ceph/qa/tasks/cephfs/test_misc.py

   1
   2 from tasks.cephfs.fuse_mount import FuseMount
   3 from tasks.cephfs.cephfs_test_case import CephFSTestCase
   4 from teuthology.orchestra.run import CommandFailedError, ConnectionLostError
   5 import errno
   6 import time
   7 import json
   8 import logging
   9
  10 log = logging.getLogger(__name__)
  11
  12 class TestMisc(CephFSTestCase):
  13     CLIENTS_REQUIRED = 2
  14
  15     def test_getattr_caps(self):
  16         """
  17         Check if MDS recognizes the 'mask' parameter of open request.
  18         The parameter allows client to request caps when opening file
  19         """
  20
  21         if not isinstance(self.mount_a, FuseMount):
  22             self.skipTest("Require FUSE client")
  23
  24         # Enable debug. Client will requests CEPH_CAP_XATTR_SHARED
  25         # on lookup/open
  26         self.mount_b.umount_wait()
  27         self.set_conf('client', 'client debug getattr caps', 'true')
  28         self.mount_b.mount_wait()
  29
  30         # create a file and hold it open. MDS will issue CEPH_CAP_EXCL_*
  31         # to mount_a
  32         p = self.mount_a.open_background("testfile")
  33         self.mount_b.wait_for_visible("testfile")
  34
  35         # this triggers a lookup request and an open request. The debug
  36         # code will check if lookup/open reply contains xattrs
  37         self.mount_b.run_shell(["cat", "testfile"])
  38
  39         self.mount_a.kill_background(p)
  40
  41     def test_root_rctime(self):
  42         """
  43         Check that the root inode has a non-default rctime on startup.
  44         """
  45
  46         t = time.time()
  47         rctime = self.mount_a.getfattr(".", "ceph.dir.rctime")
  48         log.info("rctime = {}".format(rctime))
  49         self.assertGreaterEqual(float(rctime), t - 10)
  50
  51     def test_fs_new(self):
  52         self.mount_a.umount_wait()
  53         self.mount_b.umount_wait()
  54
  55         data_pool_name = self.fs.get_data_pool_name()
  56
  57         self.fs.mds_stop()
  58         self.fs.mds_fail()
  59
  60         self.fs.mon_manager.raw_cluster_cmd('fs', 'rm', self.fs.name,
  61                                             '--yes-i-really-mean-it')
  62
  63         self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'delete',
  64                                             self.fs.metadata_pool_name,
  65                                             self.fs.metadata_pool_name,
  66                                             '--yes-i-really-really-mean-it')
  67         self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'create',
  68                                             self.fs.metadata_pool_name,
  69                                             self.fs.pgs_per_fs_pool.__str__())
  70
  71         dummyfile = '/etc/fstab'
  72
  73         self.fs.put_metadata_object_raw("key", dummyfile)
  74
  75         def get_pool_df(fs, name):
  76             try:
  77                 return fs.get_pool_df(name)['objects'] > 0
  78             except RuntimeError:
  79                 return False
  80
  81         self.wait_until_true(lambda: get_pool_df(self.fs, self.fs.metadata_pool_name), timeout=30)
  82
  83         try:
  84             self.fs.mon_manager.raw_cluster_cmd('fs', 'new', self.fs.name,
  85                                                 self.fs.metadata_pool_name,
  86                                                 data_pool_name)
  87         except CommandFailedError as e:
  88             self.assertEqual(e.exitstatus, errno.EINVAL)
  89         else:
  90             raise AssertionError("Expected EINVAL")
  91
  92         self.fs.mon_manager.raw_cluster_cmd('fs', 'new', self.fs.name,
  93                                             self.fs.metadata_pool_name,
  94                                             data_pool_name, "--force")
  95
  96         self.fs.mon_manager.raw_cluster_cmd('fs', 'rm', self.fs.name,
  97                                             '--yes-i-really-mean-it')
  98
  99
 100         self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'delete',
 101                                             self.fs.metadata_pool_name,
 102                                             self.fs.metadata_pool_name,
 103                                             '--yes-i-really-really-mean-it')
 104         self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'create',
 105                                             self.fs.metadata_pool_name,
 106                                             self.fs.pgs_per_fs_pool.__str__())
 107         self.fs.mon_manager.raw_cluster_cmd('fs', 'new', self.fs.name,
 108                                             self.fs.metadata_pool_name,
 109                                             data_pool_name)
 110
 111     def test_cap_revoke_nonresponder(self):
 112         """
 113         Check that a client is evicted if it has not responded to cap revoke
 114         request for configured number of seconds.
