def _collect_logs()

in azurelinuxagent/ga/collect_logs.py [0:0]

• 46 lines of code
• 6 McCabe index (conditional complexity)

    def _collect_logs(self):
        logger.info("Starting log collection...")

        # Invoke the command line tool in the agent to collect logs. The --scope option starts the process as a systemd
        # transient scope unit. The --property option is used to set systemd memory and cpu properties on the scope.
        systemd_cmd = [
            "systemd-run",
            "--unit={0}".format(logcollector.CGROUPS_UNIT),
            "--slice={0}".format(cgroupconfigurator.LOGCOLLECTOR_SLICE), "--scope"
        ] + CGroupConfigurator.get_instance().get_logcollector_unit_properties()

        # The log tool is invoked from the current agent's egg with the command line option
        collect_logs_cmd = [sys.executable, "-u", sys.argv[0], "-collect-logs"]
        final_command = systemd_cmd + collect_logs_cmd

        def exec_command():
            start_time = datetime.datetime.utcnow()
            success = False
            msg = None
            try:
                shellutil.run_command(final_command, log_error=False)
                duration = elapsed_milliseconds(start_time)
                archive_size = os.path.getsize(COMPRESSED_ARCHIVE_PATH)

                msg = "Successfully collected logs. Archive size: {0} b, elapsed time: {1} ms.".format(archive_size,
                                                                                                       duration)
                logger.info(msg)
                success = True

                return True
            except Exception as e:
                duration = elapsed_milliseconds(start_time)
                err_msg = ustr(e)

                if isinstance(e, CommandError):
                    # pylint has limited (i.e. no) awareness of control flow w.r.t. typing. we disable=no-member
                    # here because we know e must be a CommandError but pylint still considers the case where
                    # e is a different type of exception.
                    err_msg = ustr("Log Collector exited with code {0}").format(e.returncode)  # pylint: disable=no-member

                    if e.returncode == logcollector.INVALID_CGROUPS_ERRCODE:  # pylint: disable=no-member
                        logger.info("Disabling periodic log collection until service restart due to process error.")
                        self.stop()

                    # When the log collector memory limit is exceeded, Agent gracefully exit the process with this error code.
                    # Stop the periodic operation because it seems to be persistent.
                    elif e.returncode == logcollector.GRACEFUL_KILL_ERRCODE:  # pylint: disable=no-member
                        logger.info("Disabling periodic log collection until service restart due to exceeded process memory limit.")
                        self.stop()
                    else:
                        logger.info(err_msg)

                msg = "Failed to collect logs. Elapsed time: {0} ms. Error: {1}".format(duration, err_msg)
                # No need to log to the local log since we logged stdout, stderr from the process.

                return False
            finally:
                add_event(
                    name=AGENT_NAME,
                    version=CURRENT_VERSION,
                    op=WALAEventOperation.LogCollection,
                    is_success=success,
                    message=msg,
                    log_event=False)

        return exec_command()