ceph/qa/tasks/cephfs/test_journal_migration.py

   1
   2 from tasks.cephfs.cephfs_test_case import CephFSTestCase
   3 from tasks.workunit import task as workunit
   4
   5 JOURNAL_FORMAT_LEGACY = 0
   6 JOURNAL_FORMAT_RESILIENT = 1
   7
   8
   9 class TestJournalMigration(CephFSTestCase):
  10     CLIENTS_REQUIRED = 1
  11     MDSS_REQUIRED = 2
  12
  13     def test_journal_migration(self):
  14         old_journal_version = JOURNAL_FORMAT_LEGACY
  15         new_journal_version = JOURNAL_FORMAT_RESILIENT
  16
  17         self.mount_a.umount_wait()
  18         self.fs.mds_stop()
  19
  20         # Create a filesystem using the older journal format.
  21         self.fs.set_ceph_conf('mds', 'mds journal format', old_journal_version)
  22         self.fs.mds_restart()
  23         self.fs.recreate()
  24
  25         # Enable standby replay, to cover the bug case #8811 where
  26         # a standby replay might mistakenly end up trying to rewrite
  27         # the journal at the same time as an active daemon.
  28         self.fs.set_allow_standby_replay(True)
  29
  30         status = self.fs.wait_for_daemons()
  31
  32         self.assertTrue(self.fs.get_replay(status=status) is not None)
  33
  34         # Do some client work so that the log is populated with something.
  35         with self.mount_a.mounted_wait():
  36             self.mount_a.create_files()
  37             self.mount_a.check_files()  # sanity, this should always pass
  38
  39             # Run a more substantial workunit so that the length of the log to be
  40             # coverted is going span at least a few segments
  41             workunit(self.ctx, {
  42                 'clients': {
  43                     "client.{0}".format(self.mount_a.client_id): ["suites/fsstress.sh"],
  44                 },
  45                 "timeout": "3h"
  46             })
  47
  48         # Modify the ceph.conf to ask the MDS to use the new journal format.
  49         self.fs.set_ceph_conf('mds', 'mds journal format', new_journal_version)
  50
  51         # Restart the MDS.
  52         self.fs.mds_fail_restart()
  53
  54         # This ensures that all daemons come up into a valid state
  55         status = self.fs.wait_for_daemons()
  56
  57         # Check that files created in the initial client workload are still visible
  58         # in a client mount.
  59         with self.mount_a.mounted_wait():
  60             self.mount_a.check_files()
  61
  62         # Verify that the journal really has been rewritten.
  63         journal_version = self.fs.get_journal_version()
  64         if journal_version != new_journal_version:
  65             raise RuntimeError("Journal was not upgraded, version should be {0} but is {1}".format(
  66                 new_journal_version, journal_version()
  67             ))
  68
  69         # Verify that cephfs-journal-tool can now read the rewritten journal
  70         inspect_out = self.fs.journal_tool(["journal", "inspect"], 0)
  71         if not inspect_out.endswith(": OK"):
  72             raise RuntimeError("Unexpected journal-tool result: '{0}'".format(
  73                 inspect_out
  74             ))
  75
  76         self.fs.journal_tool(["event", "get", "json",
  77                               "--path", "/tmp/journal.json"], 0)
  78         p = self.fs.tool_remote.sh([
  79                 "python3",
  80                 "-c",
  81                 "import json; print(len(json.load(open('/tmp/journal.json'))))"
  82             ])
  83         event_count = int(p.strip())
  84         if event_count < 1000:
  85             # Approximate value of "lots", expected from having run fsstress
  86             raise RuntimeError("Unexpectedly few journal events: {0}".format(event_count))
  87
  88         # Do some client work to check that writing the log is still working
  89         with self.mount_a.mounted_wait():
  90             workunit(self.ctx, {
  91                 'clients': {
  92                     "client.{0}".format(self.mount_a.client_id): ["fs/misc/trivial_sync.sh"],
  93                 },
  94                 "timeout": "3h"
  95             })
  96
  97         # Check that both an active and a standby replay are still up
  98         status = self.fs.status()
  99         self.assertEqual(len(list(self.fs.get_replays(status=status))), 1)
 100         self.assertEqual(len(list(self.fs.get_ranks(status=status))), 1)