 115         """
 116         session_timeout = self.fs.get_var("session_timeout")
 117         eviction_timeout = session_timeout / 2.0
 118
 119         self.fs.mds_asok(['config', 'set', 'mds_cap_revoke_eviction_timeout',
 120                           str(eviction_timeout)])
 121
 122         cap_holder = self.mount_a.open_background()
 123
 124         # Wait for the file to be visible from another client, indicating
 125         # that mount_a has completed its network ops
 126         self.mount_b.wait_for_visible()
 127
 128         # Simulate client death
 129         self.mount_a.kill()
 130
 131         try:
 132             # The waiter should get stuck waiting for the capability
 133             # held on the MDS by the now-dead client A
 134             cap_waiter = self.mount_b.write_background()
 135
 136             a = time.time()
 137             time.sleep(eviction_timeout)
 138             cap_waiter.wait()
 139             b = time.time()
 140             cap_waited = b - a
 141             log.info("cap_waiter waited {0}s".format(cap_waited))
 142
 143             # check if the cap is transferred before session timeout kicked in.
 144             # this is a good enough check to ensure that the client got evicted
 145             # by the cap auto evicter rather than transitioning to stale state
 146             # and then getting evicted.
 147             self.assertLess(cap_waited, session_timeout,
 148                             "Capability handover took {0}, expected less than {1}".format(
 149                                 cap_waited, session_timeout
 150                             ))
 151
 152             self.assertTrue(self.mount_a.is_blacklisted())
 153             cap_holder.stdin.close()
 154             try:
 155                 cap_holder.wait()
 156             except (CommandFailedError, ConnectionLostError):
 157                 # We killed it (and possibly its node), so it raises an error
 158                 pass
 159         finally:
 160             self.mount_a.kill_cleanup()
 161
 162         self.mount_a.mount()
 163         self.mount_a.wait_until_mounted()
 164
 165     def test_filtered_df(self):
 166         pool_name = self.fs.get_data_pool_name()
 167         raw_df = self.fs.get_pool_df(pool_name)
 168         raw_avail = float(raw_df["max_avail"])
 169         out = self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'get',
 170                                                   pool_name, 'size',
 171                                                   '-f', 'json-pretty')
 172         _ = json.loads(out)
 173
 174         proc = self.mount_a.run_shell(['df', '.'])
 175         output = proc.stdout.getvalue()
 176         fs_avail = output.split('\n')[1].split()[3]
 177         fs_avail = float(fs_avail) * 1024
 178
 179         ratio = raw_avail / fs_avail
 180         assert 0.9 < ratio < 1.1
 181
 182     def test_dump_inode(self):
 183         info = self.fs.mds_asok(['dump', 'inode', '1'])
 184         assert(info['path'] == "/")
 185
 186     def test_dump_inode_hexademical(self):
 187         self.mount_a.run_shell(["mkdir", "-p", "foo"])
 188         ino = self.mount_a.path_to_ino("foo")
 189         assert type(ino) is int
 190         info = self.fs.mds_asok(['dump', 'inode', hex(ino)])
 191         assert info['path'] == "/foo"
 192
 193
 194 class TestCacheDrop(CephFSTestCase):
 195     CLIENTS_REQUIRED = 1
 196
 197     def _run_drop_cache_cmd(self, timeout=None):
 198         result = None
 199         mds_id = self.fs.get_lone_mds_id()
 200         if timeout is not None:
 201             result = self.fs.mon_manager.raw_cluster_cmd("tell", "mds.{0}".format(mds_id),
 202                                                     "cache", "drop", str(timeout))
 203         else:
 204             result = self.fs.mon_manager.raw_cluster_cmd("tell", "mds.{0}".format(mds_id),
 205                                                     "cache", "drop")
 206         return json.loads(result)
 207
 208     def _setup(self, max_caps=20, threshold=400):
 209         # create some files
 210         self.mount_a.create_n_files("dc-dir/dc-file", 1000, sync=True)
 211
 212         # Reduce this so the MDS doesn't rkcall the maximum for simple tests
 213         self.fs.rank_asok(['config', 'set', 'mds_recall_max_caps', str(max_caps)])
 214         self.fs.rank_asok(['config', 'set', 'mds_recall_max_decay_threshold', str(threshold)])
 215
 216     def test_drop_cache_command(self):
 217         """
