Upgrade RMM version to 23.06 (#346)

Contributes to nv-morpheus/Morpheus#995 Authors: - Christopher Harris (https://github.com/cwharris) - Michael Demoret (https://github.com/mdemoret-nv) Approvers: - Michael Demoret (https://github.com/mdemoret-nv) URL: #346
nv-morpheus · Jul 7, 2023 · 17a7c07 · 17a7c07
1 parent d96910b
commit 17a7c07
Show file tree

Hide file tree

Showing 28 changed files with 87 additions and 80 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -36,7 +36,7 @@ option(MRC_USE_CONDA "Enables finding dependencies via conda. All dependencies m
   environment" ON)
 option(MRC_USE_IWYU "Enable running include-what-you-use as part of the build process" OFF)
 
-set(MRC_RAPIDS_VERSION "23.02" CACHE STRING "Which version of RAPIDS to build for. Sets default versions for RAPIDS CMake and RMM.")
+set(MRC_RAPIDS_VERSION "23.06" CACHE STRING "Which version of RAPIDS to build for. Sets default versions for RAPIDS CMake and RMM.")
 
 set(MRC_CACHE_DIR "${CMAKE_SOURCE_DIR}/.cache" CACHE PATH "Directory to contain all CPM and CCache data")
 mark_as_advanced(MRC_CACHE_DIR)

diff --git a/ci/conda/environments/dev_env.yml b/ci/conda/environments/dev_env.yml
@@ -38,17 +38,17 @@ dependencies:
   - gflags=2.2
   - git>=2.35.3 # Needed for wildcards on safe.directory
   - glog=0.6
-  - gmock=1.10
+  - gmock=1.13
   - graphviz=3.0
-  - grpc-cpp=1.48
-  - gtest=1.10
+  - grpc-cpp=1.51
+  - gtest=1.13
   - gxx=11.2 # Specifies which versions of GXX and GCC to use
   - isort
   - jinja2=3.0
   - lcov=1.15
   - libhwloc=2.5
   - libprotobuf=3.21
-  - librmm=23.02
+  - librmm=23.06
   - libtool
   - ninja=1.10
   - nlohmann_json=3.9
@@ -61,9 +61,8 @@ dependencies:
   - pytest-timeout
   - python=3.10
   - scikit-build>=0.17
-  - spdlog=1.8.5
   - sysroot_linux-64=2.17
-  - ucx=1.13
+  - ucx=1.14
   - yapf
 
   # Remove once `mamba repoquery whoneeds cudatoolkit` is empty. For now, we need to specify a version

diff --git a/ci/conda/recipes/libmrc/build.sh b/ci/conda/recipes/libmrc/build.sh
@@ -53,7 +53,7 @@ export CXX=${GXX}
 # Common CMake args
 CMAKE_ARGS="-DBUILD_SHARED_LIBS=ON ${CMAKE_ARGS}"
 CMAKE_ARGS="-DCMAKE_BUILD_TYPE=${MRC_BUILD_TYPE} ${CMAKE_ARGS}"
-CMAKE_ARGS="-DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES=-"ALL"} ${CMAKE_ARGS}"
+CMAKE_ARGS="-DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES=-"RAPIDS"} ${CMAKE_ARGS}"
 CMAKE_ARGS="-DCMAKE_INSTALL_LIBDIR=lib ${CMAKE_ARGS}"
 CMAKE_ARGS="-DCMAKE_INSTALL_PREFIX=$PREFIX ${CMAKE_ARGS}"
 CMAKE_ARGS="-DCMAKE_MESSAGE_CONTEXT_SHOW=ON ${CMAKE_ARGS}"

diff --git a/ci/conda/recipes/libmrc/conda_build_config.yaml b/ci/conda/recipes/libmrc/conda_build_config.yaml
@@ -28,30 +28,36 @@ cuda_compiler_version:
 python:
   - 3.8
   - 3.10
+  - 3.10
 
 # Setup the dependencies to build with multiple versions of RAPIDS
-rapids_version: # Keep around compatibility with current version -1
-  - 22.12
+rapids_version: # Keep around compatibility with current version -2
   - 23.02
+  - 23.04
+  - 23.06
 
