]>
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 | #ifndef CEPH_FSMAP_H | |
17 | #define CEPH_FSMAP_H | |
18 | ||
19 | #include <errno.h> | |
20 | ||
21 | #include "include/types.h" | |
22 | #include "common/Clock.h" | |
23 | #include "msg/Message.h" | |
24 | #include "mds/MDSMap.h" | |
25 | ||
26 | #include <set> | |
27 | #include <map> | |
28 | #include <string> | |
29 | ||
30 | #include "common/config.h" | |
31 | ||
32 | #include "include/CompatSet.h" | |
33 | #include "include/ceph_features.h" | |
34 | #include "common/Formatter.h" | |
35 | #include "mds/mdstypes.h" | |
36 | ||
37 | class CephContext; | |
224ce89b | 38 | class health_check_map_t; |
7c673cae FG |
39 | |
40 | #define MDS_FEATURE_INCOMPAT_BASE CompatSet::Feature(1, "base v0.20") | |
41 | #define MDS_FEATURE_INCOMPAT_CLIENTRANGES CompatSet::Feature(2, "client writeable ranges") | |
42 | #define MDS_FEATURE_INCOMPAT_FILELAYOUT CompatSet::Feature(3, "default file layouts on dirs") | |
43 | #define MDS_FEATURE_INCOMPAT_DIRINODE CompatSet::Feature(4, "dir inode in separate object") | |
44 | #define MDS_FEATURE_INCOMPAT_ENCODING CompatSet::Feature(5, "mds uses versioned encoding") | |
45 | #define MDS_FEATURE_INCOMPAT_OMAPDIRFRAG CompatSet::Feature(6, "dirfrag is stored in omap") | |
46 | #define MDS_FEATURE_INCOMPAT_INLINE CompatSet::Feature(7, "mds uses inline data") | |
47 | #define MDS_FEATURE_INCOMPAT_NOANCHOR CompatSet::Feature(8, "no anchor table") | |
48 | ||
49 | #define MDS_FS_NAME_DEFAULT "cephfs" | |
50 | ||
51 | /** | |
52 | * The MDSMap and any additional fields describing a particular | |
53 | * filesystem (a unique fs_cluster_id_t). | |
54 | */ | |
55 | class Filesystem | |
56 | { | |
57 | public: | |
58 | fs_cluster_id_t fscid; | |
59 | MDSMap mds_map; | |
60 | ||
61 | void encode(bufferlist& bl, uint64_t features) const; | |
62 | void decode(bufferlist::iterator& p); | |
63 | ||
64 | Filesystem() | |
65 | : | |
66 | fscid(FS_CLUSTER_ID_NONE) | |
67 | { | |
68 | } | |
69 | ||
70 | void dump(Formatter *f) const; | |
71 | void print(std::ostream& out) const; | |
72 | ||
73 | /** | |
74 | * Return true if a daemon is already assigned as | |
75 | * STANDBY_REPLAY for the gid `who` | |
76 | */ | |
77 | bool has_standby_replay(mds_gid_t who) const | |
78 | { | |
79 | for (const auto &i : mds_map.mds_info) { | |
80 | const auto &info = i.second; | |
81 | if (info.state == MDSMap::STATE_STANDBY_REPLAY | |
82 | && info.rank == mds_map.mds_info.at(who).rank) { | |
83 | return true; | |
84 | } | |
85 | } | |
86 | ||
87 | return false; | |
88 | } | |
89 | }; | |
90 | WRITE_CLASS_ENCODER_FEATURES(Filesystem) | |
91 | ||
92 | class FSMap { | |
93 | protected: | |
94 | epoch_t epoch; | |
95 | uint64_t next_filesystem_id; | |
96 | fs_cluster_id_t legacy_client_fscid; | |
97 | CompatSet compat; | |
98 | bool enable_multiple; | |
99 | bool ever_enabled_multiple; // < the cluster had multiple MDSes enabled once | |
100 | ||
101 | std::map<fs_cluster_id_t, std::shared_ptr<Filesystem> > filesystems; | |
102 | ||
103 | // Remember which Filesystem an MDS daemon's info is stored in | |
104 | // (or in standby_daemons for FS_CLUSTER_ID_NONE) | |
105 | std::map<mds_gid_t, fs_cluster_id_t> mds_roles; | |
106 | ||
107 | // For MDS daemons not yet assigned to a Filesystem | |
108 | std::map<mds_gid_t, MDSMap::mds_info_t> standby_daemons; | |
109 | std::map<mds_gid_t, epoch_t> standby_epochs; | |
110 | ||
