]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | |
2 | from unittest import SkipTest | |
3 | from tasks.cephfs.fuse_mount import FuseMount | |
4 | from tasks.cephfs.cephfs_test_case import CephFSTestCase | |
91327a77 | 5 | from teuthology.orchestra.run import CommandFailedError, ConnectionLostError |
7c673cae FG |
6 | import errno |
7 | import time | |
d2e6a577 | 8 | import json |
91327a77 | 9 | import logging |
f64942e4 | 10 | import time |
7c673cae | 11 | |
91327a77 | 12 | log = logging.getLogger(__name__) |
31f18b77 | 13 | |
7c673cae FG |
14 | class TestMisc(CephFSTestCase): |
15 | CLIENTS_REQUIRED = 2 | |
16 | ||
7c673cae FG |
17 | def test_getattr_caps(self): |
18 | """ | |
19 | Check if MDS recognizes the 'mask' parameter of open request. | |
11fdf7f2 | 20 | The parameter allows client to request caps when opening file |
7c673cae FG |
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 | ||
11fdf7f2 | 38 | # this triggers a lookup request and an open request. The debug |
7c673cae FG |
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 | ||
f64942e4 AA |
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 | ||
7c673cae | 54 | def test_fs_new(self): |
a8e16298 TL |
55 | self.mount_a.umount_wait() |
56 | self.mount_b.umount_wait() | |
57 | ||
7c673cae FG |
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 | ||
224ce89b WB |
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 | |
7c673cae | 83 | |
224ce89b | 84 | self.wait_until_true(lambda: get_pool_df(self.fs, self.fs.metadata_pool_name), timeout=30) |
7c673cae FG |
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 | ||
f64942e4 AA |
120 | session_autoclose = self.fs.get_var("session_autoclose") |
121 | ||
31f18b77 | 122 | self.mount_b.umount_wait() |
7c673cae FG |
123 | ls_data = self.fs.mds_asok(['session', 'ls']) |
124 | self.assert_session_count(1, ls_data) | |
125 | ||
81eedcae | 126 | mount_a_client_id = self.mount_a.get_global_id() |
31f18b77 FG |
127 | self.mount_a.kill() |
128 | self.mount_a.kill_cleanup() | |
7c673cae | 129 | |
f64942e4 | 130 | time.sleep(session_autoclose * 1.5) |
7c673cae FG |
131 | ls_data = self.fs.mds_asok(['session', 'ls']) |
132 | self.assert_session_count(1, ls_data) | |
133 | ||
81eedcae TL |
134 | self.fs.mds_asok(['session', 'evict', "%s" % mount_a_client_id]) |
135 | ||
7c673cae FG |
136 | self.mount_a.mount() |
137 | self.mount_a.wait_until_mounted() | |
138 | self.mount_b.mount() | |
139 | self.mount_b.wait_until_mounted() | |
140 | ||
141 | ls_data = self._session_list() | |
142 | self.assert_session_count(2, ls_data) | |
143 | ||
144 | self.mount_a.kill() | |
181888fb | 145 | self.mount_a.kill_cleanup() |
7c673cae | 146 | |
f64942e4 | 147 | time.sleep(session_autoclose * 1.5) |
7c673cae FG |
148 | ls_data = self.fs.mds_asok(['session', 'ls']) |
149 | self.assert_session_count(1, ls_data) | |
d2e6a577 | 150 | |
91327a77 AA |
151 | def test_cap_revoke_nonresponder(self): |
152 | """ | |
153 | Check that a client is evicted if it has not responded to cap revoke | |
154 | request for configured number of seconds. | |
155 | """ | |
156 | session_timeout = self.fs.get_var("session_timeout") | |
157 | eviction_timeout = session_timeout / 2.0 | |
158 | ||
159 | self.fs.mds_asok(['config', 'set', 'mds_cap_revoke_eviction_timeout', | |
160 | str(eviction_timeout)]) | |
161 | ||
162 | cap_holder = self.mount_a.open_background() | |
163 | ||
164 | # Wait for the file to be visible from another client, indicating | |
165 | # that mount_a has completed its network ops | |
