]>
git.proxmox.com Git - ceph.git/blob - ceph/qa/tasks/cephfs/test_misc.py
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
12 log
= logging
.getLogger(__name__
)
14 class TestMisc(CephFSTestCase
):
17 def test_getattr_caps(self
):
19 Check if MDS recognizes the 'mask' parameter of open request.
20 The parameter allows client to request caps when opening file
23 if not isinstance(self
.mount_a
, FuseMount
):
24 raise SkipTest("Require FUSE client")
26 # Enable debug. Client will requests CEPH_CAP_XATTR_SHARED
28 self
.mount_b
.umount_wait()
29 self
.set_conf('client', 'client debug getattr caps', 'true')
31 self
.mount_b
.wait_until_mounted()
33 # create a file and hold it open. MDS will issue CEPH_CAP_EXCL_*
35 p
= self
.mount_a
.open_background("testfile")
36 self
.mount_b
.wait_for_visible("testfile")
38 # this triggers 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"])
42 self
.mount_a
.kill_background(p
)
44 def test_root_rctime(self
):
46 Check that the root inode has a non-default rctime on startup.
50 rctime
= self
.mount_a
.getfattr(".", "ceph.dir.rctime")
51 log
.info("rctime = {}".format(rctime
))
52 self
.assertGreaterEqual(rctime
, t
-10)
54 def test_fs_new(self
):
55 self
.mount_a
.umount_wait()
56 self
.mount_b
.umount_wait()
58 data_pool_name
= self
.fs
.get_data_pool_name()
63 self
.fs
.mon_manager
.raw_cluster_cmd('fs', 'rm', self
.fs
.name
,
64 '--yes-i-really-mean-it')
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
__())
74 dummyfile
= '/etc/fstab'
76 self
.fs
.put_metadata_object_raw("key", dummyfile
)
78 def get_pool_df(fs
, name
):
80 return fs
.get_pool_df(name
)['objects'] > 0
81 except RuntimeError as e
:
84 self
.wait_until_true(lambda: get_pool_df(self
.fs
, self
.fs
.metadata_pool_name
), timeout
=30)
87 self
.fs
.mon_manager
.raw_cluster_cmd('fs', 'new', self
.fs
.name
,
88 self
.fs
.metadata_pool_name
,
90 except CommandFailedError
as e
:
91 self
.assertEqual(e
.exitstatus
, errno
.EINVAL
)
93 raise AssertionError("Expected EINVAL")
95 self
.fs
.mon_manager
.raw_cluster_cmd('fs', 'new', self
.fs
.name
,
96 self
.fs
.metadata_pool_name
,
97 data_pool_name
, "--force")
99 self
.fs
.mon_manager
.raw_cluster_cmd('fs', 'rm', self
.fs
.name
,
100 '--yes-i-really-mean-it')
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
,
114 def test_evict_client(self
):
116 Check that a slow client session won't get evicted if it's the
120 session_autoclose
= self
.fs
.get_var("session_autoclose")
122 self
.mount_b
.umount_wait()
123 ls_data
= self
.fs
.mds_asok(['session', 'ls'])
124 self
.assert_session_count(1, ls_data
)
127 self
.mount_a
.kill_cleanup()
129 time
.sleep(session_autoclose
* 1.5)
130 ls_data
= self
.fs
.mds_asok(['session', 'ls'])
131 self
.assert_session_count(1, ls_data
)
134 self
.mount_a
.wait_until_mounted()
136 self
.mount_b
.wait_until_mounted()
138 ls_data
= self
._session
_list
()
139 self
.assert_session_count(2, ls_data
)
142 self
.mount_a
.kill_cleanup()
144 time
.sleep(session_autoclose
* 1.5)
145 ls_data
= self
.fs
.mds_asok(['session', 'ls'])
146 self
.assert_session_count(1, ls_data
)
148 def test_cap_revoke_nonresponder(self
):
150 Check that a client is evicted if it has not responded to cap revoke
151 request for configured number of seconds.
153 session_timeout
= self
.fs
.get_var("session_timeout")
154 eviction_timeout
= session_timeout
/ 2.0
156 self
.fs
.mds_asok(['config', 'set', 'mds_cap_revoke_eviction_timeout',
157 str(eviction_timeout
)])
159 cap_holder
= self
.mount_a
.open_background()
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()
165 # Simulate client death
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()
174 time
.sleep(eviction_timeout
)
178 log
.info("cap_waiter waited {0}s".format(cap_waited
))
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
189 self
.assertTrue(self
.mount_a
.is_blacklisted())
190 cap_holder
.stdin
.close()
193 except (CommandFailedError
, ConnectionLostError
):
194 # We killed it (and possibly its node), so it raises an error
197 self
.mount_a
.kill_cleanup()
200 self
.mount_a
.wait_until_mounted()
202 def test_filtered_df(self
):
203 pool_name
= self
.fs
.get_data_pool_name()
204 raw_df
= self
.fs
.get_pool_df(pool_name
)
205 raw_avail
= float(raw_df
["max_avail"])
206 out
= self
.fs
.mon_manager
.raw_cluster_cmd('osd', 'pool', 'get',
210 pool_size
= int(j
['size'])
212 proc
= self
.mount_a
.run_shell(['df', '.'])