111 | public: | |
112 | ||
113 | friend class MDSMonitor; | |
114 | ||
115 | FSMap() | |
116 | : epoch(0), | |
117 | next_filesystem_id(FS_CLUSTER_ID_ANONYMOUS + 1), | |
118 | legacy_client_fscid(FS_CLUSTER_ID_NONE), | |
119 | compat(get_mdsmap_compat_set_default()), | |
120 | enable_multiple(false), ever_enabled_multiple(false) | |
121 | { } | |
122 | ||
123 | FSMap(const FSMap &rhs) | |
124 | : | |
125 | epoch(rhs.epoch), | |
126 | next_filesystem_id(rhs.next_filesystem_id), | |
127 | legacy_client_fscid(rhs.legacy_client_fscid), | |
128 | compat(rhs.compat), | |
129 | enable_multiple(rhs.enable_multiple), | |
130 | ever_enabled_multiple(rhs.ever_enabled_multiple), | |
131 | mds_roles(rhs.mds_roles), | |
132 | standby_daemons(rhs.standby_daemons), | |
133 | standby_epochs(rhs.standby_epochs) | |
134 | { | |
135 | for (const auto &i : rhs.filesystems) { | |
136 | const auto &fs = i.second; | |
137 | filesystems[fs->fscid] = std::make_shared<Filesystem>(*fs); | |
138 | } | |
139 | } | |
140 | ||
141 | FSMap &operator=(const FSMap &rhs) | |
142 | { | |
143 | epoch = rhs.epoch; | |
144 | next_filesystem_id = rhs.next_filesystem_id; | |
145 | legacy_client_fscid = rhs.legacy_client_fscid; | |
146 | compat = rhs.compat; | |
147 | enable_multiple = rhs.enable_multiple; | |
148 | mds_roles = rhs.mds_roles; | |
149 | standby_daemons = rhs.standby_daemons; | |
150 | standby_epochs = rhs.standby_epochs; | |
151 | ||
152 | for (const auto &i : rhs.filesystems) { | |
153 | const auto &fs = i.second; | |
154 | filesystems[fs->fscid] = std::make_shared<Filesystem>(*fs); | |
155 | } | |
156 | ||
157 | return *this; | |
158 | } | |
159 | ||
160 | const CompatSet &get_compat() const {return compat;} | |
161 | ||
162 | void set_enable_multiple(const bool v) | |
163 | { | |
164 | enable_multiple = v; | |
165 | if (true == v) { | |
166 | ever_enabled_multiple = true; | |
167 | } | |
168 | } | |
169 | ||
170 | bool get_enable_multiple() const | |
171 | { | |
172 | return enable_multiple; | |
173 | } | |
174 | ||
175 | void set_legacy_client_fscid(fs_cluster_id_t fscid) | |
176 | { | |
177 | assert(fscid == FS_CLUSTER_ID_NONE || filesystems.count(fscid)); | |
178 | legacy_client_fscid = fscid; | |
179 | } | |
180 | ||
181 | fs_cluster_id_t get_legacy_client_fscid() const | |
182 | { | |
183 | return legacy_client_fscid; | |
184 | } | |
185 | ||
186 | /** | |
187 | * Get state of all daemons (for all filesystems, including all standbys) | |
188 | */ | |
189 | std::map<mds_gid_t, MDSMap::mds_info_t> get_mds_info() const | |
190 | { | |
191 | std::map<mds_gid_t, MDSMap::mds_info_t> result; | |
192 | for (const auto &i : standby_daemons) { | |
193 | result[i.first] = i.second; | |
194 | } | |
195 | ||
196 | for (const auto &i : filesystems) { | |
197 | const auto &fs_info = i.second->mds_map.get_mds_info(); | |
198 | for (const auto &j : fs_info) { | |
199 | result[j.first] = j.second; | |
200 | } | |
201 | } | |
202 | ||
203 | return result; | |
204 | } | |
205 | ||
206 | /** | |
207 | * Resolve daemon name to GID | |
208 | */ | |
209 | mds_gid_t find_mds_gid_by_name(const std::string& s) const | |
210 | { | |
211 | const auto info = get_mds_info(); | |
212 | for (const auto &p : info) { | |
213 | if (p.second.name == s) { | |
214 | return p.first; | |
215 | } | |
216 | } | |
217 | return MDS_GID_NONE; | |
218 | } | |
219 | ||
220 | /** | |
221 | * Resolve daemon name to status | |
222 | */ | |
