ceph/qa/tasks/reg11184.py

   1 """
   2 Special regression test for tracker #11184
   3
   4 Synopsis: osd/SnapMapper.cc: 282: FAILED assert(check(oid))
   5
   6 This is accomplished by moving a pg that wasn't part of split and still include
   7 divergent priors.
   8 """
   9 import logging
  10 import time
  11 from cStringIO import StringIO
  12
  13 from teuthology import misc as teuthology
  14 from util.rados import rados
  15 import os
  16
  17
  18 log = logging.getLogger(__name__)
  19
  20
  21 def task(ctx, config):
  22     """
  23     Test handling of divergent entries during export / import
  24     to regression test tracker #11184
  25
  26     overrides:
  27       ceph:
  28         conf:
  29           osd:
  30             debug osd: 5
  31
  32     Requires 3 osds on a single test node.
  33     """
  34     if config is None:
  35         config = {}
  36     assert isinstance(config, dict), \
  37         'divergent_priors task only accepts a dict for configuration'
  38
  39     manager = ctx.managers['ceph']
  40
  41     while len(manager.get_osd_status()['up']) < 3:
  42         time.sleep(10)
  43     manager.flush_pg_stats([0, 1, 2])
  44     manager.raw_cluster_cmd('osd', 'set', 'noout')
  45     manager.raw_cluster_cmd('osd', 'set', 'noin')
  46     manager.raw_cluster_cmd('osd', 'set', 'nodown')
  47     manager.wait_for_clean()
  48
  49     # something that is always there
  50     dummyfile = '/etc/fstab'
  51     dummyfile2 = '/etc/resolv.conf'
  52     testdir = teuthology.get_testdir(ctx)
  53
  54     # create 1 pg pool
  55     log.info('creating foo')
  56     manager.raw_cluster_cmd('osd', 'pool', 'create', 'foo', '1')
  57
  58     osds = [0, 1, 2]
  59     for i in osds:
  60         manager.set_config(i, osd_min_pg_log_entries=10)
  61         manager.set_config(i, osd_max_pg_log_entries=10)
  62         manager.set_config(i, osd_pg_log_trim_min=5)
  63
  64     # determine primary
  65     divergent = manager.get_pg_primary('foo', 0)
  66     log.info("primary and soon to be divergent is %d", divergent)
  67     non_divergent = list(osds)
  68     non_divergent.remove(divergent)
  69
  70     log.info('writing initial objects')
  71     first_mon = teuthology.get_first_mon(ctx, config)
  72     (mon,) = ctx.cluster.only(first_mon).remotes.iterkeys()
  73     # write 100 objects
  74     for i in range(100):
  75         rados(ctx, mon, ['-p', 'foo', 'put', 'existing_%d' % i, dummyfile])
  76
  77     manager.wait_for_clean()
  78
  79     # blackhole non_divergent
  80     log.info("blackholing osds %s", str(non_divergent))
  81     for i in non_divergent:
  82         manager.set_config(i, objectstore_blackhole=1)
  83
  84     DIVERGENT_WRITE = 5
  85     DIVERGENT_REMOVE = 5
  86     # Write some soon to be divergent
  87     log.info('writing divergent objects')
  88     for i in range(DIVERGENT_WRITE):
  89         rados(ctx, mon, ['-p', 'foo', 'put', 'existing_%d' % i,
  90                          dummyfile2], wait=False)
  91     # Remove some soon to be divergent
  92     log.info('remove divergent objects')
  93     for i in range(DIVERGENT_REMOVE):
  94         rados(ctx, mon, ['-p', 'foo', 'rm',
  95                          'existing_%d' % (i + DIVERGENT_WRITE)], wait=False)
  96     time.sleep(10)
  97     mon.run(
  98         args=['killall', '-9', 'rados'],
  99         wait=True,
 100         check_status=False)
 101
 102     # kill all the osds but leave divergent in
 103     log.info('killing all the osds')
 104     for i in osds:
 105         manager.kill_osd(i)
 106     for i in osds:
 107         manager.mark_down_osd(i)
 108     for i in non_divergent:
 109         manager.mark_out_osd(i)
 110
 111     # bring up non-divergent
 112     log.info("bringing up non_divergent %s", str(non_divergent))
 113     for i in non_divergent:
 114         manager.revive_osd(i)
 115     for i in non_divergent:
 116         manager.mark_in_osd(i)
 117
 118     # write 1 non-divergent object (ensure that old divergent one is divergent)
 119     objname = "existing_%d" % (DIVERGENT_WRITE + DIVERGENT_REMOVE)
 120     log.info('writing non-divergent object ' + objname)
 121     rados(ctx, mon, ['-p', 'foo', 'put', objname, dummyfile2])
 122
 123     manager.wait_for_recovery()
 124
 125     # ensure no recovery of up osds first
 126     log.info('delay recovery')
 127     for i in non_divergent:
