]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
2 | // vim: ts=8 sw=2 smarttab | |
3 | /* | |
4 | * Ceph - scalable distributed file system | |
5 | * | |
6 | * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net> | |
7 | * | |
8 | * This is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
10 | * License version 2.1, as published by the Free Software | |
11 | * Foundation. See file COPYING. | |
12 | * | |
13 | */ | |
14 | ||
f67539c2 | 15 | #include <ostream> |
7c673cae FG |
16 | |
17 | #include "FSMap.h" | |
18 | ||
11fdf7f2 | 19 | #include "common/StackStringStream.h" |
7c673cae | 20 | |
11fdf7f2 TL |
21 | #ifdef WITH_SEASTAR |
22 | #include "crimson/common/config_proxy.h" | |
23 | #else | |
24 | #include "common/config_proxy.h" | |
25 | #endif | |
26 | #include "global/global_context.h" | |
224ce89b WB |
27 | #include "mon/health_check.h" |
28 | ||
f67539c2 TL |
29 | using std::list; |
30 | using std::pair; | |
31 | using std::ostream; | |
32 | using std::string; | |
33 | ||
34 | using ceph::bufferlist; | |
35 | using ceph::Formatter; | |
36 | ||
37 | void ClusterInfo::encode(ceph::buffer::list &bl) const { | |
38 | ENCODE_START(1, 1, bl); | |
39 | encode(client_name, bl); | |
40 | encode(cluster_name, bl); | |
41 | encode(fs_name, bl); | |
42 | ENCODE_FINISH(bl); | |
43 | } | |
44 | ||
45 | void ClusterInfo::decode(ceph::buffer::list::const_iterator &iter) { | |
46 | DECODE_START(1, iter); | |
47 | decode(client_name, iter); | |
48 | decode(cluster_name, iter); | |
49 | decode(fs_name, iter); | |
50 | DECODE_FINISH(iter); | |
51 | } | |
52 | ||
53 | void ClusterInfo::dump(ceph::Formatter *f) const { | |
54 | f->dump_string("client_name", client_name); | |
55 | f->dump_string("cluster_name", cluster_name); | |
56 | f->dump_string("fs_name", fs_name); | |
57 | } | |
58 | ||
59 | void ClusterInfo::print(std::ostream& out) const { | |
60 | out << "[client_name=" << client_name << ", cluster_name=" << cluster_name | |
61 | << ", fs_name=" << fs_name << "]" << std::endl; | |
62 | } | |
63 | ||
64 | void Peer::encode(ceph::buffer::list &bl) const { | |
65 | ENCODE_START(1, 1, bl); | |
66 | encode(uuid, bl); | |
67 | encode(remote, bl); | |
68 | ENCODE_FINISH(bl); | |
69 | } | |
70 | ||
71 | void Peer::decode(ceph::buffer::list::const_iterator &iter) { | |
72 | DECODE_START(1, iter); | |
73 | decode(uuid, iter); | |
74 | decode(remote, iter); | |
75 | DECODE_FINISH(iter); | |
76 | } | |
77 | ||
78 | void Peer::dump(ceph::Formatter *f) const { | |
79 | f->open_object_section(uuid); | |
80 | f->dump_object("remote", remote); | |
81 | f->close_section(); | |
82 | } | |
83 | ||
84 | void Peer::print(std::ostream& out) const { | |
85 | out << "[uuid=" << uuid << ", remote=" << remote << "]" << std::endl; | |
86 | } | |
87 | ||
88 | void MirrorInfo::encode(ceph::buffer::list &bl) const { | |
89 | ENCODE_START(1, 1, bl); | |
90 | encode(mirrored, bl); | |
91 | encode(peers, bl); | |
92 | ENCODE_FINISH(bl); | |
93 | } | |
94 | ||
95 | void MirrorInfo::decode(ceph::buffer::list::const_iterator &iter) { | |
96 | DECODE_START(1, iter); | |
97 | decode(mirrored, iter); | |
98 | decode(peers, iter); | |
99 | DECODE_FINISH(iter); | |
100 | } | |
101 | ||
102 | void MirrorInfo::dump(ceph::Formatter *f) const { | |
103 | f->open_object_section("peers"); | |
104 | for (auto &peer : peers) { | |
105 | peer.dump(f); | |
106 | } | |
107 | f->close_section(); // peers | |
108 | } | |
109 | ||
110 | void MirrorInfo::print(std::ostream& out) const { | |
111 | out << "[peers=" << peers << "]" << std::endl; | |
112 | } | |
7c673cae FG |
113 | |
114 | void Filesystem::dump(Formatter *f) const | |
115 | { | |
116 | f->open_object_section("mdsmap"); | |
117 | mds_map.dump(f); | |
118 | f->close_section(); | |
119 | f->dump_int("id", fscid); | |
f67539c2 TL |
120 | if (mirror_info.is_mirrored()) { |
121 | f->open_object_section("mirror_info"); | |
122 | mirror_info.dump(f); | |
123 | f->close_section(); // mirror_info | |
124 | } | |
7c673cae FG |
125 | } |
126 | ||
127 | void FSMap::dump(Formatter *f) const | |
128 | { | |
129 | f->dump_int("epoch", epoch); | |
11fdf7f2 TL |
130 | // Use 'default' naming to match 'set-default' CLI |
131 | f->dump_int("default_fscid", legacy_client_fscid); | |
7c673cae FG |
132 | |
133 | f->open_object_section("compat"); | |
522d829b | 134 | default_compat.dump(f); |
7c673cae FG |
135 | f->close_section(); |
136 | ||
137 | f->open_object_section("feature_flags"); | |
138 | f->dump_bool("enable_multiple", enable_multiple); | |
139 | f->dump_bool("ever_enabled_multiple", ever_enabled_multiple); | |
140 | f->close_section(); | |
141 | ||
142 | f->open_array_section("standbys"); | |
9f95a23c | 143 | for (const auto& [gid, info] : standby_daemons) { |
7c673cae | 144 | f->open_object_section("info"); |
9f95a23c TL |
145 | info.dump(f); |
146 | f->dump_int("epoch", standby_epochs.at(gid)); | |
7c673cae FG |
147 | f->close_section(); |
148 | } | |
149 | f->close_section(); | |
150 | ||
151 | f->open_array_section("filesystems"); | |
152 | for (const auto &fs : filesystems) { | |
153 | f->open_object_section("filesystem"); | |
154 | fs.second->dump(f); | |
155 | f->close_section(); | |
156 | } | |
157 | f->close_section(); | |
158 | } | |
159 | ||
9f95a23c TL |
160 | FSMap &FSMap::operator=(const FSMap &rhs) |
161 | { | |
162 | epoch = rhs.epoch; | |
163 | next_filesystem_id = rhs.next_filesystem_id; | |
164 | legacy_client_fscid = rhs.legacy_client_fscid; | |
522d829b | 165 | default_compat = rhs.default_compat; |
9f95a23c TL |
166 | enable_multiple = rhs.enable_multiple; |
167 | mds_roles = rhs.mds_roles; | |
168 | standby_daemons = rhs.standby_daemons; | |
