Fixes to docker detached mode

.github/workflows/test-nvidia-mlperf-inference-implementations.yml

@@ -2,7 +2,7 @@ name: MLPerf Inference Nvidia implementations
 
 on:
   schedule:
-    - cron: "50 17 * * *" #to be adjusted
+    - cron: "38 18 * * *" #to be adjusted
 
 jobs:
   run_nvidia:

automation/script/docker_utils.py

@@ -110,7 +110,7 @@ def prepare_docker_inputs(input_params, docker_settings,
         keys += [
             "skip_run_cmd", "pre_run_cmds", "run_cmd_prefix", "all_gpus", "num_gpus", "device", "gh_token",
             "port_maps", "shm_size", "pass_user_id", "pass_user_group", "extra_run_args", "detached", "interactive",
-            "dt", "it", "use_host_group_id", "use_host_user_id"
+            "dt", "it", "use_host_group_id", "use_host_user_id", "keep_detached", "reuse_existing"
         ]
     # Collect Dockerfile inputs
     docker_inputs = {
@@ -377,6 +377,8 @@ def get_docker_default(key):
         "port_maps": [],
         "use_host_user_id": True,
         "use_host_group_id": True,
+        "keep_detached": False,
+        "reuse_existing": True
     }
     if key in defaults:
         return defaults[key]

script/get-gh-actions-runner/meta.yaml

@@ -22,7 +22,7 @@ new_env_keys:
 
 deps:
   - tags: detect-os
-  - tags: download-and-extract,_extract,_url.https://github.com/actions/runner/releases/download/v2.320.0/actions-runner-linux-x64-2.320.0.tar.gz
+  - tags: download-and-extract,_extract,_url.https://github.com/actions/runner/releases/download/v2.321.0/actions-runner-linux-x64-2.321.0.tar.gz
     force_cache: yes
     extra_cache_tags: gh-actions-runner-code,gh-actions,code
     env:

script/run-docker-container/customize.py

@@ -16,7 +16,7 @@ def preprocess(i):
 
     interactive = env.get('MLC_DOCKER_INTERACTIVE_MODE', '')
 
-    if str(interactive).lower() in ['yes', 'true', '1']:
+    if is_true(interactive):
         env['MLC_DOCKER_DETACHED_MODE'] = 'no'
 
     if 'MLC_DOCKER_RUN_SCRIPT_TAGS' not in env:
@@ -54,7 +54,9 @@ def preprocess(i):
     print('')
     print('Checking existing Docker container:')
     print('')
-    CMD = f"""{env['MLC_CONTAINER_TOOL']} ps --format=json  --filter "ancestor={DOCKER_CONTAINER}" """
+    # CMD = f"""{env['MLC_CONTAINER_TOOL']} ps --format=json  --filter "ancestor={DOCKER_CONTAINER}" """
+    CMD = f"""{env['MLC_CONTAINER_TOOL']} ps --format """ + \
+        "'{{ .ID }},'" + f"""  --filter "ancestor={DOCKER_CONTAINER}" """
     if os_info['platform'] == 'windows':
         CMD += " 2> nul"
     else:
@@ -71,27 +73,21 @@ def preprocess(i):
             'error': 'Unexpected error occurred with docker run:\n{}'.format(e)
         }
 
-    if len(out) > 0 and str(env.get('MLC_DOCKER_REUSE_EXISTING_CONTAINER',
-                                    '')).lower() in ["1", "true", "yes"]:  # container exists
-        # print(out)
-        out_split = out.splitlines()
+    existing_container_id = None
+    if len(out) > 0:
+        out_split = out.split(",")
         if len(out_split) > 0:
-            try:
-                out_json = json.loads(out_split[0])
-                # print("JSON successfully loaded:", out_json)
-            except json.JSONDecodeError as e:
-                print(f"Error: First line of 'out' is not valid JSON: {e}")
-                return {
-                    'return': 1, 'error': f"Error: First line of 'out' is not valid JSON: {e}"}
-    else:
-        out_json = []
+            existing_container_id = out_split[0].strip()
 
-    if isinstance(out_json, list) and len(out_json) > 0:
-        existing_container_id = out_json[0]['Id']
+    if existing_container_id and is_true(
+            env.get('MLC_DOCKER_REUSE_EXISTING_CONTAINER', '')):
         print(f"Reusing existing container {existing_container_id}")
         env['MLC_DOCKER_CONTAINER_ID'] = existing_container_id
 
     else:
+        if existing_container_id:
+            print(
+                f"""Not using existing container {existing_container_id} as env['MLC_DOCKER_REUSE_EXISTING_CONTAINER'] is not set""")
         if env.get('MLC_DOCKER_CONTAINER_ID', '') != '':
             del (env['MLC_DOCKER_CONTAINER_ID'])  # not valid ID
 
