dcgm: 3.3.5 -> 3.3.9; cudaPackages_10{,_0,_1,_2}: drop (#357655)

    After you run ALTER EXTENSION, you must run [this SQL script](https://github.com/timescale/timescaledb-extras/blob/master/utils/2.15.X-fix_hypertable_foreign_keys.sql). For more details, see the following pull requests [#6797](https://github.com/timescale/timescaledb/pull/6797).

    PostgreSQL 13 is no longer supported in TimescaleDB v2.16.

- Support for CUDA 10 has been dropped, as announced in the 24.11 release notes.

- `kanata` was updated to v1.7.0, which introduces several breaking changes.

  See the release notes of

  [v1.7.0](https://github.com/jtroo/kanata/releases/tag/v1.7.0)

, Accelerate, CoreGraphics, CoreVideo

, lmdbSupport ? true, lmdb

, leveldbSupport ? true, leveldb, snappy

, cudaSupport ? config.cudaSupport, cudaPackages ? { }

, cudnnSupport ? cudaSupport

, ncclSupport ? false

, pythonSupport ? false, python ? null, numpy ? null

, substituteAll

}:

let

  inherit (cudaPackages) backendStdenv cudatoolkit nccl;

  # The default for cudatoolkit 10.1 is CUDNN 8.0.5, the last version to support CUDA 10.1.

  # However, this caffe does not build with CUDNN 8.x, so we use CUDNN 7.6.5 instead.

  # Earlier versions of cudatoolkit use pre-8.x CUDNN, so we use the default.

  hasCudnn =

    if lib.versionOlder cudatoolkit.version "10.1"

    then cudaPackages ? cudnn

    else cudaPackages ? cudnn_7_6;

  toggle = bool: if bool then "ON" else "OFF";

  test_model_weights = fetchurl {

    # boost_python expects

    [ (if pythonSupport then "-Dpython_version=${python.pythonVersion}" else "-DBUILD_python=OFF")

      "-DBLAS=open"

    ] ++ (if cudaSupport then [

           "-DCUDA_ARCH_NAME=All"

           "-DCUDA_HOST_COMPILER=${backendStdenv.cc}/bin/cc"

         ] else [ "-DCPU_ONLY=ON" ])

      ++ ["-DUSE_NCCL=${toggle ncclSupport}"]

      ++ ["-DUSE_LEVELDB=${toggle leveldbSupport}"]

      "-DCPU_ONLY=ON"

    ] ++ ["-DUSE_LEVELDB=${toggle leveldbSupport}"]

      ++ ["-DUSE_LMDB=${toggle lmdbSupport}"];

  buildInputs = [ boost gflags glog protobuf hdf5-cpp opencv4 blas ]

                ++ lib.optional cudaSupport cudatoolkit

                ++ lib.optional (lib.versionOlder cudatoolkit.version "10.1" && hasCudnn) cudaPackages.cudnn

                ++ lib.optional (lib.versionAtLeast cudatoolkit.version "10.1" && hasCudnn) cudaPackages.cudnn_7_6

                ++ lib.optional lmdbSupport lmdb

                ++ lib.optional ncclSupport nccl

                ++ lib.optionals leveldbSupport [ leveldb snappy ]

                ++ lib.optionals pythonSupport [ python numpy ]

                ++ lib.optionals stdenv.hostPlatform.isDarwin [ Accelerate CoreGraphics CoreVideo ]

    substituteInPlace src/caffe/util/io.cpp --replace \

      'SetTotalBytesLimit(kProtoReadBytesLimit, 536870912)' \

      'SetTotalBytesLimit(kProtoReadBytesLimit)'

  '' + lib.optionalString (cudaSupport && lib.versionAtLeast cudatoolkit.version "9.0") ''

    # CUDA 9.0 doesn't support sm_20

    sed -i 's,20 21(20) ,,' cmake/Cuda.cmake

  '';

  preConfigure = lib.optionalString pythonSupport ''

    maintainers = [ ];

    broken =

      (pythonSupport && (python.isPy310))

      || cudaSupport

      || !(leveldbSupport -> (leveldb != null && snappy != null))

      || !(cudnnSupport -> (hasCudnn && cudaSupport))

      || !(ncclSupport -> (cudaSupport && !nccl.meta.unsupported))

      || !(pythonSupport -> (python != null && numpy != null))

    ;

    license = licenses.bsd2;

diff --git a/cmake/FindJsoncpp.cmake b/cmake/FindJsoncpp.cmake

index abedf7bb85..d7a52f5c7b 100644

--- a/cmake/FindJsoncpp.cmake

+++ b/cmake/FindJsoncpp.cmake

@@ -16,8 +16,8 @@

 if (NOT TARGET JsonCpp::JsonCpp)

     find_package(jsoncpp REQUIRED CONFIG)