169 | standby_epochs = rhs.standby_epochs; | |
170 | ||
171 | filesystems.clear(); | |
172 | for (const auto &i : rhs.filesystems) { | |
173 | const auto &fs = i.second; | |
174 | filesystems[fs->fscid] = std::make_shared<Filesystem>(*fs); | |
175 | } | |
176 | ||
177 | return *this; | |
178 | } | |
179 | ||
180 | void FSMap::generate_test_instances(std::list<FSMap*>& ls) | |
7c673cae FG |
181 | { |
182 | FSMap *m = new FSMap(); | |
183 | ||
184 | std::list<MDSMap*> mds_map_instances; | |
185 | MDSMap::generate_test_instances(mds_map_instances); | |
186 | ||
187 | int k = 20; | |
188 | for (auto i : mds_map_instances) { | |
11fdf7f2 | 189 | auto fs = Filesystem::create(); |
7c673cae FG |
190 | fs->fscid = k++; |
191 | fs->mds_map = *i; | |
192 | delete i; | |
193 | m->filesystems[fs->fscid] = fs; | |
194 | } | |
195 | mds_map_instances.clear(); | |
196 | ||
197 | ls.push_back(m); | |
198 | } | |
199 | ||
200 | void FSMap::print(ostream& out) const | |
201 | { | |
202 | out << "e" << epoch << std::endl; | |
203 | out << "enable_multiple, ever_enabled_multiple: " << enable_multiple << "," | |
204 | << ever_enabled_multiple << std::endl; | |
522d829b | 205 | out << "default compat: " << default_compat << std::endl; |
7c673cae FG |
206 | out << "legacy client fscid: " << legacy_client_fscid << std::endl; |
207 | out << " " << std::endl; | |
208 | ||
209 | if (filesystems.empty()) { | |
210 | out << "No filesystems configured" << std::endl; | |
7c673cae FG |
211 | } |
212 | ||
11fdf7f2 TL |
213 | for (const auto& p : filesystems) { |
214 | p.second->print(out); | |
7c673cae FG |
215 | out << " " << std::endl << " " << std::endl; // Space out a bit |
216 | } | |
217 | ||
218 | if (!standby_daemons.empty()) { | |
219 | out << "Standby daemons:" << std::endl << " " << std::endl; | |
220 | } | |
221 | ||
9f95a23c TL |
222 | for (const auto& p : standby_daemons) { |
223 | out << p.second << std::endl; | |
7c673cae FG |
224 | } |
225 | } | |
226 | ||
f67539c2 TL |
227 | void FSMap::print_daemon_summary(ostream& out) const |
228 | { | |
229 | // this appears in the "services:" section of "ceph status" | |
230 | int num_up = 0, num_in = 0, num_failed = 0; | |
231 | int num_standby_replay = 0; | |
232 | for (auto& [fscid, fs] : filesystems) { | |
233 | num_up += fs->mds_map.get_num_up_mds(); | |
234 | num_in += fs->mds_map.get_num_in_mds(); | |
235 | num_failed += fs->mds_map.get_num_failed_mds(); | |
236 | num_standby_replay += fs->mds_map.get_num_standby_replay_mds(); | |
237 | } | |
238 | int num_standby = standby_daemons.size(); | |
239 | out << num_up << "/" << num_in << " daemons up"; | |
240 | if (num_failed) { | |
241 | out << " (" << num_failed << " failed)"; | |
242 | } | |
243 | if (num_standby) { | |
244 | out << ", " << num_standby << " standby"; | |
245 | } | |
246 | if (num_standby_replay) { | |
247 | out << ", " << num_standby_replay << " hot standby"; | |
248 | } | |
249 | } | |
250 | ||
251 | void FSMap::print_fs_summary(ostream& out) const | |
252 | { | |
253 | // this appears in the "data:" section of "ceph status" | |
254 | if (!filesystems.empty()) { | |
255 | int num_failed = 0, num_recovering = 0, num_stopped = 0, num_healthy = 0; | |
256 | int num_damaged = 0; | |
257 | for (auto& [fscid, fs] : filesystems) { | |
258 | if (fs->mds_map.is_any_damaged()) { | |
259 | ++num_damaged; | |
260 | } | |
261 | if (fs->mds_map.is_any_failed()) { | |
262 | ++num_failed; | |
263 | } else if (fs->mds_map.is_degraded()) { | |
264 | ++num_recovering; | |
265 | } else if (fs->mds_map.get_max_mds() == 0) { | |
266 | ++num_stopped; | |
267 | } else { | |
268 | ++num_healthy; | |
269 | } | |
270 | } | |
271 | out << " volumes: " | |
272 | << num_healthy << "/" << filesystems.size() << " healthy"; | |
273 | if (num_recovering) { | |
274 | out << ", " << num_recovering << " recovering"; | |
275 | } | |
276 | if (num_failed) { | |
277 | out << ", " << num_failed << " failed"; | |
278 | } | |
279 | if (num_stopped) { | |
280 | out << ", " << num_stopped << " stopped"; | |
281 | } | |
282 | if (num_damaged) { | |
283 | out << "; " << num_damaged << " damaged"; | |
284 | } | |
285 | out << "\n"; | |
286 | } | |
287 | } | |
288 | ||
7c673cae FG |
289 | void FSMap::print_summary(Formatter *f, ostream *out) const |
290 | { | |
7c673cae FG |
291 | if (f) { |
292 | f->dump_unsigned("epoch", get_epoch()); | |
11fdf7f2 TL |
293 | for (const auto &p : filesystems) { |
294 | auto& fs = p.second; | |
7c673cae FG |
295 | f->dump_unsigned("id", fs->fscid); |
296 | f->dump_unsigned("up", fs->mds_map.up.size()); | |
297 | f->dump_unsigned("in", fs->mds_map.in.size()); | |
298 | f->dump_unsigned("max", fs->mds_map.max_mds); | |
299 | } | |
300 | } else { | |
11fdf7f2 TL |
301 | auto count = filesystems.size(); |
302 | if (count <= 3) { | |
303 | bool first = true; | |
304 | for (const auto& p : filesystems) { | |
305 | const auto& fs = p.second; | |
306 | if (!first) { | |
307 | *out << " "; | |
308 | } | |
309 | if (fs->mds_map.is_degraded()) { | |
310 | *out << fs->mds_map.fs_name << ":" << fs->mds_map.up.size() << "/" << fs->mds_map.in.size(); | |
311 | } else { | |
312 | *out << fs->mds_map.fs_name << ":" << fs->mds_map.in.size(); | |
313 | } | |
314 | first = false; | |
315 | } | |
316 | } else { | |
317 | *out << count << " fs"; | |
318 | unsigned degraded = 0; | |
319 | CachedStackStringStream css; | |
320 | *css << " (degraded: "; | |
321 | for (const auto& p : filesystems) { | |
322 | const auto& fs = p.second; | |
323 | if (fs->mds_map.is_degraded()) { | |
324 | degraded++; | |
325 | if (degraded <= 3) { | |
326 | *css << fs->mds_map.fs_name << ":" << fs->mds_map.up.size() << "/" << fs->mds_map.in.size(); | |
327 | } | |
