[V1.8.x] Attemp to fix cd for v1.8.x

cd/python/docker/Dockerfile

@@ -23,11 +23,15 @@
 ARG BASE_IMAGE
 FROM ${BASE_IMAGE}
 
-ARG PYTHON_CMD=python3
-RUN apt-get update && \
-    apt-get install -y wget ${PYTHON_CMD}-dev gcc && \
-    wget https://bootstrap.pypa.io/get-pip.py && \
-    ${PYTHON_CMD} get-pip.py
+RUN apt-get update || true
+RUN apt-get install -y software-properties-common
+RUN add-apt-repository -y ppa:deadsnakes/ppa
+RUN apt-get update || true
+RUN apt-get install -y python3.7-dev python3.7-distutils virtualenv wget
+RUN ln -sf /usr/bin/python3.7 /usr/local/bin/python3
+
+RUN wget -nv https://bootstrap.pypa.io/get-pip.py
+RUN python3 get-pip.py
 
 ARG MXNET_COMMIT_ID
 ENV MXNET_COMMIT_ID=${MXNET_COMMIT_ID}

ci/Jenkinsfile_docker_cache

@@ -37,7 +37,7 @@ core_logic: {
       ws('workspace/docker_cache') {
         timeout(time: total_timeout, unit: 'MINUTES') {
           utils.init_git()
-          sh "ci/docker_cache.py --docker-registry ${env.DOCKER_CACHE_REGISTRY}"
+          sh "ci/docker_cache.py --docker-registry ${env.DOCKER_ECR_REGISTRY}"
         }
       }
     }

ci/Jenkinsfile_utils.groovy

@@ -112,20 +112,7 @@ def get_git_commit_hash() {
 }
 
 def publish_test_coverage() {
-    // CodeCovs auto detection has trouble with our CIs PR validation due the merging strategy
-    git_commit_hash = get_git_commit_hash()
-
-    if (env.CHANGE_ID) {
-      // PR execution
-      codecovArgs = "-B ${env.CHANGE_TARGET} -C ${git_commit_hash} -P ${env.CHANGE_ID}"
-    } else {
-      // Branch execution
-      codecovArgs = "-B ${env.BRANCH_NAME} -C ${git_commit_hash}"
-    }
-
-    // To make sure we never fail because test coverage reporting is not available
-    // Fall back to our own copy of the bash helper if it failed to download the public version
-    sh "(curl --retry 10 -s https://codecov.io/bash | bash -s - ${codecovArgs}) || (curl --retry 10 -s https://s3-us-west-2.amazonaws.com/mxnet-ci-prod-slave-data/codecov-bash.txt | bash -s - ${codecovArgs}) || true"
+    sh "curl -s https://codecov.io/bash | bash"
 }
 
 def collect_test_results_unix(original_file_name, new_file_name) {
@@ -160,7 +147,7 @@ def collect_test_results_windows(original_file_name, new_file_name) {
 
 
 def docker_run(platform, function_name, use_nvidia, shared_mem = '500m', env_vars = "") {
-  def command = "ci/build.py %ENV_VARS% --docker-registry ${env.DOCKER_CACHE_REGISTRY} %USE_NVIDIA% --platform %PLATFORM% --docker-build-retries 3 --shm-size %SHARED_MEM% /work/runtime_functions.sh %FUNCTION_NAME%"
+  def command = "ci/build.py %ENV_VARS% --docker-registry ${env.DOCKER_ECR_REGISTRY} %USE_NVIDIA% --platform %PLATFORM% --docker-build-retries 3 --shm-size %SHARED_MEM% /work/runtime_functions.sh %FUNCTION_NAME%"
   command = command.replaceAll('%ENV_VARS%', env_vars.length() > 0 ? "-e ${env_vars}" : '')
   command = command.replaceAll('%USE_NVIDIA%', use_nvidia ? '--nvidiadocker' : '')
   command = command.replaceAll('%PLATFORM%', platform)

ci/build.py

@@ -27,6 +27,8 @@
 
 import argparse
 import glob
+import hashlib
+import os
 import pprint
 import re
 import shutil
@@ -36,7 +38,6 @@
 from subprocess import check_call, check_output
 from typing import *
 
-from safe_docker_run import SafeDockerClient
 from util import *
 
 
@@ -52,13 +53,41 @@ def get_platforms(path: str = get_dockerfiles_path()) -> List[str]:
     platforms = list(map(lambda x: os.path.split(x)[1], sorted(files)))
     return platforms
 