-    set(JSONCPP_STATIC_LIBS jsoncpp_static)

-    set(JSONCPP_INCLUDE_PATH $<TARGET_PROPERTY:jsoncpp_static,INTERFACE_INCLUDE_DIRECTORIES>)

+    set(JSONCPP_STATIC_LIBS jsoncpp)

+    set(JSONCPP_INCLUDE_PATH $<TARGET_PROPERTY:jsoncpp,INTERFACE_INCLUDE_DIRECTORIES>)

 endif()

 # set(Jsoncpp_PATH_PREFIXES /usr/local "${Jsoncpp_ROOT}" "$ENV{HOME}")

 # foreach(prefix ${Jsoncpp_PATH_PREFIXES})

diff --git a/cmake/FindLibevent.cmake b/cmake/FindLibevent.cmake

index 354d6f9b99..38aca9534e 100644

--- a/cmake/FindLibevent.cmake

+++ b/cmake/FindLibevent.cmake

@@ -26,16 +26,12 @@

 endforeach()

 find_path(LIBEVENT_INCLUDE_DIR evhttp.h event.h PATHS ${Libevent_INCLUDE_PATHS})

-find_library(LIBEVENT_STATIC_LIB NAMES libevent.a libevent_core.a libevent_extra.a PATHS ${Libevent_LIB_PATHS})

-find_library(LIBEVENT_PTHREAD_STATIC_LIB NAMES libevent_pthreads.a PATHS ${Libevent_LIB_PATHS})

+find_library(LIBEVENT_STATIC_LIB NAMES libevent.so libevent_core.so libevent_extra.so PATHS ${Libevent_LIB_PATHS})

+find_library(LIBEVENT_PTHREAD_STATIC_LIB NAMES libevent_pthreads.so PATHS ${Libevent_LIB_PATHS})

-if (LIBEVENT_INCLUDE_DIR AND LIBEVENT_STATIC_LIB AND LIBEVENT_PTHREAD_STATIC_LIB)

+if (LIBEVENT_INCLUDE_DIR AND LIBEVENT_STATIC_LIB)

     set(Libevent_FOUND TRUE)

-    add_library(libevent_event_static STATIC IMPORTED)

-    set_target_properties(libevent_event_static PROPERTIES IMPORTED_LOCATION ${LIBEVENT_STATIC_LIB})

-    add_library(libevent_event_pthread STATIC IMPORTED)

-    set_target_properties(libevent_event_pthread PROPERTIES IMPORTED_LOCATION ${LIBEVENT_PTHREAD_STATIC_LIB})

-    set(LIBEVENT_STATIC_LIBS libevent_event_static libevent_event_pthread)

+    set(LIBEVENT_STATIC_LIBS ${LIBEVENT_STATIC_LIB} ${LIBEVENT_PTHREAD_STATIC_LIB})

 else ()

     set(Libevent_FOUND FALSE)

 endif ()

diff --git a/common/CudaWorker/DcgmDgemm.cpp b/common/CudaWorker/DcgmDgemm.cpp

index 8d33a3256e..6b3284258d 100644

--- a/common/CudaWorker/DcgmDgemm.cpp

+++ b/common/CudaWorker/DcgmDgemm.cpp

@@ -17,6 +17,7 @@

 #include <exception>

 #include <stdexcept>

+#include <cinttypes>

 #define CU_CHK(op)                                               \

     if (auto const status = op; status != CUBLAS_STATUS_SUCCESS) \

@@ -122,4 +123,4 @@

     return CUBLAS_STATUS_SUCCESS;

 }

-} // namespace DcgmNs

\ No newline at end of file

+} // namespace DcgmNs

diff --git a/common/DcgmError.h b/common/DcgmError.h

index 8638cdceb1..e8d817c0d4 100644

--- a/common/DcgmError.h

+++ b/common/DcgmError.h

@@ -17,6 +17,7 @@

 #include <sstream>

 #include <string>

+#include <cinttypes>

 #include <dcgm_agent.h>

 #include <dcgm_errors.h>

diff --git a/common/DcgmStringHelpers.cpp b/common/DcgmStringHelpers.cpp

index b41917e3b7..1fe63980c7 100644

--- a/common/DcgmStringHelpers.cpp

+++ b/common/DcgmStringHelpers.cpp

@@ -17,6 +17,7 @@

 #include <cstring>

 #include <string>

+#include <algorithm>

 /*****************************************************************************/

 void dcgmTokenizeString(const std::string &src, const std::string &delimiter, std::vector<std::string> &tokens)

diff --git a/dcgmi/CommandOutputController.cpp b/dcgmi/CommandOutputController.cpp

index 5057205564..8520171efa 100644

--- a/dcgmi/CommandOutputController.cpp

+++ b/dcgmi/CommandOutputController.cpp

@@ -24,6 +24,7 @@

 #include "dcgm_agent.h"

