]>
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 | ||
15 | ||
16 | #include "FSMap.h" | |
17 | ||
18 | #include <sstream> | |
19 | using std::stringstream; | |
20 | ||
21 | ||
22 | void Filesystem::dump(Formatter *f) const | |
23 | { | |
24 | f->open_object_section("mdsmap"); | |
25 | mds_map.dump(f); | |
26 | f->close_section(); | |
27 | f->dump_int("id", fscid); | |
28 | } | |
29 | ||
30 | void FSMap::dump(Formatter *f) const | |
31 | { | |
32 | f->dump_int("epoch", epoch); | |
33 | ||
34 | f->open_object_section("compat"); | |
35 | compat.dump(f); | |
36 | f->close_section(); | |
37 | ||
38 | f->open_object_section("feature_flags"); | |
39 | f->dump_bool("enable_multiple", enable_multiple); | |
40 | f->dump_bool("ever_enabled_multiple", ever_enabled_multiple); | |
41 | f->close_section(); | |
42 | ||
43 | f->open_array_section("standbys"); | |
44 | for (const auto &i : standby_daemons) { | |
45 | f->open_object_section("info"); | |
46 | i.second.dump(f); | |
47 | f->dump_int("epoch", standby_epochs.at(i.first)); | |
48 | f->close_section(); | |
49 | } | |
50 | f->close_section(); | |
51 | ||
52 | f->open_array_section("filesystems"); | |
53 | for (const auto &fs : filesystems) { | |
54 | f->open_object_section("filesystem"); | |
55 | fs.second->dump(f); | |
56 | f->close_section(); | |
57 | } | |
58 | f->close_section(); | |
59 | } | |
60 | ||
61 | void FSMap::generate_test_instances(list<FSMap*>& ls) | |
62 | { | |
63 | FSMap *m = new FSMap(); | |
64 | ||
65 | std::list<MDSMap*> mds_map_instances; | |
66 | MDSMap::generate_test_instances(mds_map_instances); | |
67 | ||
68 | int k = 20; | |
69 | for (auto i : mds_map_instances) { | |
70 | auto fs = std::make_shared<Filesystem>(); | |
71 | fs->fscid = k++; | |
72 | fs->mds_map = *i; | |
73 | delete i; | |
74 | m->filesystems[fs->fscid] = fs; | |
75 | } | |
76 | mds_map_instances.clear(); | |
77 | ||
78 | ls.push_back(m); | |
79 | } | |
80 | ||
81 | void FSMap::print(ostream& out) const | |
82 | { | |
83 | out << "e" << epoch << std::endl; | |
84 | out << "enable_multiple, ever_enabled_multiple: " << enable_multiple << "," | |
85 | << ever_enabled_multiple << std::endl; | |
86 | out << "compat: " << compat << std::endl; | |
87 | out << "legacy client fscid: " << legacy_client_fscid << std::endl; | |
88 | out << " " << std::endl; | |
89 | ||
90 | if (filesystems.empty()) { | |
91 | out << "No filesystems configured" << std::endl; | |
92 | return; | |
93 | } | |
94 | ||
95 | for (const auto &fs : filesystems) { | |
96 | fs.second->print(out); | |
97 | out << " " << std::endl << " " << std::endl; // Space out a bit | |
98 | } | |
99 | ||
100 | if (!standby_daemons.empty()) { | |
101 | out << "Standby daemons:" << std::endl << " " << std::endl; | |
102 | } | |
103 | ||
104 | for (const auto &p : standby_daemons) { | |
105 | p.second.print_summary(out); | |
106 | out << std::endl; | |
107 | } | |
108 | } | |
109 | ||
110 | ||
111 | ||
112 | void FSMap::print_summary(Formatter *f, ostream *out) const | |
113 | { | |
114 | map<mds_role_t,string> by_rank; | |
115 | map<string,int> by_state; | |
116 | ||
117 | if (f) { | |
118 | f->dump_unsigned("epoch", get_epoch()); | |
119 | for (auto i : filesystems) { | |
120 | auto fs = i.second; | |
121 | f->dump_unsigned("id", fs->fscid); | |
122 | f->dump_unsigned("up", fs->mds_map.up.size()); | |
123 | f->dump_unsigned("in", fs->mds_map.in.size()); | |
124 | f->dump_unsigned("max", fs->mds_map.max_mds); | |
125 | } | |
126 | } else { | |
127 | *out << "e" << get_epoch() << ":"; | |
128 | if (filesystems.size() == 1) { | |
129 | auto fs = filesystems.begin()->second; | |
130 | *out << " " << fs->mds_map.up.size() << "/" << fs->mds_map.in.size() << "/" | |
131 | << fs->mds_map.max_mds << " up"; | |
132 | } else { | |
133 | for (auto i : filesystems) { | |
134 | auto fs = i.second; | |
135 | *out << " " << fs->mds_map.fs_name << "-" << fs->mds_map.up.size() << "/" | |
136 | << fs->mds_map.in.size() << "/" << fs->mds_map.max_mds << " up"; | |
137 | } | |
138 | } | |
139 | } | |
140 | ||
141 | if (f) { | |
142 | f->open_array_section("by_rank"); | |
143 | } | |
144 | ||
145 | const auto all_info = get_mds_info(); | |
146 | for (const auto &p : all_info) { | |
147 | const auto &info = p.second; | |
148 | string s = ceph_mds_state_name(info.state); | |
149 | if (info.laggy()) { | |
150 | s += "(laggy or crashed)"; | |
151 | } | |
152 | ||
153 | const fs_cluster_id_t fscid = mds_roles.at(info.global_id); | |