223 | const MDSMap::mds_info_t* find_by_name(const std::string& name) const | |
224 | { | |
225 | std::map<mds_gid_t, MDSMap::mds_info_t> result; | |
226 | for (const auto &i : standby_daemons) { | |
227 | if (i.second.name == name) { | |
228 | return &(i.second); | |
229 | } | |
230 | } | |
231 | ||
232 | for (const auto &i : filesystems) { | |
233 | const auto &fs_info = i.second->mds_map.get_mds_info(); | |
234 | for (const auto &j : fs_info) { | |
235 | if (j.second.name == name) { | |
236 | return &(j.second); | |
237 | } | |
238 | } | |
239 | } | |
240 | ||
241 | return nullptr; | |
242 | } | |
243 | ||
244 | /** | |
245 | * Does a daemon exist with this GID? | |
246 | */ | |
247 | bool gid_exists(mds_gid_t gid) const | |
248 | { | |
249 | return mds_roles.count(gid) > 0; | |
250 | } | |
251 | ||
252 | /** | |
253 | * Does a daemon with this GID exist, *and* have an MDS rank assigned? | |
254 | */ | |
255 | bool gid_has_rank(mds_gid_t gid) const | |
256 | { | |
257 | return gid_exists(gid) && mds_roles.at(gid) != FS_CLUSTER_ID_NONE; | |
258 | } | |
259 | ||
260 | /** | |
261 | * Insert a new MDS daemon, as a standby | |
262 | */ | |
263 | void insert(const MDSMap::mds_info_t &new_info); | |
264 | ||
265 | /** | |
266 | * Assign an MDS cluster standby replay rank to a standby daemon | |
267 | */ | |
268 | void assign_standby_replay( | |
269 | const mds_gid_t standby_gid, | |
270 | const fs_cluster_id_t leader_ns, | |
271 | const mds_rank_t leader_rank); | |
272 | ||
273 | /** | |
274 | * Assign an MDS cluster rank to a standby daemon | |
275 | */ | |
276 | void promote( | |
277 | mds_gid_t standby_gid, | |
278 | const std::shared_ptr<Filesystem> &filesystem, | |
279 | mds_rank_t assigned_rank); | |
280 | ||
281 | /** | |
282 | * A daemon reports that it is STATE_STOPPED: remove it, | |
283 | * and the rank it held. | |
284 | * | |
285 | * @returns a list of any additional GIDs that were removed from the map | |
286 | * as a side effect (like standby replays) | |
287 | */ | |
288 | std::list<mds_gid_t> stop(mds_gid_t who); | |
289 | ||
290 | /** | |
291 | * The rank held by 'who', if any, is to be relinquished, and | |
292 | * the state for the daemon GID is to be forgotten. | |
293 | */ | |
294 | void erase(mds_gid_t who, epoch_t blacklist_epoch); | |
295 | ||
296 | /** | |
297 | * Update to indicate that the rank held by 'who' is damaged | |
298 | */ | |
299 | void damaged(mds_gid_t who, epoch_t blacklist_epoch); | |
300 | ||
301 | /** | |
302 | * Update to indicate that the rank `rank` is to be removed | |
303 | * from the damaged list of the filesystem `fscid` | |
304 | */ | |
305 | bool undamaged(const fs_cluster_id_t fscid, const mds_rank_t rank); | |
306 | ||
307 | /** | |
308 | * Initialize a Filesystem and assign a fscid. Update legacy_client_fscid | |
309 | * to point to the new filesystem if it's the only one. | |
310 | * | |
311 | * Caller must already have validated all arguments vs. the existing | |
312 | * FSMap and OSDMap contents. | |
313 | */ | |
314 | void create_filesystem(const std::string &name, | |
315 | int64_t metadata_pool, int64_t data_pool, | |
316 | uint64_t features); | |
317 | ||
318 | /** | |
319 | * Remove the filesystem (it must exist). Caller should already | |
320 | * have failed out any MDSs that were assigned to the filesystem. | |
321 | */ | |
322 | void erase_filesystem(fs_cluster_id_t fscid) | |
323 | { | |