328 | } | |
329 | } | |
330 | if (degraded > 0) { | |
331 | if (degraded <= 3) { | |
332 | *css << ")"; | |
333 | *out << css->strv(); | |
334 | } else { | |
335 | *out << " (degraded: " << degraded << " fs)"; | |
336 | } | |
337 | } | |
7c673cae FG |
338 | } |
339 | } | |
340 | ||
341 | if (f) { | |
342 | f->open_array_section("by_rank"); | |
343 | } | |
344 | ||
11fdf7f2 TL |
345 | std::map<MDSMap::DaemonState,unsigned> by_state; |
346 | std::map<mds_role_t, std::pair<MDSMap::DaemonState, std::string>> by_rank; | |
347 | by_state[MDSMap::DaemonState::STATE_STANDBY] = standby_daemons.size(); | |
348 | for (const auto& [gid, fscid] : mds_roles) { | |
349 | if (fscid == FS_CLUSTER_ID_NONE) | |
350 | continue; | |
351 | ||
352 | const auto& info = filesystems.at(fscid)->mds_map.get_info_gid(gid); | |
353 | auto s = std::string(ceph_mds_state_name(info.state)); | |
7c673cae FG |
354 | if (info.laggy()) { |
355 | s += "(laggy or crashed)"; | |
356 | } | |
357 | ||
11fdf7f2 TL |
358 | if (f) { |
359 | f->open_object_section("mds"); | |
360 | f->dump_unsigned("filesystem_id", fscid); | |
361 | f->dump_unsigned("rank", info.rank); | |
362 | f->dump_string("name", info.name); | |
363 | f->dump_string("status", s); | |
364 | f->dump_unsigned("gid", gid); | |
365 | f->close_section(); | |
366 | } else if (info.state != MDSMap::DaemonState::STATE_STANDBY_REPLAY) { | |
367 | by_rank[mds_role_t(fscid, info.rank)] = std::make_pair(info.state, info.name + "=" + s); | |
7c673cae | 368 | } |
11fdf7f2 | 369 | by_state[info.state]++; |
7c673cae FG |
370 | } |
371 | ||
372 | if (f) { | |
373 | f->close_section(); | |
374 | } else { | |
11fdf7f2 | 375 | if (0 < by_rank.size() && by_rank.size() < 5) { |
7c673cae FG |
376 | if (filesystems.size() > 1) { |
377 | // Disambiguate filesystems | |
378 | std::map<std::string, std::string> pretty; | |
11fdf7f2 TL |
379 | for (const auto& [role,status] : by_rank) { |
380 | const auto &fs_name = filesystems.at(role.fscid)->mds_map.fs_name; | |
381 | CachedStackStringStream css; | |
382 | *css << fs_name << ":" << role.rank; | |
383 | pretty.emplace(std::piecewise_construct, std::forward_as_tuple(css->strv()), std::forward_as_tuple(status.second)); | |
384 | --by_state[status.first]; /* already printed! */ | |
7c673cae FG |
385 | } |
386 | *out << " " << pretty; | |
387 | } else { | |
388 | // Omit FSCID in output when only one filesystem exists | |
389 | std::map<mds_rank_t, std::string> shortened; | |
11fdf7f2 TL |
390 | for (const auto& [role,status] : by_rank) { |
391 | shortened[role.rank] = status.second; | |
392 | --by_state[status.first]; /* already printed! */ | |
7c673cae FG |
393 | } |
394 | *out << " " << shortened; | |
395 | } | |
396 | } | |
11fdf7f2 TL |
397 | for (const auto& [state, count] : by_state) { |
398 | if (count > 0) { | |
399 | auto s = std::string_view(ceph_mds_state_name(state)); | |
400 | *out << " " << count << " " << s; | |
401 | } | |
402 | } | |
7c673cae FG |
403 | } |
404 | ||
11fdf7f2 TL |
405 | if (f) { |
406 | const auto state = MDSMap::DaemonState::STATE_STANDBY; | |
407 | auto&& name = ceph_mds_state_name(state); | |
408 | auto count = standby_daemons.size(); | |
409 | f->dump_unsigned(name, count); | |
7c673cae FG |
410 | } |
411 | ||
412 | size_t failed = 0; | |
413 | size_t damaged = 0; | |
11fdf7f2 TL |
414 | for (const auto& p : filesystems) { |
415 | auto& fs = p.second; | |
7c673cae FG |
416 | failed += fs->mds_map.failed.size(); |
417 | damaged += fs->mds_map.damaged.size(); | |
418 | } | |
419 | ||
420 | if (failed > 0) { | |
421 | if (f) { | |
422 | f->dump_unsigned("failed", failed); | |
423 | } else { | |
424 | *out << ", " << failed << " failed"; | |
425 | } | |
426 | } | |
427 | ||
428 | if (damaged > 0) { | |
429 | if (f) { | |
430 | f->dump_unsigned("damaged", damaged); | |
431 | } else { | |
432 | *out << ", " << damaged << " damaged"; | |
433 | } | |
434 | } | |
435 | //if (stopped.size()) | |
436 | //out << ", " << stopped.size() << " stopped"; | |
437 | } | |
438 | ||
9f95a23c TL |
439 | mds_gid_t Filesystem::get_standby_replay(mds_gid_t who) const |
440 | { | |
441 | for (const auto &i : mds_map.mds_info) { | |
442 | const auto &info = i.second; | |
443 | if (info.state == MDSMap::STATE_STANDBY_REPLAY | |
444 | && info.rank == mds_map.mds_info.at(who).rank) { | |
445 | return info.global_id; | |
446 | } | |
447 | } | |
448 | return MDS_GID_NONE; | |
449 | } | |
7c673cae | 450 | |
11fdf7f2 | 451 | Filesystem::ref FSMap::create_filesystem(std::string_view name, |
522d829b TL |
452 | int64_t metadata_pool, int64_t data_pool, uint64_t features, |
453 | fs_cluster_id_t fscid) | |
7c673cae | 454 | { |
11fdf7f2 | 455 | auto fs = Filesystem::create(); |
28e407b8 | 456 | fs->mds_map.epoch = epoch; |
11fdf7f2 | 457 | fs->mds_map.fs_name = name; |
31f18b77 | 458 | fs->mds_map.data_pools.push_back(data_pool); |
7c673cae FG |
459 | fs->mds_map.metadata_pool = metadata_pool; |
460 | fs->mds_map.cas_pool = -1; | |
522d829b | 461 | fs->mds_map.compat = default_compat; |
7c673cae FG |
462 | fs->mds_map.created = ceph_clock_now(); |
463 | fs->mds_map.modified = ceph_clock_now(); | |
7c673cae | 464 | fs->mds_map.enabled = true; |
522d829b TL |
465 | if (fscid == FS_CLUSTER_ID_NONE) { |
466 | fs->fscid = next_filesystem_id++; | |
467 | } else { | |
468 | fs->fscid = fscid; | |
469 | next_filesystem_id = std::max(fscid, (fs_cluster_id_t)next_filesystem_id) + 1; | |
470 | } | |
471 | ||
472 | // File system's ID can be FS_CLUSTER_ID_ANONYMOUS if we're recovering | |
473 | // a legacy file system by passing FS_CLUSTER_ID_ANONYMOUS as the desired | |
474 | // file system ID | |
475 | if (fscid != FS_CLUSTER_ID_ANONYMOUS) { | |