 218         Basic test for checking drop cache command.
 219         Confirm it halts without a timeout.
 220         Note that the cache size post trimming is not checked here.
 221         """
 222         mds_min_caps_per_client = int(self.fs.get_config("mds_min_caps_per_client"))
 223         self._setup()
 224         result = self._run_drop_cache_cmd()
 225         self.assertEqual(result['client_recall']['return_code'], 0)
 226         self.assertEqual(result['flush_journal']['return_code'], 0)
 227         # It should take at least 1 second
 228         self.assertGreater(result['duration'], 1)
 229         self.assertGreaterEqual(result['trim_cache']['trimmed'], 1000-2*mds_min_caps_per_client)
 230
 231     def test_drop_cache_command_timeout(self):
 232         """
 233         Basic test for checking drop cache command.
 234         Confirm recall halts early via a timeout.
 235         Note that the cache size post trimming is not checked here.
 236         """
 237         self._setup()
 238         result = self._run_drop_cache_cmd(timeout=10)
 239         self.assertEqual(result['client_recall']['return_code'], -errno.ETIMEDOUT)
 240         self.assertEqual(result['flush_journal']['return_code'], 0)
 241         self.assertGreater(result['duration'], 10)
 242         self.assertGreaterEqual(result['trim_cache']['trimmed'], 100) # we did something, right?
 243
 244     def test_drop_cache_command_dead_timeout(self):
 245         """
 246         Check drop cache command with non-responding client using tell
 247         interface. Note that the cache size post trimming is not checked
 248         here.
 249         """
 250         self._setup()
 251         self.mount_a.kill()
 252         # Note: recall is subject to the timeout. The journal flush will
 253         # be delayed due to the client being dead.
 254         result = self._run_drop_cache_cmd(timeout=5)
 255         self.assertEqual(result['client_recall']['return_code'], -errno.ETIMEDOUT)
 256         self.assertEqual(result['flush_journal']['return_code'], 0)
 257         self.assertGreater(result['duration'], 5)
 258         self.assertLess(result['duration'], 120)
 259         # Note: result['trim_cache']['trimmed'] may be >0 because dropping the
 260         # cache now causes the Locker to drive eviction of stale clients (a
 261         # stale session will be autoclosed at mdsmap['session_timeout']). The
 262         # particular operation causing this is journal flush which causes the
 263         # MDS to wait wait for cap revoke.
 264         #self.assertEqual(0, result['trim_cache']['trimmed'])
 265         self.mount_a.kill_cleanup()
 266         self.mount_a.mount()
 267         self.mount_a.wait_until_mounted()
 268
 269     def test_drop_cache_command_dead(self):
 270         """
 271         Check drop cache command with non-responding client using tell
 272         interface. Note that the cache size post trimming is not checked
 273         here.
 274         """
 275         self._setup()
 276         self.mount_a.kill()
 277         result = self._run_drop_cache_cmd()
 278         self.assertEqual(result['client_recall']['return_code'], 0)
 279         self.assertEqual(result['flush_journal']['return_code'], 0)
 280         self.assertGreater(result['duration'], 5)
 281         self.assertLess(result['duration'], 120)
 282         # Note: result['trim_cache']['trimmed'] may be >0 because dropping the
 283         # cache now causes the Locker to drive eviction of stale clients (a
 284         # stale session will be autoclosed at mdsmap['session_timeout']). The
 285         # particular operation causing this is journal flush which causes the
 286         # MDS to wait wait for cap revoke.
 287         self.mount_a.kill_cleanup()
 288         self.mount_a.mount()
 289         self.mount_a.wait_until_mounted()