154 | ||
155 | if (info.rank != MDS_RANK_NONE && | |
156 | info.state != MDSMap::STATE_STANDBY_REPLAY) { | |
157 | if (f) { | |
158 | f->open_object_section("mds"); | |
159 | f->dump_unsigned("filesystem_id", fscid); | |
160 | f->dump_unsigned("rank", info.rank); | |
161 | f->dump_string("name", info.name); | |
162 | f->dump_string("status", s); | |
163 | f->close_section(); | |
164 | } else { | |
165 | by_rank[mds_role_t(fscid, info.rank)] = info.name + "=" + s; | |
166 | } | |
167 | } else { | |
168 | by_state[s]++; | |
169 | } | |
170 | } | |
171 | ||
172 | if (f) { | |
173 | f->close_section(); | |
174 | } else { | |
175 | if (!by_rank.empty()) { | |
176 | if (filesystems.size() > 1) { | |
177 | // Disambiguate filesystems | |
178 | std::map<std::string, std::string> pretty; | |
179 | for (auto i : by_rank) { | |
180 | const auto &fs_name = filesystems.at(i.first.fscid)->mds_map.fs_name; | |
181 | std::ostringstream o; | |
182 | o << "[" << fs_name << ":" << i.first.rank << "]"; | |
183 | pretty[o.str()] = i.second; | |
184 | } | |
185 | *out << " " << pretty; | |
186 | } else { | |
187 | // Omit FSCID in output when only one filesystem exists | |
188 | std::map<mds_rank_t, std::string> shortened; | |
189 | for (auto i : by_rank) { | |
190 | shortened[i.first.rank] = i.second; | |
191 | } | |
192 | *out << " " << shortened; | |
193 | } | |
194 | } | |
195 | } | |
196 | ||
197 | for (map<string,int>::reverse_iterator p = by_state.rbegin(); p != by_state.rend(); ++p) { | |
198 | if (f) { | |
199 | f->dump_unsigned(p->first.c_str(), p->second); | |
200 | } else { | |
201 | *out << ", " << p->second << " " << p->first; | |
202 | } | |
203 | } | |
204 | ||
205 | size_t failed = 0; | |
206 | size_t damaged = 0; | |
207 | for (auto i : filesystems) { | |
208 | auto fs = i.second; | |
209 | failed += fs->mds_map.failed.size(); | |
210 | damaged += fs->mds_map.damaged.size(); | |
211 | } | |
212 | ||
213 | if (failed > 0) { | |
214 | if (f) { | |
215 | f->dump_unsigned("failed", failed); | |
216 | } else { | |
217 | *out << ", " << failed << " failed"; | |
218 | } | |
219 | } | |
220 | ||
221 | if (damaged > 0) { | |
222 | if (f) { | |
223 | f->dump_unsigned("damaged", damaged); | |
224 | } else { | |
225 | *out << ", " << damaged << " damaged"; | |
226 | } | |
227 | } | |
228 | //if (stopped.size()) | |
229 | //out << ", " << stopped.size() << " stopped"; | |
230 | } | |
231 | ||
232 | ||
233 | void FSMap::create_filesystem(const std::string &name, | |
234 | int64_t metadata_pool, int64_t data_pool, | |
235 | uint64_t features) | |
236 | { | |
237 | auto fs = std::make_shared<Filesystem>(); | |
238 | fs->mds_map.fs_name = name; | |
239 | fs->mds_map.max_mds = 1; | |
240 | fs->mds_map.data_pools.insert(data_pool); | |
241 | fs->mds_map.metadata_pool = metadata_pool; | |
242 | fs->mds_map.cas_pool = -1; | |
243 | fs->mds_map.max_file_size = g_conf->mds_max_file_size; | |
244 | fs->mds_map.compat = compat; | |
245 | fs->mds_map.created = ceph_clock_now(); | |
246 | fs->mds_map.modified = ceph_clock_now(); | |
247 | fs->mds_map.session_timeout = g_conf->mds_session_timeout; | |
248 | fs->mds_map.session_autoclose = g_conf->mds_session_autoclose; | |
249 | fs->mds_map.enabled = true; | |
250 | if (features & CEPH_FEATURE_SERVER_JEWEL) { | |
251 | fs->fscid = next_filesystem_id++; | |
252 | // ANONYMOUS is only for upgrades from legacy mdsmaps, we should | |
253 | // have initialized next_filesystem_id such that it's never used here. | |
254 | assert(fs->fscid != FS_CLUSTER_ID_ANONYMOUS); | |
255 | } else { | |
256 | // Use anon fscid because this will get thrown away when encoding | |
257 | // as legacy MDSMap for legacy mons. | |
258 | assert(filesystems.empty()); | |
259 | fs->fscid = FS_CLUSTER_ID_ANONYMOUS; | |
260 | } | |
261 | filesystems[fs->fscid] = fs; | |
262 | ||
263 | // Created first filesystem? Set it as the one | |
264 | // for legacy clients to use | |
265 | if (filesystems.size() == 1) { | |
266 | legacy_client_fscid = fs->fscid; | |
267 | } | |
268 | } | |
269 | ||
270 | void FSMap::reset_filesystem(fs_cluster_id_t fscid) | |
271 | { | |
272 | auto fs = get_filesystem(fscid); | |
273 | auto new_fs = std::make_shared<Filesystem>(); | |
274 | ||
275 | // Populate rank 0 as existing (so don't go into CREATING) | |
276 | // but failed (so that next available MDS is assigned the rank) | |
277 | new_fs->mds_map.in.insert(mds_rank_t(0)); | |