476 | // ANONYMOUS is only for upgrades from legacy mdsmaps, we should | |
477 | // have initialized next_filesystem_id such that it's never used here. | |
478 | ceph_assert(fs->fscid != FS_CLUSTER_ID_ANONYMOUS); | |
479 | } | |
7c673cae FG |
480 | filesystems[fs->fscid] = fs; |
481 | ||
482 | // Created first filesystem? Set it as the one | |
483 | // for legacy clients to use | |
484 | if (filesystems.size() == 1) { | |
485 | legacy_client_fscid = fs->fscid; | |
486 | } | |
11fdf7f2 TL |
487 | |
488 | return fs; | |
7c673cae FG |
489 | } |
490 | ||
9f95a23c TL |
491 | Filesystem::const_ref FSMap::get_filesystem(std::string_view name) const |
492 | { | |
493 | for (const auto& p : filesystems) { | |
494 | if (p.second->mds_map.fs_name == name) { | |
495 | return p.second; | |
496 | } | |
497 | } | |
498 | return nullptr; | |
499 | } | |
500 | ||
501 | std::vector<Filesystem::const_ref> FSMap::get_filesystems(void) const | |
502 | { | |
503 | std::vector<Filesystem::const_ref> ret; | |
504 | for (const auto& p : filesystems) { | |
505 | ret.push_back(p.second); | |
506 | } | |
507 | return ret; | |
508 | } | |
509 | ||
7c673cae FG |
510 | void FSMap::reset_filesystem(fs_cluster_id_t fscid) |
511 | { | |
512 | auto fs = get_filesystem(fscid); | |
11fdf7f2 | 513 | auto new_fs = Filesystem::create(); |
7c673cae FG |
514 | |
515 | // Populate rank 0 as existing (so don't go into CREATING) | |
516 | // but failed (so that next available MDS is assigned the rank) | |
517 | new_fs->mds_map.in.insert(mds_rank_t(0)); | |
518 | new_fs->mds_map.failed.insert(mds_rank_t(0)); | |
519 | ||
520 | // Carry forward what makes sense | |
521 | new_fs->fscid = fs->fscid; | |
522 | new_fs->mds_map.inline_data_enabled = fs->mds_map.inline_data_enabled; | |
7c673cae FG |
523 | new_fs->mds_map.data_pools = fs->mds_map.data_pools; |
524 | new_fs->mds_map.metadata_pool = fs->mds_map.metadata_pool; | |
525 | new_fs->mds_map.cas_pool = fs->mds_map.cas_pool; | |
526 | new_fs->mds_map.fs_name = fs->mds_map.fs_name; | |
522d829b | 527 | new_fs->mds_map.compat = default_compat; |
7c673cae FG |
528 | new_fs->mds_map.created = ceph_clock_now(); |
529 | new_fs->mds_map.modified = ceph_clock_now(); | |
7c673cae FG |
530 | new_fs->mds_map.standby_count_wanted = fs->mds_map.standby_count_wanted; |
531 | new_fs->mds_map.enabled = true; | |
532 | ||
c07f9fc5 FG |
533 | // Remember mds ranks that have ever started. (They should load old inotable |
534 | // instead of creating new one if they start again.) | |
535 | new_fs->mds_map.stopped.insert(fs->mds_map.in.begin(), fs->mds_map.in.end()); | |
536 | new_fs->mds_map.stopped.insert(fs->mds_map.stopped.begin(), fs->mds_map.stopped.end()); | |
537 | new_fs->mds_map.stopped.erase(mds_rank_t(0)); | |
538 | ||
7c673cae FG |
539 | // Persist the new FSMap |
540 | filesystems[new_fs->fscid] = new_fs; | |
541 | } | |
542 | ||
543 | void FSMap::get_health(list<pair<health_status_t,string> >& summary, | |
544 | list<pair<health_status_t,string> > *detail) const | |
545 | { | |
546 | mds_rank_t standby_count_wanted = 0; | |
547 | for (const auto &i : filesystems) { | |
548 | const auto &fs = i.second; | |
549 | ||
550 | // TODO: move get_health up into here so that we can qualify | |
551 | // all the messages with what filesystem they're talking about | |
552 | fs->mds_map.get_health(summary, detail); | |
553 | ||
554 | standby_count_wanted = std::max(standby_count_wanted, fs->mds_map.get_standby_count_wanted((mds_rank_t)standby_daemons.size())); | |
555 | } | |
556 | ||
557 | if (standby_count_wanted) { | |
f67539c2 TL |
558 | CachedStackStringStream css; |
559 | *css << "insufficient standby daemons available: have " << standby_daemons.size() << "; want " << standby_count_wanted << " more"; | |
560 | summary.push_back(make_pair(HEALTH_WARN, css->str())); | |
7c673cae FG |
561 | } |
562 | } | |
563 | ||
564 | bool FSMap::check_health(void) | |
565 | { | |
566 | bool changed = false; | |
567 | for (auto &i : filesystems) { | |
568 | changed |= i.second->mds_map.check_health((mds_rank_t)standby_daemons.size()); | |
569 | } | |
570 | return changed; | |
571 | } | |
572 | ||
224ce89b WB |
573 | void FSMap::get_health_checks(health_check_map_t *checks) const |
574 | { | |
575 | mds_rank_t standby_count_wanted = 0; | |
576 | for (const auto &i : filesystems) { | |
577 | const auto &fs = i.second; | |
578 | health_check_map_t fschecks; | |
d2e6a577 | 579 | |
224ce89b | 580 | fs->mds_map.get_health_checks(&fschecks); |
d2e6a577 FG |
581 | |
582 | // Some of the failed ranks might be transient (i.e. there are standbys | |
583 | // ready to replace them). We will report only on "stuck" failed, i.e. | |
584 | // ranks which are failed and have no standby replacement available. | |
585 | std::set<mds_rank_t> stuck_failed; | |
586 | ||
587 | for (const auto &rank : fs->mds_map.failed) { | |
9f95a23c TL |
588 | auto rep_info = find_replacement_for({fs->fscid, rank}); |
589 | if (!rep_info) { | |
d2e6a577 FG |
590 | stuck_failed.insert(rank); |
591 | } | |
592 | } | |
593 | ||
594 | // FS_WITH_FAILED_MDS | |
595 | if (!stuck_failed.empty()) { | |
596 | health_check_t& fscheck = checks->get_or_add( | |
597 | "FS_WITH_FAILED_MDS", HEALTH_WARN, | |
9f95a23c | 598 | "%num% filesystem%plurals% %hasorhave% a failed mds daemon", 1); |
f67539c2 TL |
599 | CachedStackStringStream css; |
600 | *css << "fs " << fs->mds_map.fs_name << " has " << stuck_failed.size() | |
d2e6a577 | 601 | << " failed mds" << (stuck_failed.size() > 1 ? "s" : ""); |
f67539c2 | 602 | fscheck.detail.push_back(css->str()); } |
d2e6a577 | 603 | |
224ce89b WB |
604 | checks->merge(fschecks); |
605 | standby_count_wanted = std::max( | |
