Support mlc experiment entries

.github/workflows/build_wheels.yml

@@ -5,6 +5,7 @@ on:
     types: [published]
   workflow_dispatch: {}
 
+
 jobs:
   build_wheels:
     if: github.repository_owner == 'mlcommons'

automation/script/experiment.py

@@ -1,5 +1,7 @@
 from collections import defaultdict
 import os
+from mlc.main import ExperimentAction
+import mlc.utils as utils
 from mlc import utils
 from utils import *
 import logging
@@ -26,6 +28,8 @@ def experiment_run(self_module, i):
     show_time = i.get('show_time', False)
     logger = self_module.logger
     env = i.get('env', {})
+    experiment_action = ExperimentAction(self_module.action_object.parent)
+
     prune_result = prune_input(
         {'input': i, 'extra_keys_starts_with': ['exp.']})
     if prune_result['return'] > 0:
@@ -80,6 +84,18 @@ def experiment_run(self_module, i):
                     if r['return'] > 0:
                         return r
 
+            experiment_meta = {}
+            exp_tags = tags
+            ii = {'action': 'update',
+                  'target': 'experiment',
+                  'script_alias': meta['alias'],
+                  'tags': ','.join(exp_tags),
+                  'meta': experiment_meta,
+                  'force': True}
+            r = experiment_action.access(ii)
+            if r['return'] > 0:
+                return r
+
     return {'return': 0}
 
 

script/save-machine-state/capture.py

@@ -0,0 +1,106 @@
+import os
+import json
+import psutil
+import platform
+import subprocess
+from datetime import datetime
+
+
+def read_file_safe(path):
+    try:
+        with open(path, 'r') as f:
+            return f.read().strip()
+    except Exception:
+        return None
+
+
+def run_command_safe(command, require_sudo=False):
+    if require_sudo and os.geteuid() != 0:
+        return "Skipped (requires sudo)"
+    try:
+        output = subprocess.check_output(command, shell=True, text=True)
+        return output.strip()
+    except subprocess.CalledProcessError:
+        return "Error running command"
+
+
+def detect_container_context():
+    context = {
+        "docker_env": os.path.exists('/.dockerenv'),
+        "cgroup_indicators": []
+    }
+    cgroup = read_file_safe('/proc/1/cgroup')
+    if cgroup:
+        for line in cgroup.splitlines():
+            if any(x in line for x in ['docker', 'kubepods', 'containerd']):
+                context["cgroup_indicators"].append(line)
+    return context
+
+
+def get_mounted_file_systems():
+    try:
+        with open("/proc/mounts", "r") as f:
+            return [line.strip() for line in f.readlines()]
+    except BaseException:
+        return []
+
+
+def capture_machine_state():
+    state = {
+        "timestamp": datetime.now().isoformat(),
+        "platform": {
+            "system": platform.system(),
+            "node": platform.node(),
+            "release": platform.release(),
+            "version": platform.version(),
+            "machine": platform.machine(),
+            "processor": platform.processor()
+        },
+        "cpu": {
+            "logical_cores": psutil.cpu_count(logical=True),
+            "physical_cores": psutil.cpu_count(logical=False),
+            "load_avg": psutil.getloadavg(),
+            "cpu_percent": psutil.cpu_percent(interval=1)
+        },
+        "memory": {
+            "virtual_memory": dict(psutil.virtual_memory()._asdict()),
+            "swap_memory": dict(psutil.swap_memory()._asdict())
+        },
+        "disk": {
+            "disk_usage": dict(psutil.disk_usage('/')._asdict()),
+            "partitions": [dict(p._asdict()) for p in psutil.disk_partitions()]
+        },
+        "bios": {
+            "vendor": run_command_safe("dmidecode -s bios-vendor", require_sudo=True),
+            "version": run_command_safe("dmidecode -s bios-version", require_sudo=True),
+            "release_date": run_command_safe("dmidecode -s bios-release-date", require_sudo=True)
+        },
+        "thp_settings": {
+            "enabled": read_file_safe("/sys/kernel/mm/transparent_hugepage/enabled") or "Skipped (requires sudo or permission)",
+            "defrag": read_file_safe("/sys/kernel/mm/transparent_hugepage/defrag") or "Skipped (requires sudo or permission)"
+        },
+        "kernel": {
+            "cmdline": read_file_safe("/proc/cmdline")
+        },
+        "uptime": read_file_safe("/proc/uptime"),
+        "process_count": len(psutil.pids()),
+        "users_sessions": [dict(u._asdict()) for u in psutil.users()],
+        "container_context": detect_container_context(),
+        "mounted_filesystems": get_mounted_file_systems()
+    }
+    return state
+
+
+def save_state_to_file(state, filename):
+    with open(filename, "w") as f:
+        json.dump(state, f, indent=4)
+
+
+# Example usage
+if __name__ == "__main__":
+
+    state = capture_machine_state()
+    save_file = os.environ.get(
+        'MLC_SYSTEM_STATE_SAVE_FILENAME',
+        'machine_state.json')
+    save_state_to_file(state, save_file)