324 | filesystems.erase(fscid); | |
325 | } | |
326 | ||
327 | /** | |
328 | * Reset all the state information (not configuration information) | |
329 | * in a particular filesystem. Caller must have verified that | |
330 | * the filesystem already exists. | |
331 | */ | |
332 | void reset_filesystem(fs_cluster_id_t fscid); | |
333 | ||
334 | /** | |
335 | * Mutator helper for Filesystem objects: expose a non-const | |
336 | * Filesystem pointer to `fn` and update epochs appropriately. | |
337 | */ | |
338 | void modify_filesystem( | |
339 | const fs_cluster_id_t fscid, | |
340 | std::function<void(std::shared_ptr<Filesystem> )> fn) | |
341 | { | |
342 | auto fs = filesystems.at(fscid); | |
343 | fn(fs); | |
344 | fs->mds_map.epoch = epoch; | |
345 | } | |
346 | ||
347 | /** | |
348 | * Apply a mutation to the mds_info_t structure for a particular | |
349 | * daemon (identified by GID), and make appropriate updates to epochs. | |
350 | */ | |
351 | void modify_daemon( | |
352 | mds_gid_t who, | |
353 | std::function<void(MDSMap::mds_info_t *info)> fn) | |
354 | { | |
355 | if (mds_roles.at(who) == FS_CLUSTER_ID_NONE) { | |
356 | auto &info = standby_daemons.at(who); | |
357 | fn(&info); | |
358 | assert(info.state == MDSMap::STATE_STANDBY); | |
359 | standby_epochs[who] = epoch; | |
360 | } else { | |
361 | const auto &fs = filesystems[mds_roles.at(who)]; | |
362 | auto &info = fs->mds_map.mds_info.at(who); | |
363 | fn(&info); | |
364 | ||
365 | fs->mds_map.epoch = epoch; | |
366 | } | |
367 | } | |
368 | ||
369 | /** | |
370 | * Given that gid exists in a filesystem or as a standby, return | |
371 | * a reference to its info. | |
372 | */ | |
373 | const MDSMap::mds_info_t& get_info_gid(mds_gid_t gid) const | |
374 | { | |
375 | auto fscid = mds_roles.at(gid); | |
376 | if (fscid == FS_CLUSTER_ID_NONE) { | |
377 | return standby_daemons.at(gid); | |
378 | } else { | |
379 | return filesystems.at(fscid)->mds_map.mds_info.at(gid); | |
380 | } | |
381 | } | |
382 | ||
383 | /** | |
384 | * A daemon has told us it's compat, and it's too new | |
385 | * for the one we had previously. Impose the new one | |
386 | * on all filesystems. | |
387 | */ | |
388 | void update_compat(const CompatSet &c) | |
389 | { | |
390 | // We could do something more complicated here to enable | |
391 | // different filesystems to be served by different MDS versions, | |
392 | // but this is a lot simpler because it doesn't require us to | |
393 | // track the compat versions for standby daemons. | |
394 | compat = c; | |
395 | for (const auto &i : filesystems) { | |
396 | MDSMap &mds_map = i.second->mds_map; | |
397 | mds_map.compat = c; | |
398 | mds_map.epoch = epoch; | |
399 | } | |
400 | } | |
401 | ||
402 | std::shared_ptr<const Filesystem> get_legacy_filesystem() | |
403 | { | |
404 | if (legacy_client_fscid == FS_CLUSTER_ID_NONE) { | |
405 | return nullptr; | |
406 | } else { | |
407 | return filesystems.at(legacy_client_fscid); | |
408 | } | |
409 | } | |
410 | ||
411 | /** | |
412 | * A daemon has informed us of its offload targets | |
413 | */ | |
414 | void update_export_targets(mds_gid_t who, const std::set<mds_rank_t> targets) | |
415 | { | |
416 | auto fscid = mds_roles.at(who); | |
417 | modify_filesystem(fscid, [who, &targets](std::shared_ptr<Filesystem> fs) { | |
418 | fs->mds_map.mds_info.at(who).export_targets = targets; | |
419 | }); | |
420 | } | |
421 | ||
422 | epoch_t get_epoch() const { return epoch; } | |
