zulip/bots/check-rabbitmq-queue

#!/usr/bin/env python

import sys
import re
import time
import os
import subprocess

WARN_THRESHOLD = 10
CRIT_THRESHOLD = 50

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

re = re.compile(r'(\w+)\t(\d+)')
output = subprocess.check_output(['/usr/sbin/rabbitmqctl', 'list_queues'], shell=False)

status = 0
max_count = 0
warn_queues = []

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"

for line in output.split("\n"):
    line = line.strip()
    m = re.match(line)
    if m:
        queue = m.group(1)
        count = int(m.group(2))
        this_status = 0
        if count > CRIT_THRESHOLD:
            this_status = 2
            warn_queues.append(queue)
        elif count > WARN_THRESHOLD:
            this_status = max(status, 1)
            warn_queues.append(queue)

        status = max(status, this_status)
        max_count = max(max_count, count)

warn_about =  ", ".join(warn_queues)
now = int(time.time())
now_struct = time.gmtime(now)

# While we are sending digest emails, at 11am each weekday, the mail queues can
# get backed up; don't alert on those.
if not set(warn_queues) - set(("missedmessage_emails", "digest_emails")) and \
        now_struct.tm_hour == 16 and now_struct.tm_min < 25:
    status = 0
    print("%s|%s|%s|processing digests, not alerting on elevated mail queues" % (
            now, status, states[status]))
    exit(0)

if status > 0:
    print("%s|%s|%s|max count %s, queues affected: %s" % (now, status, states[status], max_count, warn_about))
else:
    print("%s|%s|%s|queues normal, max count %s" % (now, status, states[status], max_count))