278 | new_fs->mds_map.failed.insert(mds_rank_t(0)); | |
279 | ||
280 | // Carry forward what makes sense | |
281 | new_fs->fscid = fs->fscid; | |
282 | new_fs->mds_map.inline_data_enabled = fs->mds_map.inline_data_enabled; | |
283 | new_fs->mds_map.max_mds = 1; | |
284 | new_fs->mds_map.data_pools = fs->mds_map.data_pools; | |
285 | new_fs->mds_map.metadata_pool = fs->mds_map.metadata_pool; | |
286 | new_fs->mds_map.cas_pool = fs->mds_map.cas_pool; | |
287 | new_fs->mds_map.fs_name = fs->mds_map.fs_name; | |
288 | new_fs->mds_map.max_file_size = g_conf->mds_max_file_size; | |
289 | new_fs->mds_map.compat = compat; | |
290 | new_fs->mds_map.created = ceph_clock_now(); | |
291 | new_fs->mds_map.modified = ceph_clock_now(); | |
292 | new_fs->mds_map.session_timeout = g_conf->mds_session_timeout; | |
293 | new_fs->mds_map.session_autoclose = g_conf->mds_session_autoclose; | |
294 | new_fs->mds_map.standby_count_wanted = fs->mds_map.standby_count_wanted; | |
295 | new_fs->mds_map.enabled = true; | |
296 | ||
297 | // Persist the new FSMap | |
298 | filesystems[new_fs->fscid] = new_fs; | |
299 | } | |
300 | ||
301 | void FSMap::get_health(list<pair<health_status_t,string> >& summary, | |
302 | list<pair<health_status_t,string> > *detail) const | |
303 | { | |
304 | mds_rank_t standby_count_wanted = 0; | |
305 | for (const auto &i : filesystems) { | |
306 | const auto &fs = i.second; | |
307 | ||
308 | // TODO: move get_health up into here so that we can qualify | |
309 | // all the messages with what filesystem they're talking about | |
310 | fs->mds_map.get_health(summary, detail); | |
311 | ||
312 | standby_count_wanted = std::max(standby_count_wanted, fs->mds_map.get_standby_count_wanted((mds_rank_t)standby_daemons.size())); | |
313 | } | |
314 | ||
315 | if (standby_count_wanted) { | |
316 | std::ostringstream oss; | |
317 | oss << "insufficient standby daemons available: have " << standby_daemons.size() << "; want " << standby_count_wanted << " more"; | |
318 | summary.push_back(make_pair(HEALTH_WARN, oss.str())); | |
319 | } | |
320 | } | |
321 | ||
322 | bool FSMap::check_health(void) | |
323 | { | |
324 | bool changed = false; | |
325 | for (auto &i : filesystems) { | |
326 | changed |= i.second->mds_map.check_health((mds_rank_t)standby_daemons.size()); | |
327 | } | |
328 | return changed; | |
329 | } | |
330 | ||
331 | void FSMap::encode(bufferlist& bl, uint64_t features) const | |
332 | { | |
333 | if (features & CEPH_FEATURE_SERVER_JEWEL) { | |
334 | ENCODE_START(7, 6, bl); | |
335 | ::encode(epoch, bl); | |
336 | ::encode(next_filesystem_id, bl); | |
337 | ::encode(legacy_client_fscid, bl); | |
338 | ::encode(compat, bl); | |
339 | ::encode(enable_multiple, bl); | |
340 | std::vector<Filesystem> fs_list; | |
341 | for (auto i : filesystems) { | |
342 | fs_list.push_back(*(i.second)); | |
343 | } | |
344 | ::encode(fs_list, bl, features); | |
345 | ::encode(mds_roles, bl); | |
346 | ::encode(standby_daemons, bl, features); | |
347 | ::encode(standby_epochs, bl); | |
348 | ::encode(ever_enabled_multiple, bl); | |
349 | ENCODE_FINISH(bl); | |
350 | } else { | |
351 | if (filesystems.empty()) { | |
352 | MDSMap disabled_map; | |
353 | disabled_map.epoch = epoch; | |
354 | disabled_map.encode(bl, features); | |
355 | } else { | |
356 | // MDSMonitor should never have created multiple filesystems | |
357 | // until the quorum features indicated Jewel | |
358 | assert(filesystems.size() == 1); | |
359 | auto fs = filesystems.begin()->second; | |
360 | ||
361 | // Take the MDSMap for the enabled filesystem, and populated its | |
362 | // mds_info with the standbys to get a pre-jewel-style mon MDSMap. | |
363 | MDSMap full_mdsmap = fs->mds_map; | |
364 | full_mdsmap.epoch = epoch; | |
365 | for (const auto &p : standby_daemons) { | |
366 | full_mdsmap.mds_info[p.first] = p.second; | |
367 | } | |
368 | ||
369 | // Old MDSMaps don't set rank on standby replay daemons | |
370 | for (auto &i : full_mdsmap.mds_info) { | |
371 | auto &info = i.second; | |
372 | if (info.state == MDSMap::STATE_STANDBY_REPLAY) { | |
373 | info.rank = MDS_RANK_NONE; | |
374 | } | |
375 | } | |
376 | ||
377 | full_mdsmap.encode(bl, features); | |
378 | } | |
379 | } | |
380 | } | |
381 | ||
382 | void FSMap::decode(bufferlist::iterator& p) | |
383 | { | |
384 | // Because the mon used to store an MDSMap where we now | |
385 | // store an FSMap, FSMap knows how to decode the legacy | |
386 | // MDSMap format (it never needs to encode it though). | |
