ceph/src/ceph-crash.in

   1 #!@Python3_EXECUTABLE@
   2 # -*- mode:python -*-
   3 # vim: ts=4 sw=4 smarttab expandtab
   4
   5 import argparse
   6 import logging
   7 import os
   8 import signal
   9 import socket
  10 import subprocess
  11 import sys
  12 import time
  13
  14 logging.basicConfig(level=logging.INFO)
  15 log = logging.getLogger('ceph-crash')
  16
  17 auth_names = ['client.crash.%s' % socket.gethostname(),
  18               'client.crash',
  19               'client.admin']
  20
  21 def parse_args():
  22     parser = argparse.ArgumentParser()
  23     parser.add_argument(
  24         '-p', '--path', default='/var/lib/ceph/crash',
  25         help='base path to monitor for crash dumps')
  26     parser.add_argument(
  27         '-d', '--delay', default=10.0, type=float,
  28         help='minutes to delay between scans (0 to exit after one)',
  29     )
  30     parser.add_argument(
  31         '--name', '-n',
  32         help='ceph name to authenticate as (default: try client.crash, client.admin)')
  33     parser.add_argument(
  34         '--log-level', '-l',
  35         help='log level output (default: INFO), support INFO or DEBUG')
  36
  37     return parser.parse_args()
  38
  39
  40 def post_crash(path):
  41     rc = 0
  42     for n in auth_names:
  43         pr = subprocess.Popen(
  44             args=['timeout', '30', 'ceph',
  45                   '-n', n,
  46                   'crash', 'post', '-i', '-'],
  47             stdin=subprocess.PIPE,
  48             stderr=subprocess.PIPE,
  49         )
  50         f = open(os.path.join(path, 'meta'), 'rb')
  51         stderr = pr.communicate(input=f.read())
  52         rc = pr.wait()
  53         f.close()
  54         if rc != 0 or stderr != "":
  55             log.warning('post %s as %s failed: %s' % (path, n, stderr))
  56         if rc == 0:
  57             break
  58     return rc
  59
  60
  61 def scrape_path(path):
  62     for p in os.listdir(path):
  63         crashpath = os.path.join(path, p)
  64         metapath = os.path.join(crashpath, 'meta')
  65         donepath = os.path.join(crashpath, 'done')
  66         if os.path.isfile(metapath):
  67             if not os.path.isfile(donepath):
  68                 # hang out just for a bit; either we interrupted the dump
  69                 # or the daemon crashed before finishing it
  70                 time.sleep(1)
  71                 if not os.path.isfile(donepath):
  72                     return
  73             # ok, we can process this one
  74             rc = post_crash(crashpath)
  75             if rc == 0:
  76                 os.rename(crashpath, os.path.join(path, 'posted/', p))
  77                 log.debug(
  78                     "posted %s and renamed %s -> %s " %
  79                     (metapath, p, os.path.join('posted/', p))
  80                 )
  81
  82 def handler(signum):
  83     print('*** Interrupted with signal %d ***' % signum)
  84     sys.exit(0)
  85
  86 def main():
  87     global auth_names
  88     # exit code 0 on SIGINT, SIGTERM
  89     signal.signal(signal.SIGINT, handler)
  90     signal.signal(signal.SIGTERM, handler)
  91
  92     args = parse_args()
  93     if args.log_level == 'DEBUG':
  94         log.setLevel(logging.DEBUG)
  95
  96     postdir = os.path.join(args.path, 'posted')
  97     if args.name:
  98         auth_names = [args.name]
  99
 100     while not os.path.isdir(postdir):
 101         log.error("directory %s does not exist; please create" % postdir)
 102         time.sleep(30)
 103
 104     log.info("monitoring path %s, delay %ds" % (args.path, args.delay * 60.0))
 105     while True:
 106         scrape_path(args.path)
 107         if args.delay == 0:
 108             sys.exit(0)
 109         time.sleep(args.delay * 60)
 110
 111
 112 if __name__ == "__main__":
 113     main()