]>
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 | * Copyright (C) 2013,2014 Cloudwatt <libre.licensing@cloudwatt.com> | |
8 | * | |
9 | * Author: Loic Dachary <loic@dachary.org> | |
10 | * | |
11 | * This is free software; you can redistribute it and/or | |
12 | * modify it under the terms of the GNU Lesser General Public | |
13 | * License version 2.1, as published by the Free Software | |
14 | * Foundation. See file COPYING. | |
15 | * | |
16 | */ | |
17 | ||
18 | /* Object Store Device (OSD) Monitor | |
19 | */ | |
20 | ||
21 | #ifndef CEPH_OSDMONITOR_H | |
22 | #define CEPH_OSDMONITOR_H | |
23 | ||
24 | #include <map> | |
25 | #include <set> | |
f67539c2 | 26 | #include <utility> |
7c673cae FG |
27 | |
28 | #include "include/types.h" | |
11fdf7f2 | 29 | #include "include/encoding.h" |
7c673cae | 30 | #include "common/simple_cache.hpp" |
eafe8130 | 31 | #include "common/PriorityCache.h" |
7c673cae FG |
32 | #include "msg/Messenger.h" |
33 | ||
34 | #include "osd/OSDMap.h" | |
35 | #include "osd/OSDMapMapping.h" | |
36 | ||
37 | #include "CreatingPGs.h" | |
38 | #include "PaxosService.h" | |
39 | ||
f67539c2 TL |
40 | #include "erasure-code/ErasureCodeInterface.h" |
41 | #include "mon/MonOpRequest.h" | |
42 | #include <boost/functional/hash.hpp> | |
43 | ||
7c673cae FG |
44 | class Monitor; |
45 | class PGMap; | |
f67539c2 | 46 | struct MonSession; |
7c673cae FG |
47 | class MOSDMap; |
48 | ||
7c673cae | 49 | |
7c673cae FG |
50 | /// information about a particular peer's failure reports for one osd |
51 | struct failure_reporter_t { | |
52 | utime_t failed_since; ///< when they think it failed | |
53 | MonOpRequestRef op; ///< failure op request | |
54 | ||
55 | failure_reporter_t() {} | |
b3b6e05e TL |
56 | failure_reporter_t(utime_t s, MonOpRequestRef op) |
57 | : failed_since(s), op(op) {} | |
7c673cae FG |
58 | ~failure_reporter_t() { } |
59 | }; | |
60 | ||
61 | /// information about all failure reports for one osd | |
62 | struct failure_info_t { | |
f67539c2 | 63 | std::map<int, failure_reporter_t> reporters; ///< reporter -> failed_since etc |
7c673cae FG |
64 | utime_t max_failed_since; ///< most recent failed_since |
65 | ||
66 | failure_info_t() {} | |
67 | ||
68 | utime_t get_failed_since() { | |
69 | if (max_failed_since == utime_t() && !reporters.empty()) { | |
70 | // the old max must have canceled; recalculate. | |
f67539c2 | 71 | for (auto p = reporters.begin(); p != reporters.end(); ++p) |
7c673cae FG |
72 | if (p->second.failed_since > max_failed_since) |
73 | max_failed_since = p->second.failed_since; | |
74 | } | |
75 | return max_failed_since; | |
76 | } | |
77 | ||
b3b6e05e TL |
78 | // set the message for the latest report. |
79 | void add_report(int who, utime_t failed_since, MonOpRequestRef op) { | |
80 | [[maybe_unused]] auto [it, new_reporter] = | |
81 | reporters.insert_or_assign(who, failure_reporter_t{failed_since, op}); | |
82 | if (new_reporter) { | |
83 | if (max_failed_since != utime_t() && max_failed_since < failed_since) { | |
7c673cae | 84 | max_failed_since = failed_since; |
b3b6e05e | 85 | } |
7c673cae | 86 | } |
7c673cae FG |
87 | } |
88 | ||
f67539c2 TL |
89 | void take_report_messages(std::list<MonOpRequestRef>& ls) { |
90 | for (auto p = reporters.begin(); p != reporters.end(); ++p) { | |
7c673cae FG |
91 | if (p->second.op) { |
92 | ls.push_back(p->second.op); | |
93 | p->second.op.reset(); | |
94 | } | |
95 | } | |
96 | } | |
97 | ||
b3b6e05e TL |
98 | void cancel_report(int who) { |
99 | reporters.erase(who); | |
91327a77 | 100 | max_failed_since = utime_t(); |
7c673cae FG |
101 | } |
102 | }; | |
103 | ||
104 | ||
105 | class LastEpochClean { | |
106 | struct Lec { | |
f67539c2 | 107 | std::vector<epoch_t> epoch_by_pg; |
7c673cae FG |
108 | ps_t next_missing = 0; |
109 | epoch_t floor = std::numeric_limits<epoch_t>::max(); | |
522d829b | 110 | void report(unsigned pg_num, ps_t pg, epoch_t last_epoch_clean); |
7c673cae FG |
111 | }; |
112 | std::map<uint64_t, Lec> report_by_pool; | |
113 | public: | |
522d829b | 114 | void report(unsigned pg_num, const pg_t& pg, epoch_t last_epoch_clean); |
7c673cae FG |
115 | void remove_pool(uint64_t pool); |
116 | epoch_t get_lower_bound(const OSDMap& latest) const; | |
1911f103 TL |
117 | |
118 | void dump(Formatter *f) const; | |
7c673cae FG |
119 | }; |
120 | ||
121 | ||
11fdf7f2 TL |
122 | struct osdmap_manifest_t { |
123 | // all the maps we have pinned -- i.e., won't be removed unless | |
124 | // they are inside a trim interval. | |
f67539c2 | 125 | std::set<version_t> pinned; |
11fdf7f2 TL |
126 | |
127 | osdmap_manifest_t() {} | |
128 | ||
129 | version_t get_last_pinned() const | |
130 | { | |
f67539c2 | 131 | auto it = pinned.crbegin(); |
11fdf7f2 TL |