387 | MDSMap legacy_mds_map; | |
388 | ||
389 | // The highest MDSMap encoding version before we changed the | |
390 | // MDSMonitor to store an FSMap instead of an MDSMap was | |
391 | // 5, so anything older than 6 is decoded as an MDSMap, | |
392 | // and anything newer is decoded as an FSMap. | |
393 | DECODE_START_LEGACY_COMPAT_LEN_16(7, 4, 4, p); | |
394 | if (struct_v < 6) { | |
395 | // Decoding an MDSMap (upgrade) | |
396 | ::decode(epoch, p); | |
397 | ::decode(legacy_mds_map.flags, p); | |
398 | ::decode(legacy_mds_map.last_failure, p); | |
399 | ::decode(legacy_mds_map.root, p); | |
400 | ::decode(legacy_mds_map.session_timeout, p); | |
401 | ::decode(legacy_mds_map.session_autoclose, p); | |
402 | ::decode(legacy_mds_map.max_file_size, p); | |
403 | ::decode(legacy_mds_map.max_mds, p); | |
404 | ::decode(legacy_mds_map.mds_info, p); | |
405 | if (struct_v < 3) { | |
406 | __u32 n; | |
407 | ::decode(n, p); | |
408 | while (n--) { | |
409 | __u32 m; | |
410 | ::decode(m, p); | |
411 | legacy_mds_map.data_pools.insert(m); | |
412 | } | |
413 | __s32 s; | |
414 | ::decode(s, p); | |
415 | legacy_mds_map.cas_pool = s; | |
416 | } else { | |
417 | ::decode(legacy_mds_map.data_pools, p); | |
418 | ::decode(legacy_mds_map.cas_pool, p); | |
419 | } | |
420 | ||
421 | // kclient ignores everything from here | |
422 | __u16 ev = 1; | |
423 | if (struct_v >= 2) | |
424 | ::decode(ev, p); | |
425 | if (ev >= 3) | |
426 | ::decode(legacy_mds_map.compat, p); | |
427 | else | |
428 | legacy_mds_map.compat = get_mdsmap_compat_set_base(); | |
429 | if (ev < 5) { | |
430 | __u32 n; | |
431 | ::decode(n, p); | |
432 | legacy_mds_map.metadata_pool = n; | |
433 | } else { | |
434 | ::decode(legacy_mds_map.metadata_pool, p); | |
435 | } | |
436 | ::decode(legacy_mds_map.created, p); | |
437 | ::decode(legacy_mds_map.modified, p); | |
438 | ::decode(legacy_mds_map.tableserver, p); | |
439 | ::decode(legacy_mds_map.in, p); | |
440 | std::map<mds_rank_t,int32_t> inc; // Legacy field, parse and drop | |
441 | ::decode(inc, p); | |
442 | ::decode(legacy_mds_map.up, p); | |
443 | ::decode(legacy_mds_map.failed, p); | |
444 | ::decode(legacy_mds_map.stopped, p); | |
445 | if (ev >= 4) | |
446 | ::decode(legacy_mds_map.last_failure_osd_epoch, p); | |
447 | if (ev >= 6) { | |
448 | if (ev < 10) { | |
449 | // previously this was a bool about snaps, not a flag map | |
450 | bool flag; | |
451 | ::decode(flag, p); | |
452 | legacy_mds_map.ever_allowed_features = flag ? | |
453 | CEPH_MDSMAP_ALLOW_SNAPS : 0; | |
454 | ::decode(flag, p); | |
455 | legacy_mds_map.explicitly_allowed_features = flag ? | |
456 | CEPH_MDSMAP_ALLOW_SNAPS : 0; | |
457 | if (legacy_mds_map.max_mds > 1) { | |
458 | legacy_mds_map.set_multimds_allowed(); | |
459 | } | |
460 | } else { | |
461 | ::decode(legacy_mds_map.ever_allowed_features, p); | |
462 | ::decode(legacy_mds_map.explicitly_allowed_features, p); | |
463 | } | |
464 | } else { | |
465 | legacy_mds_map.ever_allowed_features = CEPH_MDSMAP_ALLOW_CLASSICS; | |
466 | legacy_mds_map.explicitly_allowed_features = 0; | |
467 | if (legacy_mds_map.max_mds > 1) { | |
468 | legacy_mds_map.set_multimds_allowed(); | |
469 | } | |
470 | } | |
471 | if (ev >= 7) | |
472 | ::decode(legacy_mds_map.inline_data_enabled, p); | |
473 | ||
474 | if (ev >= 8) { | |
475 | assert(struct_v >= 5); | |
476 | ::decode(legacy_mds_map.enabled, p); | |
477 | ::decode(legacy_mds_map.fs_name, p); | |
478 | } else { | |
479 | legacy_mds_map.fs_name = "default"; | |
480 | if (epoch > 1) { | |
481 | // If an MDS has ever been started, epoch will be greater than 1, | |
482 | // assume filesystem is enabled. | |
483 | legacy_mds_map.enabled = true; | |
484 | } else { | |
485 | // Upgrading from a cluster that never used an MDS, switch off | |
486 | // filesystem until it's explicitly enabled. | |
487 | legacy_mds_map.enabled = false; | |
488 | } | |
489 | } | |
490 | ||
491 | if (ev >= 9) { | |
492 | ::decode(legacy_mds_map.damaged, p); | |
493 | } | |
494 | ||
495 | // We're upgrading, populate filesystems from the legacy fields | |
496 | filesystems.clear(); | |
497 | standby_daemons.clear(); | |
498 | standby_epochs.clear(); | |
499 | mds_roles.clear(); | |
500 | compat = legacy_mds_map.compat; | |
501 | enable_multiple = false; | |
502 | ||
503 | // Synthesise a Filesystem from legacy_mds_map, if enabled | |
504 | if (legacy_mds_map.enabled) { | |