606 | standby_count_wanted, | |
607 | fs->mds_map.get_standby_count_wanted((mds_rank_t)standby_daemons.size())); | |
608 | } | |
609 | ||
610 | // MDS_INSUFFICIENT_STANDBY | |
611 | if (standby_count_wanted) { | |
f67539c2 TL |
612 | CachedStackStringStream css1, css2; |
613 | *css1 << "insufficient standby MDS daemons available"; | |
614 | auto& d = checks->get_or_add("MDS_INSUFFICIENT_STANDBY", HEALTH_WARN, css1->str(), 1); | |
615 | *css2 << "have " << standby_daemons.size() << "; want " << standby_count_wanted | |
616 | << " more"; | |
617 | d.detail.push_back(css2->str()); | |
224ce89b WB |
618 | } |
619 | } | |
620 | ||
9f95a23c TL |
621 | void FSMap::encode(bufferlist& bl, uint64_t features) const |
622 | { | |
522d829b | 623 | ENCODE_START(STRUCT_VERSION, 6, bl); |
9f95a23c TL |
624 | encode(epoch, bl); |
625 | encode(next_filesystem_id, bl); | |
626 | encode(legacy_client_fscid, bl); | |
522d829b | 627 | encode(default_compat, bl); |
9f95a23c TL |
628 | encode(enable_multiple, bl); |
629 | { | |
630 | std::vector<Filesystem::ref> v; | |
631 | v.reserve(filesystems.size()); | |
632 | for (auto& p : filesystems) v.emplace_back(p.second); | |
633 | encode(v, bl, features); | |
7c673cae | 634 | } |
9f95a23c TL |
635 | encode(mds_roles, bl); |
636 | encode(standby_daemons, bl, features); | |
637 | encode(standby_epochs, bl); | |
638 | encode(ever_enabled_multiple, bl); | |
639 | ENCODE_FINISH(bl); | |
7c673cae FG |
640 | } |
641 | ||
11fdf7f2 | 642 | void FSMap::decode(bufferlist::const_iterator& p) |
7c673cae | 643 | { |
a4b75251 | 644 | struct_version = 0; |
522d829b TL |
645 | DECODE_START(STRUCT_VERSION, p); |
646 | DECODE_OLDEST(7); | |
647 | struct_version = struct_v; | |
f67539c2 TL |
648 | decode(epoch, p); |
649 | decode(next_filesystem_id, p); | |
650 | decode(legacy_client_fscid, p); | |
522d829b | 651 | decode(default_compat, p); |
f67539c2 TL |
652 | decode(enable_multiple, p); |
653 | { | |
654 | std::vector<Filesystem::ref> v; | |
655 | decode(v, p); | |
7c673cae | 656 | filesystems.clear(); |
f67539c2 TL |
657 | for (auto& ref : v) { |
658 | auto em = filesystems.emplace(std::piecewise_construct, std::forward_as_tuple(ref->fscid), std::forward_as_tuple(std::move(ref))); | |
659 | ceph_assert(em.second); | |
7c673cae FG |
660 | } |
661 | } | |
f67539c2 TL |
662 | decode(mds_roles, p); |
663 | decode(standby_daemons, p); | |
664 | decode(standby_epochs, p); | |
665 | if (struct_v >= 7) { | |
666 | decode(ever_enabled_multiple, p); | |
667 | } | |
7c673cae FG |
668 | DECODE_FINISH(p); |
669 | } | |
670 | ||
11fdf7f2 | 671 | void FSMap::sanitize(const std::function<bool(int64_t pool)>& pool_exists) |
3efd9988 FG |
672 | { |
673 | for (auto &fs : filesystems) { | |
674 | fs.second->mds_map.sanitize(pool_exists); | |
675 | } | |
676 | } | |
7c673cae FG |
677 | |
678 | void Filesystem::encode(bufferlist& bl, uint64_t features) const | |
679 | { | |
f67539c2 | 680 | ENCODE_START(2, 1, bl); |
11fdf7f2 | 681 | encode(fscid, bl); |
7c673cae FG |
682 | bufferlist mdsmap_bl; |
683 | mds_map.encode(mdsmap_bl, features); | |
11fdf7f2 | 684 | encode(mdsmap_bl, bl); |
f67539c2 | 685 | encode(mirror_info, bl); |
7c673cae FG |
686 | ENCODE_FINISH(bl); |
687 | } | |
688 | ||
11fdf7f2 | 689 | void Filesystem::decode(bufferlist::const_iterator& p) |
7c673cae | 690 | { |
f67539c2 | 691 | DECODE_START(2, p); |
11fdf7f2 | 692 | decode(fscid, p); |
7c673cae | 693 | bufferlist mdsmap_bl; |
11fdf7f2 TL |
694 | decode(mdsmap_bl, p); |
695 | auto mdsmap_bl_iter = mdsmap_bl.cbegin(); | |
7c673cae | 696 | mds_map.decode(mdsmap_bl_iter); |
f67539c2 TL |
697 | if (struct_v >= 2) { |
698 | decode(mirror_info, p); | |
699 | } | |
7c673cae FG |
700 | DECODE_FINISH(p); |
701 | } | |
702 | ||
703 | int FSMap::parse_filesystem( | |
11fdf7f2 TL |
704 | std::string_view ns_str, |
705 | Filesystem::const_ref* result | |
7c673cae FG |
706 | ) const |
707 | { | |
708 | std::string ns_err; | |
94b18763 FG |
709 | std::string s(ns_str); |
710 | fs_cluster_id_t fscid = strict_strtol(s.c_str(), 10, &ns_err); | |
7c673cae FG |
711 | if (!ns_err.empty() || filesystems.count(fscid) == 0) { |
712 | for (auto &fs : filesystems) { | |
94b18763 | 713 | if (fs.second->mds_map.fs_name == s) { |
7c673cae FG |
714 | *result = std::const_pointer_cast<const Filesystem>(fs.second); |
715 | return 0; | |
716 | } | |
717 | } | |
f67539c2 | 718 | return -CEPHFS_ENOENT; |
7c673cae FG |
719 | } else { |
720 | *result = get_filesystem(fscid); | |
721 | return 0; | |
722 | } | |
723 | } | |
724 | ||
725 | void Filesystem::print(std::ostream &out) const | |
726 | { | |
727 | out << "Filesystem '" << mds_map.fs_name | |
728 | << "' (" << fscid << ")" << std::endl; | |
729 | mds_map.print(out); | |
f67539c2 TL |
730 | if (mirror_info.is_mirrored()) { |
731 | mirror_info.print(out); | |
732 | } | |
7c673cae FG |
733 | } |
734 | ||
9f95a23c | 735 | bool FSMap::is_any_degraded() const |
7c673cae | 736 | { |
9f95a23c TL |
737 | for (auto& i : filesystems) { |
738 | if (i.second->mds_map.is_degraded()) { | |
739 | return true; | |
740 | } | |
741 | } | |
742 | return false; | |
743 | } | |
744 | ||
745 | std::map<mds_gid_t, MDSMap::mds_info_t> FSMap::get_mds_info() const | |
746 | { | |
747 | std::map<mds_gid_t, mds_info_t> result; | |
748 | for (const auto &i : standby_daemons) { | |
749 | result[i.first] = i.second; | |
750 | } | |
751 | ||
752 | for (const auto &i : filesystems) { | |
753 | const auto &fs_info = i.second->mds_map.get_mds_info(); | |
754 | for (const auto &j : fs_info) { | |
755 | result[j.first] = j.second; | |
756 | } | |
757 | } | |
758 | ||
759 | return result; | |
760 | } | |
761 | ||