423 | void inc_epoch() { epoch++; } | |
424 | ||
425 | size_t filesystem_count() const {return filesystems.size();} | |
426 | bool filesystem_exists(fs_cluster_id_t fscid) const {return filesystems.count(fscid) > 0;} | |
427 | std::shared_ptr<const Filesystem> get_filesystem(fs_cluster_id_t fscid) const {return std::const_pointer_cast<const Filesystem>(filesystems.at(fscid));} | |
428 | std::shared_ptr<const Filesystem> get_filesystem(void) const {return std::const_pointer_cast<const Filesystem>(filesystems.begin()->second);} | |
429 | std::shared_ptr<const Filesystem> get_filesystem(const std::string &name) const | |
430 | { | |
431 | for (const auto &i : filesystems) { | |
432 | if (i.second->mds_map.fs_name == name) { | |
433 | return std::const_pointer_cast<const Filesystem>(i.second); | |
434 | } | |
435 | } | |
436 | return nullptr; | |
437 | } | |
438 | std::list<std::shared_ptr<const Filesystem> > get_filesystems(void) const | |
439 | { | |
440 | std::list<std::shared_ptr<const Filesystem> > ret; | |
441 | for (const auto &i : filesystems) { | |
442 | ret.push_back(std::const_pointer_cast<const Filesystem>(i.second)); | |
443 | } | |
444 | return ret; | |
445 | } | |
446 | ||
447 | int parse_filesystem( | |
448 | std::string const &ns_str, | |
449 | std::shared_ptr<const Filesystem> *result | |
450 | ) const; | |
451 | ||
452 | int parse_role( | |
453 | const std::string &role_str, | |
454 | mds_role_t *role, | |
455 | std::ostream &ss) const; | |
456 | ||
457 | /** | |
458 | * Return true if this pool is in use by any of the filesystems | |
459 | */ | |
460 | bool pool_in_use(int64_t poolid) const { | |
461 | for (auto const &i : filesystems) { | |
462 | if (i.second->mds_map.is_data_pool(poolid) | |
463 | || i.second->mds_map.metadata_pool == poolid) { | |
464 | return true; | |
465 | } | |
466 | } | |
467 | return false; | |
468 | } | |
469 | ||
470 | mds_gid_t find_standby_for(mds_role_t mds, const std::string& name) const; | |
471 | ||
31f18b77 | 472 | mds_gid_t find_unused_for(mds_role_t mds, bool force_standby_active) const; |
7c673cae FG |
473 | |
474 | mds_gid_t find_replacement_for(mds_role_t mds, const std::string& name, | |
475 | bool force_standby_active) const; | |
476 | ||
477 | void get_health(list<pair<health_status_t,std::string> >& summary, | |
478 | list<pair<health_status_t,std::string> > *detail) const; | |
479 | ||
224ce89b WB |
480 | void get_health_checks(health_check_map_t *checks) const; |
481 | ||
7c673cae FG |
482 | bool check_health(void); |
483 | ||
484 | /** | |
485 | * Assert that the FSMap, Filesystem, MDSMap, mds_info_t relations are | |
486 | * all self-consistent. | |
487 | */ | |
488 | void sanity() const; | |
489 | ||
490 | void encode(bufferlist& bl, uint64_t features) const; | |
491 | void decode(bufferlist::iterator& p); | |
492 | void decode(bufferlist& bl) { | |
493 | bufferlist::iterator p = bl.begin(); | |
494 | decode(p); | |
495 | } | |
496 | ||
497 | void print(ostream& out) const; | |
498 | void print_summary(Formatter *f, ostream *out) const; | |
499 | ||
500 | void dump(Formatter *f) const; | |
501 | static void generate_test_instances(list<FSMap*>& ls); | |
502 | }; | |
503 | WRITE_CLASS_ENCODER_FEATURES(FSMap) | |
504 | ||
505 | inline ostream& operator<<(ostream& out, const FSMap& m) { | |
506 | m.print_summary(NULL, &out); | |
507 | return out; | |
508 | } | |
509 | ||
510 | #endif |