+def _find_copied_files(dockerfile):
+    """
+    Creates a list of files copied into given dockerfile.
+    """
+    copied_files = []
+    basedir = os.path.dirname(dockerfile)
+    with open(dockerfile, "r") as f:
+        for line in f.readlines():
+            if line.startswith("COPY "):
+                copied_files.append(os.path.join(basedir, line.split(" ")[1]))
+    return copied_files
+
+def _hash_file(ctx, filename):
+    """
+    Add contents of passed file into passed hash context.
+    """
+    bufsiz = 16384
+    with open(filename,"rb") as f:
+        while True:
+            d = f.read(bufsiz)
+            if not d:
+                break
+            ctx.update(d)
 
 def get_docker_tag(platform: str, registry: str) -> str:
     """:return: docker tag to be used for the container"""
     platform = platform if any(x in platform for x in ['build.', 'publish.']) else 'build.{}'.format(platform)
     if not registry:
         registry = "mxnet_local"
-    return "{0}/{1}".format(registry, platform)
+    dockerfile = get_dockerfile(platform)
+    sha256 = hashlib.sha256()
+    _hash_file(sha256, dockerfile)
+    for f in _find_copied_files(dockerfile):
+        _hash_file(sha256, f)
+    return "{0}:{1}-{2}".format(registry, platform, sha256.hexdigest()[:12])
 
 
 def get_dockerfile(platform: str, path=get_dockerfiles_path()) -> str:
@@ -67,7 +96,7 @@ def get_dockerfile(platform: str, path=get_dockerfiles_path()) -> str:
 
 
 def build_docker(platform: str, registry: str, num_retries: int, no_cache: bool,
-                 cache_intermediate: bool) -> str:
+                 cache_intermediate: bool=False) -> str:
     """
     Build a container for the given platform
     :param platform: Platform
@@ -157,8 +186,7 @@ def default_ccache_dir() -> str:
     return os.path.join(os.path.expanduser("~"), ".ccache")
 
 
-def container_run(docker_client: SafeDockerClient,
-                  platform: str,
+def container_run(platform: str,
                   nvidia_runtime: bool,
                   docker_registry: str,
                   shared_memory_size: str,
@@ -167,17 +195,12 @@ def container_run(docker_client: SafeDockerClient,
                   environment: Dict[str, str],
                   dry_run: bool = False) -> int:
     """Run command in a container"""
-    container_wait_s = 600
-    #
-    # Environment setup
-    #
+    # set default environment variables
     environment.update({
         'CCACHE_MAXSIZE': '500G',
         'CCACHE_TEMPDIR': '/tmp/ccache',  # temp dir should be local and not shared
-        'CCACHE_DIR': '/work/ccache',  # this path is inside the container as /work/ccache is
-                                       # mounted
-        'CCACHE_LOGFILE': '/tmp/ccache.log',  # a container-scoped log, useful for ccache
-                                              # verification.
+        'CCACHE_DIR': '/work/ccache',  # this path is inside the container as /work/ccache is mounted
+        'CCACHE_LOGFILE': '/tmp/ccache.log',  # a container-scoped log, useful for ccache verification.
     })
     environment.update({k: os.environ[k] for k in ['CCACHE_MAXSIZE'] if k in os.environ})
 
@@ -189,13 +212,9 @@ def container_run(docker_client: SafeDockerClient,
     os.makedirs(local_ccache_dir, exist_ok=True)
     logging.info("Using ccache directory: %s", local_ccache_dir)
 
-    # Equivalent command
-    docker_cmd_list = [
-        "docker",
-        'run',
-        "--gpus all" if nvidia_runtime else "",
-        "--cap-add",
-        "SYS_PTRACE", # Required by ASAN
+    # Build docker command
+    docker_arg_list = [
+        "--cap-add", "SYS_PTRACE", # Required by ASAN
         '--rm',
         '--shm-size={}'.format(shared_memory_size),
         # mount mxnet root
@@ -211,40 +230,27 @@ def container_run(docker_client: SafeDockerClient,
         '-e', "CCACHE_DIR={}".format(environment['CCACHE_DIR']),
         # a container-scoped log, useful for ccache verification.
         '-e', "CCACHE_LOGFILE={}".format(environment['CCACHE_LOGFILE']),
-        '-ti',
-        tag]
-    docker_cmd_list.extend(command)
-    docker_cmd = ' \\\n\t'.join(docker_cmd_list)
-    logging.info("Running %s in container %s", command, tag)
-    logging.info("Executing the equivalent of:\n%s\n", docker_cmd)
+    ]
+    docker_arg_list += [tag]
+    docker_arg_list.extend(command)
+
+    def docker_run_cmd(cmd):
+        logging.info("Running %s in container %s", command, tag)
+        logging.info("Executing command:\n%s\n", ' \\\n\t'.join(cmd))
+        subprocess.run(cmd, stdout=sys.stdout, stderr=sys.stderr, check=True)
 