522d829b | 762 | const MDSMap::mds_info_t* FSMap::get_available_standby(const Filesystem& fs) const |
9f95a23c | 763 | { |
522d829b | 764 | const bool upgradeable = fs.is_upgradeable(); |
9f95a23c | 765 | const mds_info_t* who = nullptr; |
11fdf7f2 TL |
766 | for (const auto& [gid, info] : standby_daemons) { |
767 | ceph_assert(info.rank == MDS_RANK_NONE); | |
768 | ceph_assert(info.state == MDSMap::STATE_STANDBY); | |
7c673cae | 769 | |
11fdf7f2 | 770 | if (info.laggy() || info.is_frozen()) { |
7c673cae | 771 | continue; |
522d829b TL |
772 | } else if (!info.compat.writeable(fs.mds_map.compat)) { |
773 | /* standby is not compatible with this fs */ | |
774 | continue; | |
775 | } else if (!upgradeable && !fs.mds_map.compat.writeable(info.compat)) { | |
776 | /* promotion would change fs.mds_map.compat and we're not upgradeable */ | |
777 | continue; | |
7c673cae FG |
778 | } |
779 | ||
522d829b | 780 | if (info.join_fscid == fs.fscid) { |
9f95a23c TL |
781 | who = &info; |
782 | break; | |
783 | } else if (info.join_fscid == FS_CLUSTER_ID_NONE) { | |
784 | who = &info; /* vanilla standby */ | |
785 | } else if (who == nullptr) { | |
786 | who = &info; /* standby for another fs, last resort */ | |
787 | } | |
788 | } | |
789 | return who; | |
790 | } | |
791 | ||
792 | mds_gid_t FSMap::find_mds_gid_by_name(std::string_view s) const | |
793 | { | |
794 | const auto info = get_mds_info(); | |
795 | for (const auto &p : info) { | |
796 | if (p.second.name == s) { | |
797 | return p.first; | |
798 | } | |
7c673cae | 799 | } |
11fdf7f2 | 800 | return MDS_GID_NONE; |
7c673cae FG |
801 | } |
802 | ||
9f95a23c TL |
803 | const MDSMap::mds_info_t* FSMap::find_by_name(std::string_view name) const |
804 | { | |
805 | std::map<mds_gid_t, mds_info_t> result; | |
806 | for (const auto &i : standby_daemons) { | |
807 | if (i.second.name == name) { | |
808 | return &(i.second); | |
809 | } | |
810 | } | |
811 | ||
812 | for (const auto &i : filesystems) { | |
813 | const auto &fs_info = i.second->mds_map.get_mds_info(); | |
814 | for (const auto &j : fs_info) { | |
815 | if (j.second.name == name) { | |
816 | return &(j.second); | |
817 | } | |
818 | } | |
819 | } | |
820 | ||
821 | return nullptr; | |
822 | } | |
823 | ||
824 | const MDSMap::mds_info_t* FSMap::find_replacement_for(mds_role_t role) const | |
11fdf7f2 TL |
825 | { |
826 | auto&& fs = get_filesystem(role.fscid); | |
7c673cae | 827 | |
11fdf7f2 TL |
828 | // First see if we have a STANDBY_REPLAY |
829 | for (const auto& [gid, info] : fs->mds_map.mds_info) { | |
830 | if (info.rank == role.rank && info.state == MDSMap::STATE_STANDBY_REPLAY) { | |
831 | if (info.is_frozen()) { | |
832 | /* the standby-replay is frozen, do nothing! */ | |
9f95a23c | 833 | return nullptr; |
11fdf7f2 | 834 | } else { |
522d829b | 835 | ceph_assert(info.compat.writeable(fs->mds_map.compat)); |
9f95a23c | 836 | return &info; |
11fdf7f2 | 837 | } |
7c673cae FG |
838 | } |
839 | } | |
7c673cae | 840 | |
522d829b | 841 | return get_available_standby(*fs); |
7c673cae FG |
842 | } |
843 | ||
a4b75251 | 844 | void FSMap::sanity(bool pending) const |
7c673cae | 845 | { |
a4b75251 TL |
846 | /* Only do some sanity checks on **new** FSMaps. Older versions may not be |
847 | * compliant. | |
848 | */ | |
849 | ||
7c673cae | 850 | if (legacy_client_fscid != FS_CLUSTER_ID_NONE) { |
11fdf7f2 | 851 | ceph_assert(filesystems.count(legacy_client_fscid) == 1); |
7c673cae FG |
852 | } |
853 | ||
522d829b TL |
854 | for (const auto& [fscid, fs] : filesystems) { |
855 | ceph_assert(fscid == fs->fscid); | |
856 | for (const auto& [gid, info] : fs->mds_map.mds_info) { | |
857 | ceph_assert(info.rank != MDS_RANK_NONE); | |
858 | ceph_assert(mds_roles.at(gid) == fscid); | |
859 | ceph_assert(standby_daemons.count(gid) == 0); | |
860 | ceph_assert(standby_epochs.count(gid) == 0); | |
861 | if (info.state != MDSMap::STATE_STANDBY_REPLAY) { | |
862 | ceph_assert(fs->mds_map.up.at(info.rank) == gid); | |
863 | ceph_assert(fs->mds_map.failed.count(info.rank) == 0); | |
864 | ceph_assert(fs->mds_map.damaged.count(info.rank) == 0); | |
865 | } else { | |
a4b75251 | 866 | ceph_assert(!pending || fs->mds_map.allows_standby_replay()); |
7c673cae | 867 | } |
522d829b | 868 | ceph_assert(info.compat.writeable(fs->mds_map.compat)); |
7c673cae FG |
869 | } |
870 | ||
871 | for (const auto &j : fs->mds_map.up) { | |
872 | mds_rank_t rank = j.first; | |
11fdf7f2 | 873 | ceph_assert(fs->mds_map.in.count(rank) == 1); |
7c673cae | 874 | mds_gid_t gid = j.second; |
11fdf7f2 | 875 | ceph_assert(fs->mds_map.mds_info.count(gid) == 1); |
7c673cae FG |
876 | } |
877 | } | |
878 | ||
879 | for (const auto &i : standby_daemons) { | |
11fdf7f2 TL |
880 | ceph_assert(i.second.state == MDSMap::STATE_STANDBY); |
881 | ceph_assert(i.second.rank == MDS_RANK_NONE); | |
882 | ceph_assert(i.second.global_id == i.first); | |
883 | ceph_assert(standby_epochs.count(i.first) == 1); | |
884 | ceph_assert(mds_roles.count(i.first) == 1); | |
885 | ceph_assert(mds_roles.at(i.first) == FS_CLUSTER_ID_NONE); | |
7c673cae FG |
886 | } |
887 | ||
888 | for (const auto &i : standby_epochs) { | |
11fdf7f2 | 889 | ceph_assert(standby_daemons.count(i.first) == 1); |
7c673cae FG |
890 | } |
891 | ||
892 | for (const auto &i : mds_roles) { | |
893 | if (i.second == FS_CLUSTER_ID_NONE) { | |
11fdf7f2 | 894 | ceph_assert(standby_daemons.count(i.first) == 1); |
7c673cae | 895 | } else { |
11fdf7f2 TL |
896 | ceph_assert(filesystems.count(i.second) == 1); |
897 | ceph_assert(filesystems.at(i.second)->mds_map.mds_info.count(i.first) == 1); | |
7c673cae FG |
898 | } |
899 | } | |
900 | } | |
901 | ||
902 | void FSMap::promote( | |
903 | mds_gid_t standby_gid, | |
11fdf7f2 | 904 | Filesystem& filesystem, |