505 | // Construct a Filesystem from the legacy MDSMap | |
506 | auto migrate_fs = std::make_shared<Filesystem>(); | |
507 | migrate_fs->fscid = FS_CLUSTER_ID_ANONYMOUS; | |
508 | migrate_fs->mds_map = legacy_mds_map; | |
509 | migrate_fs->mds_map.epoch = epoch; | |
510 | filesystems[migrate_fs->fscid] = migrate_fs; | |
511 | ||
512 | // List of GIDs that had invalid states | |
513 | std::set<mds_gid_t> drop_gids; | |
514 | ||
515 | // Construct mds_roles, standby_daemons, and remove | |
516 | // standbys from the MDSMap in the Filesystem. | |
517 | for (auto &p : migrate_fs->mds_map.mds_info) { | |
518 | if (p.second.state == MDSMap::STATE_STANDBY_REPLAY) { | |
519 | // In legacy MDSMap, standby replay daemons don't have | |
520 | // rank set, but since FSMap they do. | |
521 | p.second.rank = p.second.standby_for_rank; | |
522 | } | |
523 | if (p.second.rank == MDS_RANK_NONE) { | |
524 | if (p.second.state != MDSMap::STATE_STANDBY) { | |
525 | // Old MDSMaps can have down:dne here, which | |
526 | // is invalid in an FSMap (#17837) | |
527 | drop_gids.insert(p.first); | |
528 | } else { | |
529 | insert(p.second); // into standby_daemons | |
530 | } | |
531 | } else { | |
532 | mds_roles[p.first] = migrate_fs->fscid; | |
533 | } | |
534 | } | |
535 | for (const auto &p : standby_daemons) { | |
536 | // Erase from this Filesystem's MDSMap, because it has | |
537 | // been copied into FSMap::Standby_daemons above | |
538 | migrate_fs->mds_map.mds_info.erase(p.first); | |
539 | } | |
540 | for (const auto &gid : drop_gids) { | |
541 | // Throw away all info for this MDS because it was identified | |
542 | // as having invalid state above. | |
543 | migrate_fs->mds_map.mds_info.erase(gid); | |
544 | } | |
545 | ||
546 | legacy_client_fscid = migrate_fs->fscid; | |
547 | } else { | |
548 | legacy_client_fscid = FS_CLUSTER_ID_NONE; | |
549 | } | |
550 | } else { | |
551 | ::decode(epoch, p); | |
552 | ::decode(next_filesystem_id, p); | |
553 | ::decode(legacy_client_fscid, p); | |
554 | ::decode(compat, p); | |
555 | ::decode(enable_multiple, p); | |
556 | std::vector<Filesystem> fs_list; | |
557 | ::decode(fs_list, p); | |
558 | filesystems.clear(); | |
559 | for (std::vector<Filesystem>::const_iterator fs = fs_list.begin(); fs != fs_list.end(); ++fs) { | |
560 | filesystems[fs->fscid] = std::make_shared<Filesystem>(*fs); | |
561 | } | |
562 | ||
563 | ::decode(mds_roles, p); | |
564 | ::decode(standby_daemons, p); | |
565 | ::decode(standby_epochs, p); | |
566 | if (struct_v >= 7) { | |
567 | ::decode(ever_enabled_multiple, p); | |
568 | } | |
569 | } | |
570 | ||
571 | DECODE_FINISH(p); | |
572 | } | |
573 | ||
574 | ||
575 | void Filesystem::encode(bufferlist& bl, uint64_t features) const | |
576 | { | |
577 | ENCODE_START(1, 1, bl); | |
578 | ::encode(fscid, bl); | |
579 | bufferlist mdsmap_bl; | |
580 | mds_map.encode(mdsmap_bl, features); | |
581 | ::encode(mdsmap_bl, bl); | |
582 | ENCODE_FINISH(bl); | |
583 | } | |
584 | ||
585 | void Filesystem::decode(bufferlist::iterator& p) | |
586 | { | |
587 | DECODE_START(1, p); | |
588 | ::decode(fscid, p); | |
589 | bufferlist mdsmap_bl; | |
590 | ::decode(mdsmap_bl, p); | |
591 | bufferlist::iterator mdsmap_bl_iter = mdsmap_bl.begin(); | |
592 | mds_map.decode(mdsmap_bl_iter); | |
593 | DECODE_FINISH(p); | |
594 | } | |
595 | ||
596 | int FSMap::parse_filesystem( | |
597 | std::string const &ns_str, | |
598 | std::shared_ptr<const Filesystem> *result | |
599 | ) const | |
600 | { | |
601 | std::string ns_err; | |
602 | fs_cluster_id_t fscid = strict_strtol(ns_str.c_str(), 10, &ns_err); | |
603 | if (!ns_err.empty() || filesystems.count(fscid) == 0) { | |
604 | for (auto &fs : filesystems) { | |
605 | if (fs.second->mds_map.fs_name == ns_str) { | |
606 | *result = std::const_pointer_cast<const Filesystem>(fs.second); | |
607 | return 0; | |
608 | } | |
609 | } | |
610 | return -ENOENT; | |
611 | } else { | |
612 | *result = get_filesystem(fscid); | |
613 | return 0; | |
614 | } | |
615 | } | |
616 | ||
617 | void Filesystem::print(std::ostream &out) const | |
618 | { | |
619 | out << "Filesystem '" << mds_map.fs_name | |
620 | << "' (" << fscid << ")" << std::endl; | |
621 | mds_map.print(out); | |
622 | } | |
623 | ||
624 | mds_gid_t FSMap::find_standby_for(mds_role_t role, const std::string& name) const | |
625 | { | |
626 | mds_gid_t result = MDS_GID_NONE; | |
627 | ||
628 | // First see if we have a STANDBY_REPLAY | |
629 | auto fs = get_filesystem(role.fscid); | |