132 | if (it == pinned.crend()) { |
133 | return 0; | |
134 | } | |
135 | return *it; | |
136 | } | |
137 | ||
138 | version_t get_first_pinned() const | |
139 | { | |
f67539c2 | 140 | auto it = pinned.cbegin(); |
11fdf7f2 TL |
141 | if (it == pinned.cend()) { |
142 | return 0; | |
143 | } | |
144 | return *it; | |
145 | } | |
146 | ||
147 | bool is_pinned(version_t v) const | |
148 | { | |
149 | return pinned.find(v) != pinned.end(); | |
150 | } | |
151 | ||
152 | void pin(version_t v) | |
153 | { | |
154 | pinned.insert(v); | |
155 | } | |
156 | ||
157 | version_t get_lower_closest_pinned(version_t v) const { | |
f67539c2 | 158 | auto p = pinned.lower_bound(v); |
11fdf7f2 TL |
159 | if (p == pinned.cend()) { |
160 | return 0; | |
161 | } else if (*p > v) { | |
162 | if (p == pinned.cbegin()) { | |
163 | return 0; | |
164 | } | |
165 | --p; | |
166 | } | |
167 | return *p; | |
168 | } | |
169 | ||
f67539c2 | 170 | void encode(ceph::buffer::list& bl) const |
11fdf7f2 TL |
171 | { |
172 | ENCODE_START(1, 1, bl); | |
173 | encode(pinned, bl); | |
174 | ENCODE_FINISH(bl); | |
175 | } | |
176 | ||
f67539c2 | 177 | void decode(ceph::buffer::list::const_iterator& bl) |
11fdf7f2 TL |
178 | { |
179 | DECODE_START(1, bl); | |
180 | decode(pinned, bl); | |
181 | DECODE_FINISH(bl); | |
182 | } | |
183 | ||
f67539c2 | 184 | void decode(ceph::buffer::list& bl) { |
11fdf7f2 TL |
185 | auto p = bl.cbegin(); |
186 | decode(p); | |
187 | } | |
188 | ||
f67539c2 | 189 | void dump(ceph::Formatter *f) { |
11fdf7f2 TL |
190 | f->dump_unsigned("first_pinned", get_first_pinned()); |
191 | f->dump_unsigned("last_pinned", get_last_pinned()); | |
192 | f->open_array_section("pinned_maps"); | |
193 | for (auto& i : pinned) { | |
194 | f->dump_unsigned("epoch", i); | |
195 | } | |
196 | f->close_section(); | |
197 | } | |
198 | }; | |
199 | WRITE_CLASS_ENCODER(osdmap_manifest_t); | |
200 | ||
eafe8130 TL |
201 | class OSDMonitor : public PaxosService, |
202 | public md_config_obs_t { | |
7c673cae FG |
203 | CephContext *cct; |
204 | ||
205 | public: | |
206 | OSDMap osdmap; | |
207 | ||
eafe8130 TL |
208 | // config observer |
209 | const char** get_tracked_conf_keys() const override; | |
210 | void handle_conf_change(const ConfigProxy& conf, | |
211 | const std::set<std::string> &changed) override; | |
7c673cae FG |
212 | // [leader] |
213 | OSDMap::Incremental pending_inc; | |
f67539c2 TL |
214 | std::map<int, ceph::buffer::list> pending_metadata; |
215 | std::set<int> pending_metadata_rm; | |
216 | std::map<int, failure_info_t> failure_info; | |
217 | std::map<int,utime_t> down_pending_out; // osd down -> out | |
81eedcae | 218 | bool priority_convert = false; |
f67539c2 | 219 | std::map<int64_t,std::set<snapid_t>> pending_pseudo_purged_snaps; |
eafe8130 TL |
220 | std::shared_ptr<PriorityCache::PriCache> rocksdb_binned_kv_cache = nullptr; |
221 | std::shared_ptr<PriorityCache::Manager> pcm = nullptr; | |
222 | ceph::mutex balancer_lock = ceph::make_mutex("OSDMonitor::balancer_lock"); | |
7c673cae | 223 | |
f67539c2 | 224 | std::map<int,double> osd_weight; |
7c673cae | 225 | |
28e407b8 AA |
226 | using osdmap_key_t = std::pair<version_t, uint64_t>; |
227 | using osdmap_cache_t = SimpleLRU<osdmap_key_t, | |
f67539c2 | 228 | ceph::buffer::list, |
28e407b8 AA |
229 | std::less<osdmap_key_t>, |
230 | boost::hash<osdmap_key_t>>; | |
231 | osdmap_cache_t inc_osd_cache; | |
232 | osdmap_cache_t full_osd_cache; | |
7c673cae | 233 | |
11fdf7f2 TL |
234 | bool has_osdmap_manifest; |
235 | osdmap_manifest_t osdmap_manifest; | |
236 | ||
7c673cae FG |
237 | bool check_failures(utime_t now); |
238 | bool check_failure(utime_t now, int target_osd, failure_info_t& fi); | |
b3b6e05e TL |
239 | utime_t get_grace_time(utime_t now, int target_osd, failure_info_t& fi) const; |
240 | bool is_failure_stale(utime_t now, failure_info_t& fi) const; | |
224ce89b | 241 | void force_failure(int target_osd, int by); |
7c673cae | 242 | |
7c673cae FG |
243 | bool _have_pending_crush(); |
244 | CrushWrapper &_get_stable_crush(); | |
245 | void _get_pending_crush(CrushWrapper& newcrush); | |
246 | ||
247 | enum FastReadType { | |
248 | FAST_READ_OFF, | |
249 | FAST_READ_ON, | |
250 | FAST_READ_DEFAULT | |
251 | }; | |
252 | ||
494da23a TL |
253 | struct CleanUpmapJob : public ParallelPGMapper::Job { |
254 | CephContext *cct; | |
255 | const OSDMap& osdmap; | |
256 | OSDMap::Incremental& pending_inc; | |
257 | // lock to protect pending_inc form changing | |
258 | // when checking is done | |
9f95a23c TL |
259 | ceph::mutex pending_inc_lock = |
260 | ceph::make_mutex("CleanUpmapJob::pending_inc_lock"); | |
494da23a TL |
261 | |
262 | CleanUpmapJob(CephContext *cct, const OSDMap& om, OSDMap::Incremental& pi) | |
