]>
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 | ||
19 | #ifndef CEPH_OSDMAP_H | |
20 | #define CEPH_OSDMAP_H | |
21 | ||
22 | /* | |
23 | * describe properties of the OSD cluster. | |
24 | * disks, disk groups, total # osds, | |
25 | * | |
26 | */ | |
27 | #include "include/types.h" | |
28 | #include "osd_types.h" | |
29 | ||
30 | //#include "include/ceph_features.h" | |
31 | #include "crush/CrushWrapper.h" | |
32 | #include <vector> | |
33 | #include <list> | |
34 | #include <set> | |
35 | #include <map> | |
36 | #include "include/memory.h" | |
94b18763 | 37 | #include "include/btree_map.h" |
7c673cae FG |
38 | using namespace std; |
39 | ||
40 | // forward declaration | |
41 | class CephContext; | |
42 | class CrushWrapper; | |
224ce89b | 43 | class health_check_map_t; |
7c673cae FG |
44 | |
45 | // FIXME C++11 does not have std::equal for two differently-typed containers. | |
46 | // use this until we move to c++14 | |
47 | template<typename A, typename B> | |
48 | bool vectors_equal(A a, B b) | |
49 | { | |
50 | return | |
51 | a.size() == b.size() && | |
52 | (a.empty() || | |
53 | memcmp((char*)&a[0], (char*)&b[0], sizeof(a[0]) * a.size()) == 0); | |
54 | } | |
55 | ||
56 | ||
57 | /* | |
58 | * we track up to two intervals during which the osd was alive and | |
59 | * healthy. the most recent is [up_from,up_thru), where up_thru is | |
60 | * the last epoch the osd is known to have _started_. i.e., a lower | |
61 | * bound on the actual osd death. down_at (if it is > up_from) is an | |
62 | * upper bound on the actual osd death. | |
63 | * | |
64 | * the second is the last_clean interval [first,last]. in that case, | |
65 | * the last interval is the last epoch known to have been either | |
66 | * _finished_, or during which the osd cleanly shut down. when | |
67 | * possible, we push this forward to the epoch the osd was eventually | |
68 | * marked down. | |
69 | * | |
70 | * the lost_at is used to allow build_prior to proceed without waiting | |
71 | * for an osd to recover. In certain cases, progress may be blocked | |
72 | * because an osd is down that may contain updates (i.e., a pg may have | |
73 | * gone rw during an interval). If the osd can't be brought online, we | |
74 | * can force things to proceed knowing that we _might_ be losing some | |
75 | * acked writes. If the osd comes back to life later, that's fine to, | |
76 | * but those writes will still be lost (the divergent objects will be | |
77 | * thrown out). | |
78 | */ | |
79 | struct osd_info_t { | |
80 | epoch_t last_clean_begin; // last interval that ended with a clean osd shutdown | |
81 | epoch_t last_clean_end; | |
82 | epoch_t up_from; // epoch osd marked up | |
83 | epoch_t up_thru; // lower bound on actual osd death (if > up_from) | |
84 | epoch_t down_at; // upper bound on actual osd death (if > up_from) | |
85 | epoch_t lost_at; // last epoch we decided data was "lost" | |
86 | ||
87 | osd_info_t() : last_clean_begin(0), last_clean_end(0), | |
88 | up_from(0), up_thru(0), down_at(0), lost_at(0) {} | |
89 | ||
90 | void dump(Formatter *f) const; | |
91 | void encode(bufferlist& bl) const; | |
92 | void decode(bufferlist::iterator& bl); | |
93 | static void generate_test_instances(list<osd_info_t*>& o); | |
94 | }; | |
95 | WRITE_CLASS_ENCODER(osd_info_t) | |
96 | ||
97 | ostream& operator<<(ostream& out, const osd_info_t& info); | |
98 | ||
99 | struct osd_xinfo_t { | |
100 | utime_t down_stamp; ///< timestamp when we were last marked down | |
101 | float laggy_probability; ///< encoded as __u32: 0 = definitely not laggy, 0xffffffff definitely laggy | |
102 | __u32 laggy_interval; ///< average interval between being marked laggy and recovering | |
103 | uint64_t features; ///< features supported by this osd we should know about | |
104 | __u32 old_weight; ///< weight prior to being auto marked out | |
105 | ||
106 | osd_xinfo_t() : laggy_probability(0), laggy_interval(0), | |
107 | features(0), old_weight(0) {} | |
108 | ||
109 | void dump(Formatter *f) const; | |
110 | void encode(bufferlist& bl) const; | |
111 | void decode(bufferlist::iterator& bl); | |
112 | static void generate_test_instances(list<osd_xinfo_t*>& o); | |
113 | }; | |
114 | WRITE_CLASS_ENCODER(osd_xinfo_t) | |
115 | ||
116 | ostream& operator<<(ostream& out, const osd_xinfo_t& xi); | |
117 | ||
118 | ||
31f18b77 FG |
119 | struct PGTempMap { |
120 | #if 1 | |
121 | bufferlist data; | |
122 | typedef btree::btree_map<pg_t,int32_t*> map_t; | |
123 | map_t map; | |
124 | ||
125 | void encode(bufferlist& bl) const { | |
126 | uint32_t n = map.size(); | |
127 | ::encode(n, bl); | |
128 | for (auto &p : map) { | |
129 | ::encode(p.first, bl); | |
130 | bl.append((char*)p.second, (*p.second + 1) * sizeof(int32_t)); | |
131 | } | |
132 | } | |
133 | void decode(bufferlist::iterator& p) { | |
134 | data.clear(); | |
135 | map.clear(); | |
136 | uint32_t n; | |
137 | ::decode(n, p); | |
138 | if (!n) | |
139 | return; | |
140 | bufferlist::iterator pstart = p; | |
141 | size_t start_off = pstart.get_off(); | |
142 | vector<pair<pg_t,size_t>> offsets; | |
143 | offsets.resize(n); | |
144 | for (unsigned i=0; i<n; ++i) { | |
145 | pg_t pgid; | |
146 | ::decode(pgid, p); | |
147 | offsets[i].first = pgid; | |
148 | offsets[i].second = p.get_off() - start_off; | |
149 | uint32_t vn; | |
150 | ::decode(vn, p); | |
151 | p.advance(vn * sizeof(int32_t)); | |
152 | } | |
153 | size_t len = p.get_off() - start_off; | |
154 | pstart.copy(len, data); | |
155 | if (data.get_num_buffers() > 1) { | |
156 | data.rebuild(); | |
157 | } | |
158 | //map.reserve(n); | |
159 | char *start = data.c_str(); | |
160 | for (auto i : offsets) { | |
