Drop Centos7 support

[NvTimLiu]

To fix: #1991

Drop Centos7 support, switch to build in a Rocky 8 Docker image

Update the script to support both amd64 and arm64 CPUs

[NvTimLiu]

build

[jlowe]

Curious why we don't build GDS on arm64? It used to be separate but now is part of the CUDA toolkit. Is it not part of the arm64 CUDA toolkit?

[NvTimLiu]

Curious why we don't build GDS on arm64? It used to be separate but now is part of the CUDA toolkit. Is it not part of the arm64 CUDA toolkit?

Yes, GDS cuFiles RDMA lib is not in the arm64 CUDA toolkit

---

[INFO] [exec] Could NOT find cuFile (missing: cuFile_LIBRARY cuFileRDMA_LIBRARY
[INFO] [exec] cuFile_INCLUDE_DIR)

---

cufaultinj links to static cupti_static is not found in arm64 CUDA toolkit

[INFO] [exec] -- Generating done (0.0s)
[INFO] [exec] CMake Error at faultinj/CMakeLists.txt:34 (target_link_libraries):
[INFO] [exec] Target "cufaultinj" links to:
[INFO] [exec]
[INFO] [exec] CUDA::cupti_static
[INFO] [exec]
[INFO] [exec] but the target was not found. Possible reasons include:
[INFO] [exec]
[INFO] [exec] * There is a typo in the target name.
[INFO] [exec] * A find_package call is missing for an IMPORTED target.
[INFO] [exec] * An ALIAS target is missing.
[INFO] [exec]
[INFO] [exec]
[INFO] [exec]
[INFO] [exec] CMake Generate step failed. Build files cannot be regenerated correctly.

---

rmm OOM issue reported as below for arm64 test if USE_SANITIZER=OFF, but I've no idea what the root cause of the issue

[ERROR] There was an error in the forked process
[ERROR] Could not allocate native memory: std::bad_alloc: out_of_memory: RMM failure at:/home/nvidia/timl/spark-rapids-jni/thirdparty/cudf/cpp/build/_deps/rmm-src/include/rmm/mr/device/pool_memory_resource.hpp:254: Maximum pool size exceeded
[ERROR] org.apache.maven.surefire.booter.SurefireBooterForkException: There was an error in the forked process
[ERROR] Could not allocate native memory: std::bad_alloc: out_of_memory: RMM failure at:/home/nvidia/timl/spark-rapids-jni/thirdparty/cudf/cpp/build/_deps/rmm-src/include/rmm/mr/device/pool_memory_resource.hpp:254: Maximum pool size exceeded
[ERROR] at org.apache.maven.plugin.surefire.booterclient.ForkStarter.fork(ForkStarter.ja

[jlowe]

We need to fix the tests on arm64, not hack the build script to run the sanitizer to hide the problem. We will not be running the sanitizer in production, so letting this slip through by hacking the build script is not what we want.

[NvTimLiu]

build

[jlowe]

We need to fix the tests on arm64, not hack the build script to run the sanitizer to hide the problem. We will not be running the sanitizer in production, so letting this slip through by hacking the build script is not what we want.

[NvTimLiu]

@jlowe I've filed an issue for the rmm OOM UT failed on arm64 if USE_SANITIZER=OFF

Should we leave it unfixed and follow it up in the issue below?
We can merge this PR first to drop centOS7 support for 24.06 release since there're only 2-3 weeks remaining before v24.06 release, thanks!

#2022

[NvTimLiu]

build

[jlowe]

If tests are failing it's a blocker for this PR. Hacking the script to pass the test is not the right answer. The goal is to produce a good build on rocky8, but the failing tests indicate there may be issues with the build. The tests are there for a reason, to indicate whether the code is working properly. If the tests used to pass when building on centos7 but fail when building on rocky8 then there may be an issue with the code produced by the rocky8 build. We need to investigate the test failures and fix them.

[NvTimLiu]

build

[jlowe]

Converted this to a draft PR, as we don't want this merged until we get to the bottom of the arm64 test failures after building on rocky8, tracked by #2022.

[NvTimLiu]

Here's the relevant fix for the issue2022 : rapidsai/cudf#15706

[NvTimLiu]

build

[ttnghia]

Unfortunately this breaks my local build:

[INFO]      [exec] /usr/local/cmake-3.26.4-linux-x86_64/bin/cmake --regenerate-during-build -S/home/nghiat/Devel/jni/1/thirdparty/cudf/cpp -B/home/nghiat/Devel/jni/1/thirdparty/cudf/cpp/build
[INFO]      [exec] CMake Error at CMakeLists.txt:27 (project):
[INFO]      [exec]   The CMAKE_C_COMPILER:
[INFO]      [exec] 
[INFO]      [exec]     /opt/rh/devtoolset-11/root/usr/bin/cc
[INFO]      [exec] 
[INFO]      [exec]   is not a full path to an existing compiler tool.
[INFO]      [exec] 
[INFO]      [exec]   Tell CMake where to find the compiler by setting either the environment
[INFO]      [exec]   variable "CC" or the CMake cache entry CMAKE_C_COMPILER to the full path to
[INFO]      [exec]   the compiler, or to the compiler name if it is in the PATH.
[INFO]      [exec] 
[INFO]      [exec] 
[INFO]      [exec] CMake Error at CMakeLists.txt:27 (project):
[INFO]      [exec]   The CMAKE_CXX_COMPILER:
[INFO]      [exec] 
[INFO]      [exec]     /opt/rh/devtoolset-11/root/usr/bin/c++
[INFO]      [exec] 
[INFO]      [exec]   is not a full path to an existing compiler tool.
[INFO]      [exec] 
[INFO]      [exec]   Tell CMake where to find the compiler by setting either the environment
[INFO]      [exec]   variable "CXX" or the CMake cache entry CMAKE_CXX_COMPILER to the full path
[INFO]      [exec]   to the compiler, or to the compiler name if it is in the PATH.
[INFO]      [exec] 
[INFO]      [exec] 
[INFO]      [exec] ninja: error: rebuilding 'build.ninja': subcommand failed

[jlowe]

I suspect if you clean out thirdparty/cudf/cpp/build it will fix that issue. The compiler changed locations between the old and new image, so anything caching the compiler location (such as an existing cmake cache file) will break.

[ttnghia]

Oh yes, that fixes it. Thanks.