7 from teuthology import misc as teuthology
8 from teuthology.orchestra import run
9 from util.rados import rados
11 log = logging.getLogger(__name__)
13 def task(ctx, config):
15 Test handling of lost objects.
17 A pretty rigid cluseter is brought up andtested by this task
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()
27 manager = ceph_manager.CephManager(
30 logger=log.getChild('ceph_manager'),
33 while len(manager.get_osd_status()['up']) < 3:
36 manager.wait_for_clean()
38 manager.create_pool(POOL)
40 # something that is always there
41 dummyfile = '/etc/fstab'
43 # take an osd out until the very end
45 manager.mark_down_osd(2)
46 manager.mark_out_osd(2)
48 # kludge to make sure they get a map
49 rados(ctx, mon, ['-p', POOL, 'put', 'dummy', dummyfile])
51 manager.flush_pg_stats([0, 1])
52 manager.wait_for_recovery()
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])
60 # delay recovery, and make the pg log very long (to prevent backfill)
61 manager.raw_cluster_cmd(
64 '--osd-recovery-delay-start 1000 --osd-min-pg-log-entries 100000000'
68 manager.mark_down_osd(0)
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])
75 # bring osd.0 back up, let it peer, but don't replicate the new
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'
83 manager.mark_in_osd(0)
84 manager.wait_till_osd_is_up(0)
86 manager.flush_pg_stats([1, 0])
87 manager.wait_till_active()
89 # take out osd.1 and the only copy of those objects.
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')
95 # bring up osd.2 so that things would otherwise, in theory, recovery fully
97 manager.mark_in_osd(2)
98 manager.wait_till_osd_is_up(2)
100 manager.flush_pg_stats([0, 2])
101 manager.wait_till_active()
102 manager.flush_pg_stats([0, 2])
104 # verify that there are unfound objects
105 unfound = manager.get_num_unfound_objects()
106 log.info("there are %d unfound objects" % unfound)
109 testdir = teuthology.get_testdir(ctx)
111 if config.get('parallel_bench', True):
112 procs.append(mon.run(
115 " ".join(['adjust-ulimits',
117 '{tdir}/archive/coverage',
119 '--no-log-to-stderr',
120 '--name', 'client.admin',
124 'bench', '240', 'write',
125 ]).format(tdir=testdir),
127 logger=log.getChild('radosbench.{id}'.format(id='client.admin')),
134 pgs = manager.get_pg_stats()
136 if pg['stat_sum']['num_objects_unfound'] > 0:
137 primary = 'osd.%d' % pg['acting'][0]
139 # verify that i can list them direct from the osd
140 log.info('listing missing/lost in %s state %s', pg['pgid'],
142 m = manager.list_pg_missing(pg['pgid'])
144 assert m['num_unfound'] == pg['stat_sum']['num_objects_unfound']
146 for o in m['objects']:
147 if len(o['locations']) == 0:
149 assert m['num_unfound'] == num_unfound
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')
155 log.info("no unfound in %s", pg['pgid'])
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()
163 for f in range(1, 10):
164 err = rados(ctx, mon, ['-p', POOL, 'get', 'new_%d' % f, '-'])
166 err = rados(ctx, mon, ['-p', POOL, 'get', 'existed_%d' % f, '-'])
168 err = rados(ctx, mon, ['-p', POOL, 'get', 'existing_%d' % f, '-'])
171 # see if osd.1 can cope
172 manager.revive_osd(1)
173 manager.mark_in_osd(1)
174 manager.wait_till_osd_is_up(1)
175 manager.wait_for_clean()