ceph/qa/tasks/mon_recovery.py

   1 """
   2 Monitor recovery
   3 """
   4 import logging
   5 from tasks import ceph_manager
   6 from teuthology import misc as teuthology
   7
   8
   9 log = logging.getLogger(__name__)
  10
  11 def task(ctx, config):
  12     """
  13     Test monitor recovery.
  14     """
  15     if config is None:
  16         config = {}
  17     assert isinstance(config, dict), \
  18         'task only accepts a dict for configuration'
  19     first_mon = teuthology.get_first_mon(ctx, config)
  20     (mon,) = ctx.cluster.only(first_mon).remotes.keys()
  21
  22     manager = ceph_manager.CephManager(
  23         mon,
  24         ctx=ctx,
  25         logger=log.getChild('ceph_manager'),
  26         )
  27
  28     mons = [f.split('.')[1] for f in teuthology.get_mon_names(ctx)]
  29     log.info("mon ids = %s" % mons)
  30
  31     manager.wait_for_mon_quorum_size(len(mons))
  32
  33     log.info('verifying all monitors are in the quorum')
  34     for m in mons:
  35         s = manager.get_mon_status(m)
  36         assert s['state'] == 'leader' or s['state'] == 'peon'
  37         assert len(s['quorum']) == len(mons)
  38
  39     log.info('restarting each monitor in turn')
  40     for m in mons:
  41         # stop a monitor
  42         manager.kill_mon(m)
  43         manager.wait_for_mon_quorum_size(len(mons) - 1)
  44
  45         # restart
  46         manager.revive_mon(m)
  47         manager.wait_for_mon_quorum_size(len(mons))
  48
  49     # in forward and reverse order,
  50     rmons = mons
  51     rmons.reverse()
  52     for mons in mons, rmons:
  53         log.info('stopping all monitors')
  54         for m in mons:
  55             manager.kill_mon(m)
  56
  57         log.info('forming a minimal quorum for %s, then adding monitors' % mons)
  58         qnum = (len(mons) // 2) + 1
  59         num = 0
  60         for m in mons:
  61             manager.revive_mon(m)
  62             num += 1
  63             if num >= qnum:
  64                 manager.wait_for_mon_quorum_size(num)
  65
  66     # on both leader and non-leader ranks...
  67     for rank in [0, 1]:
  68         # take one out
  69         log.info('removing mon %s' % mons[rank])
  70         manager.kill_mon(mons[rank])
  71         manager.wait_for_mon_quorum_size(len(mons) - 1)
  72
  73         log.info('causing some monitor log activity')
  74         m = 30
  75         for n in range(1, m):
  76             manager.raw_cluster_cmd('log', '%d of %d' % (n, m))
  77
  78         log.info('adding mon %s back in' % mons[rank])
  79         manager.revive_mon(mons[rank])
  80         manager.wait_for_mon_quorum_size(len(mons))