ceph/qa/tasks/osd_backfill.py

   1 """
   2 Osd backfill test
   3 """
   4 import logging
   5 import ceph_manager
   6 import time
   7 from teuthology import misc as teuthology
   8
   9
  10 log = logging.getLogger(__name__)
  11
  12
  13 def rados_start(ctx, remote, cmd):
  14     """
  15     Run a remote rados command (currently used to only write data)
  16     """
  17     log.info("rados %s" % ' '.join(cmd))
  18     testdir = teuthology.get_testdir(ctx)
  19     pre = [
  20         'adjust-ulimits',
  21         'ceph-coverage',
  22         '{tdir}/archive/coverage'.format(tdir=testdir),
  23         'rados',
  24         ];
  25     pre.extend(cmd)
  26     proc = remote.run(
  27         args=pre,
  28         wait=False,
  29         )
  30     return proc
  31
  32 def task(ctx, config):
  33     """
  34     Test backfill
  35     """
  36     if config is None:
  37         config = {}
  38     assert isinstance(config, dict), \
  39         'thrashosds task only accepts a dict for configuration'
  40     first_mon = teuthology.get_first_mon(ctx, config)
  41     (mon,) = ctx.cluster.only(first_mon).remotes.iterkeys()
  42
  43     num_osds = teuthology.num_instances_of_type(ctx.cluster, 'osd')
  44     log.info('num_osds is %s' % num_osds)
  45     assert num_osds == 3
  46
  47     manager = ceph_manager.CephManager(
  48         mon,
  49         ctx=ctx,
  50         logger=log.getChild('ceph_manager'),
  51         )
  52
  53     while len(manager.get_osd_status()['up']) < 3:
  54         time.sleep(10)
  55     manager.flush_pg_stats([0, 1, 2])
  56     manager.wait_for_clean()
  57
  58     # write some data
  59     p = rados_start(ctx, mon, ['-p', 'rbd', 'bench', '15', 'write', '-b', '4096',
  60                           '--no-cleanup'])
  61     err = p.wait()
  62     log.info('err is %d' % err)
  63
  64     # mark osd.0 out to trigger a rebalance/backfill
  65     manager.mark_out_osd(0)
  66
  67     # also mark it down to it won't be included in pg_temps
  68     manager.kill_osd(0)
  69     manager.mark_down_osd(0)
  70
  71     # wait for everything to peer and be happy...
  72     manager.flush_pg_stats([1, 2])
  73     manager.wait_for_recovery()
  74
  75     # write some new data
  76     p = rados_start(ctx, mon, ['-p', 'rbd', 'bench', '30', 'write', '-b', '4096',
  77                           '--no-cleanup'])
  78
  79     time.sleep(15)
  80
  81     # blackhole + restart osd.1
  82     # this triggers a divergent backfill target
  83     manager.blackhole_kill_osd(1)
  84     time.sleep(2)
  85     manager.revive_osd(1)
  86
  87     # wait for our writes to complete + succeed
  88     err = p.wait()
  89     log.info('err is %d' % err)
  90
  91     # wait for osd.1 and osd.2 to be up
  92     manager.wait_till_osd_is_up(1)
  93     manager.wait_till_osd_is_up(2)
  94
  95     # cluster must recover
  96     manager.flush_pg_stats([1, 2])
  97     manager.wait_for_recovery()
  98
  99     # re-add osd.0
 100     manager.revive_osd(0)
 101     manager.flush_pg_stats([1, 2])
 102     manager.wait_for_clean()
 103
 104