sonic-buildimage/files/image_config/monit/memory_checker

#!/usr/bin/env python3

"""
memory_checker

This script is part of the feature which will restart the container if memory
usage of it is larger than the threshold value.

This script is used to check the memory usage of specified cotnainer and
is intended to be run by Monit. It will write an alerting message into
syslog if memory usage of the container is larger than the threshold value for X
times within Y cycles/minutes. Note that if print(...) statement in this script
was executed, the string in it will be appended to Monit syslog messages.

The following is an example in Monit configuration file to show how Monit will run
this script:

check program container_memory_<container_name> with path "/usr/bin/memory_checker <container_name> <threshold_value>"
    if status == 3 for X times within Y cycles exec "/usr/bin/restart_service <container_name>"
"""

import argparse
import subprocess
import sys
import syslog
import re
import time

import docker

from swsscommon import swsscommon

EVENTS_PUBLISHER_SOURCE = "sonic-events-host"
EVENTS_PUBLISHER_TAG = "mem-threshold-exceeded"

def get_command_result(command):
    """Executes the command and return the resulting output.

    Args:
        command: A string contains the command to be executed.

    Returns:
        A string which contains the output of command.
    """
    command_stdout = ""

    try:
        proc_instance = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE,
                                         universal_newlines=True)
        command_stdout, command_stderr = proc_instance.communicate()
        if proc_instance.returncode != 0:
            syslog.syslog(syslog.LOG_ERR, "[memory_checker] Failed to execute the command '{}'. Return code: '{}'"
                          .format(command, proc_instance.returncode))
            sys.exit(1)
    except (OSError, ValueError) as err:
        syslog.syslog(syslog.LOG_ERR, "[memory_checker] Failed to execute the command '{}'. Error: '{}'"
                      .format(command, err))
        sys.exit(2)

    return command_stdout.strip()


def publish_events(container_name, mem_usage_bytes, threshold_value):
    events_handle = swsscommon.events_init_publisher(EVENTS_PUBLISHER_SOURCE)
    params = swsscommon.FieldValueMap()
    params["ctr_name"] = container_name
    params["mem_usage"] = mem_usage_bytes
    params["threshold"] = threshold_value
    swsscommon.event_publish(events_handle, EVENTS_PUBLISHER_TAG, params)
    swsscommon.events_deinit_publisher(events_handle)


def check_memory_usage(container_name, threshold_value):
    """Checks the memory usage of a container and writes an alerting messages into
    the syslog if the memory usage is larger than the threshold value.

    Args:
        container_name: A string represtents name of a container
        threshold_value: An integer indicates the threshold value (Bytes) of memory usage.

    Returns:
        None.
    """
    command = ["docker", "stats", "--no-stream", "--format", "{{.MemUsage}}", container_name]
    command_stdout = get_command_result(command)
    mem_usage = command_stdout.split("/")[0].strip()
    match_obj = re.match(r"\d+\.?\d*", mem_usage)
    if match_obj:
        mem_usage_value = float(mem_usage[match_obj.start():match_obj.end()])
        mem_usage_unit = mem_usage[match_obj.end():]

        mem_usage_bytes = 0.0
        if mem_usage_unit == "B":
            mem_usage_bytes = mem_usage_value
        elif mem_usage_unit == "KiB":
            mem_usage_bytes = mem_usage_value * 1024
        elif mem_usage_unit == "MiB":
            mem_usage_bytes = mem_usage_value * 1024 ** 2
        elif mem_usage_unit == "GiB":
            mem_usage_bytes = mem_usage_value * 1024 ** 3

        if mem_usage_bytes > threshold_value:
            print("[{}]: Memory usage ({} Bytes) is larger than the threshold ({} Bytes)!"
                  .format(container_name, mem_usage_bytes, threshold_value))
            syslog.syslog(syslog.LOG_INFO, "[{}]: Memory usage ({} Bytes) is larger than the threshold ({} Bytes)!"
                          .format(container_name, mem_usage_bytes, threshold_value))
            # publish event
            publish_events(container_name, "{:.2f}".format(mem_usage_bytes), str(threshold_value))
            sys.exit(3)
    else:
        syslog.syslog(syslog.LOG_ERR, "[memory_checker] Failed to retrieve memory value from '{}'"
                      .format(mem_usage))
        sys.exit(4)


def is_service_active(service_name):
    """Test if service is running.

    Args:
        service_name: A string contains the service name

    Returns:
        True if service is running, False otherwise
    """
    status = subprocess.run(["systemctl", "is-active", "--quiet", service_name])
    return status.returncode == 0


def get_running_container_names():
    """Retrieves names of running containers by talking to the docker daemon.

    Args:
        None.

    Returns:
        running_container_names: A list indicates names of running containers.
    """
    try:
        docker_client = docker.DockerClient(base_url='unix://var/run/docker.sock')
        running_container_list = docker_client.containers.list(filters={"status": "running"})
        running_container_names = [ container.name for container in running_container_list ]
    except (docker.errors.APIError, docker.errors.DockerException) as err:
        syslog.syslog(syslog.LOG_ERR,
                      "Failed to retrieve the running container list from docker daemon! Error message is: '{}'"
                      .format(err))
        sys.exit(5)

    return running_container_names


def main():
    parser = argparse.ArgumentParser(description="Check memory usage of a container \
            and an alerting message will be written into syslog if memory usage \
            is larger than the threshold value", usage="/usr/bin/memory_checker <container_name> <threshold_value_in_bytes>")
    parser.add_argument("container_name", help="container name")
    # TODO: Currently the threshold value is hard coded as a command line argument and will
    # remove this in the new version since we want to read this value from 'CONFIG_DB'.
    parser.add_argument("threshold_value", type=int, help="threshold value in bytes")
    args = parser.parse_args()

    if not is_service_active("docker"):
        syslog.syslog(syslog.LOG_INFO,
                      "[memory_checker] Exits without checking memory usage of container '{}' since docker daemon is not running!"
                      .format(args.container_name))
        sys.exit(0)

    running_container_names = get_running_container_names()
    if args.container_name in running_container_names:
        check_memory_usage(args.container_name, args.threshold_value)
    else:
        syslog.syslog(syslog.LOG_INFO,
                      "[memory_checker] Exits without checking memory usage since container '{}' is not running!"
                      .format(args.container_name))


if __name__ == "__main__":
    main()