7c673cae FG |
905 | mds_rank_t assigned_rank) |
906 | { | |
11fdf7f2 | 907 | ceph_assert(gid_exists(standby_gid)); |
7c673cae FG |
908 | bool is_standby_replay = mds_roles.at(standby_gid) != FS_CLUSTER_ID_NONE; |
909 | if (!is_standby_replay) { | |
11fdf7f2 TL |
910 | ceph_assert(standby_daemons.count(standby_gid)); |
911 | ceph_assert(standby_daemons.at(standby_gid).state == MDSMap::STATE_STANDBY); | |
7c673cae FG |
912 | } |
913 | ||
11fdf7f2 | 914 | MDSMap &mds_map = filesystem.mds_map; |
7c673cae FG |
915 | |
916 | // Insert daemon state to Filesystem | |
917 | if (!is_standby_replay) { | |
918 | mds_map.mds_info[standby_gid] = standby_daemons.at(standby_gid); | |
919 | } else { | |
11fdf7f2 TL |
920 | ceph_assert(mds_map.mds_info.count(standby_gid)); |
921 | ceph_assert(mds_map.mds_info.at(standby_gid).state == MDSMap::STATE_STANDBY_REPLAY); | |
922 | ceph_assert(mds_map.mds_info.at(standby_gid).rank == assigned_rank); | |
7c673cae | 923 | } |
522d829b | 924 | auto& info = mds_map.mds_info.at(standby_gid); |
7c673cae | 925 | |
a4b75251 TL |
926 | if (!filesystem.mds_map.compat.writeable(info.compat)) { |
927 | ceph_assert(filesystem.is_upgradeable()); | |
928 | filesystem.mds_map.compat.merge(info.compat); | |
929 | } | |
930 | ||
7c673cae FG |
931 | if (mds_map.stopped.erase(assigned_rank)) { |
932 | // The cluster is being expanded with a stopped rank | |
933 | info.state = MDSMap::STATE_STARTING; | |
934 | } else if (!mds_map.is_in(assigned_rank)) { | |
935 | // The cluster is being expanded with a new rank | |
936 | info.state = MDSMap::STATE_CREATING; | |
937 | } else { | |
938 | // An existing rank is being assigned to a replacement | |
939 | info.state = MDSMap::STATE_REPLAY; | |
940 | mds_map.failed.erase(assigned_rank); | |
941 | } | |
942 | info.rank = assigned_rank; | |
943 | info.inc = epoch; | |
522d829b | 944 | mds_roles.at(standby_gid) = filesystem.fscid; |
7c673cae FG |
945 | |
946 | // Update the rank state in Filesystem | |
947 | mds_map.in.insert(assigned_rank); | |
948 | mds_map.up[assigned_rank] = standby_gid; | |
949 | ||
950 | // Remove from the list of standbys | |
951 | if (!is_standby_replay) { | |
952 | standby_daemons.erase(standby_gid); | |
953 | standby_epochs.erase(standby_gid); | |
954 | } | |
955 | ||
956 | // Indicate that Filesystem has been modified | |
957 | mds_map.epoch = epoch; | |
958 | } | |
959 | ||
960 | void FSMap::assign_standby_replay( | |
961 | const mds_gid_t standby_gid, | |
962 | const fs_cluster_id_t leader_ns, | |
963 | const mds_rank_t leader_rank) | |
964 | { | |
11fdf7f2 TL |
965 | ceph_assert(mds_roles.at(standby_gid) == FS_CLUSTER_ID_NONE); |
966 | ceph_assert(gid_exists(standby_gid)); | |
967 | ceph_assert(!gid_has_rank(standby_gid)); | |
968 | ceph_assert(standby_daemons.count(standby_gid)); | |
7c673cae FG |
969 | |
970 | // Insert to the filesystem | |
971 | auto fs = filesystems.at(leader_ns); | |
972 | fs->mds_map.mds_info[standby_gid] = standby_daemons.at(standby_gid); | |
973 | fs->mds_map.mds_info[standby_gid].rank = leader_rank; | |
974 | fs->mds_map.mds_info[standby_gid].state = MDSMap::STATE_STANDBY_REPLAY; | |
975 | mds_roles[standby_gid] = leader_ns; | |
976 | ||
977 | // Remove from the list of standbys | |
978 | standby_daemons.erase(standby_gid); | |
979 | standby_epochs.erase(standby_gid); | |
980 | ||
981 | // Indicate that Filesystem has been modified | |
982 | fs->mds_map.epoch = epoch; | |
983 | } | |
984 | ||
f67539c2 | 985 | void FSMap::erase(mds_gid_t who, epoch_t blocklist_epoch) |
7c673cae FG |
986 | { |
987 | if (mds_roles.at(who) == FS_CLUSTER_ID_NONE) { | |
988 | standby_daemons.erase(who); | |
989 | standby_epochs.erase(who); | |
990 | } else { | |
991 | auto &fs = filesystems.at(mds_roles.at(who)); | |
992 | const auto &info = fs->mds_map.mds_info.at(who); | |
993 | if (info.state != MDSMap::STATE_STANDBY_REPLAY) { | |
994 | if (info.state == MDSMap::STATE_CREATING) { | |
995 | // If this gid didn't make it past CREATING, then forget | |
996 | // the rank ever existed so that next time it's handed out | |
997 | // to a gid it'll go back into CREATING. | |
998 | fs->mds_map.in.erase(info.rank); | |
999 | } else { | |
1000 | // Put this rank into the failed list so that the next available | |
1001 | // STANDBY will pick it up. | |
1002 | fs->mds_map.failed.insert(info.rank); | |
1003 | } | |
11fdf7f2 | 1004 | ceph_assert(fs->mds_map.up.at(info.rank) == info.global_id); |
7c673cae FG |
1005 | fs->mds_map.up.erase(info.rank); |
1006 | } | |
1007 | fs->mds_map.mds_info.erase(who); | |
f67539c2 | 1008 | fs->mds_map.last_failure_osd_epoch = blocklist_epoch; |
7c673cae FG |
1009 | fs->mds_map.epoch = epoch; |
1010 | } | |
1011 | ||
1012 | mds_roles.erase(who); | |
1013 | } | |
1014 | ||
f67539c2 | 1015 | void FSMap::damaged(mds_gid_t who, epoch_t blocklist_epoch) |
7c673cae | 1016 | { |
11fdf7f2 | 1017 | ceph_assert(mds_roles.at(who) != FS_CLUSTER_ID_NONE); |
7c673cae | 1018 | auto fs = filesystems.at(mds_roles.at(who)); |
a4b75251 | 1019 | mds_rank_t rank = fs->mds_map.mds_info.at(who).rank; |
7c673cae | 1020 | |
f67539c2 | 1021 | erase(who, blocklist_epoch); |
7c673cae FG |
1022 | fs->mds_map.failed.erase(rank); |
1023 | fs->mds_map.damaged.insert(rank); | |
1024 | ||
11fdf7f2 | 1025 | ceph_assert(fs->mds_map.epoch == epoch); |
7c673cae FG |
1026 | } |
1027 | ||
1028 | /** | |
1029 | * Update to indicate that the rank `rank` is to be removed | |
1030 | * from the damaged list of the filesystem `fscid` | |
1031 | */ | |
1032 | bool FSMap::undamaged(const fs_cluster_id_t fscid, const mds_rank_t rank) | |
1033 | { | |
1034 | auto fs = filesystems.at(fscid); | |
1035 | ||