 # Multiple versions of abseil are required to satisfy the solver for some
 # environments. RAPIDS 22.06 only works with gRPC 1.45 and 22.08 only works with
 # 1.46. For each version of gRPC, support 2 abseil versions. Zip all of the keys
 # together to avoid impossible combinations
 libabseil:
-  - 20220623.0
+  - 20230125.0
+  - 20230125.0
   - 20230125.0
 
 grpc_cpp:
-  - 1.46
+  - 1.51
+  - 1.51
   - 1.51
 
 ucx:
   - 1.13
-  - 1.13
+  - 1.14
+  - 1.14
 
 libprotobuf:
-  - 3.20
+  - 3.21
+  - 3.21
   - 3.21
 
 zip_keys:

diff --git a/ci/conda/recipes/libmrc/meta.yaml b/ci/conda/recipes/libmrc/meta.yaml
@@ -29,7 +29,7 @@ source:
 build:
   number: {{ GIT_DESCRIBE_NUMBER }}
   script_env:
-    - CMAKE_CUDA_ARCHITECTURES=ALL
+    - CMAKE_CUDA_ARCHITECTURES=RAPIDS
     - MRC_CACHE_DIR
     - PARALLEL_LEVEL
   include_recipe: False
@@ -58,9 +58,9 @@ requirements:
     - flatbuffers 2.0.*
     - gflags
     - glog
-    - gmock 1.10.*
+    - gmock 1.13.*
     - grpc-cpp
-    - gtest 1.10.*
+    - gtest 1.13.*
     - libabseil
     - libhwloc 2.5.*
     - libprotobuf
@@ -70,7 +70,6 @@ requirements:
     - pybind11-stubgen 0.10.5
     - python {{ python }}
     - scikit-build >=0.17
-    - spdlog 1.8.5
     - ucx
     # Need to specify cudatoolkit to get correct version. Remove once all libraries migrate to cuda-toolkit
     - cudatoolkit {{ cuda_version }}.*

diff --git a/ci/conda/recipes/run_conda_build.sh b/ci/conda/recipes/run_conda_build.sh
@@ -92,7 +92,7 @@ fi
 # Choose default variants
 if hasArg quick; then
    # For quick build, just do most recent version of rapids
-   CONDA_ARGS_ARRAY+=("--variants" "{rapids_version: 23.02}")
+   CONDA_ARGS_ARRAY+=("--variants" "{rapids_version: 23.06}")
 fi
 
 # And default channels

diff --git a/ci/iwyu/mappings.imp b/ci/iwyu/mappings.imp
@@ -27,7 +27,10 @@
 
 # gprc
 { "include": ["@<grpcpp/impl/.*>", "private", "<grpcpp/grpcpp.h>", "public"] },
+{ "include": ["@<grpcpp/.*>", "private", "<grpcpp/grpcpp.h>", "public"] },
 { "include": ["@<grpc/impl/.*>", "private", "<grpcpp/grpcpp.h>", "public"] },
+{ "include": ["@<grpc/.*>", "private", "<grpcpp/grpcpp.h>", "public"] },
+{ "symbol": ["@grpc::.*", "private", "<grpcpp/grpcpp.h>", "public"] },
 
 # nlohmann json
 { "include": ["<nlohmann/detail/iterators/iter_impl.hpp>", "private", "<nlohmann/json.hpp>", "public"] },

diff --git a/cpp/mrc/include/mrc/memory/resources/arena_resource.hpp b/cpp/mrc/include/mrc/memory/resources/arena_resource.hpp
@@ -23,8 +23,8 @@
 #include <cuda_runtime_api.h>
 #include <glog/logging.h>
 #include <rmm/detail/error.hpp>
+#include <rmm/detail/logging_assert.hpp>
 #include <rmm/logger.hpp>
-#include <spdlog/common.h>
 
 #include <cstddef>
 #include <map>
@@ -107,12 +107,7 @@ class arena_resource final : public adaptor<Upstream>
       adaptor<Upstream>(std::move(upstream_mr)),
       global_arena_(std::make_shared<global_arena>(&this->resource(), initial_size, maximum_size)),
       dump_log_on_failure_(dump_log_on_failure)
-    {
-        if (dump_log_on_failure_)
-        {
-            logger_ = spdlog::basic_logger_mt("arena_memory_dump", "rmm_arena_memory_dump.log");
-        }
-    }
+    {}
 
     ~arena_resource() override = default;
 
@@ -316,8 +311,6 @@ class arena_resource final : public adaptor<Upstream>
     // std::map<cudaStream_t, arena> stream_arenas_;
     /// If true, dump memory information to log on allocation failure.
     bool dump_log_on_failure_;  // NOLINT
-    /// The logger for memory dump.
-    std::shared_ptr<spdlog::logger> logger_{};  // NOLINT
     /// Mutex for read and write locks.
     mutable std::shared_timed_mutex mtx_;  // NOLINT
 };

diff --git a/cpp/mrc/include/mrc/memory/resources/detail/arena.hpp b/cpp/mrc/include/mrc/memory/resources/detail/arena.hpp
@@ -21,8 +21,8 @@
 #include <rmm/cuda_stream_view.hpp>
 #include <rmm/detail/aligned.hpp>
 #include <rmm/detail/error.hpp>
+#include <rmm/detail/logging_assert.hpp>
 #include <rmm/logger.hpp>
-#include <spdlog/common.h>
 
