summaryrefslogtreecommitdiffstats
path: root/qa/tasks/rep_lost_unfound_delete.py
blob: 8e99ade27c1654bdbeee51caf85e813aea768603 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
"""
Lost_unfound
"""
import logging
import time

from tasks import ceph_manager
from tasks.util.rados import rados
from teuthology import misc as teuthology
from teuthology.orchestra import run

log = logging.getLogger(__name__)

def task(ctx, config):
    """
    Test handling of lost objects.

    A pretty rigid cluster is brought up and tested by this task
    """
    POOL = 'unfounddel_pool'
    if config is None:
        config = {}
    assert isinstance(config, dict), \
        'lost_unfound task only accepts a dict for configuration'
    first_mon = teuthology.get_first_mon(ctx, config)
    (mon,) = ctx.cluster.only(first_mon).remotes.keys()

    manager = ceph_manager.CephManager(
        mon,
        ctx=ctx,
        logger=log.getChild('ceph_manager'),
        )

    while len(manager.get_osd_status()['up']) < 3:
        time.sleep(10)
    manager.flush_pg_stats([0, 1, 2])
    manager.wait_for_clean()

    manager.create_pool(POOL)

    # something that is always there
    dummyfile = '/etc/fstab'

    # take an osd out until the very end
    manager.kill_osd(2)
    manager.mark_down_osd(2)
    manager.mark_out_osd(2)

    # kludge to make sure they get a map
    rados(ctx, mon, ['-p', POOL, 'put', 'dummy', dummyfile])

    manager.flush_pg_stats([0, 1])
    manager.wait_for_recovery()

    # create old objects
    for f in range(1, 10):
        rados(ctx, mon, ['-p', POOL, 'put', 'existing_%d' % f, dummyfile])
        rados(ctx, mon, ['-p', POOL, 'put', 'existed_%d' % f, dummyfile])
        rados(ctx, mon, ['-p', POOL, 'rm', 'existed_%d' % f])

    # delay recovery, and make the pg log very long (to prevent backfill)
    manager.raw_cluster_cmd(
            'tell', 'osd.1',
            'injectargs',
            '--osd-recovery-delay-start 1000 --osd-min-pg-log-entries 100000000'
            )

    manager.kill_osd(0)
    manager.mark_down_osd(0)
    
    for f in range(1, 10):
        rados(ctx, mon, ['-p', POOL, 'put', 'new_%d' % f, dummyfile])
        rados(ctx, mon, ['-p', POOL, 'put', 'existed_%d' % f, dummyfile])
        rados(ctx, mon, ['-p', POOL, 'put', 'existing_%d' % f, dummyfile])

    # bring osd.0 back up, let it peer, but don't replicate the new
    # objects...
    log.info('osd.0 command_args is %s' % 'foo')
    log.info(ctx.daemons.get_daemon('osd', 0).command_args)
    ctx.daemons.get_daemon('osd', 0).command_kwargs['args'].extend([
            '--osd-recovery-delay-start', '1000'
            ])
    manager.revive_osd(0)
    manager.mark_in_osd(0)
    manager.wait_till_osd_is_up(0)

    manager.flush_pg_stats([0, 1])
    manager.wait_till_active()

    # take out osd.1 and the only copy of those objects.
    manager.kill_osd(1)
    manager.mark_down_osd(1)
    manager.mark_out_osd(1)
    manager.raw_cluster_cmd('osd', 'lost', '1', '--yes-i-really-mean-it')

    # bring up osd.2 so that things would otherwise, in theory, recovery fully
    manager.revive_osd(2)
    manager.mark_in_osd(2)
    manager.wait_till_osd_is_up(2)

    manager.flush_pg_stats([0, 2])
    manager.wait_till_active()
    manager.flush_pg_stats([0, 2])

    # verify that there are unfound objects
    unfound = manager.get_num_unfound_objects()
    log.info("there are %d unfound objects" % unfound)
    assert unfound

    testdir = teuthology.get_testdir(ctx)
    procs = []
    if config.get('parallel_bench', True):
        procs.append(mon.run(
            args=[
                "/bin/sh", "-c",
                " ".join(['adjust-ulimits',
                          'ceph-coverage',
                          '{tdir}/archive/coverage',
                          'rados',
                          '--no-log-to-stderr',
                          '--name', 'client.admin',
                          '-b', str(4<<10),
                          '-p' , POOL,
                          '-t', '20',
                          'bench', '240', 'write',
                      ]).format(tdir=testdir),
            ],
            logger=log.getChild('radosbench.{id}'.format(id='client.admin')),
            stdin=run.PIPE,
            wait=False
        ))
    time.sleep(10)

    # mark stuff lost
    pgs = manager.get_pg_stats()
    for pg in pgs:
        if pg['stat_sum']['num_objects_unfound'] > 0:
            primary = 'osd.%d' % pg['acting'][0]

            # verify that i can list them direct from the osd
            log.info('listing missing/lost in %s state %s', pg['pgid'],
                     pg['state']);
            m = manager.list_pg_unfound(pg['pgid'])
            #log.info('%s' % m)
            assert m['num_unfound'] == pg['stat_sum']['num_objects_unfound']
            num_unfound=0
            for o in m['objects']:
                if len(o['locations']) == 0:
                    num_unfound += 1
            assert m['num_unfound'] == num_unfound

            log.info("reverting unfound in %s on %s", pg['pgid'], primary)
            manager.raw_cluster_cmd('pg', pg['pgid'],
                                    'mark_unfound_lost', 'delete')
        else:
            log.info("no unfound in %s", pg['pgid'])

    manager.raw_cluster_cmd('tell', 'osd.0', 'debug', 'kick_recovery_wq', '5')
    manager.raw_cluster_cmd('tell', 'osd.2', 'debug', 'kick_recovery_wq', '5')
    manager.flush_pg_stats([0, 2])
    manager.wait_for_recovery()

    # verify result
    for f in range(1, 10):
        err = rados(ctx, mon, ['-p', POOL, 'get', 'new_%d' % f, '-'])
        assert err
        err = rados(ctx, mon, ['-p', POOL, 'get', 'existed_%d' % f, '-'])
        assert err
        err = rados(ctx, mon, ['-p', POOL, 'get', 'existing_%d' % f, '-'])
        assert err

    # see if osd.1 can cope
    manager.mark_in_osd(1)
    manager.revive_osd(1)
    manager.wait_till_osd_is_up(1)
    manager.wait_for_clean()
    run.wait(procs)
    manager.wait_for_clean()