263 | : ParallelPGMapper::Job(&om), | |
264 | cct(cct), | |
265 | osdmap(om), | |
266 | pending_inc(pi) {} | |
267 | ||
f67539c2 TL |
268 | void process(const std::vector<pg_t>& to_check) override { |
269 | std::vector<pg_t> to_cancel; | |
270 | std::map<pg_t, mempool::osdmap::vector<std::pair<int,int>>> to_remap; | |
494da23a TL |
271 | osdmap.check_pg_upmaps(cct, to_check, &to_cancel, &to_remap); |
272 | // don't bother taking lock if nothing changes | |
273 | if (!to_cancel.empty() || !to_remap.empty()) { | |
274 | std::lock_guard l(pending_inc_lock); | |
275 | osdmap.clean_pg_upmaps(cct, &pending_inc, to_cancel, to_remap); | |
276 | } | |
277 | } | |
278 | ||
279 | void process(int64_t poolid, unsigned ps_begin, unsigned ps_end) override {} | |
280 | void complete() override {} | |
281 | }; // public as this will need to be accessible from TestTestOSDMap.cc | |
282 | ||
7c673cae | 283 | // svc |
11fdf7f2 | 284 | public: |
7c673cae | 285 | void create_initial() override; |
f67539c2 | 286 | void get_store_prefixes(std::set<std::string>& s) const override; |
7c673cae FG |
287 | |
288 | private: | |
289 | void update_from_paxos(bool *need_bootstrap) override; | |
290 | void create_pending() override; // prepare a new pending | |
291 | void encode_pending(MonitorDBStore::TransactionRef t) override; | |
292 | void on_active() override; | |
293 | void on_restart() override; | |
294 | void on_shutdown() override; | |
11fdf7f2 TL |
295 | |
296 | /* osdmap full map prune */ | |
297 | void load_osdmap_manifest(); | |
298 | bool should_prune() const; | |
299 | void _prune_update_trimmed( | |
300 | MonitorDBStore::TransactionRef tx, | |
301 | version_t first); | |
302 | void prune_init(osdmap_manifest_t& manifest); | |
303 | bool _prune_sanitize_options() const; | |
304 | bool is_prune_enabled() const; | |
305 | bool is_prune_supported() const; | |
306 | bool do_prune(MonitorDBStore::TransactionRef tx); | |
307 | ||
eafe8130 TL |
308 | // Priority cache control |
309 | uint32_t mon_osd_cache_size = 0; ///< Number of cached OSDMaps | |
310 | uint64_t rocksdb_cache_size = 0; ///< Cache for kv Db | |
311 | double cache_kv_ratio = 0; ///< Cache ratio dedicated to kv | |
312 | double cache_inc_ratio = 0; ///< Cache ratio dedicated to inc | |
313 | double cache_full_ratio = 0; ///< Cache ratio dedicated to full | |
314 | uint64_t mon_memory_base = 0; ///< Mon base memory for cache autotuning | |
315 | double mon_memory_fragmentation = 0; ///< Expected memory fragmentation | |
316 | uint64_t mon_memory_target = 0; ///< Mon target memory for cache autotuning | |
317 | uint64_t mon_memory_min = 0; ///< Min memory to cache osdmaps | |
318 | bool mon_memory_autotune = false; ///< Cache auto tune setting | |
319 | int register_cache_with_pcm(); | |
320 | int _set_cache_sizes(); | |
321 | int _set_cache_ratios(); | |
322 | void _set_new_cache_sizes(); | |
323 | void _set_cache_autotuning(); | |
324 | int _update_mon_cache_settings(); | |
325 | ||
326 | friend struct OSDMemCache; | |
327 | friend struct IncCache; | |
328 | friend struct FullCache; | |
329 | ||
7c673cae FG |
330 | /** |
331 | * we haven't delegated full version stashing to paxosservice for some time | |
332 | * now, making this function useless in current context. | |
333 | */ | |
334 | void encode_full(MonitorDBStore::TransactionRef t) override { } | |
335 | /** | |
336 | * do not let paxosservice periodically stash full osdmaps, or we will break our | |
337 | * locally-managed full maps. (update_from_paxos loads the latest and writes them | |
338 | * out going forward from there, but if we just synced that may mean we skip some.) | |
339 | */ | |
340 | bool should_stash_full() override { | |
341 | return false; | |
342 | } | |
343 | ||
344 | /** | |
345 | * hook into trim to include the oldest full map in the trim transaction | |
346 | * | |
347 | * This ensures that anyone post-sync will have enough to rebuild their | |
348 | * full osdmaps. | |
349 | */ | |
350 | void encode_trim_extra(MonitorDBStore::TransactionRef tx, version_t first) override; | |
351 | ||
352 | void update_msgr_features(); | |
7c673cae FG |
353 | /** |
354 | * check if the cluster supports the features required by the | |
355 | * given crush map. Outputs the daemons which don't support it | |
356 | * to the stringstream. | |
357 | * | |
358 | * @returns true if the map is passable, false otherwise | |
359 | */ | |
360 | bool validate_crush_against_features(const CrushWrapper *newcrush, | |
f67539c2 | 361 | std::stringstream &ss); |
7c673cae FG |
362 | void check_osdmap_subs(); |
363 | void share_map_with_random_osd(); | |
364 | ||
9f95a23c TL |
365 | ceph::mutex prime_pg_temp_lock = |
366 | ceph::make_mutex("OSDMonitor::prime_pg_temp_lock"); | |