 #include <algorithm>
 #include <cstddef>
@@ -400,29 +400,30 @@ class global_arena final
         }
     }
 
-    /**
-     * @brief Dump memory to log.
-     *
-     * @param logger the spdlog logger to use
-     */
-    void dump_memory_log(std::shared_ptr<spdlog::logger> const& logger) const
-    {
-        lock_guard lock(mtx_);
+    // /**
+    //  * @brief Dump memory to log.
+    //  *
+    //  * @param logger the spdlog logger to use
+    //  */
+    // void dump_memory_log(std::shared_ptr<spdlog::logger> const& logger) const
+    // {
+    //     lock_guard lock(mtx_);
 
-        logger->info("  Maximum size: {}", rmm::detail::bytes{maximum_size_});
-        logger->info("  Current size: {}", rmm::detail::bytes{current_size_});
+    //     logger->info("  Maximum size: {}", rmm::detail::bytes{maximum_size_});
+    //     logger->info("  Current size: {}", rmm::detail::bytes{current_size_});
 
-        logger->info("  # free blocks: {}", free_blocks_.size());
-        if (!free_blocks_.empty())
-        {
-            logger->info("  Total size of free blocks: {}", rmm::detail::bytes{total_block_size(free_blocks_)});
-            auto const largest_free = *std::max_element(free_blocks_.begin(), free_blocks_.end(), block_size_compare);
-            logger->info("  Size of largest free block: {}", rmm::detail::bytes{largest_free.size()});
-        }
+    //     logger->info("  # free blocks: {}", free_blocks_.size());
+    //     if (!free_blocks_.empty())
+    //     {
+    //         logger->info("  Total size of free blocks: {}", rmm::detail::bytes{total_block_size(free_blocks_)});
+    //         auto const largest_free = *std::max_element(free_blocks_.begin(), free_blocks_.end(),
+    //         block_size_compare); logger->info("  Size of largest free block: {}",
+    //         rmm::detail::bytes{largest_free.size()});
+    //     }
 
-        logger->info("  # upstream blocks={}", upstream_blocks_.size());
-        logger->info("  Total size of upstream blocks: {}", rmm::detail::bytes{total_block_size(upstream_blocks_)});
-    }
+    //     logger->info("  # upstream blocks={}", upstream_blocks_.size());
+    //     logger->info("  Total size of upstream blocks: {}", rmm::detail::bytes{total_block_size(upstream_blocks_)});
+    // }
 
   private:
     using lock_guard = std::lock_guard<std::mutex>;
@@ -565,22 +566,23 @@ class arena
         free_blocks_.clear();
     }
 
-    /**
-     * Dump memory to log.
-     *
-     * @param logger the spdlog logger to use
-     */
-    void dump_memory_log(std::shared_ptr<spdlog::logger> const& logger) const
-    {
-        lock_guard lock(mtx_);
-        logger->info("    # free blocks: {}", free_blocks_.size());
-        if (!free_blocks_.empty())
-        {
-            logger->info("    Total size of free blocks: {}", rmm::detail::bytes{total_block_size(free_blocks_)});
-            auto const largest_free = *std::max_element(free_blocks_.begin(), free_blocks_.end(), block_size_compare);
-            logger->info("    Size of largest free block: {}", rmm::detail::bytes{largest_free.size()});
-        }
-    }
+    // /**
+    //  * Dump memory to log.
+    //  *
+    //  * @param logger the spdlog logger to use
+    //  */
+    // void dump_memory_log(std::shared_ptr<spdlog::logger> const& logger) const
+    // {
+    //     lock_guard lock(mtx_);
+    //     logger->info("    # free blocks: {}", free_blocks_.size());
+    //     if (!free_blocks_.empty())
+    //     {
+    //         logger->info("    Total size of free blocks: {}", rmm::detail::bytes{total_block_size(free_blocks_)});
+    //         auto const largest_free = *std::max_element(free_blocks_.begin(), free_blocks_.end(),
+    //         block_size_compare); logger->info("    Size of largest free block: {}",
+    //         rmm::detail::bytes{largest_free.size()});
+    //     }
+    // }
 
   private:
     using lock_guard = std::lock_guard<std::mutex>;

diff --git a/cpp/mrc/src/internal/control_plane/client.cpp b/cpp/mrc/src/internal/control_plane/client.cpp
@@ -35,9 +35,7 @@
 #include "mrc/runnable/runner.hpp"
 
 #include <google/protobuf/any.pb.h>
-#include <grpcpp/completion_queue.h>
 #include <grpcpp/grpcpp.h>
-#include <grpcpp/security/credentials.h>
 #include <rxcpp/rx.hpp>
 
 #include <ostream>

diff --git a/cpp/mrc/src/internal/grpc/progress_engine.cpp b/cpp/mrc/src/internal/grpc/progress_engine.cpp
@@ -19,7 +19,6 @@
 