161 | map.insert(map.end(), make_pair(i.first, (int32_t*)(start + i.second))); | |
162 | } | |
163 | } | |
164 | void rebuild() { | |
165 | bufferlist bl; | |
166 | encode(bl); | |
167 | auto p = bl.begin(); | |
168 | decode(p); | |
169 | } | |
170 | friend bool operator==(const PGTempMap& l, const PGTempMap& r) { | |
171 | return | |
172 | l.map.size() == r.map.size() && | |
173 | l.data.contents_equal(r.data); | |
174 | } | |
175 | ||
176 | class iterator { | |
177 | map_t::const_iterator it; | |
178 | map_t::const_iterator end; | |
179 | pair<pg_t,vector<int32_t>> current; | |
180 | void init_current() { | |
181 | if (it != end) { | |
182 | current.first = it->first; | |
183 | assert(it->second); | |
184 | current.second.resize(*it->second); | |
185 | int32_t *p = it->second + 1; | |
186 | for (int n = 0; n < *it->second; ++n, ++p) { | |
187 | current.second[n] = *p; | |
188 | } | |
189 | } | |
190 | } | |
191 | public: | |
192 | iterator(map_t::const_iterator p, | |
193 | map_t::const_iterator e) | |
194 | : it(p), end(e) { | |
195 | init_current(); | |
196 | } | |
197 | ||
198 | const pair<pg_t,vector<int32_t>>& operator*() const { | |
199 | return current; | |
200 | } | |
201 | const pair<pg_t,vector<int32_t>>* operator->() const { | |
202 | return ¤t; | |
203 | } | |
204 | friend bool operator==(const iterator& l, const iterator& r) { | |
205 | return l.it == r.it; | |
206 | } | |
207 | friend bool operator!=(const iterator& l, const iterator& r) { | |
208 | return l.it != r.it; | |
209 | } | |
210 | iterator& operator++() { | |
211 | ++it; | |
212 | if (it != end) | |
213 | init_current(); | |
214 | return *this; | |
215 | } | |
216 | iterator operator++(int) { | |
217 | iterator r = *this; | |
218 | ++it; | |
219 | if (it != end) | |
220 | init_current(); | |
221 | return r; | |
222 | } | |
223 | }; | |
224 | iterator begin() const { | |
225 | return iterator(map.begin(), map.end()); | |
226 | } | |
227 | iterator end() const { | |
228 | return iterator(map.end(), map.end()); | |
229 | } | |
230 | iterator find(pg_t pgid) const { | |
231 | return iterator(map.find(pgid), map.end()); | |
232 | } | |
233 | size_t size() const { | |
234 | return map.size(); | |
235 | } | |
236 | size_t count(pg_t pgid) const { | |
237 | return map.count(pgid); | |
238 | } | |
239 | void erase(pg_t pgid) { | |
240 | map.erase(pgid); | |
241 | } | |
242 | void clear() { | |
243 | map.clear(); | |
244 | data.clear(); | |
245 | } | |
246 | void set(pg_t pgid, const mempool::osdmap::vector<int32_t>& v) { | |
247 | size_t need = sizeof(int32_t) * (1 + v.size()); | |
248 | if (need < data.get_append_buffer_unused_tail_length()) { | |
249 | bufferptr z(data.get_append_buffer_unused_tail_length()); | |
250 | z.zero(); | |
251 | data.append(z.c_str(), z.length()); | |
252 | } | |
253 | ::encode(v, data); | |
254 | map[pgid] = (int32_t*)(data.back().end_c_str()) - (1 + v.size()); | |
255 | } | |
256 | mempool::osdmap::vector<int32_t> get(pg_t pgid) { | |
257 | mempool::osdmap::vector<int32_t> v; | |
258 | int32_t *p = map[pgid]; | |
259 | size_t n = *p++; | |
260 | v.resize(n); | |
261 | for (size_t i = 0; i < n; ++i, ++p) { | |
262 | v[i] = *p; | |
263 | } | |
264 | return v; | |
265 | } | |
266 | #else | |
267 | // trivial implementation | |
268 | mempool::osdmap::map<pg_t,mempool::osdmap::vector<int32_t> > pg_temp; | |
269 | ||
270 | void encode(bufferlist& bl) const { | |
271 | ::encode(pg_temp, bl); | |
272 | } | |
273 | void decode(bufferlist::iterator& p) { | |
274 | ::decode(pg_temp, p); | |
275 | } | |
276 | friend bool operator==(const PGTempMap& l, const PGTempMap& r) { | |
277 | return | |
278 | l.pg_temp.size() == r.pg_temp.size() && | |
279 | l.pg_temp == r.pg_temp; | |
280 | } | |
281 | ||
282 | class iterator { | |
283 | mempool::osdmap::map<pg_t,mempool::osdmap::vector<int32_t> >::const_iterator it; | |
284 | public: | |
285 | iterator(mempool::osdmap::map<pg_t, | |
286 | mempool::osdmap::vector<int32_t> >::const_iterator p) | |
287 | : it(p) {} | |
288 | ||
289 | pair<pg_t,const mempool::osdmap::vector<int32_t>&> operator*() const { | |
290 | return *it; | |
291 | } | |
292 | const pair<const pg_t,mempool::osdmap::vector<int32_t>>* operator->() const { | |
293 | return &*it; | |
294 | } | |
295 | friend bool operator==(const iterator& l, const iterator& r) { | |
296 | return l.it == r.it; | |
297 | } | |
298 | friend bool operator!=(const iterator& l, const iterator& r) { | |
299 | return l.it != r.it; | |
300 | } | |
301 | iterator& operator++() { | |
302 | ++it; | |
303 | return *this; | |
304 | } | |
305 | iterator operator++(int) { | |
306 | iterator r = *this; | |
307 | ++it; | |
308 | return r; | |
309 | } | |
310 | }; | |
311 | iterator begin() const { | |
312 | return iterator(pg_temp.cbegin()); | |
313 | } | |
314 | iterator end() const { | |
315 | return iterator(pg_temp.cend()); | |
316 | } | |
317 | iterator find(pg_t pgid) const { | |
318 | return iterator(pg_temp.find(pgid)); | |
319 | } | |
320 | size_t size() const { | |
321 | return pg_temp.size(); | |
322 | } | |
323 | size_t count(pg_t pgid) const { | |
324 | return pg_temp.count(pgid); | |
325 | } | |
326 | void erase(pg_t pgid) { | |
327 | pg_temp.erase(pgid); | |
328 | } | |
329 | void clear() { | |
330 | pg_temp.clear(); | |
331 | } | |
332 | void set(pg_t pgid, const mempool::osdmap::vector<int32_t>& v) { | |
333 | pg_temp[pgid] = v; | |
334 | } | |
335 | const mempool::osdmap::vector<int32_t>& get(pg_t pgid) { | |
336 | return pg_temp.at(pgid); | |
337 | } | |
338 | #endif | |
339 | void dump(Formatter *f) const { | |
340 | for (const auto &pg : *this) { | |
341 | f->open_object_section("osds"); | |
342 | f->dump_stream("pgid") << pg.first; | |