7c673cae FG |
367 | struct PrimeTempJob : public ParallelPGMapper::Job { |
368 | OSDMonitor *osdmon; | |
369 | PrimeTempJob(const OSDMap& om, OSDMonitor *m) | |
370 | : ParallelPGMapper::Job(&om), osdmon(m) {} | |
371 | void process(int64_t pool, unsigned ps_begin, unsigned ps_end) override { | |
372 | for (unsigned ps = ps_begin; ps < ps_end; ++ps) { | |
373 | pg_t pgid(ps, pool); | |
374 | osdmon->prime_pg_temp(*osdmap, pgid); | |
375 | } | |
376 | } | |
f67539c2 | 377 | void process(const std::vector<pg_t>& pgs) override {} |
7c673cae FG |
378 | void complete() override {} |
379 | }; | |
380 | void maybe_prime_pg_temp(); | |
381 | void prime_pg_temp(const OSDMap& next, pg_t pgid); | |
382 | ||
383 | ParallelPGMapper mapper; ///< for background pg work | |
384 | OSDMapMapping mapping; ///< pg <-> osd mappings | |
f67539c2 | 385 | std::unique_ptr<ParallelPGMapper::Job> mapping_job; ///< background mapping job |
7c673cae FG |
386 | void start_mapping(); |
387 | ||
388 | void update_logger(); | |
389 | ||
390 | void handle_query(PaxosServiceMessage *m); | |
391 | bool preprocess_query(MonOpRequestRef op) override; // true if processed. | |
392 | bool prepare_update(MonOpRequestRef op) override; | |
393 | bool should_propose(double &delay) override; | |
394 | ||
11fdf7f2 | 395 | version_t get_trim_to() const override; |
7c673cae FG |
396 | |
397 | bool can_mark_down(int o); | |
398 | bool can_mark_up(int o); | |
399 | bool can_mark_out(int o); | |
400 | bool can_mark_in(int o); | |
401 | ||
402 | // ... | |
28e407b8 AA |
403 | MOSDMap *build_latest_full(uint64_t features); |
404 | MOSDMap *build_incremental(epoch_t first, epoch_t last, uint64_t features); | |
7c673cae FG |
405 | void send_full(MonOpRequestRef op); |
406 | void send_incremental(MonOpRequestRef op, epoch_t first); | |
407 | public: | |
f67539c2 TL |
408 | /** |
409 | * Make sure the existing (up) OSDs support the given features | |
410 | * @return 0 on success, or an error code if any OSDs re missing features. | |
411 | * @param ss Filled in with ane explanation of failure, if any | |
412 | */ | |
413 | int check_cluster_features(uint64_t features, std::stringstream &ss); | |
7c673cae FG |
414 | // @param req an optional op request, if the osdmaps are replies to it. so |
415 | // @c Monitor::send_reply() can mark_event with it. | |
416 | void send_incremental(epoch_t first, MonSession *session, bool onetime, | |
417 | MonOpRequestRef req = MonOpRequestRef()); | |
418 | ||
419 | private: | |
f67539c2 | 420 | void print_utilization(std::ostream &out, ceph::Formatter *f, bool tree) const; |
7c673cae | 421 | |
11fdf7f2 | 422 | bool check_source(MonOpRequestRef op, uuid_d fsid); |
7c673cae FG |
423 | |
424 | bool preprocess_get_osdmap(MonOpRequestRef op); | |
425 | ||
426 | bool preprocess_mark_me_down(MonOpRequestRef op); | |
427 | ||
428 | friend class C_AckMarkedDown; | |
429 | bool preprocess_failure(MonOpRequestRef op); | |
430 | bool prepare_failure(MonOpRequestRef op); | |
431 | bool prepare_mark_me_down(MonOpRequestRef op); | |
432 | void process_failures(); | |
f67539c2 | 433 | void take_all_failures(std::list<MonOpRequestRef>& ls); |
7c673cae | 434 | |
9f95a23c TL |
435 | bool preprocess_mark_me_dead(MonOpRequestRef op); |
436 | bool prepare_mark_me_dead(MonOpRequestRef op); | |
437 | ||
7c673cae FG |
438 | bool preprocess_full(MonOpRequestRef op); |
439 | bool prepare_full(MonOpRequestRef op); | |
440 | ||
441 | bool preprocess_boot(MonOpRequestRef op); | |
442 | bool prepare_boot(MonOpRequestRef op); | |
443 | void _booted(MonOpRequestRef op, bool logit); | |
444 | ||
445 | void update_up_thru(int from, epoch_t up_thru); | |
446 | bool preprocess_alive(MonOpRequestRef op); | |
447 | bool prepare_alive(MonOpRequestRef op); | |
448 | void _reply_map(MonOpRequestRef op, epoch_t e); | |
449 | ||
450 | bool preprocess_pgtemp(MonOpRequestRef op); | |
451 | bool prepare_pgtemp(MonOpRequestRef op); | |
452 | ||
453 | bool preprocess_pg_created(MonOpRequestRef op); | |
454 | bool prepare_pg_created(MonOpRequestRef op); | |
455 | ||
11fdf7f2 TL |
456 | bool preprocess_pg_ready_to_merge(MonOpRequestRef op); |
457 | bool prepare_pg_ready_to_merge(MonOpRequestRef op); | |
458 | ||
f67539c2 | 459 | int _check_remove_pool(int64_t pool_id, const pg_pool_t &pool, std::ostream *ss); |
7c673cae FG |
460 | bool _check_become_tier( |
461 | int64_t tier_pool_id, const pg_pool_t *tier_pool, | |
462 | int64_t base_pool_id, const pg_pool_t *base_pool, | |
f67539c2 | 463 | int *err, std::ostream *ss) const; |
7c673cae FG |
464 | bool _check_remove_tier( |
465 | int64_t base_pool_id, const pg_pool_t *base_pool, const pg_pool_t *tier_pool, | |
f67539c2 | 466 | int *err, std::ostream *ss) const; |