 #include <boost/fiber/operations.hpp>
 #include <glog/logging.h>
-#include <grpc/support/time.h>
 #include <grpcpp/grpcpp.h>
 
 #include <chrono>

diff --git a/cpp/mrc/src/internal/grpc/server.cpp b/cpp/mrc/src/internal/grpc/server.cpp
@@ -26,8 +26,6 @@
 #include "mrc/runnable/launcher.hpp"
 #include "mrc/runnable/runner.hpp"
 
-#include <grpcpp/security/server_credentials.h>
-
 #include <memory>
 #include <utility>
 

diff --git a/cpp/mrc/src/internal/memory/callback_adaptor.hpp b/cpp/mrc/src/internal/memory/callback_adaptor.hpp
@@ -20,6 +20,7 @@
 #include "mrc/memory/adaptors.hpp"
 
 #include <atomic>
+#include <functional>
 
 namespace mrc::memory {
 

diff --git a/cpp/mrc/src/internal/memory/device_resources.cpp b/cpp/mrc/src/internal/memory/device_resources.cpp
@@ -37,7 +37,6 @@
 
 #include <boost/fiber/future/future.hpp>
 #include <glog/logging.h>
-#include <spdlog/sinks/basic_file_sink.h>
 
 #include <map>
 #include <set>

diff --git a/cpp/mrc/src/internal/memory/host_resources.cpp b/cpp/mrc/src/internal/memory/host_resources.cpp
@@ -37,7 +37,6 @@
 
 #include <boost/fiber/future/future.hpp>
 #include <glog/logging.h>
-#include <spdlog/sinks/basic_file_sink.h>
 
 #include <map>
 #include <memory>

diff --git a/cpp/mrc/src/internal/ucx/context.cpp b/cpp/mrc/src/internal/ucx/context.cpp
@@ -19,6 +19,7 @@
 
 #include <glog/logging.h>
 #include <ucp/api/ucp.h>
+#include <ucp/api/ucp_compat.h>
 #include <ucp/api/ucp_def.h>
 #include <ucs/type/status.h>  // for ucs_status_string, UCS_OK
 

diff --git a/cpp/mrc/src/internal/ucx/endpoint.cpp b/cpp/mrc/src/internal/ucx/endpoint.cpp
@@ -24,7 +24,8 @@
 #include "mrc/types.hpp"
 
 #include <glog/logging.h>
-#include <ucp/api/ucp.h>      // for ucp_ep_close_nb, ucp_ep_create, UCP_EP_...
+#include <ucp/api/ucp.h>  // for ucp_ep_close_nb, ucp_ep_create, UCP_EP_...
+#include <ucp/api/ucp_compat.h>
 #include <ucp/api/ucp_def.h>  // for ucp_ep_params_t, ucp_address_t, ucp_ep
 #include <ucs/type/status.h>  // for ucs_status_string, UCS_OK, UCS_PTR_STATUS
 

diff --git a/cpp/mrc/src/internal/ucx/worker.cpp b/cpp/mrc/src/internal/ucx/worker.cpp
@@ -24,7 +24,8 @@
 #include "mrc/types.hpp"
 
 #include <glog/logging.h>
-#include <ucp/api/ucp.h>           // for ucp_*
+#include <ucp/api/ucp.h>  // for ucp_*
+#include <ucp/api/ucp_compat.h>
 #include <ucp/api/ucp_def.h>       // for ucp_worker_h
 #include <ucs/type/status.h>       // for ucs_status_string, UCS_OK
 #include <ucs/type/thread_mode.h>  // for UCS_THREAD_MODE_MULTI

diff --git a/cpp/mrc/src/tests/test_grpc.cpp b/cpp/mrc/src/tests/test_grpc.cpp
@@ -46,7 +46,6 @@
 #include <boost/fiber/future/future.hpp>
 #include <glog/logging.h>
 #include <grpcpp/grpcpp.h>
-#include <grpcpp/security/credentials.h>
 #include <gtest/gtest.h>
 #include <rxcpp/rx.hpp>
 
@@ -55,10 +54,14 @@
 #include <map>
 #include <memory>
 #include <ostream>
+#include <string>
 #include <thread>
 #include <utility>
 #include <vector>
 
+// Avoid forward declaring template specialization base classes
+// IWYU pragma: no_forward_declare grpc::ServerAsyncReaderWriter
+
 using namespace mrc;
 using namespace mrc::codable;
 

diff --git a/cpp/mrc/src/tests/test_memory.cpp b/cpp/mrc/src/tests/test_memory.cpp
@@ -34,7 +34,6 @@
 
 #include <glog/logging.h>
 #include <gtest/gtest.h>
-#include <spdlog/sinks/basic_file_sink.h>
 
 #include <array>
 #include <atomic>