1036 | if (fs->mds_map.damaged.erase(rank)) { | |
1037 | fs->mds_map.failed.insert(rank); | |
1038 | fs->mds_map.epoch = epoch; | |
1039 | return true; | |
1040 | } else { | |
1041 | return false; | |
1042 | } | |
1043 | } | |
1044 | ||
1045 | void FSMap::insert(const MDSMap::mds_info_t &new_info) | |
1046 | { | |
a4b75251 TL |
1047 | static const CompatSet empty; |
1048 | ||
11fdf7f2 TL |
1049 | ceph_assert(new_info.state == MDSMap::STATE_STANDBY); |
1050 | ceph_assert(new_info.rank == MDS_RANK_NONE); | |
7c673cae | 1051 | mds_roles[new_info.global_id] = FS_CLUSTER_ID_NONE; |
a4b75251 TL |
1052 | auto& info = standby_daemons[new_info.global_id]; |
1053 | info = new_info; | |
1054 | if (empty.compare(info.compat) == 0) { | |
1055 | // bootstrap old compat: boot beacon contains empty compat on old (v16.2.4 | |
1056 | // or older) MDS. | |
1057 | info.compat = MDSMap::get_compat_set_v16_2_4(); | |
1058 | } | |
7c673cae FG |
1059 | standby_epochs[new_info.global_id] = epoch; |
1060 | } | |
1061 | ||
9f95a23c | 1062 | std::vector<mds_gid_t> FSMap::stop(mds_gid_t who) |
7c673cae | 1063 | { |
11fdf7f2 | 1064 | ceph_assert(mds_roles.at(who) != FS_CLUSTER_ID_NONE); |
7c673cae FG |
1065 | auto fs = filesystems.at(mds_roles.at(who)); |
1066 | const auto &info = fs->mds_map.mds_info.at(who); | |
1067 | fs->mds_map.up.erase(info.rank); | |
1068 | fs->mds_map.in.erase(info.rank); | |
1069 | fs->mds_map.stopped.insert(info.rank); | |
1070 | ||
1071 | // Also drop any standby replays that were following this rank | |
9f95a23c | 1072 | std::vector<mds_gid_t> standbys; |
7c673cae FG |
1073 | for (const auto &i : fs->mds_map.mds_info) { |
1074 | const auto &other_gid = i.first; | |
1075 | const auto &other_info = i.second; | |
1076 | if (other_info.rank == info.rank | |
1077 | && other_info.state == MDSMap::STATE_STANDBY_REPLAY) { | |
1078 | standbys.push_back(other_gid); | |
1079 | erase(other_gid, 0); | |
1080 | } | |
1081 | } | |
1082 | ||
1083 | fs->mds_map.mds_info.erase(who); | |
1084 | mds_roles.erase(who); | |
1085 | ||
1086 | fs->mds_map.epoch = epoch; | |
1087 | ||
1088 | return standbys; | |
1089 | } | |
1090 | ||
1091 | ||
1092 | /** | |
1093 | * Given one of the following forms: | |
1094 | * <fs name>:<rank> | |
1095 | * <fs id>:<rank> | |
1096 | * <rank> | |
1097 | * | |
1098 | * Parse into a mds_role_t. The rank-only form is only valid | |
1099 | * if legacy_client_ns is set. | |
1100 | */ | |
f67539c2 TL |
1101 | |
1102 | int FSMap::parse_role( | |
1103 | std::string_view role_str, | |
1104 | mds_role_t *role, | |
1105 | std::ostream &ss, | |
1106 | const std::vector<string> &filter) const | |
1107 | { | |
1108 | int r = parse_role(role_str, role, ss); | |
1109 | if (r < 0) return r; | |
1110 | ||
1111 | string_view fs_name = get_filesystem(role->fscid)->mds_map.get_fs_name(); | |
1112 | ||
1113 | if (!filter.empty() && | |
1114 | std::find(filter.begin(), filter.end(), fs_name) == filter.end()) { | |
1115 | if (r >= 0) { | |
1116 | ss << "Invalid file system"; | |
1117 | } | |
1118 | return -CEPHFS_ENOENT; | |
1119 | } | |
1120 | ||
1121 | return r; | |
1122 | } | |
1123 | ||
7c673cae | 1124 | int FSMap::parse_role( |
11fdf7f2 | 1125 | std::string_view role_str, |
7c673cae FG |
1126 | mds_role_t *role, |
1127 | std::ostream &ss) const | |
1128 | { | |
1129 | size_t colon_pos = role_str.find(":"); | |
1130 | size_t rank_pos; | |
11fdf7f2 | 1131 | Filesystem::const_ref fs; |
7c673cae FG |
1132 | if (colon_pos == std::string::npos) { |
1133 | if (legacy_client_fscid == FS_CLUSTER_ID_NONE) { | |
1134 | ss << "No filesystem selected"; | |
f67539c2 | 1135 | return -CEPHFS_ENOENT; |
7c673cae FG |
1136 | } |
1137 | fs = get_filesystem(legacy_client_fscid); | |
1138 | rank_pos = 0; | |
1139 | } else { | |
1140 | if (parse_filesystem(role_str.substr(0, colon_pos), &fs) < 0) { | |
1141 | ss << "Invalid filesystem"; | |
f67539c2 | 1142 | return -CEPHFS_ENOENT; |
7c673cae FG |
1143 | } |
1144 | rank_pos = colon_pos+1; | |
1145 | } | |
1146 | ||
1147 | mds_rank_t rank; | |
1148 | std::string err; | |
94b18763 | 1149 | std::string rank_str(role_str.substr(rank_pos)); |
7c673cae FG |
1150 | long rank_i = strict_strtol(rank_str.c_str(), 10, &err); |
1151 | if (rank_i < 0 || !err.empty()) { | |
1152 | ss << "Invalid rank '" << rank_str << "'"; | |
f67539c2 | 1153 | return -CEPHFS_EINVAL; |
7c673cae FG |
1154 | } else { |
1155 | rank = rank_i; | |
1156 | } | |
1157 | ||
1158 | if (fs->mds_map.in.count(rank) == 0) { | |
1159 | ss << "Rank '" << rank << "' not found"; | |
f67539c2 | 1160 | return -CEPHFS_ENOENT; |
7c673cae FG |
1161 | } |
1162 | ||
1163 | *role = {fs->fscid, rank}; | |
1164 | ||
1165 | return 0; | |
1166 | } | |
9f95a23c TL |
1167 | |
1168 | bool FSMap::pool_in_use(int64_t poolid) const | |
1169 | { | |
1170 | for (auto const &i : filesystems) { | |
1171 | if (i.second->mds_map.is_data_pool(poolid) | |
1172 | || i.second->mds_map.metadata_pool == poolid) { | |
1173 | return true; | |
1174 | } | |
1175 | } | |
1176 | return false; | |
1177 | } | |
1178 | ||
1179 | void FSMap::erase_filesystem(fs_cluster_id_t fscid) | |
1180 | { | |
1181 | filesystems.erase(fscid); | |
1182 | for (auto& [gid, info] : standby_daemons) { | |
1183 | if (info.join_fscid == fscid) { | |
1184 | modify_daemon(gid, [](auto& info) { | |
1185 | info.join_fscid = FS_CLUSTER_ID_NONE; | |
1186 | }); | |
1187 | } | |
1188 | } | |
1189 | for (auto& p : filesystems) { | |
1190 | for (auto& [gid, info] : p.second->mds_map.get_mds_info()) { | |
1191 | if (info.join_fscid == fscid) { | |
1192 | modify_daemon(gid, [](auto& info) { | |
1193 | info.join_fscid = FS_CLUSTER_ID_NONE; | |
1194 | }); | |
1195 | } | |
1196 | } | |
1197 | } | |
1198 | } |