7c673cae | 467 | |
f67539c2 TL |
468 | int _prepare_remove_pool(int64_t pool, std::ostream *ss, bool no_fake); |
469 | int _prepare_rename_pool(int64_t pool, std::string newname); | |
7c673cae | 470 | |
28e407b8 | 471 | bool enforce_pool_op_caps(MonOpRequestRef op); |
7c673cae FG |
472 | bool preprocess_pool_op (MonOpRequestRef op); |
473 | bool preprocess_pool_op_create (MonOpRequestRef op); | |
474 | bool prepare_pool_op (MonOpRequestRef op); | |
475 | bool prepare_pool_op_create (MonOpRequestRef op); | |
476 | bool prepare_pool_op_delete(MonOpRequestRef op); | |
f67539c2 TL |
477 | int crush_rename_bucket(const std::string& srcname, |
478 | const std::string& dstname, | |
479 | std::ostream *ss); | |
480 | void check_legacy_ec_plugin(const std::string& plugin, | |
481 | const std::string& profile) const; | |
482 | int normalize_profile(const std::string& profilename, | |
483 | ceph::ErasureCodeProfile &profile, | |
7c673cae | 484 | bool force, |
f67539c2 TL |
485 | std::ostream *ss); |
486 | int crush_rule_create_erasure(const std::string &name, | |
487 | const std::string &profile, | |
31f18b77 | 488 | int *rule, |
f67539c2 TL |
489 | std::ostream *ss); |
490 | int get_crush_rule(const std::string &rule_name, | |
491 | int *crush_rule, | |
492 | std::ostream *ss); | |
493 | int get_erasure_code(const std::string &erasure_code_profile, | |
494 | ceph::ErasureCodeInterfaceRef *erasure_code, | |
495 | std::ostream *ss) const; | |
31f18b77 | 496 | int prepare_pool_crush_rule(const unsigned pool_type, |
f67539c2 TL |
497 | const std::string &erasure_code_profile, |
498 | const std::string &rule_name, | |
499 | int *crush_rule, | |
500 | std::ostream *ss); | |
7c673cae FG |
501 | bool erasure_code_profile_in_use( |
502 | const mempool::osdmap::map<int64_t, pg_pool_t> &pools, | |
f67539c2 TL |
503 | const std::string &profile, |
504 | std::ostream *ss); | |
505 | int parse_erasure_code_profile(const std::vector<std::string> &erasure_code_profile, | |
506 | std::map<std::string,std::string> *erasure_code_profile_map, | |
507 | std::ostream *ss); | |
7c673cae | 508 | int prepare_pool_size(const unsigned pool_type, |
f67539c2 | 509 | const std::string &erasure_code_profile, |
11fdf7f2 | 510 | uint8_t repl_size, |
7c673cae | 511 | unsigned *size, unsigned *min_size, |
f67539c2 | 512 | std::ostream *ss); |
7c673cae | 513 | int prepare_pool_stripe_width(const unsigned pool_type, |
f67539c2 | 514 | const std::string &erasure_code_profile, |
7c673cae | 515 | unsigned *stripe_width, |
f67539c2 TL |
516 | std::ostream *ss); |
517 | int check_pg_num(int64_t pool, int pg_num, int size, std::ostream* ss); | |
518 | int prepare_new_pool(std::string& name, | |
31f18b77 | 519 | int crush_rule, |
f67539c2 | 520 | const std::string &crush_rule_name, |
7c673cae | 521 | unsigned pg_num, unsigned pgp_num, |
11fdf7f2 TL |
522 | unsigned pg_num_min, |
523 | uint64_t repl_size, | |
524 | const uint64_t target_size_bytes, | |
525 | const float target_size_ratio, | |
f67539c2 | 526 | const std::string &erasure_code_profile, |
7c673cae FG |
527 | const unsigned pool_type, |
528 | const uint64_t expected_num_objects, | |
529 | FastReadType fast_read, | |
f67539c2 TL |
530 | const std::string& pg_autoscale_mode, |
531 | std::ostream *ss); | |
7c673cae FG |
532 | int prepare_new_pool(MonOpRequestRef op); |
533 | ||
3efd9988 FG |
534 | void set_pool_flags(int64_t pool_id, uint64_t flags); |
535 | void clear_pool_flags(int64_t pool_id, uint64_t flags); | |
7c673cae | 536 | bool update_pools_status(); |
7c673cae | 537 | |
9f95a23c TL |
538 | bool _is_removed_snap(int64_t pool_id, snapid_t snapid); |
539 | bool _is_pending_removed_snap(int64_t pool_id, snapid_t snapid); | |
540 | ||
f67539c2 TL |
541 | std::string make_purged_snap_epoch_key(epoch_t epoch); |
542 | std::string make_purged_snap_key(int64_t pool, snapid_t snap); | |
543 | std::string make_purged_snap_key_value(int64_t pool, snapid_t snap, snapid_t num, | |
544 | epoch_t epoch, ceph::buffer::list *v); | |
9f95a23c | 545 | |
11fdf7f2 | 546 | bool try_prune_purged_snaps(); |
9f95a23c | 547 | int lookup_purged_snap(int64_t pool, snapid_t snap, |
11fdf7f2 TL |
548 | snapid_t *begin, snapid_t *end); |
549 | ||
9f95a23c TL |
550 | void insert_purged_snap_update( |
551 | int64_t pool, | |
552 | snapid_t start, snapid_t end, | |
553 | epoch_t epoch, | |
554 | MonitorDBStore::TransactionRef t); | |
555 | ||
7c673cae FG |
556 | bool prepare_set_flag(MonOpRequestRef op, int flag); |
557 | bool prepare_unset_flag(MonOpRequestRef op, int flag); | |
558 | ||
559 | void _pool_op_reply(MonOpRequestRef op, | |
f67539c2 | 560 | int ret, epoch_t epoch, ceph::buffer::list *blp=NULL); |
7c673cae FG |
561 | |
562 | struct C_Booted : public C_MonOp { | |
