]>
git.proxmox.com Git - ceph.git/blob - ceph/qa/tasks/rep_lost_unfound_delete.py
7 from tasks
import ceph_manager
8 from tasks
.util
.rados
import rados
9 from teuthology
import misc
as teuthology
10 from teuthology
.orchestra
import run
12 log
= logging
.getLogger(__name__
)
14 def task(ctx
, config
):
16 Test handling of lost objects.
18 A pretty rigid cluster is brought up and tested by this task
20 POOL
= 'unfounddel_pool'
23 assert isinstance(config
, dict), \
24 'lost_unfound task only accepts a dict for configuration'
25 first_mon
= teuthology
.get_first_mon(ctx
, config
)
26 (mon
,) = ctx
.cluster
.only(first_mon
).remotes
.keys()
28 manager
= ceph_manager
.CephManager(
31 logger
=log
.getChild('ceph_manager'),
34 while len(manager
.get_osd_status()['up']) < 3:
36 manager
.flush_pg_stats([0, 1, 2])
37 manager
.wait_for_clean()
39 manager
.create_pool(POOL
)
41 # something that is always there
42 dummyfile
= '/etc/fstab'
44 # take an osd out until the very end
46 manager
.mark_down_osd(2)
47 manager
.mark_out_osd(2)
49 # kludge to make sure they get a map
50 rados(ctx
, mon
, ['-p', POOL
, 'put', 'dummy', dummyfile
])
52 manager
.flush_pg_stats([0, 1])
53 manager
.wait_for_recovery()
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
])
61 # delay recovery, and make the pg log very long (to prevent backfill)
62 manager
.raw_cluster_cmd(
65 '--osd-recovery-delay-start 1000 --osd-min-pg-log-entries 100000000'
69 manager
.mark_down_osd(0)
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
])
76 # bring osd.0 back up, let it peer, but don't replicate the new
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'
84 manager
.mark_in_osd(0)
85 manager
.wait_till_osd_is_up(0)
87 manager
.flush_pg_stats([0, 1])
88 manager
.wait_till_active()
90 # take out osd.1 and the only copy of those objects.
92 manager
.mark_down_osd(1)
93 manager
.mark_out_osd(1)
94 manager
.raw_cluster_cmd('osd', 'lost', '1', '--yes-i-really-mean-it')
96 # bring up osd.2 so that things would otherwise, in theory, recovery fully
98 manager
.mark_in_osd(2)
99 manager
.wait_till_osd_is_up(2)
101 manager
.flush_pg_stats([0, 2])
102 manager
.wait_till_active()
103 manager
.flush_pg_stats([0, 2])
105 # verify that there are unfound objects
106 unfound
= manager
.get_num_unfound_objects()
107 log
.info("there are %d unfound objects" % unfound
)
110 testdir
= teuthology
.get_testdir(ctx
)
112 if config
.get('parallel_bench', True):
113 procs
.append(mon
.run(
116 " ".join(['adjust-ulimits',
118 '{tdir}/archive/coverage',
120 '--no-log-to-stderr',
121 '--name', 'client.admin',
125 'bench', '240', 'write',
126 ]).format(tdir
=testdir
),
128 logger
=log
.getChild('radosbench.{id}'.format(id='client.admin')),
135 pgs
= manager
.get_pg_stats()
137 if pg
['stat_sum']['num_objects_unfound'] > 0:
138 primary
= 'osd.%d' % pg
['acting'][0]
140 # verify that i can list them direct from the osd
141 log
.info('listing missing/lost in %s state %s', pg
['pgid'],
143 m
= manager
.list_pg_unfound(pg
['pgid'])
145 assert m
['num_unfound'] == pg
['stat_sum']['num_objects_unfound']
147 for o
in m
['objects']:
148 if len(o
['locations']) == 0:
150 assert m
['num_unfound'] == num_unfound
152 log
.info("reverting unfound in %s on %s", pg
['pgid'], primary
)
153 manager
.raw_cluster_cmd('pg', pg
['pgid'],
154 'mark_unfound_lost', 'delete')
156 log
.info("no unfound in %s", pg
['pgid'])
158 manager
.raw_cluster_cmd('tell', 'osd.0', 'debug', 'kick_recovery_wq', '5')
159 manager
.raw_cluster_cmd('tell', 'osd.2', 'debug', 'kick_recovery_wq', '5')
160 manager
.flush_pg_stats([0, 2])
161 manager
.wait_for_recovery()
164 for f
in range(1, 10):
165 err
= rados(ctx
, mon
, ['-p', POOL
, 'get', 'new_%d' % f
, '-'])
167 err
= rados(ctx
, mon
, ['-p', POOL
, 'get', 'existed_%d' % f
, '-'])
169 err
= rados(ctx
, mon
, ['-p', POOL
, 'get', 'existing_%d' % f
, '-'])
172 # see if osd.1 can cope
173 manager
.mark_in_osd(1)
174 manager
.revive_osd(1)
175 manager
.wait_till_osd_is_up(1)
176 manager
.wait_for_clean()
178 manager
.wait_for_clean()