ceph/qa/tasks/rep_lost_unfound_delete.py

   1 """
   2 Lost_unfound
   3 """
   4 import logging
   5 from teuthology.orchestra import run
   6 import ceph_manager
   7 import time
   8 from teuthology import misc as teuthology
   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 = 'unfounddel_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     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
  36     manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
  37     manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
  38     manager.wait_for_clean()
  39
  40     manager.create_pool(POOL)
  41
  42     # something that is always there
  43     dummyfile = '/etc/fstab'
  44
  45     # take an osd out until the very end
  46     manager.kill_osd(2)
  47     manager.mark_down_osd(2)
  48     manager.mark_out_osd(2)
  49
  50     # kludge to make sure they get a map
  51     rados(ctx, mon, ['-p', POOL, 'put', 'dummy', dummyfile])
  52
  53     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
  54     manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
  55     manager.wait_for_recovery()
  56
  57     # create old objects
  58     for f in range(1, 10):
  59         rados(ctx, mon, ['-p', POOL, 'put', 'existing_%d' % f, dummyfile])
  60         rados(ctx, mon, ['-p', POOL, 'put', 'existed_%d' % f, dummyfile])
  61         rados(ctx, mon, ['-p', POOL, 'rm', 'existed_%d' % f])
  62
  63     # delay recovery, and make the pg log very long (to prevent backfill)
  64     manager.raw_cluster_cmd(
  65             'tell', 'osd.1',
  66             'injectargs',
  67             '--osd-recovery-delay-start 1000 --osd-min-pg-log-entries 100000000'
  68             )
  69
  70     manager.kill_osd(0)
  71     manager.mark_down_osd(0)
  72
  73     for f in range(1, 10):
  74         rados(ctx, mon, ['-p', POOL, 'put', 'new_%d' % f, dummyfile])
  75         rados(ctx, mon, ['-p', POOL, 'put', 'existed_%d' % f, dummyfile])
  76         rados(ctx, mon, ['-p', POOL, 'put', 'existing_%d' % f, dummyfile])
  77
  78     # bring osd.0 back up, let it peer, but don't replicate the new
  79     # objects...
  80     log.info('osd.0 command_args is %s' % 'foo')
  81     log.info(ctx.daemons.get_daemon('osd', 0).command_args)
  82     ctx.daemons.get_daemon('osd', 0).command_kwargs['args'].extend([
  83             '--osd-recovery-delay-start', '1000'
  84             ])
  85     manager.revive_osd(0)
  86     manager.mark_in_osd(0)
  87     manager.wait_till_osd_is_up(0)
  88
  89     manager.raw_cluster_cmd('tell', 'osd.1', 'flush_pg_stats')
  90     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
  91     manager.wait_till_active()
  92
  93     # take out osd.1 and the only copy of those objects.
  94     manager.kill_osd(1)
  95     manager.mark_down_osd(1)
  96     manager.mark_out_osd(1)
  97     manager.raw_cluster_cmd('osd', 'lost', '1', '--yes-i-really-mean-it')
  98
  99     # bring up osd.2 so that things would otherwise, in theory, recovery fully
 100     manager.revive_osd(2)
 101     manager.mark_in_osd(2)
 102     manager.wait_till_osd_is_up(2)
 103
 104     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
 105     manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
 106     manager.wait_till_active()
 107     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
 108     manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
 109
 110     # verify that there are unfound objects
 111     unfound = manager.get_num_unfound_objects()
 112     log.info("there are %d unfound objects" % unfound)
 113     assert unfound
 114
 115     testdir = teuthology.get_testdir(ctx)
 116     procs = []
 117     if config.get('parallel_bench', True):
 118         procs.append(mon.run(
 119             args=[
 120                 "/bin/sh", "-c",
 121                 " ".join(['adjust-ulimits',
 122                           'ceph-coverage',
 123                           '{tdir}/archive/coverage',
 124                           'rados',
 125                           '--no-log-to-stderr',
 126                           '--name', 'client.admin',
 127                           '-b', str(4<<10),
 128                           '-p' , POOL,
 129                           '-t', '20',
 130                           'bench', '240', 'write',
 131                       ]).format(tdir=testdir),
 132             ],
 133             logger=log.getChild('radosbench.{id}'.format(id='client.admin')),
 134             stdin=run.PIPE,
 135             wait=False
 136         ))
 137     time.sleep(10)
 138
 139     # mark stuff lost
 140     pgs = manager.get_pg_stats()
 141     for pg in pgs:
 142         if pg['stat_sum']['num_objects_unfound'] > 0:
 143             primary = 'osd.%d' % pg['acting'][0]
 144
 145             # verify that i can list them direct from the osd
 146             log.info('listing missing/lost in %s state %s', pg['pgid'],
 147                      pg['state']);
 148             m = manager.list_pg_missing(pg['pgid'])
 149             #log.info('%s' % m)
 150             assert m['num_unfound'] == pg['stat_sum']['num_objects_unfound']
 151             num_unfound=0
 152             for o in m['objects']:
 153                 if len(o['locations']) == 0:
 154                     num_unfound += 1
 155             assert m['num_unfound'] == num_unfound
 156
 157             log.info("reverting unfound in %s on %s", pg['pgid'], primary)
 158             manager.raw_cluster_cmd('pg', pg['pgid'],
 159                                     'mark_unfound_lost', 'delete')
 160         else:
 161             log.info("no unfound in %s", pg['pgid'])
 162
 163     manager.raw_cluster_cmd('tell', 'osd.0', 'debug', 'kick_recovery_wq', '5')
 164     manager.raw_cluster_cmd('tell', 'osd.2', 'debug', 'kick_recovery_wq', '5')
 165     manager.raw_cluster_cmd('tell', 'osd.0', 'flush_pg_stats')
 166     manager.raw_cluster_cmd('tell', 'osd.2', 'flush_pg_stats')
 167     manager.wait_for_recovery()
 168
 169     # verify result
 170     for f in range(1, 10):
 171         err = rados(ctx, mon, ['-p', POOL, 'get', 'new_%d' % f, '-'])
 172         assert err
 173         err = rados(ctx, mon, ['-p', POOL, 'get', 'existed_%d' % f, '-'])
 174         assert err
 175         err = rados(ctx, mon, ['-p', POOL, 'get', 'existing_%d' % f, '-'])
 176         assert err
 177
 178     # see if osd.1 can cope
 179     manager.revive_osd(1)
 180     manager.mark_in_osd(1)
 181     manager.wait_till_osd_is_up(1)
 182     manager.wait_for_clean()
 183     run.wait(procs)
 184