563 | OSDMonitor *cmon; | |
564 | bool logit; | |
565 | C_Booted(OSDMonitor *cm, MonOpRequestRef op_, bool l=true) : | |
566 | C_MonOp(op_), cmon(cm), logit(l) {} | |
567 | void _finish(int r) override { | |
568 | if (r >= 0) | |
569 | cmon->_booted(op, logit); | |
570 | else if (r == -ECANCELED) | |
571 | return; | |
572 | else if (r == -EAGAIN) | |
573 | cmon->dispatch(op); | |
574 | else | |
11fdf7f2 | 575 | ceph_abort_msg("bad C_Booted return value"); |
7c673cae FG |
576 | } |
577 | }; | |
578 | ||
579 | struct C_ReplyMap : public C_MonOp { | |
580 | OSDMonitor *osdmon; | |
581 | epoch_t e; | |
582 | C_ReplyMap(OSDMonitor *o, MonOpRequestRef op_, epoch_t ee) | |
583 | : C_MonOp(op_), osdmon(o), e(ee) {} | |
584 | void _finish(int r) override { | |
585 | if (r >= 0) | |
586 | osdmon->_reply_map(op, e); | |
587 | else if (r == -ECANCELED) | |
588 | return; | |
589 | else if (r == -EAGAIN) | |
590 | osdmon->dispatch(op); | |
591 | else | |
11fdf7f2 | 592 | ceph_abort_msg("bad C_ReplyMap return value"); |
7c673cae FG |
593 | } |
594 | }; | |
595 | struct C_PoolOp : public C_MonOp { | |
596 | OSDMonitor *osdmon; | |
597 | int replyCode; | |
598 | int epoch; | |
f67539c2 TL |
599 | ceph::buffer::list reply_data; |
600 | C_PoolOp(OSDMonitor * osd, MonOpRequestRef op_, int rc, int e, ceph::buffer::list *rd=NULL) : | |
7c673cae FG |
601 | C_MonOp(op_), osdmon(osd), replyCode(rc), epoch(e) { |
602 | if (rd) | |
603 | reply_data = *rd; | |
604 | } | |
605 | void _finish(int r) override { | |
606 | if (r >= 0) | |
607 | osdmon->_pool_op_reply(op, replyCode, epoch, &reply_data); | |
608 | else if (r == -ECANCELED) | |
609 | return; | |
610 | else if (r == -EAGAIN) | |
611 | osdmon->dispatch(op); | |
612 | else | |
11fdf7f2 | 613 | ceph_abort_msg("bad C_PoolOp return value"); |
7c673cae FG |
614 | } |
615 | }; | |
616 | ||
617 | bool preprocess_remove_snaps(MonOpRequestRef op); | |
618 | bool prepare_remove_snaps(MonOpRequestRef op); | |
619 | ||
9f95a23c TL |
620 | bool preprocess_get_purged_snaps(MonOpRequestRef op); |
621 | ||
f67539c2 TL |
622 | int load_metadata(int osd, std::map<std::string, std::string>& m, |
623 | std::ostream *err); | |
624 | void count_metadata(const std::string& field, ceph::Formatter *f); | |
28e407b8 | 625 | |
f67539c2 TL |
626 | void reencode_incremental_map(ceph::buffer::list& bl, uint64_t features); |
627 | void reencode_full_map(ceph::buffer::list& bl, uint64_t features); | |
c07f9fc5 | 628 | public: |
f67539c2 TL |
629 | void count_metadata(const std::string& field, std::map<std::string,int> *out); |
630 | void get_versions(std::map<std::string, std::list<std::string>> &versions); | |
c07f9fc5 | 631 | protected: |
7c673cae FG |
632 | int get_osd_objectstore_type(int osd, std::string *type); |
633 | bool is_pool_currently_all_bluestore(int64_t pool_id, const pg_pool_t &pool, | |
f67539c2 | 634 | std::ostream *err); |
7c673cae | 635 | |
f67539c2 TL |
636 | // when we last received PG stats from each osd and the osd's osd_beacon_report_interval |
637 | std::map<int, std::pair<utime_t, int>> last_osd_report; | |
7c673cae FG |
638 | // TODO: use last_osd_report to store the osd report epochs, once we don't |
639 | // need to upgrade from pre-luminous releases. | |
f67539c2 | 640 | std::map<int,epoch_t> osd_epochs; |
7c673cae FG |
641 | LastEpochClean last_epoch_clean; |
642 | bool preprocess_beacon(MonOpRequestRef op); | |
643 | bool prepare_beacon(MonOpRequestRef op); | |
644 | epoch_t get_min_last_epoch_clean() const; | |
645 | ||
646 | friend class C_UpdateCreatingPGs; | |
11fdf7f2 | 647 | std::map<int, std::map<epoch_t, std::set<spg_t>>> creating_pgs_by_osd_epoch; |
7c673cae FG |
648 | std::vector<pg_t> pending_created_pgs; |
649 | // the epoch when the pg mapping was calculated | |
650 | epoch_t creating_pgs_epoch = 0; | |
651 | creating_pgs_t creating_pgs; | |
c07f9fc5 | 652 | mutable std::mutex creating_pgs_lock; |
7c673cae | 653 | |
94b18763 FG |
654 | creating_pgs_t update_pending_pgs(const OSDMap::Incremental& inc, |
655 | const OSDMap& nextmap); | |
31f18b77 | 656 | unsigned scan_for_creating_pgs( |
7c673cae FG |
657 | const mempool::osdmap::map<int64_t,pg_pool_t>& pools, |
658 | const mempool::osdmap::set<int64_t>& removed_pools, | |
659 | utime_t modified, | |
660 | creating_pgs_t* creating_pgs) const; | |
f67539c2 | 661 | std::pair<int32_t, pg_t> get_parent_pg(pg_t pgid) const; |
7c673cae FG |
662 | void update_creating_pgs(); |
663 | void check_pg_creates_subs(); | |
c07f9fc5 | 664 | epoch_t send_pg_creates(int osd, Connection *con, epoch_t next) const; |
7c673cae | 665 | |
31f18b77 FG |
666 | int32_t _allocate_osd_id(int32_t* existing_id); |
667 | ||
f6b5b4d7 TL |
668 | int get_grace_interval_threshold(); |