630 | for (const auto &i : fs->mds_map.mds_info) { | |
631 | const auto &info = i.second; | |
632 | if (info.rank == role.rank && info.state == MDSMap::STATE_STANDBY_REPLAY) { | |
633 | return info.global_id; | |
634 | } | |
635 | } | |
636 | ||
637 | // See if there are any STANDBY daemons available | |
638 | for (const auto &i : standby_daemons) { | |
639 | const auto &gid = i.first; | |
640 | const auto &info = i.second; | |
641 | assert(info.state == MDSMap::STATE_STANDBY); | |
642 | assert(info.rank == MDS_RANK_NONE); | |
643 | ||
644 | if (info.laggy()) { | |
645 | continue; | |
646 | } | |
647 | ||
648 | // The mds_info_t may or may not tell us exactly which filesystem | |
649 | // the standby_for_rank refers to: lookup via legacy_client_fscid | |
650 | mds_role_t target_role = { | |
651 | info.standby_for_fscid == FS_CLUSTER_ID_NONE ? | |
652 | legacy_client_fscid : info.standby_for_fscid, | |
653 | info.standby_for_rank}; | |
654 | ||
655 | if ((target_role.rank == role.rank && target_role.fscid == role.fscid) | |
656 | || (name.length() && info.standby_for_name == name)) { | |
657 | // It's a named standby for *me*, use it. | |
658 | return gid; | |
659 | } else if ( | |
660 | info.standby_for_rank < 0 && info.standby_for_name.length() == 0 && | |
661 | (info.standby_for_fscid == FS_CLUSTER_ID_NONE || | |
662 | info.standby_for_fscid == role.fscid)) { | |
663 | // It's not a named standby for anyone, use it if we don't find | |
664 | // a named standby for me later, unless it targets another FSCID. | |
665 | result = gid; | |
666 | } | |
667 | } | |
668 | ||
669 | return result; | |
670 | } | |
671 | ||
672 | mds_gid_t FSMap::find_unused(fs_cluster_id_t fscid, | |
673 | bool force_standby_active) const { | |
674 | for (const auto &i : standby_daemons) { | |
675 | const auto &gid = i.first; | |
676 | const auto &info = i.second; | |
677 | assert(info.state == MDSMap::STATE_STANDBY); | |
678 | ||
679 | if (info.laggy() || info.rank >= 0) | |
680 | continue; | |
681 | ||
682 | if (info.standby_for_fscid != FS_CLUSTER_ID_NONE && | |
683 | info.standby_for_fscid != fscid) | |
684 | continue; | |
685 | ||
686 | // To be considered 'unused' a daemon must either not | |
687 | // be selected for standby-replay or the force_standby_active | |
688 | // setting must be enabled to use replay daemons anyway. | |
689 | if (!info.standby_replay || force_standby_active) { | |
690 | return gid; | |
691 | } | |
692 | } | |
693 | return MDS_GID_NONE; | |
694 | } | |
695 | ||
696 | mds_gid_t FSMap::find_replacement_for(mds_role_t role, const std::string& name, | |
697 | bool force_standby_active) const { | |
698 | const mds_gid_t standby = find_standby_for(role, name); | |
699 | if (standby) | |
700 | return standby; | |
701 | else | |
702 | return find_unused(role.fscid, force_standby_active); | |
703 | } | |
704 | ||
705 | void FSMap::sanity() const | |
706 | { | |
707 | if (legacy_client_fscid != FS_CLUSTER_ID_NONE) { | |
708 | assert(filesystems.count(legacy_client_fscid) == 1); | |
709 | } | |
710 | ||
711 | for (const auto &i : filesystems) { | |
712 | auto fs = i.second; | |
713 | assert(fs->mds_map.compat.compare(compat) == 0); | |
714 | assert(fs->fscid == i.first); | |
715 | for (const auto &j : fs->mds_map.mds_info) { | |
716 | assert(j.second.rank != MDS_RANK_NONE); | |
717 | assert(mds_roles.count(j.first) == 1); | |
718 | assert(standby_daemons.count(j.first) == 0); | |
719 | assert(standby_epochs.count(j.first) == 0); | |
720 | assert(mds_roles.at(j.first) == i.first); | |
721 | if (j.second.state != MDSMap::STATE_STANDBY_REPLAY) { | |
722 | assert(fs->mds_map.up.at(j.second.rank) == j.first); | |
723 | assert(fs->mds_map.failed.count(j.second.rank) == 0); | |
724 | assert(fs->mds_map.damaged.count(j.second.rank) == 0); | |
725 | } | |
726 | } | |
727 | ||
728 | for (const auto &j : fs->mds_map.up) { | |
729 | mds_rank_t rank = j.first; | |
730 | assert(fs->mds_map.in.count(rank) == 1); | |
731 | mds_gid_t gid = j.second; | |
732 | assert(fs->mds_map.mds_info.count(gid) == 1); | |
733 | } | |
734 | } | |
735 | ||
736 | for (const auto &i : standby_daemons) { | |
737 | assert(i.second.state == MDSMap::STATE_STANDBY); | |
738 | assert(i.second.rank == MDS_RANK_NONE); | |
739 | assert(i.second.global_id == i.first); | |
740 | assert(standby_epochs.count(i.first) == 1); | |
741 | assert(mds_roles.count(i.first) == 1); | |
742 | assert(mds_roles.at(i.first) == FS_CLUSTER_ID_NONE); | |