 #include <DcgmStringHelpers.h>

 #include <algorithm>

+#include <functional>

 #include <cstdarg>

 #include <iostream>

 #include <string>

diff --git a/dcgmi/Diag.h b/dcgmi/Diag.h

index a326f7b949..563fb3c9c0 100755

--- a/dcgmi/Diag.h

+++ b/dcgmi/Diag.h

@@ -24,6 +24,7 @@

 #define DIAG_H_

 #include <optional>

+#include <functional>

 #include "Command.h"

 #include "CommandOutputController.h"

diff --git a/hostengine/src/HostEngineOutput.cpp b/hostengine/src/HostEngineOutput.cpp

index 23c6ca9f54..798b83b3e4 100644

--- a/hostengine/src/HostEngineOutput.cpp

+++ b/hostengine/src/HostEngineOutput.cpp

@@ -20,6 +20,7 @@

 #include <iostream>

 #include <limits>

 #include <string_view>

+#include <unordered_map>

 namespace

 {

@@ -365,4 +366,4 @@

         }

     }

     os << std::endl;

-}

\ No newline at end of file

+}

diff --git a/nvvs/src/NvvsCommon.cpp b/nvvs/src/NvvsCommon.cpp

index 8f7888649b..1604d9dabe 100644

--- a/nvvs/src/NvvsCommon.cpp

+++ b/nvvs/src/NvvsCommon.cpp

@@ -15,6 +15,7 @@

  */

 #include <sstream>

 #include <stdexcept>

+#include <algorithm>

 #include <sys/stat.h>

 #include <sys/types.h>

diff --git a/sdk/nvidia/nvml/nvml_loader/nvml_loader.cpp b/sdk/nvidia/nvml/nvml_loader/nvml_loader.cpp

index 9eebeaf1c4..6e21201229 100644

--- a/sdk/nvidia/nvml/nvml_loader/nvml_loader.cpp

+++ b/sdk/nvidia/nvml/nvml_loader/nvml_loader.cpp

@@ -20,6 +20,7 @@

 #include <atomic>

 #include <mutex>

+#include <cstdlib>

 static void *g_nvmlLib                                     = 0;

 static std::atomic_uint32_t g_nvmlStaticLibResetHooksCount = 1;

{ lib

, gcc11Stdenv

, stdenv

, fetchFromGitHub

, autoAddDriverRunpath

, catch2

, cmake

, cudaPackages_10_2

, ninja

, cudaPackages_11_8

, cudaPackages_12

, boost

, fmt_9

, git

, jsoncpp

, symlinkJoin

, tclap_1_4

, yaml-cpp

, static ? gcc11Stdenv.hostPlatform.isStatic

}:

let

  # DCGM depends on 3 different versions of CUDA at the same time.

  # The runtime closure, thankfully, is quite small because most things

  # are statically linked.

  cudaPackageSetByVersion = [

    {

      version = "10";

      # Nixpkgs cudaPackages_10 doesn't have redist packages broken out.

      pkgSet = [

        cudaPackages_10_2.cudatoolkit

        cudaPackages_10_2.cudatoolkit.lib

      ];

    }

    {

      version = "11";

      pkgSet = getCudaPackages cudaPackages_11_8;

    }

    {

      version = "12";

      pkgSet = getCudaPackages cudaPackages_12;

    }

  # DCGM depends on 2 different versions of CUDA at the same time.

  # The runtime closure, thankfully, is quite small as it does not

  # include the CUDA libraries.

  cudaPackageSets = [

    cudaPackages_11_8

    cudaPackages_12

  ];

  # Select needed redist packages from cudaPackages

    libcurand

  ];

  # Builds CMake code to add CUDA paths for include and lib.

  mkAppendCudaPaths = { version, pkgSet }:

  # Builds CMake flags to add CUDA paths for include and lib.

  mkCudaFlags = cudaPackages:

    let

      version = cudaPackages.cudaMajorVersion;

      # The DCGM CMake assumes that the folder containing cuda.h contains all headers, so we must

      # combine everything together for headers to work.

      # It would be more convenient to use symlinkJoin on *just* the include subdirectories

      # of each package, but not all of them have an include directory and making that work

      # is more effort than it's worth for this temporary, build-time package.

      combined = symlinkJoin {

        name = "cuda-combined-${version}";

        paths = pkgSet;

      headers = symlinkJoin {

        name = "cuda-headers-combined-${version}";

        paths = lib.map (pkg: "${lib.getInclude pkg}/include") (getCudaPackages cudaPackages);

      };

