コード例 #1
0
ファイル: __main__.py プロジェクト: mesosphere/cloudkeeper
def main() -> None:
    setup_logger("resotoworker")
    # Try to run in a new process group and
    # ignore if not possible for whatever reason
    try:
        os.setpgid(0, 0)
    except Exception:
        pass

    resotolib.proc.parent_pid = os.getpid()

    arg_parser = ArgumentParser(
        description="resoto worker",
        env_args_prefix="RESOTOWORKER_",
    )
    add_args(arg_parser)
    jwt_add_args(arg_parser)
    logging_add_args(arg_parser)
    core_add_args(arg_parser)
    Config.add_args(arg_parser)
    TLSData.add_args(arg_parser)

    # Find resoto Plugins in the resoto.plugins module
    plugin_loader = PluginLoader()
    plugin_loader.add_plugin_args(arg_parser)

    # At this point the CLI, all Plugins as well as the WebServer have
    # added their args to the arg parser
    arg_parser.parse_args()

    try:
        wait_for_resotocore(resotocore.http_uri)
    except TimeoutError as e:
        log.fatal(f"Failed to connect to resotocore: {e}")
        sys.exit(1)

    tls_data = None
    if resotocore.is_secure:
        tls_data = TLSData(
            common_name=ArgumentParser.args.subscriber_id,
            resotocore_uri=resotocore.http_uri,
        )
        tls_data.start()
    config = Config(
        ArgumentParser.args.subscriber_id,
        resotocore_uri=resotocore.http_uri,
        tls_data=tls_data,
    )
    add_config(config)
    plugin_loader.add_plugin_config(config)
    config.load_config()

    def send_request(request: requests.Request) -> requests.Response:
        prepared = request.prepare()
        s = requests.Session()
        verify = None
        if tls_data:
            verify = tls_data.verify
        return s.send(request=prepared, verify=verify)

    core = Resotocore(send_request, config)

    collector = Collector(core.send_to_resotocore, config)

    # Handle Ctrl+c and other means of termination/shutdown
    resotolib.proc.initializer()
    add_event_listener(EventType.SHUTDOWN, shutdown, blocking=False)

    # Try to increase nofile and nproc limits
    increase_limits()

    web_server_args = {}
    if tls_data:
        web_server_args = {
            "ssl_cert": tls_data.cert_path,
            "ssl_key": tls_data.key_path,
        }
    web_server = WebServer(
        WebApp(mountpoint=Config.resotoworker.web_path),
        web_host=Config.resotoworker.web_host,
        web_port=Config.resotoworker.web_port,
        **web_server_args,
    )
    web_server.daemon = True
    web_server.start()

    core_actions = CoreActions(
        identifier=f"{ArgumentParser.args.subscriber_id}-collector",
        resotocore_uri=resotocore.http_uri,
        resotocore_ws_uri=resotocore.ws_uri,
        actions={
            "collect": {
                "timeout": Config.resotoworker.timeout,
                "wait_for_completion": True,
            },
            "cleanup": {
                "timeout": Config.resotoworker.timeout,
                "wait_for_completion": True,
            },
        },
        message_processor=partial(core_actions_processor, plugin_loader, tls_data, collector),
        tls_data=tls_data,
    )

    task_queue_filter = {}
    if len(Config.resotoworker.collector) > 0:
        task_queue_filter = {"cloud": list(Config.resotoworker.collector)}
    core_tasks = CoreTasks(
        identifier=f"{ArgumentParser.args.subscriber_id}-tagger",
        resotocore_ws_uri=resotocore.ws_uri,
        tasks=["tag"],
        task_queue_filter=task_queue_filter,
        message_processor=core_tag_tasks_processor,
        tls_data=tls_data,
    )
    core_actions.start()
    core_tasks.start()

    for Plugin in plugin_loader.plugins(PluginType.ACTION):
        try:
            log.debug(f"Starting action plugin {Plugin}")
            plugin = Plugin(tls_data=tls_data)
            plugin.start()
        except Exception as e:
            log.exception(f"Caught unhandled persistent Plugin exception {e}")

    # We wait for the shutdown Event to be set() and then end the program
    # While doing so we print the list of active threads once per 15 minutes
    shutdown_event.wait()
    web_server.shutdown()
    time.sleep(1)  # everything gets 1000ms to shutdown gracefully before we force it
    resotolib.proc.kill_children(resotolib.proc.SIGTERM, ensure_death=True)
    log.info("Shutdown complete")
    os._exit(0)
コード例 #2
0
ファイル: __main__.py プロジェクト: mesosphere/cloudkeeper
def main() -> None:
    setup_logger("resotometrics")
    resotolib.proc.parent_pid = os.getpid()

    add_event_listener(EventType.SHUTDOWN, shutdown)
    arg_parser = ArgumentParser(description="resoto metrics exporter",
                                env_args_prefix="RESOTOMETRICS_")
    add_args(arg_parser)
    Config.add_args(arg_parser)
    resotocore_add_args(arg_parser)
    logging_add_args(arg_parser)
    jwt_add_args(arg_parser)
    TLSData.add_args(arg_parser)
    arg_parser.parse_args()

    try:
        wait_for_resotocore(resotocore.http_uri)
    except TimeoutError as e:
        log.fatal(f"Failed to connect to resotocore: {e}")
        sys.exit(1)

    tls_data = None
    if resotocore.is_secure:
        tls_data = TLSData(
            common_name=ArgumentParser.args.subscriber_id,
            resotocore_uri=resotocore.http_uri,
        )
        tls_data.start()
    config = Config(
        ArgumentParser.args.subscriber_id,
        resotocore_uri=resotocore.http_uri,
        tls_data=tls_data,
    )
    config.add_config(ResotoMetricsConfig)
    config.load_config()

    resotolib.proc.initializer()

    metrics = Metrics()
    graph_collector = GraphCollector(metrics)
    REGISTRY.register(graph_collector)

    resotocore_graph = Config.resotometrics.graph
    graph_uri = f"{resotocore.http_uri}/graph/{resotocore_graph}"
    search_uri = f"{graph_uri}/search/aggregate?section=reported"

    message_processor = partial(core_actions_processor, metrics, search_uri,
                                tls_data)
    core_actions = CoreActions(
        identifier=ArgumentParser.args.subscriber_id,
        resotocore_uri=resotocore.http_uri,
        resotocore_ws_uri=resotocore.ws_uri,
        actions={
            "generate_metrics": {
                "timeout": Config.resotometrics.timeout,
                "wait_for_completion": True,
            },
        },
        message_processor=message_processor,
        tls_data=tls_data,
    )
    web_server_args = {}
    if tls_data:
        web_server_args = {
            "ssl_cert": tls_data.cert_path,
            "ssl_key": tls_data.key_path,
        }
    web_server = WebServer(
        WebApp(mountpoint=Config.resotometrics.web_path),
        web_host=Config.resotometrics.web_host,
        web_port=Config.resotometrics.web_port,
        **web_server_args,
    )
    web_server.daemon = True
    web_server.start()
    core_actions.start()
    shutdown_event.wait()
    web_server.shutdown()
    core_actions.shutdown()
    resotolib.proc.kill_children(resotolib.proc.SIGTERM, ensure_death=True)
    log.info("Shutdown complete")
    sys.exit(0)