743 | } | |
744 | ||
745 | for (const auto &i : standby_epochs) { | |
746 | assert(standby_daemons.count(i.first) == 1); | |
747 | } | |
748 | ||
749 | for (const auto &i : mds_roles) { | |
750 | if (i.second == FS_CLUSTER_ID_NONE) { | |
751 | assert(standby_daemons.count(i.first) == 1); | |
752 | } else { | |
753 | assert(filesystems.count(i.second) == 1); | |
754 | assert(filesystems.at(i.second)->mds_map.mds_info.count(i.first) == 1); | |
755 | } | |
756 | } | |
757 | } | |
758 | ||
759 | void FSMap::promote( | |
760 | mds_gid_t standby_gid, | |
761 | const std::shared_ptr<Filesystem> &filesystem, | |
762 | mds_rank_t assigned_rank) | |
763 | { | |
764 | assert(gid_exists(standby_gid)); | |
765 | bool is_standby_replay = mds_roles.at(standby_gid) != FS_CLUSTER_ID_NONE; | |
766 | if (!is_standby_replay) { | |
767 | assert(standby_daemons.count(standby_gid)); | |
768 | assert(standby_daemons.at(standby_gid).state == MDSMap::STATE_STANDBY); | |
769 | } | |
770 | ||
771 | MDSMap &mds_map = filesystem->mds_map; | |
772 | ||
773 | // Insert daemon state to Filesystem | |
774 | if (!is_standby_replay) { | |
775 | mds_map.mds_info[standby_gid] = standby_daemons.at(standby_gid); | |
776 | } else { | |
777 | assert(mds_map.mds_info.count(standby_gid)); | |
778 | assert(mds_map.mds_info.at(standby_gid).state == MDSMap::STATE_STANDBY_REPLAY); | |
779 | assert(mds_map.mds_info.at(standby_gid).rank == assigned_rank); | |
780 | } | |
781 | MDSMap::mds_info_t &info = mds_map.mds_info[standby_gid]; | |
782 | ||
783 | if (mds_map.stopped.erase(assigned_rank)) { | |
784 | // The cluster is being expanded with a stopped rank | |
785 | info.state = MDSMap::STATE_STARTING; | |
786 | } else if (!mds_map.is_in(assigned_rank)) { | |
787 | // The cluster is being expanded with a new rank | |
788 | info.state = MDSMap::STATE_CREATING; | |
789 | } else { | |
790 | // An existing rank is being assigned to a replacement | |
791 | info.state = MDSMap::STATE_REPLAY; | |
792 | mds_map.failed.erase(assigned_rank); | |
793 | } | |
794 | info.rank = assigned_rank; | |
795 | info.inc = epoch; | |
796 | mds_roles[standby_gid] = filesystem->fscid; | |
797 | ||
798 | // Update the rank state in Filesystem | |
799 | mds_map.in.insert(assigned_rank); | |
800 | mds_map.up[assigned_rank] = standby_gid; | |
801 | ||
802 | // Remove from the list of standbys | |
803 | if (!is_standby_replay) { | |
804 | standby_daemons.erase(standby_gid); | |
805 | standby_epochs.erase(standby_gid); | |
806 | } | |
807 | ||
808 | // Indicate that Filesystem has been modified | |
809 | mds_map.epoch = epoch; | |
810 | } | |
811 | ||
812 | void FSMap::assign_standby_replay( | |
813 | const mds_gid_t standby_gid, | |
814 | const fs_cluster_id_t leader_ns, | |
815 | const mds_rank_t leader_rank) | |
816 | { | |
817 | assert(mds_roles.at(standby_gid) == FS_CLUSTER_ID_NONE); | |
818 | assert(gid_exists(standby_gid)); | |
819 | assert(!gid_has_rank(standby_gid)); | |
820 | assert(standby_daemons.count(standby_gid)); | |
821 | ||
822 | // Insert to the filesystem | |
823 | auto fs = filesystems.at(leader_ns); | |
824 | fs->mds_map.mds_info[standby_gid] = standby_daemons.at(standby_gid); | |
825 | fs->mds_map.mds_info[standby_gid].rank = leader_rank; | |
826 | fs->mds_map.mds_info[standby_gid].state = MDSMap::STATE_STANDBY_REPLAY; | |
827 | mds_roles[standby_gid] = leader_ns; | |
828 | ||
829 | // Remove from the list of standbys | |
830 | standby_daemons.erase(standby_gid); | |
831 | standby_epochs.erase(standby_gid); | |
832 | ||
833 | // Indicate that Filesystem has been modified | |
834 | fs->mds_map.epoch = epoch; | |
835 | } | |
836 | ||
837 | void FSMap::erase(mds_gid_t who, epoch_t blacklist_epoch) | |
838 | { | |
839 | if (mds_roles.at(who) == FS_CLUSTER_ID_NONE) { | |
840 | standby_daemons.erase(who); | |
841 | standby_epochs.erase(who); | |
842 | } else { | |
843 | auto &fs = filesystems.at(mds_roles.at(who)); | |
844 | const auto &info = fs->mds_map.mds_info.at(who); | |
845 | if (info.state != MDSMap::STATE_STANDBY_REPLAY) { | |
846 | if (info.state == MDSMap::STATE_CREATING) { | |
847 | // If this gid didn't make it past CREATING, then forget | |
848 | // the rank ever existed so that next time it's handed out | |
849 | // to a gid it'll go back into CREATING. | |
850 | fs->mds_map.in.erase(info.rank); | |
851 | } else { | |
852 | // Put this rank into the failed list so that the next available | |