343 | f->open_array_section("osds"); | |
344 | for (const auto osd : pg.second) | |
345 | f->dump_int("osd", osd); | |
346 | f->close_section(); | |
347 | f->close_section(); | |
348 | } | |
349 | } | |
350 | }; | |
351 | WRITE_CLASS_ENCODER(PGTempMap) | |
352 | ||
7c673cae FG |
353 | /** OSDMap |
354 | */ | |
355 | class OSDMap { | |
356 | public: | |
357 | MEMPOOL_CLASS_HELPERS(); | |
358 | ||
359 | class Incremental { | |
360 | public: | |
361 | MEMPOOL_CLASS_HELPERS(); | |
362 | ||
363 | /// feature bits we were encoded with. the subsequent OSDMap | |
364 | /// encoding should match. | |
365 | uint64_t encode_features; | |
366 | uuid_d fsid; | |
367 | epoch_t epoch; // new epoch; we are a diff from epoch-1 to epoch | |
368 | utime_t modified; | |
369 | int64_t new_pool_max; //incremented by the OSDMonitor on each pool create | |
370 | int32_t new_flags; | |
31f18b77 | 371 | int8_t new_require_osd_release = -1; |
7c673cae FG |
372 | |
373 | // full (rare) | |
374 | bufferlist fullmap; // in lieu of below. | |
375 | bufferlist crush; | |
376 | ||
377 | // incremental | |
378 | int32_t new_max_osd; | |
379 | mempool::osdmap::map<int64_t,pg_pool_t> new_pools; | |
380 | mempool::osdmap::map<int64_t,string> new_pool_names; | |
381 | mempool::osdmap::set<int64_t> old_pools; | |
382 | mempool::osdmap::map<string,map<string,string> > new_erasure_code_profiles; | |
383 | mempool::osdmap::vector<string> old_erasure_code_profiles; | |
384 | mempool::osdmap::map<int32_t,entity_addr_t> new_up_client; | |
385 | mempool::osdmap::map<int32_t,entity_addr_t> new_up_cluster; | |
31f18b77 | 386 | mempool::osdmap::map<int32_t,uint32_t> new_state; // XORed onto previous state. |
7c673cae FG |
387 | mempool::osdmap::map<int32_t,uint32_t> new_weight; |
388 | mempool::osdmap::map<pg_t,mempool::osdmap::vector<int32_t> > new_pg_temp; // [] to remove | |
389 | mempool::osdmap::map<pg_t, int32_t> new_primary_temp; // [-1] to remove | |
390 | mempool::osdmap::map<int32_t,uint32_t> new_primary_affinity; | |
391 | mempool::osdmap::map<int32_t,epoch_t> new_up_thru; | |
392 | mempool::osdmap::map<int32_t,pair<epoch_t,epoch_t> > new_last_clean_interval; | |
393 | mempool::osdmap::map<int32_t,epoch_t> new_lost; | |
394 | mempool::osdmap::map<int32_t,uuid_d> new_uuid; | |
395 | mempool::osdmap::map<int32_t,osd_xinfo_t> new_xinfo; | |
396 | ||
397 | mempool::osdmap::map<entity_addr_t,utime_t> new_blacklist; | |
398 | mempool::osdmap::vector<entity_addr_t> old_blacklist; | |
399 | mempool::osdmap::map<int32_t, entity_addr_t> new_hb_back_up; | |
400 | mempool::osdmap::map<int32_t, entity_addr_t> new_hb_front_up; | |
401 | ||
402 | mempool::osdmap::map<pg_t,mempool::osdmap::vector<int32_t>> new_pg_upmap; | |
403 | mempool::osdmap::map<pg_t,mempool::osdmap::vector<pair<int32_t,int32_t>>> new_pg_upmap_items; | |
404 | mempool::osdmap::set<pg_t> old_pg_upmap, old_pg_upmap_items; | |
405 | ||
406 | string cluster_snapshot; | |
407 | ||
408 | float new_nearfull_ratio = -1; | |
409 | float new_backfillfull_ratio = -1; | |
410 | float new_full_ratio = -1; | |
411 | ||
31f18b77 | 412 | int8_t new_require_min_compat_client = -1; |
7c673cae FG |
413 | |
414 | mutable bool have_crc; ///< crc values are defined | |
415 | uint32_t full_crc; ///< crc of the resulting OSDMap | |
416 | mutable uint32_t inc_crc; ///< crc of this incremental | |
417 | ||
418 | int get_net_marked_out(const OSDMap *previous) const; | |
419 | int get_net_marked_down(const OSDMap *previous) const; | |
420 | int identify_osd(uuid_d u) const; | |
421 | ||
422 | void encode_client_old(bufferlist& bl) const; | |
423 | void encode_classic(bufferlist& bl, uint64_t features) const; | |
424 | void encode(bufferlist& bl, uint64_t features=CEPH_FEATURES_ALL) const; | |
425 | void decode_classic(bufferlist::iterator &p); | |
426 | void decode(bufferlist::iterator &bl); | |
427 | void dump(Formatter *f) const; | |
428 | static void generate_test_instances(list<Incremental*>& o); | |
429 | ||
430 | explicit Incremental(epoch_t e=0) : | |
431 | encode_features(0), | |
432 | epoch(e), new_pool_max(-1), new_flags(-1), new_max_osd(-1), | |
433 | have_crc(false), full_crc(0), inc_crc(0) { | |
434 | memset(&fsid, 0, sizeof(fsid)); | |
435 | } | |
436 | explicit Incremental(bufferlist &bl) { | |
437 | bufferlist::iterator p = bl.begin(); | |
438 | decode(p); | |
439 | } | |
440 | explicit Incremental(bufferlist::iterator &p) { | |
441 | decode(p); | |
442 | } | |
443 | ||
444 | pg_pool_t *get_new_pool(int64_t pool, const pg_pool_t *orig) { | |
445 | if (new_pools.count(pool) == 0) | |
446 | new_pools[pool] = *orig; | |
447 | return &new_pools[pool]; | |
448 | } | |
449 | bool has_erasure_code_profile(const string &name) const { | |
450 | auto i = new_erasure_code_profiles.find(name); | |
451 | return i != new_erasure_code_profiles.end(); | |
452 | } | |
453 | void set_erasure_code_profile(const string &name, | |
454 | const map<string,string>& profile) { | |
455 | new_erasure_code_profiles[name] = profile; | |
456 | } | |
457 | ||
458 | /// propage update pools' snap metadata to any of their tiers | |
459 | int propagate_snaps_to_tiers(CephContext *cct, const OSDMap &base); | |
31f18b77 FG |
460 | |
461 | /// filter out osds with any pending state changing | |
462 | size_t get_pending_state_osds(vector<int> *osds) { | |
463 | assert(osds); | |
464 | osds->clear(); | |
465 | ||
466 | for (auto &p : new_state) { | |
467 | osds->push_back(p.first); | |
468 | } | |
469 | ||
470 | return osds->size(); | |
471 | } | |
472 | ||
473 | bool pending_osd_has_state(int osd, unsigned state) { | |
474 | return new_state.count(osd) && (new_state[osd] & state) != 0; | |
475 | } | |
476 | ||