 128         manager.wait_run_admin_socket(
 129             'osd', i, ['set_recovery_delay', '100000'])
 130
 131     # bring in our divergent friend
 132     log.info("revive divergent %d", divergent)
 133     manager.raw_cluster_cmd('osd', 'set', 'noup')
 134     manager.revive_osd(divergent)
 135
 136     log.info('delay recovery divergent')
 137     manager.wait_run_admin_socket(
 138         'osd', divergent, ['set_recovery_delay', '100000'])
 139
 140     manager.raw_cluster_cmd('osd', 'unset', 'noup')
 141     while len(manager.get_osd_status()['up']) < 3:
 142         time.sleep(10)
 143
 144     log.info('wait for peering')
 145     rados(ctx, mon, ['-p', 'foo', 'put', 'foo', dummyfile])
 146
 147     # At this point the divergent_priors should have been detected
 148
 149     log.info("killing divergent %d", divergent)
 150     manager.kill_osd(divergent)
 151
 152     # Split pgs for pool foo
 153     manager.raw_cluster_cmd('osd', 'pool', 'set', 'foo', 'pg_num', '2')
 154     time.sleep(5)
 155
 156     # Export a pg
 157     (exp_remote,) = ctx.\
 158         cluster.only('osd.{o}'.format(o=divergent)).remotes.iterkeys()
 159     FSPATH = manager.get_filepath()
 160     JPATH = os.path.join(FSPATH, "journal")
 161     prefix = ("sudo adjust-ulimits ceph-objectstore-tool "
 162               "--data-path {fpath} --journal-path {jpath} "
 163               "--log-file="
 164               "/var/log/ceph/objectstore_tool.$$.log ".
 165               format(fpath=FSPATH, jpath=JPATH))
 166     pid = os.getpid()
 167     expfile = os.path.join(testdir, "exp.{pid}.out".format(pid=pid))
 168     cmd = ((prefix + "--op export --pgid 1.0 --file {file}").
 169            format(id=divergent, file=expfile))
 170     proc = exp_remote.run(args=cmd, wait=True,
 171                           check_status=False, stdout=StringIO())
 172     assert proc.exitstatus == 0
 173
 174     # Remove the same pg that was exported
 175     cmd = ((prefix + "--op remove --pgid 1.0").
 176            format(id=divergent, file=expfile))
 177     proc = exp_remote.run(args=cmd, wait=True,
 178                           check_status=False, stdout=StringIO())
 179     assert proc.exitstatus == 0
 180
 181     # Kill one of non-divergent OSDs
 182     log.info('killing osd.%d' % non_divergent[1])
 183     manager.kill_osd(non_divergent[1])
 184     manager.mark_down_osd(non_divergent[1])
 185     # manager.mark_out_osd(non_divergent[1])
 186
 187     cmd = ((prefix + "--op import --file {file}").
 188            format(id=non_divergent[1], file=expfile))
 189     proc = exp_remote.run(args=cmd, wait=True,
 190                           check_status=False, stdout=StringIO())
 191     assert proc.exitstatus == 0
 192
 193     # bring in our divergent friend and other node
 194     log.info("revive divergent %d", divergent)
 195     manager.revive_osd(divergent)
 196     manager.mark_in_osd(divergent)
 197     log.info("revive %d", non_divergent[1])
 198     manager.revive_osd(non_divergent[1])
 199
 200     while len(manager.get_osd_status()['up']) < 3:
 201         time.sleep(10)
 202
 203     log.info('delay recovery divergent')
 204     manager.set_config(divergent, osd_recovery_delay_start=100000)
 205     log.info('mark divergent in')
 206     manager.mark_in_osd(divergent)
 207
 208     log.info('wait for peering')
 209     rados(ctx, mon, ['-p', 'foo', 'put', 'foo', dummyfile])
 210
 211     log.info("killing divergent %d", divergent)
 212     manager.kill_osd(divergent)
 213     log.info("reviving divergent %d", divergent)
 214     manager.revive_osd(divergent)
 215     time.sleep(3)
 216
 217     log.info('allowing recovery')
 218     # Set osd_recovery_delay_start back to 0 and kick the queue
 219     for i in osds:
 220         manager.raw_cluster_cmd('tell', 'osd.%d' % i, 'debug',
 221                                 'kick_recovery_wq', ' 0')
 222
 223     log.info('reading divergent objects')
 224     for i in range(DIVERGENT_WRITE + DIVERGENT_REMOVE):
 225         exit_status = rados(ctx, mon, ['-p', 'foo', 'get', 'existing_%d' % i,
 226                                        '/tmp/existing'])
 227         assert exit_status is 0
 228
 229     (remote,) = ctx.\
 230         cluster.only('osd.{o}'.format(o=divergent)).remotes.iterkeys()
 231     cmd = 'rm {file}'.format(file=expfile)
 232     remote.run(args=cmd, wait=True)
 233     log.info("success")