166 | self.mount_b.wait_for_visible() | |
167 | ||
168 | # Simulate client death | |
169 | self.mount_a.kill() | |
170 | ||
171 | try: | |
172 | # The waiter should get stuck waiting for the capability | |
173 | # held on the MDS by the now-dead client A | |
174 | cap_waiter = self.mount_b.write_background() | |
175 | ||
176 | a = time.time() | |
177 | time.sleep(eviction_timeout) | |
178 | cap_waiter.wait() | |
179 | b = time.time() | |
180 | cap_waited = b - a | |
181 | log.info("cap_waiter waited {0}s".format(cap_waited)) | |
182 | ||
183 | # check if the cap is transferred before session timeout kicked in. | |
184 | # this is a good enough check to ensure that the client got evicted | |
185 | # by the cap auto evicter rather than transitioning to stale state | |
186 | # and then getting evicted. | |
187 | self.assertLess(cap_waited, session_timeout, | |
188 | "Capability handover took {0}, expected less than {1}".format( | |
189 | cap_waited, session_timeout | |
190 | )) | |
191 | ||
11fdf7f2 | 192 | self.assertTrue(self.mount_a.is_blacklisted()) |
91327a77 AA |
193 | cap_holder.stdin.close() |
194 | try: | |
195 | cap_holder.wait() | |
196 | except (CommandFailedError, ConnectionLostError): | |
197 | # We killed it (and possibly its node), so it raises an error | |
198 | pass | |
199 | finally: | |
200 | self.mount_a.kill_cleanup() | |
201 | ||
202 | self.mount_a.mount() | |
203 | self.mount_a.wait_until_mounted() | |
204 | ||
d2e6a577 FG |
205 | def test_filtered_df(self): |
206 | pool_name = self.fs.get_data_pool_name() | |
207 | raw_df = self.fs.get_pool_df(pool_name) | |
208 | raw_avail = float(raw_df["max_avail"]) | |
209 | out = self.fs.mon_manager.raw_cluster_cmd('osd', 'pool', 'get', | |
210 | pool_name, 'size', | |
211 | '-f', 'json-pretty') | |
212 | j = json.loads(out) | |
213 | pool_size = int(j['size']) | |
214 | ||
215 | proc = self.mount_a.run_shell(['df', '.']) | |
216 | output = proc.stdout.getvalue() | |
217 | fs_avail = output.split('\n')[1].split()[3] | |
218 | fs_avail = float(fs_avail) * 1024 | |
219 | ||
181888fb | 220 | ratio = raw_avail / fs_avail |
d2e6a577 | 221 | assert 0.9 < ratio < 1.1 |
f64942e4 | 222 | |
11fdf7f2 TL |
223 | def test_dump_inode(self): |
224 | info = self.fs.mds_asok(['dump', 'inode', '1']) | |
225 | assert(info['path'] == "/") | |
f64942e4 | 226 | |
11fdf7f2 TL |
227 | def test_dump_inode_hexademical(self): |
228 | self.mount_a.run_shell(["mkdir", "-p", "foo"]) | |
229 | ino = self.mount_a.path_to_ino("foo") | |
230 | assert type(ino) is int | |
231 | info = self.fs.mds_asok(['dump', 'inode', hex(ino)]) | |
232 | assert info['path'] == "/foo" | |
f64942e4 | 233 | |
f64942e4 | 234 | |
11fdf7f2 TL |
235 | class TestCacheDrop(CephFSTestCase): |
236 | CLIENTS_REQUIRED = 1 | |
f64942e4 | 237 | |
11fdf7f2 TL |
238 | def _run_drop_cache_cmd(self, timeout=None): |
239 | result = None | |
240 | mds_id = self.fs.get_lone_mds_id() | |
241 | if timeout is not None: | |
242 | result = self.fs.mon_manager.raw_cluster_cmd("tell", "mds.{0}".format(mds_id), | |
243 | "cache", "drop", str(timeout)) | |
244 | else: | |
245 | result = self.fs.mon_manager.raw_cluster_cmd("tell", "mds.{0}".format(mds_id), | |
246 | "cache", "drop") | |
247 | return json.loads(result) | |
f64942e4 | 248 | |
11fdf7f2 | 249 | def _setup(self, max_caps=20, threshold=400): |
f64942e4 | 250 | # create some files |
11fdf7f2 | 251 | self.mount_a.create_n_files("dc-dir/dc-file", 1000, sync=True) |
f64942e4 | 252 | |
11fdf7f2 TL |