669 | bool grace_interval_threshold_exceeded(int last_failed); | |
670 | void set_default_laggy_params(int target_osd); | |
671 | ||
7c673cae | 672 | public: |
f67539c2 | 673 | OSDMonitor(CephContext *cct, Monitor &mn, Paxos &p, const std::string& service_name); |
7c673cae FG |
674 | |
675 | void tick() override; // check state, take actions | |
676 | ||
7c673cae FG |
677 | bool preprocess_command(MonOpRequestRef op); |
678 | bool prepare_command(MonOpRequestRef op); | |
11fdf7f2 | 679 | bool prepare_command_impl(MonOpRequestRef op, const cmdmap_t& cmdmap); |
7c673cae | 680 | |
31f18b77 FG |
681 | int validate_osd_create( |
682 | const int32_t id, | |
683 | const uuid_d& uuid, | |
684 | const bool check_osd_exists, | |
685 | int32_t* existing_id, | |
f67539c2 | 686 | std::stringstream& ss); |
31f18b77 FG |
687 | int prepare_command_osd_create( |
688 | const int32_t id, | |
689 | const uuid_d& uuid, | |
690 | int32_t* existing_id, | |
f67539c2 | 691 | std::stringstream& ss); |
3a9019d9 | 692 | void do_osd_create(const int32_t id, const uuid_d& uuid, |
f67539c2 | 693 | const std::string& device_class, |
3a9019d9 | 694 | int32_t* new_id); |
f67539c2 TL |
695 | int prepare_command_osd_purge(int32_t id, std::stringstream& ss); |
696 | int prepare_command_osd_destroy(int32_t id, std::stringstream& ss); | |
31f18b77 FG |
697 | int _prepare_command_osd_crush_remove( |
698 | CrushWrapper &newcrush, | |
699 | int32_t id, | |
700 | int32_t ancestor, | |
701 | bool has_ancestor, | |
702 | bool unlink_only); | |
703 | void do_osd_crush_remove(CrushWrapper& newcrush); | |
704 | int prepare_command_osd_crush_remove( | |
705 | CrushWrapper &newcrush, | |
706 | int32_t id, | |
707 | int32_t ancestor, | |
708 | bool has_ancestor, | |
709 | bool unlink_only); | |
710 | int prepare_command_osd_remove(int32_t id); | |
711 | int prepare_command_osd_new( | |
712 | MonOpRequestRef op, | |
11fdf7f2 | 713 | const cmdmap_t& cmdmap, |
f67539c2 TL |
714 | const std::map<std::string,std::string>& secrets, |
715 | std::stringstream &ss, | |
716 | ceph::Formatter *f); | |
31f18b77 | 717 | |
11fdf7f2 | 718 | int prepare_command_pool_set(const cmdmap_t& cmdmap, |
f67539c2 | 719 | std::stringstream& ss); |
11fdf7f2 | 720 | |
f67539c2 | 721 | int prepare_command_pool_application(const std::string &prefix, |
11fdf7f2 | 722 | const cmdmap_t& cmdmap, |
f67539c2 TL |
723 | std::stringstream& ss); |
724 | int preprocess_command_pool_application(const std::string &prefix, | |
11fdf7f2 | 725 | const cmdmap_t& cmdmap, |
f67539c2 | 726 | std::stringstream& ss, |
11fdf7f2 | 727 | bool *modified); |
f67539c2 TL |
728 | int _command_pool_application(const std::string &prefix, |
729 | const cmdmap_t& cmdmap, | |
730 | std::stringstream& ss, | |
731 | bool *modified, | |
732 | bool preparing); | |
7c673cae FG |
733 | |
734 | bool handle_osd_timeouts(const utime_t &now, | |
f67539c2 | 735 | std::map<int, std::pair<utime_t, int>> &last_osd_report); |
7c673cae FG |
736 | |
737 | void send_latest(MonOpRequestRef op, epoch_t start=0); | |
738 | void send_latest_now_nodelete(MonOpRequestRef op, epoch_t start=0) { | |
739 | op->mark_osdmon_event(__func__); | |
740 | send_incremental(op, start); | |
741 | } | |
742 | ||
f67539c2 TL |
743 | int get_version(version_t ver, ceph::buffer::list& bl) override; |
744 | int get_version(version_t ver, uint64_t feature, ceph::buffer::list& bl); | |
28e407b8 | 745 | |
f67539c2 TL |
746 | int get_version_full(version_t ver, uint64_t feature, ceph::buffer::list& bl); |
747 | int get_version_full(version_t ver, ceph::buffer::list& bl) override; | |
11fdf7f2 | 748 | int get_inc(version_t ver, OSDMap::Incremental& inc); |
f67539c2 | 749 | int get_full_from_pinned_map(version_t ver, ceph::buffer::list& bl); |
7c673cae | 750 | |
f67539c2 TL |
751 | epoch_t blocklist(const entity_addrvec_t& av, utime_t until); |
752 | epoch_t blocklist(entity_addr_t a, utime_t until); | |
7c673cae | 753 | |
f67539c2 TL |
754 | void dump_info(ceph::Formatter *f); |
755 | int dump_osd_metadata(int osd, ceph::Formatter *f, std::ostream *err); | |
756 | void print_nodes(ceph::Formatter *f); | |
7c673cae FG |
757 | |
758 | void check_osdmap_sub(Subscription *sub); | |
759 | void check_pg_creates_sub(Subscription *sub); | |
760 | ||
11fdf7f2 TL |
761 | void do_application_enable(int64_t pool_id, const std::string &app_name, |
762 | const std::string &app_key="", | |
1911f103 TL |
763 | const std::string &app_value="", |
764 | bool force=false); | |
494da23a TL |
765 | void do_set_pool_opt(int64_t pool_id, pool_opts_t::key_t opt, |
766 | pool_opts_t::value_t); | |
c07f9fc5 | 767 | |
7c673cae FG |
768 | void add_flag(int flag) { |
769 | if (!(osdmap.flags & flag)) { | |
770 | if (pending_inc.new_flags < 0) | |
