ceph/qa/tasks/cephfs/test_misc.py

   1
   2 from unittest import SkipTest
   3 from tasks.cephfs.fuse_mount import FuseMount
   4 from tasks.cephfs.cephfs_test_case import CephFSTestCase
   5 from teuthology.orchestra.run import CommandFailedError, ConnectionLostError
   6 import errno
   7 import time
   8 import json
   9 import logging
  10 import time
  11
  12 log = logging.getLogger(__name__)
  13
  14 class TestMisc(CephFSTestCase):
  15     CLIENTS_REQUIRED = 2
  16
  17     def test_getattr_caps(self):
  18         """
  19         Check if MDS recognizes the 'mask' parameter of open request.
  20         The paramter allows client to request caps when opening file
  21         """
  22
  23         if not isinstance(self.mount_a, FuseMount):
  24             raise SkipTest("Require FUSE client")
  25
  26         # Enable debug. Client will requests CEPH_CAP_XATTR_SHARED
  27         # on lookup/open
  28         self.mount_b.umount_wait()
  29         self.set_conf('client', 'client debug getattr caps', 'true')
  30         self.mount_b.mount()
  31         self.mount_b.wait_until_mounted()
  32
  33         # create a file and hold it open. MDS will issue CEPH_CAP_EXCL_*
  34         # to mount_a
  35         p = self.mount_a.open_background("testfile")
  36         self.mount_b.wait_for_visible("testfile")
  37
  38         # this tiggers a lookup request and an open request. The debug
  39         # code will check if lookup/open reply contains xattrs
  40         self.mount_b.run_shell(["cat", "testfile"])
  41
  42         self.mount_a.kill_background(p)
  43
  44     def test_root_rctime(self):
  45         """
  46         Check that the root inode has a non-default rctime on startup.
  47         """
  48
  49         t = time.time()
  50         rctime = self.mount_a.getfattr(".", "ceph.dir.rctime")
  51         log.info("rctime = {}".format(rctime))
  52         self.assertGreaterEqual(rctime, t-10)
  53
  54     def test_fs_new(self):
  55         self.mount_a.umount_wait()
  56         self.mount_b.umount_wait()
  57
  58         data_pool_name = self.fs.get_data_pool_name()
  59
  60         self.fs.mds_stop()
  61         self.fs.mds_fail()
  62
  63         self.fs.mon_manager.raw_cluster_cmd('fs', 'rm', self.fs.name,
  64                                             '--yes-i-really-mean-it')
  65
  66         self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'delete',
  67                                             self.fs.metadata_pool_name,
  68                                             self.fs.metadata_pool_name,
  69                                             '--yes-i-really-really-mean-it')
  70         self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'create',
  71                                             self.fs.metadata_pool_name,
  72                                             self.fs.get_pgs_per_fs_pool().__str__())
  73
  74         dummyfile = '/etc/fstab'
  75
  76         self.fs.put_metadata_object_raw("key", dummyfile)
  77
  78         def get_pool_df(fs, name):
  79             try:
  80                 return fs.get_pool_df(name)['objects'] > 0
  81             except RuntimeError as e:
  82                 return False
  83
  84         self.wait_until_true(lambda: get_pool_df(self.fs, self.fs.metadata_pool_name), timeout=30)
  85
  86         try:
  87             self.fs.mon_manager.raw_cluster_cmd('fs', 'new', self.fs.name,
  88                                                 self.fs.metadata_pool_name,
  89                                                 data_pool_name)
  90         except CommandFailedError as e:
  91             self.assertEqual(e.exitstatus, errno.EINVAL)
  92         else:
  93             raise AssertionError("Expected EINVAL")
  94
  95         self.fs.mon_manager.raw_cluster_cmd('fs', 'new', self.fs.name,
  96                                             self.fs.metadata_pool_name,
  97                                             data_pool_name, "--force")
  98
  99         self.fs.mon_manager.raw_cluster_cmd('fs', 'rm', self.fs.name,
 100                                             '--yes-i-really-mean-it')
 101
 102
 103         self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'delete',
 104                                             self.fs.metadata_pool_name,
 105                                             self.fs.metadata_pool_name,
 106                                             '--yes-i-really-really-mean-it')
 107         self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'create',
 108                                             self.fs.metadata_pool_name,
 109                                             self.fs.get_pgs_per_fs_pool().__str__())
 110         self.fs.mon_manager.raw_cluster_cmd('fs', 'new', self.fs.name,
 111                                             self.fs.metadata_pool_name,
 112                                             data_pool_name)
 113
 114     def test_evict_client(self):
 115         """
 116         Check that a slow client session won't get evicted if it's the
 117         only session
 118         """
 119
 120         session_autoclose = self.fs.get_var("session_autoclose")
 121
 122         self.mount_b.umount_wait()
 123         ls_data = self.fs.mds_asok(['session', 'ls'])
 124         self.assert_session_count(1, ls_data)
 125
 126         self.mount_a.kill()
 127         self.mount_a.kill_cleanup()
 128
 129         time.sleep(session_autoclose * 1.5)
 130         ls_data = self.fs.mds_asok(['session', 'ls'])
 131         self.assert_session_count(1, ls_data)
 132
 133         self.mount_a.mount()
 134         self.mount_a.wait_until_mounted()
 135         self.mount_b.mount()
 136         self.mount_b.wait_until_mounted()
 137
 138         ls_data = self._session_list()
 139         self.assert_session_count(2, ls_data)
 140
 141         self.mount_a.kill()
 142         self.mount_a.kill_cleanup()
 143
 144         time.sleep(session_autoclose * 1.5)
 145         ls_data = self.fs.mds_asok(['session', 'ls'])
 146         self.assert_session_count(1, ls_data)
 147
 148     def test_cap_revoke_nonresponder(self):
 149         """
 150         Check that a client is evicted if it has not responded to cap revoke
 151         request for configured number of seconds.
