ceph/qa/tasks/lost_unfound.py

   1 """
   2 Lost_unfound
   3 """
   4 import logging
   5 import time
   6 from tasks import ceph_manager
   7 from tasks.util.rados import rados
   8 from teuthology import misc as teuthology
   9 from teuthology.orchestra import run
  10
  11 log = logging.getLogger(__name__)
  12
  13 def task(ctx, config):
  14     """
  15     Test handling of lost objects.
  16
  17     A pretty rigid cluseter is brought up andtested by this task
  18     """
  19     POOL = 'unfound_pool'
  20     if config is None:
  21         config = {}
  22     assert isinstance(config, dict), \
  23         'lost_unfound task only accepts a dict for configuration'
  24     first_mon = teuthology.get_first_mon(ctx, config)
  25     (mon,) = ctx.cluster.only(first_mon).remotes.keys()
  26
  27     manager = ceph_manager.CephManager(
  28         mon,
  29         ctx=ctx,
  30         logger=log.getChild('ceph_manager'),
  31         )
  32
  33     while len(manager.get_osd_status()['up']) < 3:
  34         time.sleep(10)
  35
  36     manager.wait_for_clean()
  37
  38     manager.create_pool(POOL)
  39
  40     # something that is always there
  41     dummyfile = '/etc/fstab'
  42
  43     # take an osd out until the very end
  44     manager.kill_osd(2)
  45     manager.mark_down_osd(2)
  46     manager.mark_out_osd(2)
  47
  48     # kludge to make sure they get a map
  49     rados(ctx, mon, ['-p', POOL, 'put', 'dummy', dummyfile])
  50
  51     manager.flush_pg_stats([0, 1])
  52     manager.wait_for_recovery()
  53
  54     # create old objects
  55     for f in range(1, 10):
  56         rados(ctx, mon, ['-p', POOL, 'put', 'existing_%d' % f, dummyfile])
  57         rados(ctx, mon, ['-p', POOL, 'put', 'existed_%d' % f, dummyfile])
  58         rados(ctx, mon, ['-p', POOL, 'rm', 'existed_%d' % f])
  59
  60     # delay recovery, and make the pg log very long (to prevent backfill)
  61     manager.raw_cluster_cmd(
  62             'tell', 'osd.1',
  63             'injectargs',
  64             '--osd-recovery-delay-start 1000 --osd-min-pg-log-entries 100000000'
  65             )
  66
  67     manager.kill_osd(0)
  68     manager.mark_down_osd(0)
  69
  70     for f in range(1, 10):
  71         rados(ctx, mon, ['-p', POOL, 'put', 'new_%d' % f, dummyfile])
  72         rados(ctx, mon, ['-p', POOL, 'put', 'existed_%d' % f, dummyfile])
  73         rados(ctx, mon, ['-p', POOL, 'put', 'existing_%d' % f, dummyfile])
  74
  75     # bring osd.0 back up, let it peer, but don't replicate the new
  76     # objects...
  77     log.info('osd.0 command_args is %s' % 'foo')
  78     log.info(ctx.daemons.get_daemon('osd', 0).command_args)
  79     ctx.daemons.get_daemon('osd', 0).command_kwargs['args'].extend([
  80             '--osd-recovery-delay-start', '1000'
  81             ])
  82     manager.revive_osd(0)
  83     manager.mark_in_osd(0)
  84     manager.wait_till_osd_is_up(0)
  85
  86     manager.flush_pg_stats([1, 0])
  87     manager.wait_till_active()
  88
  89     # take out osd.1 and the only copy of those objects.
  90     manager.kill_osd(1)
  91     manager.mark_down_osd(1)
  92     manager.mark_out_osd(1)
  93     manager.raw_cluster_cmd('osd', 'lost', '1', '--yes-i-really-mean-it')
  94
  95     # bring up osd.2 so that things would otherwise, in theory, recovery fully
  96     manager.revive_osd(2)
  97     manager.mark_in_osd(2)
  98     manager.wait_till_osd_is_up(2)
  99
 100     manager.flush_pg_stats([0, 2])
 101     manager.wait_till_active()
 102     manager.flush_pg_stats([0, 2])
 103
 104     # verify that there are unfound objects
 105     unfound = manager.get_num_unfound_objects()
 106     log.info("there are %d unfound objects" % unfound)
 107     assert unfound
 108
 109     testdir = teuthology.get_testdir(ctx)
 110     procs = []
 111     if config.get('parallel_bench', True):
 112         procs.append(mon.run(
 113             args=[
 114                 "/bin/sh", "-c",
 115                 " ".join(['adjust-ulimits',
 116                           'ceph-coverage',
 117                           '{tdir}/archive/coverage',
 118                           'rados',
 119                           '--no-log-to-stderr',
 120                           '--name', 'client.admin',
 121                           '-b', str(4<<10),
 122                           '-p' , POOL,
 123                           '-t', '20',
 124                           'bench', '240', 'write',
 125                       ]).format(tdir=testdir),
 126             ],
 127             logger=log.getChild('radosbench.{id}'.format(id='client.admin')),
 128             stdin=run.PIPE,
 129             wait=False
 130         ))
 131     time.sleep(10)
 132
 133     # mark stuff lost
 134     pgs = manager.get_pg_stats()
 135     for pg in pgs:
 136         if pg['stat_sum']['num_objects_unfound'] > 0:
 137             primary = 'osd.%d' % pg['acting'][0]
 138
 139             # verify that i can list them direct from the osd
 140             log.info('listing missing/lost in %s state %s', pg['pgid'],
 141                      pg['state']);
 142             m = manager.list_pg_unfound(pg['pgid'])
 143             #log.info('%s' % m)
 144             assert m['num_unfound'] == pg['stat_sum']['num_objects_unfound']
 145             num_unfound=0
 146             for o in m['objects']:
 147                 if len(o['locations']) == 0:
 148                     num_unfound += 1
 149             assert m['num_unfound'] == num_unfound
 150
 151             log.info("reverting unfound in %s on %s", pg['pgid'], primary)
 152             manager.raw_cluster_cmd('pg', pg['pgid'],
 153                                     'mark_unfound_lost', 'revert')
 154         else:
 155             log.info("no unfound in %s", pg['pgid'])
 156
 157     manager.raw_cluster_cmd('tell', 'osd.0', 'debug', 'kick_recovery_wq', '5')
 158     manager.raw_cluster_cmd('tell', 'osd.2', 'debug', 'kick_recovery_wq', '5')
 159     manager.flush_pg_stats([0, 2])
 160     manager.wait_for_recovery()
 161
 162     # verify result
 163     for f in range(1, 10):
 164         err = rados(ctx, mon, ['-p', POOL, 'get', 'new_%d' % f, '-'])
 165         assert err
 166         err = rados(ctx, mon, ['-p', POOL, 'get', 'existed_%d' % f, '-'])
 167         assert err
 168         err = rados(ctx, mon, ['-p', POOL, 'get', 'existing_%d' % f, '-'])
 169         assert not err
 170
 171     # see if osd.1 can cope
 172     manager.mark_in_osd(1)
 173     manager.revive_osd(1)
 174     manager.wait_till_osd_is_up(1)
 175     manager.wait_for_clean()
 176     run.wait(procs)