1#!/usr/bin/python
2#
3# Copyright (c) 2015 The Chromium OS Authors. All rights reserved.
4# Use of this source code is governed by a BSD-style license that can be
5# found in the LICENSE file.
6
7
8"""Script to check the number of long-running processes.
9
10This script gets the number of processes for "gsutil" and "autoserv"
11that are running more than 24 hours, and throws the number to stats
12dashboard.
13
14This script depends on the "etimes" user-defined format of "ps".
15Goobuntu 14.04 has the version of ps that supports etimes, but not
16Goobuntu 12.04.
17"""
18
19
20import subprocess
21
22from autotest_lib.server import site_utils
23
24try:
25    from chromite.lib import metrics
26except ImportError:
27    metrics = site_utils.metrics_mock
28
29
30PROGRAM_TO_CHECK_SET = set(['gsutil', 'autoserv'])
31
32def check_proc(prog, max_elapsed_sec):
33    """Check the number of long-running processes for a given program.
34
35    Finds out the number of processes for a given program that have run
36    more than a given elapsed time.
37    Sends out the number to stats dashboard.
38
39    @param prog: Program name.
40    @param max_elapsed_sec: Max elapsed time in seconds. Processes that
41                            have run more than this value will be caught.
42    """
43    cmd = ('ps -eo etimes,args | grep "%s" | awk \'{if($1 > %d) print $0}\' | '
44           'wc -l' % (prog, max_elapsed_sec))
45    count = int(subprocess.check_output(cmd, shell = True))
46
47    if prog not in PROGRAM_TO_CHECK_SET:
48        prog = 'unknown'
49
50    metrics.Gauge('chromeos/autotest/hung_processes').set(
51            count, fields={'program': prog}
52    )
53
54
55def main():
56    """Main script. """
57    with site_utils.SetupTsMonGlobalState('check_hung_proc', short_lived=True):
58        for p in PROGRAM_TO_CHECK_SET:
59            check_proc(p, 86400)
60
61
62if __name__ == '__main__':
63    main()
64