ceph/qa/tasks/lost_unfound.py

   1 """
   2 Lost_unfound
   3 """
   4 import logging
   5 import time
   6 import ceph_manager
   7 from teuthology import misc as teuthology
   8 from teuthology.orchestra import run
   9 from util.rados import rados
  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.iterkeys()
  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.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
  52     manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
  53     manager.wait_for_recovery()
  54
  55     # create old objects
  56     for f in range(1, 10):
  57         rados(ctx, mon, ['-p', POOL, 'put', 'existing_%d' % f, dummyfile])
  58         rados(ctx, mon, ['-p', POOL, 'put', 'existed_%d' % f, dummyfile])
  59         rados(ctx, mon, ['-p', POOL, 'rm', 'existed_%d' % f])
  60
  61     # delay recovery, and make the pg log very long (to prevent backfill)
  62     manager.raw_cluster_cmd(
  63             'tell', 'osd.1',
  64             'injectargs',
  65             '--osd-recovery-delay-start 1000 --osd-min-pg-log-entries 100000000'
  66             )
  67
  68     manager.kill_osd(0)
  69     manager.mark_down_osd(0)
  70
  71     for f in range(1, 10):
  72         rados(ctx, mon, ['-p', POOL, 'put', 'new_%d' % f, dummyfile])
  73         rados(ctx, mon, ['-p', POOL, 'put', 'existed_%d' % f, dummyfile])
  74         rados(ctx, mon, ['-p', POOL, 'put', 'existing_%d' % f, dummyfile])
  75
  76     # bring osd.0 back up, let it peer, but don't replicate the new
  77     # objects...
  78     log.info('osd.0 command_args is %s' % 'foo')
  79     log.info(ctx.daemons.get_daemon('osd', 0).command_args)
  80     ctx.daemons.get_daemon('osd', 0).command_kwargs['args'].extend([
  81             '--osd-recovery-delay-start', '1000'
  82             ])
  83     manager.revive_osd(0)
  84     manager.mark_in_osd(0)
  85     manager.wait_till_osd_is_up(0)
  86
  87     manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
  88     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
  89     manager.wait_till_active()
  90
  91     # take out osd.1 and the only copy of those objects.
  92     manager.kill_osd(1)
  93     manager.mark_down_osd(1)
  94     manager.mark_out_osd(1)
  95     manager.raw_cluster_cmd('osd', 'lost', '1', '--yes-i-really-mean-it')
  96
  97     # bring up osd.2 so that things would otherwise, in theory, recovery fully
  98     manager.revive_osd(2)
  99     manager.mark_in_osd(2)
 100     manager.wait_till_osd_is_up(2)
 101
 102     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
 103     manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
 104     manager.wait_till_active()
 105     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
 106     manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
 107
 108     # verify that there are unfound objects
 109     unfound = manager.get_num_unfound_objects()
 110     log.info("there are %d unfound objects" % unfound)
 111     assert unfound
 112
 113     testdir = teuthology.get_testdir(ctx)
 114     procs = []
 115     if config.get('parallel_bench', True):
 116         procs.append(mon.run(
 117             args=[
 118                 "/bin/sh", "-c",
 119                 " ".join(['adjust-ulimits',
 120                           'ceph-coverage',
 121                           '{tdir}/archive/coverage',
 122                           'rados',
 123                           '--no-log-to-stderr',
 124                           '--name', 'client.admin',
 125                           '-b', str(4<<10),
 126                           '-p' , POOL,
 127                           '-t', '20',
 128                           'bench', '240', 'write',
 129                       ]).format(tdir=testdir),
 130             ],
 131             logger=log.getChild('radosbench.{id}'.format(id='client.admin')),
 132             stdin=run.PIPE,
 133             wait=False
 134         ))
 135     time.sleep(10)
 136
 137     # mark stuff lost
 138     pgs = manager.get_pg_stats()
 139     for pg in pgs:
 140         if pg['stat_sum']['num_objects_unfound'] > 0:
 141             primary = 'osd.%d' % pg['acting'][0]
 142
 143             # verify that i can list them direct from the osd
 144             log.info('listing missing/lost in %s state %s', pg['pgid'],
 145                      pg['state']);
 146             m = manager.list_pg_missing(pg['pgid'])
 147             #log.info('%s' % m)
 148             assert m['num_unfound'] == pg['stat_sum']['num_objects_unfound']
 149             num_unfound=0
 150             for o in m['objects']:
 151                 if len(o['locations']) == 0:
 152                     num_unfound += 1
 153             assert m['num_unfound'] == num_unfound
 154
 155             log.info("reverting unfound in %s on %s", pg['pgid'], primary)
 156             manager.raw_cluster_cmd('pg', pg['pgid'],
 157                                     'mark_unfound_lost', 'revert')
 158         else:
 159             log.info("no unfound in %s", pg['pgid'])
 160
 161     manager.raw_cluster_cmd('tell', 'osd.0', 'debug', 'kick_recovery_wq', '5')
 162     manager.raw_cluster_cmd('tell', 'osd.2', 'debug', 'kick_recovery_wq', '5')
 163     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
 164     manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
 165     manager.wait_for_recovery()
 166
 167     # verify result
 168     for f in range(1, 10):
 169         err = rados(ctx, mon, ['-p', POOL, 'get', 'new_%d' % f, '-'])
 170         assert err
 171         err = rados(ctx, mon, ['-p', POOL, 'get', 'existed_%d' % f, '-'])
 172         assert err
 173         err = rados(ctx, mon, ['-p', POOL, 'get', 'existing_%d' % f, '-'])
 174         assert not err
 175
 176     # see if osd.1 can cope
 177     manager.revive_osd(1)
 178     manager.mark_in_osd(1)
 179     manager.wait_till_osd_is_up(1)
 180     manager.wait_for_clean()
 181     run.wait(procs)