      # The combined package above breaks the build for some reason so we just configure

      # each package's library path.

      libs = lib.concatMapStringsSep " " (x: ''"${x}/lib"'') pkgSet;

    in ''

      list(APPEND Cuda${version}_INCLUDE_PATHS "${combined}/include")

      list(APPEND Cuda${version}_LIB_PATHS ${libs})

    '';

# gcc11 is required by DCGM's very particular build system

# C.f. https://github.com/NVIDIA/DCGM/blob/7e1012302679e4bb7496483b32dcffb56e528c92/dcgmbuild/build.sh#L22

in gcc11Stdenv.mkDerivation rec {

    in [

      (lib.cmakeFeature "CUDA${version}_INCLUDE_DIR" "${headers}")

      (lib.cmakeFeature "CUDA${version}_LIBS" "${cudaPackages.cuda_cudart.stubs}/lib/stubs/libcuda.so")

      (lib.cmakeFeature "CUDA${version}_STATIC_LIBS" "${lib.getLib cudaPackages.cuda_cudart}/lib/libcudart.so")

      (lib.cmakeFeature "CUDA${version}_STATIC_CUBLAS_LIBS" (lib.concatStringsSep ";" [

        "${lib.getLib cudaPackages.libcublas}/lib/libcublas.so"

        "${lib.getLib cudaPackages.libcublas}/lib/libcublasLt.so"

      ]))

    ];

in stdenv.mkDerivation rec {

  pname = "dcgm";

  version = "3.3.5"; # N.B: If you change this, be sure prometheus-dcgm-exporter supports this version.

  version = "3.3.9"; # N.B: If you change this, be sure prometheus-dcgm-exporter supports this version.

  src = fetchFromGitHub {

    owner = "NVIDIA";

    repo = "DCGM";

    rev = "refs/tags/v${version}";

    hash = "sha256-n/uWvgvxAGfr1X51XgtHfFGDOO5AMBSV5UWQQpsylpg=";

    hash = "sha256-PysxuN5WT7GB0oOvT5ezYeOau6AMVDDWE5HOAcmqw/Y=";

  };

  # Add our paths to the CUDA paths so FindCuda.cmake can find them.

  EXTRA_CUDA_PATHS = lib.concatMapStringsSep "\n" mkAppendCudaPaths cudaPackageSetByVersion;

  prePatch = ''

    echo "$EXTRA_CUDA_PATHS"$'\n'"$(cat cmake/FindCuda.cmake)" > cmake/FindCuda.cmake

  '';

  patches = [

    ./fix-includes.patch

    ./dynamic-libs.patch

  ];

  hardeningDisable = [ "all" ];

    autoAddDriverRunpath

    cmake

    ninja

    git

    python3

  ];

  buildInputs = [

    # Header-only

    boost

    catch2

    plog.dev

    tclap_1_4

    # Dependencies that can be either static or dynamic.

    (fmt_9.override { enableShared = !static; }) # DCGM's build uses the static outputs regardless of enableShared

    (yaml-cpp.override { inherit static; stdenv = gcc11Stdenv; })

    # TODO: Dependencies that DCGM's CMake hard-codes to be static-only.

    (jsoncpp.override { enableStatic = true; })

    (libevent.override { sslSupport = false; static = true; })

    fmt_9

    yaml-cpp

    jsoncpp

    libevent

  ];

  disallowedReferences = lib.concatMap (x: x.pkgSet) cudaPackageSetByVersion;

  # Add our paths to the CMake flags so FindCuda.cmake can find them.

  cmakeFlags = lib.concatMap mkCudaFlags cudaPackageSets;

  # Lots of dodgy C++.

  env.NIX_CFLAGS_COMPILE = "-Wno-error";

  doCheck = true;

  checkPhase = ''

    runHook preCheck

    ctest -j $NIX_BUILD_CORES --output-on-failure --exclude-regex ${

      lib.escapeShellArg (

        lib.concatMapStringsSep "|" (test: "^${lib.escapeRegex test}$") [

          "DcgmModuleSysmon Watches"

          "DcgmModuleSysmon maxSampleAge"

          "DcgmModuleSysmon::CalculateCoreUtilization"

          "DcgmModuleSysmon::ParseProcStatCpuLine"

          "DcgmModuleSysmon::ParseThermalFileContentsAndStore"

          "DcgmModuleSysmon::PopulateTemperatureFileMap"

          "DcgmModuleSysmon::ReadCoreSpeed"

          "DcgmModuleSysmon::ReadTemperature"

          "Sysmon: initialize module"

        ]

      )

    }

    runHook postCheck

  '';

  disallowedReferences = lib.concatMap getCudaPackages cudaPackageSets;

  meta = with lib; {

    description = "Data Center GPU Manager (DCGM) is a daemon that allows users to monitor NVIDIA data-center GPUs";