     if not dry_run:
-        #############################
-        #
-        signal.pthread_sigmask(signal.SIG_BLOCK, {signal.SIGINT, signal.SIGTERM})
-        # noinspection PyShadowingNames
-        runtime = None
-        if nvidia_runtime:
-            # noinspection PyShadowingNames
-            # runc is default (docker info | grep -i runtime)
-            runtime = 'nvidia'
-
-        return docker_client.run(
-            tag,
-            runtime=runtime,
-            command=command,
-            shm_size=shared_memory_size,
-            user='{}:{}'.format(os.getuid(), os.getgid()),
-            cap_add='SYS_PTRACE',
-            volumes={
-                mx_root:
-                    {'bind': '/work/mxnet', 'mode': 'rw'},
-                local_build_folder:
-                    {'bind': '/work/build', 'mode': 'rw'},
-                local_ccache_dir:
-                    {'bind': '/work/ccache', 'mode': 'rw'},
-            },
-            environment=environment)
+        if not nvidia_runtime:
+            docker_run_cmd(['docker', 'run'] + docker_arg_list)
+        else:
+            try:
+                docker_run_cmd(['docker', 'run', '--gpus', 'all'] + docker_arg_list)
+            except subprocess.CalledProcessError as e:
+                if e.returncode == 125:
+                    docker_run_cmd(['docker', 'run', '--runtime', 'nvidia'] + docker_arg_list)
+                else:
+                    raise
+
     return 0
 
 
@@ -348,7 +354,6 @@ def main() -> int:
     args = parser.parse_args()
 
     command = list(chain(*args.command))
-    docker_client = SafeDockerClient()
 
     environment = dict([(e.split('=')[:2] if '=' in e else (e, os.environ[e]))
                         for e in args.environment])
@@ -375,21 +380,21 @@ def main() -> int:
         ret = 0
         if command:
             ret = container_run(
-                docker_client=docker_client, platform=platform, nvidia_runtime=args.nvidiadocker,
+                platform=platform, nvidia_runtime=args.nvidiadocker,
                 shared_memory_size=args.shared_memory_size, command=command, docker_registry=args.docker_registry,
                 local_ccache_dir=args.ccache_dir, environment=environment)
         elif args.print_docker_run:
             command = []
             ret = container_run(
-                docker_client=docker_client, platform=platform, nvidia_runtime=args.nvidiadocker,
+                platform=platform, nvidia_runtime=args.nvidiadocker,
                 shared_memory_size=args.shared_memory_size, command=command, docker_registry=args.docker_registry,
                 local_ccache_dir=args.ccache_dir, dry_run=True, environment=environment)
         else:
             # With no commands, execute a build function for the target platform
             command = ["/work/mxnet/ci/docker/runtime_functions.sh", "build_{}".format(platform)]
             logging.info("No command specified, trying default build: %s", ' '.join(command))
             ret = container_run(
-                docker_client=docker_client, platform=platform, nvidia_runtime=args.nvidiadocker,
+                platform=platform, nvidia_runtime=args.nvidiadocker,
                 shared_memory_size=args.shared_memory_size, command=command, docker_registry=args.docker_registry,
                 local_ccache_dir=args.ccache_dir, environment=environment)
 
@@ -406,7 +411,8 @@ def main() -> int:
             tag = get_docker_tag(platform=platform, registry=args.docker_registry)
             load_docker_cache(tag=tag, docker_registry=args.docker_registry)
             build_docker(platform, registry=args.docker_registry,
-                         num_retries=args.docker_build_retries, no_cache=args.no_cache)
+                         num_retries=args.docker_build_retries, no_cache=args.no_cache,
+                         cache_intermediate=args.cache_intermediate)
             if args.build_only:
                 continue
             shutil.rmtree(buildir(), ignore_errors=True)
@@ -418,7 +424,7 @@ def main() -> int:
                 continue
             command = ["/work/mxnet/ci/docker/runtime_functions.sh", build_platform]
             container_run(
-                docker_client=docker_client, platform=platform, nvidia_runtime=args.nvidiadocker,
+                platform=platform, nvidia_runtime=args.nvidiadocker,
                 shared_memory_size=args.shared_memory_size, command=command, docker_registry=args.docker_registry,
                 local_ccache_dir=args.ccache_dir, environment=environment)
             shutil.move(buildir(), plat_buildir)

