos")
         set(HYBRIDSQL_ASSERTS_URL "${HYBRIDSQL_ASSERTS_HOME}/releases/download/v${HYBRIDSQL_ASSERTS_VERSION}/thirdparty-${HYBRIDSQL_ASSERTS_VERSION}-linux-gnu-x86_64-centos.tar.gz" PARENT_SCOPE)

diff --git a/third-party/README.md b/third-party/README.md
@@ -82,6 +82,17 @@ The pre-compiled thirdparty for Linux is built on Centos7 with gcc8, and macOS i
 - make, autoreconf
 - pkg-config
 
+
+For Debian:
+```sh
+sudo apt-get install bison python3-dev libcppunit-dev build-essential cmake autoconf tcl pkg-config git curl patch libtool-bin unzip 
+# ensure python3 is the default, you may skip if it already is
+sudo update-alternatives --install /usr/bin/python python /usr/bin/python3 100
+
+curl --create-dirs -SLo /usr/local/bin/bazel https://github.com/bazelbuild/bazelisk/releases/download/v1.19.0/bazelisk-linux-amd64
+chmod +x /usr/local/bin/bazel
+```
+
 ### Build thirdparty
 
 ```bash

diff --git a/third-party/cmake/FetchZetasql.cmake b/third-party/cmake/FetchZetasql.cmake
@@ -13,16 +13,22 @@
 # limitations under the License.
 
 set(ZETASQL_HOME https://github.com/4paradigm/zetasql)
-set(ZETASQL_VERSION 0.3.4-b0)
-set(ZETASQL_HASH_DARWIN 0cd74b469396f877751cf7efd39685cc4fc544608aef4a901f3f640c3d8aa947)
-set(ZETASQL_HASH_LINUX_UBUNTU f94e97ad0b70b34d876cd3820f64756a164b061f3cb40a089e179f2997995bdc)
-set(ZETASQL_HASH_LINUX_CENTOS 8e853c4dd9b6638092357fa250ee29e634a1bc8b85ab8c829fbafac1b43ee8b8)
+set(ZETASQL_VERSION 0.3.4)
+set(ZETASQL_HASH_DARWIN a8d18f1595fa6a78d09c5feca813e9a332cd4d156a95124165f7c1a1ebcb86b2)
+set(ZETASQL_HASH_LINUX_UBUNTU 11988ad45de027566386a8e832e7385755f9367d78243b8d2aa6ef3493c0ee3d)
+set(ZETASQL_HASH_LINUX_CENTOS 92c9eb4d1ffb7d96fcb33150366906c52f27db4cf8dfb442c06e5fe5c738cd4a)
+set(ZETASQL_HASH_LINUX_AARCH64 e20e03938ec108b1c08203c238841ae781498dd66c5d791f4ecc326440287cdb)
 set(ZETASQL_TAG v${ZETASQL_VERSION})
 
 function(init_zetasql_urls)
   if (CMAKE_SYSTEM_NAME STREQUAL "Linux")
     get_linux_lsb_release_information()
 
+    if (CMAKE_SYSTEM_PROCESSOR MATCHES "(arm64)|(ARM64)|(aarch64)|(AARCH64)")
+      set(ZETASQL_URL "${ZETASQL_HOME}/releases/download/v${ZETASQL_VERSION}/libzetasql-${ZETASQL_VERSION}-linux-gnu-aarch64.tar.gz" PARENT_SCOPE)
+      set(ZETASQL_HASH ${ZETASQL_HASH_LINUX_AARCH64} PARENT_SCOPE)
+    endif()
+
     if (LSB_RELEASE_ID_SHORT STREQUAL "centos")
       set(ZETASQL_URL "${ZETASQL_HOME}/releases/download/v${ZETASQL_VERSION}/libzetasql-${ZETASQL_VERSION}-linux-gnu-x86_64-centos.tar.gz" PARENT_SCOPE)
       set(ZETASQL_HASH ${ZETASQL_HASH_LINUX_CENTOS} PARENT_SCOPE)
@@ -42,9 +48,6 @@ endfunction()
 if (NOT BUILD_BUNDLED_ZETASQL)
   init_zetasql_urls()
 
-  if (CMAKE_SYSTEM_PROCESSOR MATCHES "(arm64)|(ARM64)|(aarch64)|(AARCH64)")
-    message(FATAL_ERROR "pre-compiled zetasql for arm64 not available, try compile zetasql from source by cmake flag: '-DBUILD_BUNDLED_ZETASQL=ON'")
-  endif()
   message(STATUS "Download pre-compiled zetasql from ${ZETASQL_URL}")
   # download pre-compiled zetasql from GitHub Release
   ExternalProject_Add(zetasql