477 | void pending_osd_state_set(int osd, unsigned state) { | |
478 | new_state[osd] |= state; | |
479 | } | |
480 | ||
481 | // cancel the specified pending osd state if there is any | |
482 | // return ture on success, false otherwise. | |
483 | bool pending_osd_state_clear(int osd, unsigned state) { | |
484 | if (!pending_osd_has_state(osd, state)) { | |
485 | // never has been set or already has been cancelled. | |
486 | return false; | |
487 | } | |
488 | ||
489 | new_state[osd] &= ~state; | |
490 | return true; | |
491 | } | |
492 | ||
7c673cae FG |
493 | }; |
494 | ||
495 | private: | |
496 | uuid_d fsid; | |
497 | epoch_t epoch; // what epoch of the osd cluster descriptor is this | |
498 | utime_t created, modified; // epoch start time | |
499 | int32_t pool_max; // the largest pool num, ever | |
500 | ||
501 | uint32_t flags; | |
502 | ||
503 | int num_osd; // not saved; see calc_num_osds | |
504 | int num_up_osd; // not saved; see calc_num_osds | |
505 | int num_in_osd; // not saved; see calc_num_osds | |
506 | ||
507 | int32_t max_osd; | |
31f18b77 | 508 | vector<uint32_t> osd_state; |
7c673cae | 509 | |
28e407b8 AA |
510 | // These features affect OSDMap[::Incremental] encoding, or the |
511 | // encoding of some type embedded therein (CrushWrapper, something | |
512 | // from osd_types, etc.). | |
513 | static constexpr uint64_t SIGNIFICANT_FEATURES = | |
514 | CEPH_FEATUREMASK_PGID64 | | |
515 | CEPH_FEATUREMASK_PGPOOL3 | | |
516 | CEPH_FEATUREMASK_OSDENC | | |
517 | CEPH_FEATUREMASK_OSDMAP_ENC | | |
518 | CEPH_FEATUREMASK_OSD_POOLRESEND | | |
519 | CEPH_FEATUREMASK_NEW_OSDOP_ENCODING | | |
520 | CEPH_FEATUREMASK_MSG_ADDR2 | | |
521 | CEPH_FEATUREMASK_CRUSH_TUNABLES5 | | |
522 | CEPH_FEATUREMASK_CRUSH_CHOOSE_ARGS | | |
523 | CEPH_FEATUREMASK_SERVER_LUMINOUS ; | |
7c673cae FG |
524 | struct addrs_s { |
525 | mempool::osdmap::vector<ceph::shared_ptr<entity_addr_t> > client_addr; | |
526 | mempool::osdmap::vector<ceph::shared_ptr<entity_addr_t> > cluster_addr; | |
527 | mempool::osdmap::vector<ceph::shared_ptr<entity_addr_t> > hb_back_addr; | |
528 | mempool::osdmap::vector<ceph::shared_ptr<entity_addr_t> > hb_front_addr; | |
529 | entity_addr_t blank; | |
530 | }; | |
531 | ceph::shared_ptr<addrs_s> osd_addrs; | |
532 | ||
533 | mempool::osdmap::vector<__u32> osd_weight; // 16.16 fixed point, 0x10000 = "in", 0 = "out" | |
534 | mempool::osdmap::vector<osd_info_t> osd_info; | |
31f18b77 | 535 | ceph::shared_ptr<PGTempMap> pg_temp; // temp pg mapping (e.g. while we rebuild) |
7c673cae FG |
536 | ceph::shared_ptr< mempool::osdmap::map<pg_t,int32_t > > primary_temp; // temp primary mapping (e.g. while we rebuild) |
537 | ceph::shared_ptr< mempool::osdmap::vector<__u32> > osd_primary_affinity; ///< 16.16 fixed point, 0x10000 = baseline | |
538 | ||
539 | // remap (post-CRUSH, pre-up) | |
540 | mempool::osdmap::map<pg_t,mempool::osdmap::vector<int32_t>> pg_upmap; ///< remap pg | |
541 | mempool::osdmap::map<pg_t,mempool::osdmap::vector<pair<int32_t,int32_t>>> pg_upmap_items; ///< remap osds in up set | |
542 | ||
543 | mempool::osdmap::map<int64_t,pg_pool_t> pools; | |
544 | mempool::osdmap::map<int64_t,string> pool_name; | |
545 | mempool::osdmap::map<string,map<string,string> > erasure_code_profiles; | |
546 | mempool::osdmap::map<string,int64_t> name_pool; | |
547 | ||
548 | ceph::shared_ptr< mempool::osdmap::vector<uuid_d> > osd_uuid; | |
549 | mempool::osdmap::vector<osd_xinfo_t> osd_xinfo; | |
550 | ||
551 | mempool::osdmap::unordered_map<entity_addr_t,utime_t> blacklist; | |
552 | ||
553 | epoch_t cluster_snapshot_epoch; | |
554 | string cluster_snapshot; | |
555 | bool new_blacklist_entries; | |
556 | ||
557 | float full_ratio = 0, backfillfull_ratio = 0, nearfull_ratio = 0; | |
558 | ||
559 | /// min compat client we want to support | |
31f18b77 | 560 | uint8_t require_min_compat_client = 0; // CEPH_RELEASE_* |
7c673cae | 561 | |
31f18b77 FG |
562 | public: |
563 | /// require osds to run at least this release | |
564 | uint8_t require_osd_release = 0; // CEPH_RELEASE_* | |
565 | ||
566 | private: | |
7c673cae FG |
567 | mutable uint64_t cached_up_osd_features; |
568 | ||
569 | mutable bool crc_defined; | |
570 | mutable uint32_t crc; | |
571 | ||
572 | void _calc_up_osd_features(); | |
573 | ||
574 | public: | |
575 | bool have_crc() const { return crc_defined; } | |
576 | uint32_t get_crc() const { return crc; } | |
577 | ||
578 | ceph::shared_ptr<CrushWrapper> crush; // hierarchical map | |
31f18b77 FG |
579 | private: |
580 | uint32_t crush_version = 1; | |
7c673cae FG |
581 | |
582 | friend class OSDMonitor; | |
583 | ||
584 | public: | |
585 | OSDMap() : epoch(0), | |
224ce89b | 586 | pool_max(0), |
7c673cae FG |
587 | flags(0), |
588 | num_osd(0), num_up_osd(0), num_in_osd(0), | |
589 | max_osd(0), | |
590 | osd_addrs(std::make_shared<addrs_s>()), | |
31f18b77 | 591 | pg_temp(std::make_shared<PGTempMap>()), |
7c673cae FG |
592 | primary_temp(std::make_shared<mempool::osdmap::map<pg_t,int32_t>>()), |
593 | osd_uuid(std::make_shared<mempool::osdmap::vector<uuid_d>>()), | |
594 | cluster_snapshot_epoch(0), | |
595 | new_blacklist_entries(false), | |
596 | cached_up_osd_features(0), | |
597 | crc_defined(false), crc(0), | |
598 | crush(std::make_shared<CrushWrapper>()) { | |
599 | memset(&fsid, 0, sizeof(fsid)); | |
600 | } | |
601 | ||
602 | // no copying | |
603 | private: | |
604 | OSDMap(const OSDMap& other) = default; | |
605 | OSDMap& operator=(const OSDMap& other) = default; | |
606 | public: | |
607 | ||
28e407b8 AA |
608 | /// return feature mask subset that is relevant to OSDMap encoding |
609 | static uint64_t get_significant_features(uint64_t features) { | |
610 | return SIGNIFICANT_FEATURES & features; | |