ci/docker/install/ubuntu_python.sh

@@ -23,7 +23,12 @@
 set -ex
 # install libraries for mxnet's python package on ubuntu
 apt-get update || true
-apt-get install -y python-dev python3-dev virtualenv wget
+apt-get install -y software-properties-common
+add-apt-repository -y ppa:deadsnakes/ppa
+apt-get update || true
+apt-get install -y python3.6-dev virtualenv wget
+# setup symlink in /usr/local/bin to override python3 version
+ln -sf /usr/bin/python3.6 /usr/local/bin/python3
 
 # the version of the pip shipped with ubuntu may be too lower, install a recent version here
 wget -nv https://bootstrap.pypa.io/get-pip.py

ci/docker/runtime_functions.sh

@@ -517,7 +517,7 @@ build_ubuntu_cpu_cmake_debug() {
     cmake \
         -DUSE_CUDA=OFF \
         -DUSE_TVM_OP=ON \
-        -DPython3_EXECUTABLE=/usr/bin/python3 \
+        -DPython3_EXECUTABLE=python3 \
         -DUSE_MKL_IF_AVAILABLE=OFF \
         -DUSE_OPENMP=OFF \
         -DUSE_OPENCV=ON \
@@ -538,7 +538,7 @@ build_ubuntu_cpu_cmake_no_tvm_op() {
     cmake \
         -DUSE_CUDA=OFF \
         -DUSE_TVM_OP=OFF \
-        -DPython3_EXECUTABLE=/usr/bin/python3 \
+        -DPython3_EXECUTABLE=python3 \
         -DUSE_MKL_IF_AVAILABLE=OFF \
         -DUSE_OPENMP=OFF \
         -DUSE_OPENCV=ON \
@@ -874,7 +874,7 @@ build_ubuntu_gpu_cmake_mkldnn() {
         -DUSE_CUDA=1                            \
         -DUSE_CUDNN=1                           \
         -DUSE_TVM_OP=0                          \
-        -DPython3_EXECUTABLE=/usr/bin/python3   \
+        -DPython3_EXECUTABLE=python3            \
         -DUSE_MKLML_MKL=1                       \
         -DCMAKE_BUILD_TYPE=Release              \
         -DMXNET_CUDA_ARCH="$CI_CMAKE_CUDA_ARCH" \
@@ -892,8 +892,8 @@ build_ubuntu_gpu_cmake() {
         -DUSE_SIGNAL_HANDLER=ON                 \
         -DUSE_CUDA=ON                           \
         -DUSE_CUDNN=ON                          \
-        -DUSE_TVM_OP=OFF                         \
-        -DPython3_EXECUTABLE=/usr/bin/python3   \
+        -DUSE_TVM_OP=OFF                        \
+        -DPython3_EXECUTABLE=python3            \
         -DUSE_MKL_IF_AVAILABLE=OFF              \
         -DUSE_MKLML_MKL=OFF                     \
         -DUSE_MKLDNN=OFF                        \
@@ -915,8 +915,8 @@ build_ubuntu_gpu_cmake_no_rtc() {
         -DUSE_SIGNAL_HANDLER=ON                 \
         -DUSE_CUDA=ON                           \
         -DUSE_CUDNN=ON                          \
-        -DUSE_TVM_OP=OFF                         \
-        -DPython3_EXECUTABLE=/usr/bin/python3   \
+        -DUSE_TVM_OP=OFF                        \
+        -DPython3_EXECUTABLE=python3            \
         -DUSE_MKL_IF_AVAILABLE=OFF              \
         -DUSE_MKLML_MKL=OFF                     \
         -DUSE_MKLDNN=ON                         \
@@ -956,8 +956,8 @@ build_ubuntu_gpu_large_tensor() {
         -DUSE_SIGNAL_HANDLER=ON                 \
         -DUSE_CUDA=ON                           \
         -DUSE_CUDNN=ON                          \
-        -DUSE_TVM_OP=OFF                         \
-        -DPython3_EXECUTABLE=/usr/bin/python3   \
+        -DUSE_TVM_OP=OFF                        \
+        -DPython3_EXECUTABLE=python3            \
         -DUSE_MKL_IF_AVAILABLE=OFF              \
         -DUSE_MKLML_MKL=OFF                     \
         -DUSE_MKLDNN=OFF                        \