script/save-machine-state/customize.py

@@ -0,0 +1,47 @@
+from mlc import utils
+import os
+import subprocess
+
+
+def check_installation(command, os_info):
+    if os_info['platform'] == "windows":
+        return subprocess.call(
+            [command, '--version'], stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=True) == 0
+    elif os_info['platform'] == "linux":
+        return subprocess.call(['which', command], stdout=subprocess.PIPE,
+                               stderr=subprocess.PIPE) == 0  # 0 means the package is there
+
+
+def preprocess(i):
+
+    os_info = i['os_info']
+    env = i['env']
+
+    if not check_installation("numactl", os_info):
+        env['MLC_INSTALL_NUMACTL'] = 'True'
+
+    # if not check_installation("cpupower",os_info):
+    env['MLC_INSTALL_CPUPOWER'] = 'True'
+
+    if env.get('MLC_PLATFORM_DETAILS_FILE_PATH', '') == '':
+        if env.get('MLC_PLATFORM_DETAILS_DIR_PATH', '') == '':
+            env['MLC_PLATFORM_DETAILS_DIR_PATH'] = os.getcwd()
+        if env.get('MLC_PLATFORM_DETAILS_FILE_NAME', '') == '':
+            env['MLC_PLATFORM_DETAILS_FILE_NAME'] = "system-info.txt"
+        env['MLC_PLATFORM_DETAILS_FILE_PATH'] = os.path.join(
+            env['MLC_PLATFORM_DETAILS_DIR_PATH'], env['MLC_PLATFORM_DETAILS_FILE_NAME'])
+
+    return {'return': 0}
+
+
+def postprocess(i):
+
+    state = i['state']
+
+    env = i['env']
+
+    os_info = i['os_info']
+
+    automation = i['automation']
+
+    return {'return': 0}

script/save-machine-state/meta.yaml

@@ -0,0 +1,67 @@
+alias: save-machine-state
+automation_alias: script
+automation_uid: 5b4e0237da074764
+cache: false
+category: Platform information
+deps:
+- tags: detect,os
+- skip_if_env:
+    MLC_HOST_OS_TYPE:
+    - windows
+  tags: detect,sudo
+- names:
+  - python
+  - python3
+  tags: get,python
+- skip_if_any_env:
+    MLC_HOST_OS_TYPE:
+    - windows
+  skip_if_env:
+    MLC_SUDO_USER:
+    - 'no'
+  tags: get,sys-util,generic,_psmisc
+- enable_if_env:
+    MLC_HOST_OS_TYPE:
+    - linux
+  skip_if_env:
+    MLC_SUDO_USER:
+    - 'no'
+  tags: get,sys-util,generic,_systemd
+- enable_if_env:
+    MLC_HOST_OS_TYPE:
+    - linux
+  skip_if_env:
+    MLC_SUDO_USER:
+    - 'no'
+  tags: get,sys-util,generic,_dmidecode
+- tags: get,generic-python-lib,_package.psutil
+input_mapping:
+  outfile: MLC_SYSTEM_STATE_SAVE_FILENAME
+prehook_deps:
+- enable_if_env:
+    MLC_HOST_OS_TYPE:
+    - linux
+    MLC_INSTALL_NUMACTL:
+    - 'True'
+  skip_if_env:
+    MLC_SUDO_USER:
+    - 'no'
+  tags: get,sys-util,generic,_numactl
+- enable_if_env:
+    MLC_HOST_OS_TYPE:
+    - linux
+    MLC_INSTALL_CPUPOWER:
+    - 'True'
+  env:
+    MLC_TMP_FAIL_SAFE: 'yes'
+  ignore_missing: true
+  skip_if_env:
+    MLC_SUDO_USER:
+    - 'no'
+  tags: get,sys-util,generic,_linux-tools
+tags:
+- machine-state
+- save
+- machine
+- state
+uid: 2f62820ed7294659

script/save-machine-state/run.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+set -e
+
+${MLC_PYTHON_BIN_WITH_PATH} ${MLC_TMP_CURRENT_SCRIPT_PATH}/capture.py