611 | } | |
612 | ||
613 | uint64_t get_encoding_features() const; | |
614 | ||
7c673cae FG |
615 | void deepish_copy_from(const OSDMap& o) { |
616 | *this = o; | |
617 | primary_temp.reset(new mempool::osdmap::map<pg_t,int32_t>(*o.primary_temp)); | |
31f18b77 | 618 | pg_temp.reset(new PGTempMap(*o.pg_temp)); |
7c673cae FG |
619 | osd_uuid.reset(new mempool::osdmap::vector<uuid_d>(*o.osd_uuid)); |
620 | ||
621 | if (o.osd_primary_affinity) | |
622 | osd_primary_affinity.reset(new mempool::osdmap::vector<__u32>(*o.osd_primary_affinity)); | |
623 | ||
624 | // NOTE: this still references shared entity_addr_t's. | |
625 | osd_addrs.reset(new addrs_s(*o.osd_addrs)); | |
626 | ||
627 | // NOTE: we do not copy crush. note that apply_incremental will | |
628 | // allocate a new CrushWrapper, though. | |
629 | } | |
630 | ||
631 | // map info | |
632 | const uuid_d& get_fsid() const { return fsid; } | |
633 | void set_fsid(uuid_d& f) { fsid = f; } | |
634 | ||
635 | epoch_t get_epoch() const { return epoch; } | |
636 | void inc_epoch() { epoch++; } | |
637 | ||
638 | void set_epoch(epoch_t e); | |
639 | ||
31f18b77 FG |
640 | uint32_t get_crush_version() const { |
641 | return crush_version; | |
642 | } | |
643 | ||
7c673cae FG |
644 | /* stamps etc */ |
645 | const utime_t& get_created() const { return created; } | |
646 | const utime_t& get_modified() const { return modified; } | |
647 | ||
648 | bool is_blacklisted(const entity_addr_t& a) const; | |
649 | void get_blacklist(list<pair<entity_addr_t,utime_t > > *bl) const; | |
31f18b77 | 650 | void get_blacklist(std::set<entity_addr_t> *bl) const; |
7c673cae FG |
651 | |
652 | string get_cluster_snapshot() const { | |
653 | if (cluster_snapshot_epoch == epoch) | |
654 | return cluster_snapshot; | |
655 | return string(); | |
656 | } | |
657 | ||
658 | float get_full_ratio() const { | |
659 | return full_ratio; | |
660 | } | |
661 | float get_backfillfull_ratio() const { | |
662 | return backfillfull_ratio; | |
663 | } | |
664 | float get_nearfull_ratio() const { | |
665 | return nearfull_ratio; | |
666 | } | |
7c673cae | 667 | void get_full_osd_util( |
31f18b77 | 668 | const mempool::pgmap::unordered_map<int32_t,osd_stat_t> &osd_stat, |
7c673cae FG |
669 | map<int, float> *full, |
670 | map<int, float> *backfill, | |
671 | map<int, float> *nearfull) const; | |
3efd9988 FG |
672 | void get_full_pools(CephContext *cct, |
673 | set<int64_t> *full, | |
674 | set<int64_t> *backfillfull, | |
675 | set<int64_t> *nearfull) const; | |
31f18b77 FG |
676 | void get_full_osd_counts(set<int> *full, set<int> *backfill, |
677 | set<int> *nearfull) const; | |
678 | ||
679 | ||
7c673cae FG |
680 | /***** cluster state *****/ |
681 | /* osds */ | |
682 | int get_max_osd() const { return max_osd; } | |
683 | void set_max_osd(int m); | |
684 | ||
685 | unsigned get_num_osds() const { | |
686 | return num_osd; | |
687 | } | |
688 | unsigned get_num_up_osds() const { | |
689 | return num_up_osd; | |
690 | } | |
691 | unsigned get_num_in_osds() const { | |
692 | return num_in_osd; | |
693 | } | |
694 | /// recalculate cached values for get_num{,_up,_in}_osds | |
695 | int calc_num_osds(); | |
696 | ||
697 | void get_all_osds(set<int32_t>& ls) const; | |
698 | void get_up_osds(set<int32_t>& ls) const; | |
31f18b77 | 699 | void get_out_osds(set<int32_t>& ls) const; |
7c673cae FG |
700 | unsigned get_num_pg_temp() const { |
701 | return pg_temp->size(); | |
702 | } | |
703 | ||
704 | int get_flags() const { return flags; } | |
705 | bool test_flag(int f) const { return flags & f; } | |
706 | void set_flag(int f) { flags |= f; } | |
707 | void clear_flag(int f) { flags &= ~f; } | |
708 | ||
709 | static void calc_state_set(int state, set<string>& st); | |
710 | ||
711 | int get_state(int o) const { | |
712 | assert(o < max_osd); | |
713 | return osd_state[o]; | |
714 | } | |
715 | int get_state(int o, set<string>& st) const { | |
716 | assert(o < max_osd); | |
717 | unsigned t = osd_state[o]; | |
718 | calc_state_set(t, st); | |
719 | return osd_state[o]; | |
720 | } | |
721 | void set_state(int o, unsigned s) { | |
722 | assert(o < max_osd); | |
723 | osd_state[o] = s; | |
724 | } | |
725 | void set_weight(int o, unsigned w) { | |
726 | assert(o < max_osd); | |
727 | osd_weight[o] = w; | |
728 | if (w) | |
729 | osd_state[o] |= CEPH_OSD_EXISTS; | |
730 | } | |
731 | unsigned get_weight(int o) const { | |
732 | assert(o < max_osd); | |
733 | return osd_weight[o]; | |
734 | } | |
735 | float get_weightf(int o) const { | |
736 | return (float)get_weight(o) / (float)CEPH_OSD_IN; | |
737 | } | |
738 | void adjust_osd_weights(const map<int,double>& weights, Incremental& inc) const; | |
739 | ||
740 | void set_primary_affinity(int o, int w) { | |
741 | assert(o < max_osd); | |
742 | if (!osd_primary_affinity) | |
743 | osd_primary_affinity.reset( | |
744 | new mempool::osdmap::vector<__u32>( | |
745 | max_osd, CEPH_OSD_DEFAULT_PRIMARY_AFFINITY)); | |
746 | (*osd_primary_affinity)[o] = w; | |
747 | } | |
748 | unsigned get_primary_affinity(int o) const { | |
749 | assert(o < max_osd); | |
750 | if (!osd_primary_affinity) | |
751 | return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY; | |
752 | return (*osd_primary_affinity)[o]; | |
753 | } | |
754 | float get_primary_affinityf(int o) const { | |
755 | return (float)get_primary_affinity(o) / (float)CEPH_OSD_MAX_PRIMARY_AFFINITY; | |
756 | } | |
757 | ||
758 | bool has_erasure_code_profile(const string &name) const { | |
759 | auto i = erasure_code_profiles.find(name); | |
760 | return i != erasure_code_profiles.end(); | |
761 | } | |
762 | int get_erasure_code_profile_default(CephContext *cct, | |
763 | map<string,string> &profile_map, | |
