zulip/scripts/nagios/check-rabbitmq-consumers

#!/usr/bin/env python3

import sys
import time
import argparse
import configparser
from collections import defaultdict
import os
import subprocess

if False:
    # See https://zulip.readthedocs.io/en/latest/testing/mypy.html#mypy-in-production-scripts
    from typing import Dict

states = {
    0: "OK",
    1: "WARNING",
    2: "CRITICAL",
    3: "UNKNOWN"
}

if 'USER' in os.environ and not os.environ['USER'] in ['root', 'rabbitmq']:
    print("This script must be run as the root or rabbitmq user")


usage = """Usage: check-rabbitmq-consumers --queue=[queue-name] --min-threshold=[min-threshold]"""

parser = argparse.ArgumentParser(usage=usage)
parser.add_argument('--min-threshold',
                    dest='min_count',
                    type=int,
                    default=1,
                    action='store')

options = parser.parse_args()

config_file = configparser.RawConfigParser()
config_file.read("/etc/zulip/zulip.conf")
def get_config(section, key, default_value):
    # type: (str, str, str) -> str
    if config_file.has_option(section, key):
        return config_file.get(section, key)
    return default_value
TORNADO_PROCESSES = int(get_config('application_server', 'tornado_processes', '1'))

output = subprocess.check_output(['/usr/sbin/rabbitmqctl', 'list_consumers'],
                                 universal_newlines=True)

consumers = defaultdict(int)  # type: Dict[str, int]

sys.path.append(os.path.join(os.path.dirname(os.path.dirname(os.path.dirname(__file__)))))
queues = {
    'deferred_work'
    'digest_emails',
    'email_mirror',
    'embed_links',
    'embedded_bots',
    'error_reports',
    'feedback_messages',
    'invites',
    'message_sender',
    'missedmessage_emails',
    'missedmessage_email_senders',
    'email_senders',
    'missedmessage_mobile_notifications',
    'outgoing_webhooks',
    'signups',
    'slow_queries',
    'user_activity'
    'user_activity_interval',
    'user_presence',
    # These queues may not be present if settings.TORNADO_PROCESSES > 1
    'notify_tornado',
    'tornado_return',
}

for queue_name in queues:
    queue_name = queue_name.strip()
    consumers[queue_name] = 0

for line in output.split('\n'):
    parts = line.split('\t')
    if len(parts) >= 2:
        queue_name = parts[0]
        if queue_name.startswith("tornado_return_"):
            queue_name = "tornado_return"
        if queue_name.startswith("notify_tornado_"):
            queue_name = "notify_tornado"
        consumers[queue_name] += 1

now = int(time.time())

for queue_name in consumers.keys():
    state_file_path = "/var/lib/nagios_state/check-rabbitmq-consumers-" + queue_name
    state_file_tmp = state_file_path + "-tmp"

    target_count = options.min_count
    if queue_name in ["tornado_return", "notify_tornado"]:
        target_count = TORNADO_PROCESSES

    if consumers[queue_name] < target_count:
        status = 2
    else:
        status = 0
    with open(state_file_tmp, "w") as f:
        f.write("%s|%s|%s|queue %s has %s consumers, needs %s\n" % (
            now, status, states[status], queue_name,
            consumers[queue_name], target_count))
    os.rename(state_file_tmp, state_file_path)