771 | pending_inc.new_flags = osdmap.flags; | |
772 | pending_inc.new_flags |= flag; | |
773 | } | |
774 | } | |
775 | ||
776 | void remove_flag(int flag) { | |
777 | if(osdmap.flags & flag) { | |
778 | if (pending_inc.new_flags < 0) | |
779 | pending_inc.new_flags = osdmap.flags; | |
780 | pending_inc.new_flags &= ~flag; | |
781 | } | |
782 | } | |
81eedcae | 783 | void convert_pool_priorities(void); |
f67539c2 TL |
784 | /** |
785 | * Find the pools which are requested to be put into stretch mode, | |
786 | * validate that they are allowed to be in stretch mode (eg, are replicated) | |
787 | * and place copies of them in the pools set. | |
788 | * This does not make any changes to the pools or state; it's just | |
789 | * a safety-check-and-collect function. | |
790 | */ | |
791 | void try_enable_stretch_mode_pools(stringstream& ss, bool *okay, | |
792 | int *errcode, | |
793 | set<pg_pool_t*>* pools, const string& new_crush_rule); | |
794 | /** | |
795 | * Check validity of inputs and OSD/CRUSH state to | |
796 | * engage stretch mode. Designed to be used with | |
797 | * MonmapMonitor::try_enable_stretch_mode() where we call both twice, | |
798 | * first with commit=false to validate. | |
799 | * @param ss: a stringstream to write errors into | |
800 | * @param okay: Filled to true if okay, false if validation fails | |
801 | * @param errcode: filled with -errno if there's a problem | |
802 | * @param commit: true if we should commit the change, false if just testing | |
803 | * @param dividing_bucket: the bucket type (eg 'dc') that divides the cluster | |
804 | * @param bucket_count: The number of buckets required in peering. | |
805 | * Currently must be 2. | |
806 | * @param pools: The pg_pool_ts which are being set to stretch mode (obtained | |
807 | * from try_enable_stretch_mode_pools()). | |
808 | * @param new_crush_rule: The crush rule to set the pools to. | |
809 | */ | |
810 | void try_enable_stretch_mode(stringstream& ss, bool *okay, | |
811 | int *errcode, bool commit, | |
812 | const string& dividing_bucket, | |
813 | uint32_t bucket_count, | |
814 | const set<pg_pool_t*>& pools, | |
815 | const string& new_crush_rule); | |
816 | /** | |
817 | * Check the input dead_buckets mapping (buckets->dead monitors) to see | |
818 | * if the OSDs are also down. If so, fill in really_down_buckets and | |
819 | * really_down_mons and return true; else return false. | |
820 | */ | |
821 | bool check_for_dead_crush_zones(const map<string,set<string>>& dead_buckets, | |
822 | set<int> *really_down_buckets, | |
823 | set<string> *really_down_mons); | |
824 | /** | |
825 | * Set degraded mode in the OSDMap, adding the given dead buckets to the dead set | |
826 | * and using the live_zones (should presently be size 1) | |
827 | */ | |
828 | void trigger_degraded_stretch_mode(const set<int>& dead_buckets, | |
829 | const set<string>& live_zones); | |
b3b6e05e TL |
830 | /** |
831 | * This is just to maintain stretch_recovery_triggered; below | |
832 | */ | |
833 | void set_degraded_stretch_mode(); | |
f67539c2 TL |
834 | /** |
835 | * Set recovery stretch mode in the OSDMap, resetting pool size back to normal | |
836 | */ | |
837 | void trigger_recovery_stretch_mode(); | |
b3b6e05e TL |
838 | /** |
839 | * This is just to maintain stretch_recovery_triggered; below | |
840 | */ | |
841 | void set_recovery_stretch_mode(); | |
842 | /** | |
843 | * This is just to maintain stretch_recovery_triggered; below | |
844 | */ | |
845 | void set_healthy_stretch_mode(); | |
f67539c2 TL |
846 | /** |
847 | * Tells the OSD there's a new pg digest, in case it's interested. | |
848 | * (It's interested when in recovering stretch mode.) | |
849 | */ | |
850 | void notify_new_pg_digest(); | |
851 | /** | |
852 | * Check if we can exit recovery stretch mode and go back to normal. | |
853 | * @param force If true, we will force the exit through once it is legal, | |
854 | * without regard to the reported PG status. | |
855 | */ | |
856 | void try_end_recovery_stretch_mode(bool force); | |
857 | /** | |
858 | * Sets the osdmap and pg_pool_t values back to healthy stretch mode status. | |
859 | */ | |
860 | void trigger_healthy_stretch_mode(); | |
522d829b TL |
861 | /** |
862 | * Obtain the crush rule being used for stretch pools. | |
863 | * Note that right now this is heuristic and simply selects the | |
864 | * most-used rule on replicated stretch pools. | |
865 | * @return the crush rule ID, or a negative errno | |
866 | */ | |
867 | int get_replicated_stretch_crush_rule(); | |
f67539c2 TL |
868 | private: |
869 | utime_t stretch_recovery_triggered; // what time we committed a switch to recovery mode | |
7c673cae FG |
870 | }; |
871 | ||
872 | #endif |