764 | ostream *ss); | |
765 | void set_erasure_code_profile(const string &name, | |
766 | const map<string,string>& profile) { | |
767 | erasure_code_profiles[name] = profile; | |
768 | } | |
769 | const map<string,string> &get_erasure_code_profile( | |
770 | const string &name) const { | |
771 | static map<string,string> empty; | |
772 | auto i = erasure_code_profiles.find(name); | |
773 | if (i == erasure_code_profiles.end()) | |
774 | return empty; | |
775 | else | |
776 | return i->second; | |
777 | } | |
778 | const mempool::osdmap::map<string,map<string,string> > &get_erasure_code_profiles() const { | |
779 | return erasure_code_profiles; | |
780 | } | |
781 | ||
782 | bool exists(int osd) const { | |
783 | //assert(osd >= 0); | |
784 | return osd >= 0 && osd < max_osd && (osd_state[osd] & CEPH_OSD_EXISTS); | |
785 | } | |
786 | ||
31f18b77 FG |
787 | bool is_destroyed(int osd) const { |
788 | return exists(osd) && (osd_state[osd] & CEPH_OSD_DESTROYED); | |
789 | } | |
790 | ||
7c673cae FG |
791 | bool is_up(int osd) const { |
792 | return exists(osd) && (osd_state[osd] & CEPH_OSD_UP); | |
793 | } | |
794 | ||
795 | bool has_been_up_since(int osd, epoch_t epoch) const { | |
796 | return is_up(osd) && get_up_from(osd) <= epoch; | |
797 | } | |
798 | ||
799 | bool is_down(int osd) const { | |
800 | return !is_up(osd); | |
801 | } | |
802 | ||
803 | bool is_out(int osd) const { | |
804 | return !exists(osd) || get_weight(osd) == CEPH_OSD_OUT; | |
805 | } | |
806 | ||
807 | bool is_in(int osd) const { | |
808 | return !is_out(osd); | |
809 | } | |
810 | ||
31f18b77 FG |
811 | bool is_noup(int osd) const { |
812 | return exists(osd) && (osd_state[osd] & CEPH_OSD_NOUP); | |
813 | } | |
814 | ||
815 | bool is_nodown(int osd) const { | |
816 | return exists(osd) && (osd_state[osd] & CEPH_OSD_NODOWN); | |
817 | } | |
818 | ||
819 | bool is_noin(int osd) const { | |
820 | return exists(osd) && (osd_state[osd] & CEPH_OSD_NOIN); | |
821 | } | |
822 | ||
823 | bool is_noout(int osd) const { | |
824 | return exists(osd) && (osd_state[osd] & CEPH_OSD_NOOUT); | |
825 | } | |
826 | ||
827 | void get_noup_osds(vector<int> *osds) const { | |
828 | assert(osds); | |
829 | osds->clear(); | |
830 | ||
831 | for (int i = 0; i < max_osd; i++) { | |
832 | if (is_noup(i)) { | |
833 | osds->push_back(i); | |
834 | } | |
835 | } | |
836 | } | |
837 | ||
838 | void get_nodown_osds(vector<int> *osds) const { | |
839 | assert(osds); | |
840 | osds->clear(); | |
841 | ||
842 | for (int i = 0; i < max_osd; i++) { | |
843 | if (is_nodown(i)) { | |
844 | osds->push_back(i); | |
845 | } | |
846 | } | |
847 | } | |
848 | ||
849 | void get_noin_osds(vector<int> *osds) const { | |
850 | assert(osds); | |
851 | osds->clear(); | |
852 | ||
853 | for (int i = 0; i < max_osd; i++) { | |
854 | if (is_noin(i)) { | |
855 | osds->push_back(i); | |
856 | } | |
857 | } | |
858 | } | |
859 | ||
860 | void get_noout_osds(vector<int> *osds) const { | |
861 | assert(osds); | |
862 | osds->clear(); | |
863 | ||
864 | for (int i = 0; i < max_osd; i++) { | |
865 | if (is_noout(i)) { | |
866 | osds->push_back(i); | |
867 | } | |
868 | } | |
869 | } | |
870 | ||
7c673cae FG |
871 | /** |
872 | * check if an entire crush subtree is down | |
873 | */ | |
874 | bool subtree_is_down(int id, set<int> *down_cache) const; | |
875 | bool containing_subtree_is_down(CephContext *cct, int osd, int subtree_type, set<int> *down_cache) const; | |
876 | ||
31f18b77 FG |
877 | bool subtree_type_is_down(CephContext *cct, int id, int subtree_type, set<int> *down_in_osds, set<int> *up_in_osds, |
878 | set<int> *subtree_up, unordered_map<int, set<int> > *subtree_type_down) const; | |
879 | ||
7c673cae FG |
880 | int identify_osd(const entity_addr_t& addr) const; |
881 | int identify_osd(const uuid_d& u) const; | |
882 | int identify_osd_on_all_channels(const entity_addr_t& addr) const; | |
883 | ||
884 | bool have_addr(const entity_addr_t& addr) const { | |
885 | return identify_osd(addr) >= 0; | |
886 | } | |
887 | int find_osd_on_ip(const entity_addr_t& ip) const; | |
888 | const entity_addr_t &get_addr(int osd) const { | |
889 | assert(exists(osd)); | |
890 | return osd_addrs->client_addr[osd] ? *osd_addrs->client_addr[osd] : osd_addrs->blank; | |
891 | } | |
892 | const entity_addr_t &get_cluster_addr(int osd) const { | |
893 | assert(exists(osd)); | |
894 | if (!osd_addrs->cluster_addr[osd] || *osd_addrs->cluster_addr[osd] == entity_addr_t()) | |
895 | return get_addr(osd); | |
896 | return *osd_addrs->cluster_addr[osd]; | |
897 | } | |
898 | const entity_addr_t &get_hb_back_addr(int osd) const { | |
899 | assert(exists(osd)); | |
900 | return osd_addrs->hb_back_addr[osd] ? *osd_addrs->hb_back_addr[osd] : osd_addrs->blank; | |
901 | } | |
902 | const entity_addr_t &get_hb_front_addr(int osd) const { | |
903 | assert(exists(osd)); | |
904 | return osd_addrs->hb_front_addr[osd] ? *osd_addrs->hb_front_addr[osd] : osd_addrs->blank; | |
905 | } | |
906 | entity_inst_t get_most_recent_inst(int osd) const { | |
907 | assert(exists(osd)); | |
908 | return entity_inst_t(entity_name_t::OSD(osd), get_addr(osd)); | |
909 | } | |
910 | entity_inst_t get_inst(int osd) const { | |
911 | assert(is_up(osd)); | |
912 | return get_most_recent_inst(osd); | |
913 | } | |
914 | entity_inst_t get_cluster_inst(int osd) const { | |
915 | assert(is_up(osd)); | |
916 | return entity_inst_t(entity_name_t::OSD(osd), get_cluster_addr(osd)); | |
917 | } | |
918 | entity_inst_t get_hb_back_inst(int osd) const { | |
919 | assert(is_up(osd)); | |
920 | return entity_inst_t(entity_name_t::OSD(osd), get_hb_back_addr(osd)); | |
921 | } | |
922 | entity_inst_t get_hb_front_inst(int osd) const { | |