@@ -237,13 +233,8 @@ def postprocess(i):
     run_opts += port_map_cmd_string
 
     # Currently have problem running Docker in detached mode on Windows:
-    detached = str(
-        env.get(
-            'MLC_DOCKER_DETACHED_MODE',
-            '')).lower() in [
-        'yes',
-        'true',
-        "1"]
+    detached = is_true(env.get('MLC_DOCKER_DETACHED_MODE', ''))
+
 #    if detached and os_info['platform'] != 'windows':
     if detached:
         if os_info['platform'] == 'windows':
@@ -257,8 +248,7 @@ def postprocess(i):
             CONTAINER = f"""{env['MLC_CONTAINER_TOOL']} run -dt {run_opts} --rm  {docker_image_repo}/{docker_image_name}:{docker_image_tag} bash"""
             CMD = f"""ID=`{CONTAINER}` && {env['MLC_CONTAINER_TOOL']} exec $ID bash -c '{run_cmd}'"""
 
-            if False and str(env.get('MLC_KEEP_DETACHED_CONTAINER', '')).lower() not in [
-                    'yes', "1", 'true']:
+            if not is_true(env.get('MLC_KEEP_DETACHED_CONTAINER', '')):
                 CMD += f""" && {env['MLC_CONTAINER_TOOL']} kill $ID >/dev/null"""
 
         CMD += ' && echo "ID=$ID"'

script/run-docker-container/meta.yaml

@@ -14,7 +14,6 @@ cache: false
 category: Docker automation
 
 default_env:
-  MLC_DOCKER_DETACHED_MODE: 'yes'
   MLC_DOCKER_REUSE_EXISTING_CONTAINER: 'no'
   MLC_DOCKER_PRIVILEGED_MODE: 'no'
   MLC_PODMAN_MAP_USER_ID: 'no'
@@ -31,6 +30,7 @@ input_mapping:
   docker_base_image: MLC_DOCKER_IMAGE_BASE
   base_image: MLC_DOCKER_IMAGE_BASE
   keep_detached: MLC_KEEP_DETACHED_CONTAINER
+  reuse_existing: MLC_DOCKER_REUSE_EXISTING_CONTAINER
   docker_os: MLC_DOCKER_OS
   docker_os_version: MLC_DOCKER_OS_VERSION
   os: MLC_DOCKER_OS

script/run-mlperf-inference-app/customize.py

@@ -5,6 +5,7 @@
 import subprocess
 import copy
 import mlperf_utils
+from utils import *
 
 summary_ext = ['.csv', '.json', '.xlsx']
 
@@ -218,8 +219,7 @@ def preprocess(i):
 
         print('=========================================================')
 
-    if str(env.get('MLC_MLPERF_USE_DOCKER', '')
-           ).lower() in ["1", "true", "yes"]:
+    if is_true(env.get('MLC_MLPERF_USE_DOCKER', '')):
         action = "docker"
         # del(env['OUTPUT_BASE_DIR'])
         state = {}
@@ -232,7 +232,7 @@ def preprocess(i):
             if k.startswith("docker_"):
                 docker_extra_input[k] = inp[k]
         inp = {}
-        if str(docker_dt).lower() in ["yes", "true", "1"]:
+        if is_true(docker_dt):
             # turning it off for the first run and after that we turn it on
             if env.get('MLC_DOCKER_REUSE_EXISTING_CONTAINER', '') == '':
                 env['MLC_DOCKER_REUSE_EXISTING_CONTAINER'] = 'no'
@@ -292,7 +292,7 @@ def preprocess(i):
                     env['OUTPUT_BASE_DIR'], f"{env['MLC_MLPERF_RUN_STYLE']}_results")
 
             if action == "docker":
-                if str(docker_dt).lower() not in ["yes", "true", "1"]:
+                if not is_true(docker_dt):
                     print(
                         f"\nStop Running loadgen scenario: {scenario} and mode: {mode}")
                     # We run commands interactively inside the docker container
@@ -320,8 +320,8 @@ def preprocess(i):
                 if state.get('docker', {}):
                     del (state['docker'])
 
-    if env.get('MLC_DOCKER_CONTAINER_ID', '') != '' and str(env.get(
-            'MLC_DOCKER_CONTAINER_KEEP_ALIVE', '')).lower() not in ["yes", "1", "true"]:
+    if env.get('MLC_DOCKER_CONTAINER_ID', '') != '' and not is_true(env.get(
+            'MLC_DOCKER_CONTAINER_KEEP_ALIVE', '')):
         container_id = env['MLC_DOCKER_CONTAINER_ID']
         CMD = f"docker kill {container_id}"
         docker_out = subprocess.check_output(CMD, shell=True).decode("utf-8")