253 | # Reduce this so the MDS doesn't rkcall the maximum for simple tests |
254 | self.fs.rank_asok(['config', 'set', 'mds_recall_max_caps', str(max_caps)]) | |
255 | self.fs.rank_asok(['config', 'set', 'mds_recall_max_decay_threshold', str(threshold)]) | |
f64942e4 | 256 | |
11fdf7f2 | 257 | def test_drop_cache_command(self): |
f64942e4 | 258 | """ |
11fdf7f2 TL |
259 | Basic test for checking drop cache command. |
260 | Confirm it halts without a timeout. | |
f64942e4 AA |
261 | Note that the cache size post trimming is not checked here. |
262 | """ | |
11fdf7f2 TL |
263 | mds_min_caps_per_client = int(self.fs.get_config("mds_min_caps_per_client")) |
264 | self._setup() | |
265 | result = self._run_drop_cache_cmd() | |
266 | self.assertTrue(result['client_recall']['return_code'] == 0) | |
267 | self.assertTrue(result['flush_journal']['return_code'] == 0) | |
268 | # It should take at least 1 second | |
269 | self.assertTrue(result['duration'] > 1) | |
270 | self.assertGreaterEqual(result['trim_cache']['trimmed'], 1000-2*mds_min_caps_per_client) | |
271 | ||
272 | def test_drop_cache_command_timeout(self): | |
f64942e4 | 273 | """ |
11fdf7f2 TL |
274 | Basic test for checking drop cache command. |
275 | Confirm recall halts early via a timeout. | |
f64942e4 AA |
276 | Note that the cache size post trimming is not checked here. |
277 | """ | |
11fdf7f2 TL |
278 | self._setup() |
279 | result = self._run_drop_cache_cmd(timeout=10) | |
280 | self.assertTrue(result['client_recall']['return_code'] == -errno.ETIMEDOUT) | |
281 | self.assertTrue(result['flush_journal']['return_code'] == 0) | |
282 | self.assertTrue(result['duration'] > 10) | |
283 | self.assertGreaterEqual(result['trim_cache']['trimmed'], 100) # we did something, right? | |
284 | ||
285 | def test_drop_cache_command_dead_timeout(self): | |
f64942e4 | 286 | """ |
11fdf7f2 TL |
287 | Check drop cache command with non-responding client using tell |
288 | interface. Note that the cache size post trimming is not checked | |
289 | here. | |
f64942e4 | 290 | """ |
11fdf7f2 TL |
291 | self._setup() |
292 | self.mount_a.kill() | |
293 | # Note: recall is subject to the timeout. The journal flush will | |
294 | # be delayed due to the client being dead. | |
295 | result = self._run_drop_cache_cmd(timeout=5) | |
296 | self.assertTrue(result['client_recall']['return_code'] == -errno.ETIMEDOUT) | |
297 | self.assertTrue(result['flush_journal']['return_code'] == 0) | |
298 | self.assertTrue(result['duration'] > 5) | |
299 | self.assertTrue(result['duration'] < 120) | |
300 | self.assertEqual(0, result['trim_cache']['trimmed']) | |
301 | self.mount_a.kill_cleanup() | |
302 | self.mount_a.mount() | |
303 | self.mount_a.wait_until_mounted() | |
f64942e4 | 304 | |
11fdf7f2 | 305 | def test_drop_cache_command_dead(self): |
f64942e4 AA |
306 | """ |
307 | Check drop cache command with non-responding client using tell | |
308 | interface. Note that the cache size post trimming is not checked | |
309 | here. | |
310 | """ | |
11fdf7f2 TL |
311 | self._setup() |
312 | self.mount_a.kill() | |
313 | result = self._run_drop_cache_cmd() | |
314 | self.assertTrue(result['client_recall']['return_code'] == 0) | |
315 | self.assertTrue(result['flush_journal']['return_code'] == 0) | |
316 | self.assertTrue(result['duration'] > 5) | |
317 | self.assertTrue(result['duration'] < 120) | |
318 | self.assertEqual(0, result['trim_cache']['trimmed']) | |
319 | self.mount_a.kill_cleanup() | |
320 | self.mount_a.mount() | |
321 | self.mount_a.wait_until_mounted() |