923 | assert(is_up(osd)); | |
924 | return entity_inst_t(entity_name_t::OSD(osd), get_hb_front_addr(osd)); | |
925 | } | |
926 | ||
927 | const uuid_d& get_uuid(int osd) const { | |
928 | assert(exists(osd)); | |
929 | return (*osd_uuid)[osd]; | |
930 | } | |
931 | ||
932 | const epoch_t& get_up_from(int osd) const { | |
933 | assert(exists(osd)); | |
934 | return osd_info[osd].up_from; | |
935 | } | |
936 | const epoch_t& get_up_thru(int osd) const { | |
937 | assert(exists(osd)); | |
938 | return osd_info[osd].up_thru; | |
939 | } | |
940 | const epoch_t& get_down_at(int osd) const { | |
941 | assert(exists(osd)); | |
942 | return osd_info[osd].down_at; | |
943 | } | |
944 | const osd_info_t& get_info(int osd) const { | |
945 | assert(osd < max_osd); | |
946 | return osd_info[osd]; | |
947 | } | |
948 | ||
949 | const osd_xinfo_t& get_xinfo(int osd) const { | |
950 | assert(osd < max_osd); | |
951 | return osd_xinfo[osd]; | |
952 | } | |
953 | ||
954 | int get_next_up_osd_after(int n) const { | |
955 | if (get_max_osd() == 0) | |
956 | return -1; | |
957 | for (int i = n + 1; i != n; ++i) { | |
958 | if (i >= get_max_osd()) | |
959 | i = 0; | |
960 | if (i == n) | |
961 | break; | |
962 | if (is_up(i)) | |
963 | return i; | |
964 | } | |
965 | return -1; | |
966 | } | |
967 | ||
968 | int get_previous_up_osd_before(int n) const { | |
969 | if (get_max_osd() == 0) | |
970 | return -1; | |
971 | for (int i = n - 1; i != n; --i) { | |
972 | if (i < 0) | |
973 | i = get_max_osd() - 1; | |
974 | if (i == n) | |
975 | break; | |
976 | if (is_up(i)) | |
977 | return i; | |
978 | } | |
979 | return -1; | |
980 | } | |
981 | ||
982 | /** | |
983 | * get feature bits required by the current structure | |
984 | * | |
985 | * @param entity_type [in] what entity type we are asking about | |
986 | * @param mask [out] set of all possible map-related features we could set | |
987 | * @return feature bits used by this map | |
988 | */ | |
989 | uint64_t get_features(int entity_type, uint64_t *mask) const; | |
990 | ||
991 | /** | |
992 | * get oldest *client* version (firefly, hammer, etc.) that can connect given | |
993 | * the feature bits required (according to get_features()). | |
994 | */ | |
31f18b77 | 995 | uint8_t get_min_compat_client() const; |
7c673cae FG |
996 | |
997 | /** | |
998 | * get intersection of features supported by up osds | |
999 | */ | |
1000 | uint64_t get_up_osd_features() const; | |
1001 | ||
94b18763 FG |
1002 | void maybe_remove_pg_upmaps(CephContext *cct, |
1003 | const OSDMap& osdmap, | |
1004 | Incremental *pending_inc); | |
1005 | ||
7c673cae FG |
1006 | int apply_incremental(const Incremental &inc); |
1007 | ||
1008 | /// try to re-use/reference addrs in oldmap from newmap | |
1009 | static void dedup(const OSDMap *oldmap, OSDMap *newmap); | |
1010 | ||
1011 | static void clean_temps(CephContext *cct, const OSDMap& osdmap, | |
1012 | Incremental *pending_inc); | |
1013 | ||
1014 | // serialize, unserialize | |
1015 | private: | |
1016 | void encode_client_old(bufferlist& bl) const; | |
1017 | void encode_classic(bufferlist& bl, uint64_t features) const; | |
1018 | void decode_classic(bufferlist::iterator& p); | |
1019 | void post_decode(); | |
1020 | public: | |
1021 | void encode(bufferlist& bl, uint64_t features=CEPH_FEATURES_ALL) const; | |
1022 | void decode(bufferlist& bl); | |
1023 | void decode(bufferlist::iterator& bl); | |
1024 | ||
1025 | ||
1026 | /**** mapping facilities ****/ | |
1027 | int map_to_pg( | |
1028 | int64_t pool, | |
1029 | const string& name, | |
1030 | const string& key, | |
1031 | const string& nspace, | |
1032 | pg_t *pg) const; | |
1033 | int object_locator_to_pg(const object_t& oid, const object_locator_t& loc, | |
1034 | pg_t &pg) const; | |
1035 | pg_t object_locator_to_pg(const object_t& oid, | |
1036 | const object_locator_t& loc) const { | |
1037 | pg_t pg; | |
1038 | int ret = object_locator_to_pg(oid, loc, pg); | |
1039 | assert(ret == 0); | |
1040 | return pg; | |
1041 | } | |
1042 | ||
1043 | ||
1044 | static object_locator_t file_to_object_locator(const file_layout_t& layout) { | |
1045 | return object_locator_t(layout.pool_id, layout.pool_ns); | |
1046 | } | |
1047 | ||
1048 | ceph_object_layout file_to_object_layout(object_t oid, | |
1049 | file_layout_t& layout) const { | |
1050 | return make_object_layout(oid, layout.pool_id, layout.pool_ns); | |
1051 | } | |
1052 | ||
1053 | ceph_object_layout make_object_layout(object_t oid, int pg_pool, | |
1054 | string nspace) const; | |
1055 | ||
1056 | int get_pg_num(int pg_pool) const | |
1057 | { | |
1058 | const pg_pool_t *pool = get_pg_pool(pg_pool); | |
1059 | assert(NULL != pool); | |
1060 | return pool->get_pg_num(); | |
1061 | } | |
1062 | ||
1063 | bool pg_exists(pg_t pgid) const { | |
1064 | const pg_pool_t *p = get_pg_pool(pgid.pool()); | |
1065 | return p && pgid.ps() < p->get_pg_num(); | |
1066 | } | |
1067 | ||
224ce89b WB |
1068 | int get_pg_pool_min_size(pg_t pgid) const { |
1069 | if (!pg_exists(pgid)) { | |
1070 | return -ENOENT; | |
1071 | } | |
1072 | const pg_pool_t *p = get_pg_pool(pgid.pool()); | |
1073 | assert(p); | |
1074 | return p->get_min_size(); | |
1075 | } | |
1076 | ||
1077 | int get_pg_pool_size(pg_t pgid) const { | |
1078 | if (!pg_exists(pgid)) { | |
1079 | return -ENOENT; | |
1080 | } | |
1081 | const pg_pool_t *p = get_pg_pool(pgid.pool()); | |
1082 | assert(p); | |
1083 | return p->get_size(); | |
1084 | } | |
1085 | ||
94b18763 FG |
1086 | int get_pg_pool_crush_rule(pg_t pgid) const { |
1087 | if (!pg_exists(pgid)) { | |
1088 | return -ENOENT; | |
1089 | } | |
1090 | const pg_pool_t *p = get_pg_pool(pgid.pool()); | |
1091 | assert(p); | |