213 output
= proc
.stdout
.getvalue()
214 fs_avail
= output
.split('\n')[1].split()[3]
215 fs_avail
= float(fs_avail
) * 1024
217 ratio
= raw_avail
/ fs_avail
218 assert 0.9 < ratio
< 1.1
220 def test_dump_inode(self
):
221 info
= self
.fs
.mds_asok(['dump', 'inode', '1'])
222 assert(info
['path'] == "/")
224 def test_dump_inode_hexademical(self
):
225 self
.mount_a
.run_shell(["mkdir", "-p", "foo"])
226 ino
= self
.mount_a
.path_to_ino("foo")
227 assert type(ino
) is int
228 info
= self
.fs
.mds_asok(['dump', 'inode', hex(ino
)])
229 assert info
['path'] == "/foo"
232 class TestCacheDrop(CephFSTestCase
):
235 def _run_drop_cache_cmd(self
, timeout
=None):
237 mds_id
= self
.fs
.get_lone_mds_id()
238 if timeout
is not None:
239 result
= self
.fs
.mon_manager
.raw_cluster_cmd("tell", "mds.{0}".format(mds_id
),
240 "cache", "drop", str(timeout
))
242 result
= self
.fs
.mon_manager
.raw_cluster_cmd("tell", "mds.{0}".format(mds_id
),
244 return json
.loads(result
)
246 def _setup(self
, max_caps
=20, threshold
=400):
248 self
.mount_a
.create_n_files("dc-dir/dc-file", 1000, sync
=True)
250 # Reduce this so the MDS doesn't rkcall the maximum for simple tests
251 self
.fs
.rank_asok(['config', 'set', 'mds_recall_max_caps', str(max_caps
)])
252 self
.fs
.rank_asok(['config', 'set', 'mds_recall_max_decay_threshold', str(threshold
)])
254 def test_drop_cache_command(self
):
256 Basic test for checking drop cache command.
257 Confirm it halts without a timeout.
258 Note that the cache size post trimming is not checked here.
260 mds_min_caps_per_client
= int(self
.fs
.get_config("mds_min_caps_per_client"))
262 result
= self
._run
_drop
_cache
_cmd
()
263 self
.assertTrue(result
['client_recall']['return_code'] == 0)
264 self
.assertTrue(result
['flush_journal']['return_code'] == 0)
265 # It should take at least 1 second
266 self
.assertTrue(result
['duration'] > 1)
267 self
.assertGreaterEqual(result
['trim_cache']['trimmed'], 1000-2*mds_min_caps_per_client
)
269 def test_drop_cache_command_timeout(self
):
271 Basic test for checking drop cache command.
272 Confirm recall halts early via a timeout.
273 Note that the cache size post trimming is not checked here.
276 result
= self
._run
_drop
_cache
_cmd
(timeout
=10)
277 self
.assertTrue(result
['client_recall']['return_code'] == -errno
.ETIMEDOUT
)
278 self
.assertTrue(result
['flush_journal']['return_code'] == 0)
279 self
.assertTrue(result
['duration'] > 10)
280 self
.assertGreaterEqual(result
['trim_cache']['trimmed'], 100) # we did something, right?
282 def test_drop_cache_command_dead_timeout(self
):
284 Check drop cache command with non-responding client using tell
285 interface. Note that the cache size post trimming is not checked
290 # Note: recall is subject to the timeout. The journal flush will
291 # be delayed due to the client being dead.
292 result
= self
._run
_drop
_cache
_cmd
(timeout
=5)
293 self
.assertTrue(result
['client_recall']['return_code'] == -errno
.ETIMEDOUT
)
294 self
.assertTrue(result
['flush_journal']['return_code'] == 0)
295 self
.assertTrue(result
['duration'] > 5)
296 self
.assertTrue(result
['duration'] < 120)
297 self
.assertEqual(0, result
['trim_cache']['trimmed'])
298 self
.mount_a
.kill_cleanup()
300 self
.mount_a
.wait_until_mounted()
302 def test_drop_cache_command_dead(self
):
304 Check drop cache command with non-responding client using tell
305 interface. Note that the cache size post trimming is not checked
310 result
= self
._run
_drop
_cache
_cmd
()
311 self
.assertTrue(result
['client_recall']['return_code'] == 0)
312 self
.assertTrue(result
['flush_journal']['return_code'] == 0)
313 self
.assertTrue(result
['duration'] > 5)
314 self
.assertTrue(result
['duration'] < 120)
315 self
.assertEqual(0, result
['trim_cache']['trimmed'])
316 self
.mount_a
.kill_cleanup()
318 self
.mount_a
.wait_until_mounted()