853 | // STANDBY will pick it up. | |
854 | fs->mds_map.failed.insert(info.rank); | |
855 | } | |
856 | assert(fs->mds_map.up.at(info.rank) == info.global_id); | |
857 | fs->mds_map.up.erase(info.rank); | |
858 | } | |
859 | fs->mds_map.mds_info.erase(who); | |
860 | fs->mds_map.last_failure_osd_epoch = blacklist_epoch; | |
861 | fs->mds_map.epoch = epoch; | |
862 | } | |
863 | ||
864 | mds_roles.erase(who); | |
865 | } | |
866 | ||
867 | void FSMap::damaged(mds_gid_t who, epoch_t blacklist_epoch) | |
868 | { | |
869 | assert(mds_roles.at(who) != FS_CLUSTER_ID_NONE); | |
870 | auto fs = filesystems.at(mds_roles.at(who)); | |
871 | mds_rank_t rank = fs->mds_map.mds_info[who].rank; | |
872 | ||
873 | erase(who, blacklist_epoch); | |
874 | fs->mds_map.failed.erase(rank); | |
875 | fs->mds_map.damaged.insert(rank); | |
876 | ||
877 | assert(fs->mds_map.epoch == epoch); | |
878 | } | |
879 | ||
880 | /** | |
881 | * Update to indicate that the rank `rank` is to be removed | |
882 | * from the damaged list of the filesystem `fscid` | |
883 | */ | |
884 | bool FSMap::undamaged(const fs_cluster_id_t fscid, const mds_rank_t rank) | |
885 | { | |
886 | auto fs = filesystems.at(fscid); | |
887 | ||
888 | if (fs->mds_map.damaged.erase(rank)) { | |
889 | fs->mds_map.failed.insert(rank); | |
890 | fs->mds_map.epoch = epoch; | |
891 | return true; | |
892 | } else { | |
893 | return false; | |
894 | } | |
895 | } | |
896 | ||
897 | void FSMap::insert(const MDSMap::mds_info_t &new_info) | |
898 | { | |
899 | assert(new_info.state == MDSMap::STATE_STANDBY); | |
900 | assert(new_info.rank == MDS_RANK_NONE); | |
901 | mds_roles[new_info.global_id] = FS_CLUSTER_ID_NONE; | |
902 | standby_daemons[new_info.global_id] = new_info; | |
903 | standby_epochs[new_info.global_id] = epoch; | |
904 | } | |
905 | ||
906 | std::list<mds_gid_t> FSMap::stop(mds_gid_t who) | |
907 | { | |
908 | assert(mds_roles.at(who) != FS_CLUSTER_ID_NONE); | |
909 | auto fs = filesystems.at(mds_roles.at(who)); | |
910 | const auto &info = fs->mds_map.mds_info.at(who); | |
911 | fs->mds_map.up.erase(info.rank); | |
912 | fs->mds_map.in.erase(info.rank); | |
913 | fs->mds_map.stopped.insert(info.rank); | |
914 | ||
915 | // Also drop any standby replays that were following this rank | |
916 | std::list<mds_gid_t> standbys; | |
917 | for (const auto &i : fs->mds_map.mds_info) { | |
918 | const auto &other_gid = i.first; | |
919 | const auto &other_info = i.second; | |
920 | if (other_info.rank == info.rank | |
921 | && other_info.state == MDSMap::STATE_STANDBY_REPLAY) { | |
922 | standbys.push_back(other_gid); | |
923 | erase(other_gid, 0); | |
924 | } | |
925 | } | |
926 | ||
927 | fs->mds_map.mds_info.erase(who); | |
928 | mds_roles.erase(who); | |
929 | ||
930 | fs->mds_map.epoch = epoch; | |
931 | ||
932 | return standbys; | |
933 | } | |
934 | ||
935 | ||
936 | /** | |
937 | * Given one of the following forms: | |
938 | * <fs name>:<rank> | |
939 | * <fs id>:<rank> | |
940 | * <rank> | |
941 | * | |
942 | * Parse into a mds_role_t. The rank-only form is only valid | |
943 | * if legacy_client_ns is set. | |
944 | */ | |
945 | int FSMap::parse_role( | |
946 | const std::string &role_str, | |
947 | mds_role_t *role, | |
948 | std::ostream &ss) const | |
949 | { | |
950 | size_t colon_pos = role_str.find(":"); | |
951 | size_t rank_pos; | |
952 | std::shared_ptr<const Filesystem> fs; | |
953 | if (colon_pos == std::string::npos) { | |
954 | if (legacy_client_fscid == FS_CLUSTER_ID_NONE) { | |
955 | ss << "No filesystem selected"; | |
956 | return -ENOENT; | |
957 | } | |
958 | fs = get_filesystem(legacy_client_fscid); | |
959 | rank_pos = 0; | |
960 | } else { | |
961 | if (parse_filesystem(role_str.substr(0, colon_pos), &fs) < 0) { | |
962 | ss << "Invalid filesystem"; | |
963 | return -ENOENT; | |
964 | } | |
965 | rank_pos = colon_pos+1; | |
966 | } | |
967 | ||
968 | mds_rank_t rank; | |
969 | std::string err; | |
970 | std::string rank_str = role_str.substr(rank_pos); | |
971 | long rank_i = strict_strtol(rank_str.c_str(), 10, &err); | |
972 | if (rank_i < 0 || !err.empty()) { | |
973 | ss << "Invalid rank '" << rank_str << "'"; | |
974 | return -EINVAL; | |
975 | } else { | |
976 | rank = rank_i; | |
977 | } | |
978 | ||
979 | if (fs->mds_map.in.count(rank) == 0) { | |
980 | ss << "Rank '" << rank << "' not found"; | |
981 | return -ENOENT; | |
982 | } | |
983 | ||
984 | *role = {fs->fscid, rank}; | |
985 | ||
986 | return 0; | |
987 | } |