 152         """
 153         session_timeout = self.fs.get_var("session_timeout")
 154         eviction_timeout = session_timeout / 2.0
 155
 156         self.fs.mds_asok(['config', 'set', 'mds_cap_revoke_eviction_timeout',
 157                           str(eviction_timeout)])
 158
 159         cap_holder = self.mount_a.open_background()
 160
 161         # Wait for the file to be visible from another client, indicating
 162         # that mount_a has completed its network ops
 163         self.mount_b.wait_for_visible()
 164
 165         # Simulate client death
 166         self.mount_a.kill()
 167
 168         try:
 169             # The waiter should get stuck waiting for the capability
 170             # held on the MDS by the now-dead client A
 171             cap_waiter = self.mount_b.write_background()
 172
 173             a = time.time()
 174             time.sleep(eviction_timeout)
 175             cap_waiter.wait()
 176             b = time.time()
 177             cap_waited = b - a
 178             log.info("cap_waiter waited {0}s".format(cap_waited))
 179
 180             # check if the cap is transferred before session timeout kicked in.
 181             # this is a good enough check to ensure that the client got evicted
 182             # by the cap auto evicter rather than transitioning to stale state
 183             # and then getting evicted.
 184             self.assertLess(cap_waited, session_timeout,
 185                             "Capability handover took {0}, expected less than {1}".format(
 186                                 cap_waited, session_timeout
 187                             ))
 188
 189             cap_holder.stdin.close()
 190             try:
 191                 cap_holder.wait()
 192             except (CommandFailedError, ConnectionLostError):
 193                 # We killed it (and possibly its node), so it raises an error
 194                 pass
 195         finally:
 196             self.mount_a.kill_cleanup()
 197
 198         self.mount_a.mount()
 199         self.mount_a.wait_until_mounted()
 200
 201     def test_filtered_df(self):
 202         pool_name = self.fs.get_data_pool_name()
 203         raw_df = self.fs.get_pool_df(pool_name)
 204         raw_avail = float(raw_df["max_avail"])
 205         out = self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'get',
 206                                                   pool_name, 'size',
 207                                                   '-f', 'json-pretty')
 208         j = json.loads(out)
 209         pool_size = int(j['size'])
 210
 211         proc = self.mount_a.run_shell(['df', '.'])
 212         output = proc.stdout.getvalue()
 213         fs_avail = output.split('\n')[1].split()[3]
 214         fs_avail = float(fs_avail) * 1024
 215
 216         ratio = raw_avail / fs_avail
 217         assert 0.9 < ratio < 1.1
 218
 219     def _run_drop_cache_cmd(self, timeout, use_tell):
 220         drop_res = None
 221         if use_tell:
 222             mds_id = self.fs.get_lone_mds_id()
 223             drop_res = json.loads(
 224                 self.fs.mon_manager.raw_cluster_cmd("tell", "mds.{0}".format(mds_id),
 225                                                     "cache", "drop", str(timeout)))
 226         else:
 227             drop_res = self.fs.mds_asok(["cache", "drop", str(timeout)])
 228         return drop_res
 229
 230     def _drop_cache_command(self, timeout, use_tell=True):
 231         self.mount_b.umount_wait()
 232         ls_data = self.fs.mds_asok(['session', 'ls'])
 233         self.assert_session_count(1, ls_data)
 234
 235         # create some files
 236         self.mount_a.create_n_files("dc-dir/dc-file", 1000)
 237         # drop cache
 238         drop_res = self._run_drop_cache_cmd(timeout, use_tell)
 239
 240         self.assertTrue(drop_res['client_recall']['return_code'] == 0)
 241         self.assertTrue(drop_res['flush_journal']['return_code'] == 0)
 242
 243     def _drop_cache_command_timeout(self, timeout, use_tell=True):
 244         self.mount_b.umount_wait()
 245         ls_data = self.fs.mds_asok(['session', 'ls'])
 246         self.assert_session_count(1, ls_data)
 247
 248         # create some files
 249         self.mount_a.create_n_files("dc-dir/dc-file-t", 1000)
 250
 251         # simulate client death and try drop cache
 252         self.mount_a.kill()
 253         drop_res = self._run_drop_cache_cmd(timeout, use_tell)
 254
 255         self.assertTrue(drop_res['client_recall']['return_code'] == -errno.ETIMEDOUT)
 256         self.assertTrue(drop_res['flush_journal']['return_code'] == 0)
 257
 258         self.mount_a.kill_cleanup()
 259         self.mount_a.mount()
 260         self.mount_a.wait_until_mounted()
 261
 262     def test_drop_cache_command_asok(self):
 263         """
 264         Basic test for checking drop cache command using admin socket.
 265         Note that the cache size post trimming is not checked here.
 266         """
 267         self._drop_cache_command(10, use_tell=False)
 268
 269     def test_drop_cache_command_tell(self):
 270         """
 271         Basic test for checking drop cache command using tell interface.
 272         Note that the cache size post trimming is not checked here.
 273         """
 274         self._drop_cache_command(10)
 275
 276     def test_drop_cache_command_timeout_asok(self):
 277         """
 278         Check drop cache command with non-responding client using admin
 279         socket. Note that the cache size post trimming is not checked here.
 280         """
 281         self._drop_cache_command_timeout(5, use_tell=False)
 282
 283     def test_drop_cache_command_timeout_tell(self):
 284         """
 285         Check drop cache command with non-responding client using tell
 286         interface. Note that the cache size post trimming is not checked
 287         here.
 288         """
 289         self._drop_cache_command_timeout(5)