1092 | return p->get_crush_rule(); | |
1093 | } | |
1094 | ||
7c673cae FG |
1095 | private: |
1096 | /// pg -> (raw osd list) | |
31f18b77 | 1097 | void _pg_to_raw_osds( |
7c673cae FG |
1098 | const pg_pool_t& pool, pg_t pg, |
1099 | vector<int> *osds, | |
1100 | ps_t *ppps) const; | |
1101 | int _pick_primary(const vector<int>& osds) const; | |
1102 | void _remove_nonexistent_osds(const pg_pool_t& pool, vector<int>& osds) const; | |
1103 | ||
1104 | void _apply_primary_affinity(ps_t seed, const pg_pool_t& pool, | |
1105 | vector<int> *osds, int *primary) const; | |
1106 | ||
1107 | /// apply pg_upmap[_items] mappings | |
224ce89b | 1108 | void _apply_upmap(const pg_pool_t& pi, pg_t pg, vector<int> *raw) const; |
7c673cae FG |
1109 | |
1110 | /// pg -> (up osd list) | |
1111 | void _raw_to_up_osds(const pg_pool_t& pool, const vector<int>& raw, | |
1112 | vector<int> *up) const; | |
1113 | ||
1114 | ||
1115 | /** | |
1116 | * Get the pg and primary temp, if they are specified. | |
1117 | * @param temp_pg [out] Will be empty or contain the temp PG mapping on return | |
1118 | * @param temp_primary [out] Will be the value in primary_temp, or a value derived | |
1119 | * from the pg_temp (if specified), or -1 if you should use the calculated (up_)primary. | |
1120 | */ | |
1121 | void _get_temp_osds(const pg_pool_t& pool, pg_t pg, | |
1122 | vector<int> *temp_pg, int *temp_primary) const; | |
1123 | ||
1124 | /** | |
1125 | * map to up and acting. Fills in whatever fields are non-NULL. | |
1126 | */ | |
1127 | void _pg_to_up_acting_osds(const pg_t& pg, vector<int> *up, int *up_primary, | |
1128 | vector<int> *acting, int *acting_primary, | |
1129 | bool raw_pg_to_pg = true) const; | |
1130 | ||
1131 | public: | |
1132 | /*** | |
1133 | * This is suitable only for looking at raw CRUSH outputs. It skips | |
1134 | * applying the temp and up checks and should not be used | |
1135 | * by anybody for data mapping purposes. | |
1136 | * raw and primary must be non-NULL | |
1137 | */ | |
31f18b77 | 1138 | void pg_to_raw_osds(pg_t pg, vector<int> *raw, int *primary) const; |
7c673cae | 1139 | /// map a pg to its acting set. @return acting set size |
31f18b77 | 1140 | void pg_to_acting_osds(const pg_t& pg, vector<int> *acting, |
7c673cae FG |
1141 | int *acting_primary) const { |
1142 | _pg_to_up_acting_osds(pg, NULL, NULL, acting, acting_primary); | |
7c673cae | 1143 | } |
31f18b77 | 1144 | void pg_to_acting_osds(pg_t pg, vector<int>& acting) const { |
7c673cae FG |
1145 | return pg_to_acting_osds(pg, &acting, NULL); |
1146 | } | |
1147 | /** | |
1148 | * This does not apply temp overrides and should not be used | |
1149 | * by anybody for data mapping purposes. Specify both pointers. | |
1150 | */ | |
1151 | void pg_to_raw_up(pg_t pg, vector<int> *up, int *primary) const; | |
1152 | /** | |
1153 | * map a pg to its acting set as well as its up set. You must use | |
1154 | * the acting set for data mapping purposes, but some users will | |
1155 | * also find the up set useful for things like deciding what to | |
1156 | * set as pg_temp. | |
1157 | * Each of these pointers must be non-NULL. | |
1158 | */ | |
1159 | void pg_to_up_acting_osds(pg_t pg, vector<int> *up, int *up_primary, | |
1160 | vector<int> *acting, int *acting_primary) const { | |
1161 | _pg_to_up_acting_osds(pg, up, up_primary, acting, acting_primary); | |
1162 | } | |
1163 | void pg_to_up_acting_osds(pg_t pg, vector<int>& up, vector<int>& acting) const { | |
1164 | int up_primary, acting_primary; | |
1165 | pg_to_up_acting_osds(pg, &up, &up_primary, &acting, &acting_primary); | |
1166 | } | |
1167 | bool pg_is_ec(pg_t pg) const { | |
1168 | auto i = pools.find(pg.pool()); | |
1169 | assert(i != pools.end()); | |
1170 | return i->second.ec_pool(); | |
1171 | } | |
1172 | bool get_primary_shard(const pg_t& pgid, spg_t *out) const { | |
1173 | auto i = get_pools().find(pgid.pool()); | |
1174 | if (i == get_pools().end()) { | |
1175 | return false; | |
1176 | } | |
1177 | if (!i->second.ec_pool()) { | |
1178 | *out = spg_t(pgid); | |
1179 | return true; | |
1180 | } | |
1181 | int primary; | |
1182 | vector<int> acting; | |
1183 | pg_to_acting_osds(pgid, &acting, &primary); | |
1184 | for (uint8_t i = 0; i < acting.size(); ++i) { | |
1185 | if (acting[i] == primary) { | |
1186 | *out = spg_t(pgid, shard_id_t(i)); | |
1187 | return true; | |
1188 | } | |
1189 | } | |
1190 | return false; | |
1191 | } | |
1192 | ||
1193 | int64_t lookup_pg_pool_name(const string& name) const { | |
1194 | auto p = name_pool.find(name); | |
1195 | if (p == name_pool.end()) | |
1196 | return -ENOENT; | |
1197 | return p->second; | |
1198 | } | |
1199 | ||
1200 | int64_t get_pool_max() const { | |
1201 | return pool_max; | |
1202 | } | |
1203 | const mempool::osdmap::map<int64_t,pg_pool_t>& get_pools() const { | |
1204 | return pools; | |
1205 | } | |
1206 | mempool::osdmap::map<int64_t,pg_pool_t>& get_pools() { | |
1207 | return pools; | |
1208 | } | |
3efd9988 FG |
1209 | void get_pool_ids_by_rule(int rule_id, set<int64_t> *pool_ids) const { |
1210 | assert(pool_ids); | |
1211 | for (auto &p: pools) { | |
1212 | if ((int)p.second.get_crush_rule() == rule_id) { | |
1213 | pool_ids->insert(p.first); | |
1214 | } | |
1215 | } | |
1216 | } | |
1217 | void get_pool_ids_by_osd(CephContext *cct, | |
1218 | int osd, | |
1219 | set<int64_t> *pool_ids) const; | |
7c673cae FG |
1220 | const string& get_pool_name(int64_t p) const { |
1221 | auto i = pool_name.find(p); | |
1222 | assert(i != pool_name.end()); | |
1223 | return i->second; | |
1224 | } | |
c07f9fc5 FG |
1225 | const mempool::osdmap::map<int64_t,string>& get_pool_names() const { |
1226 | return pool_name; | |
1227 | } | |
7c673cae FG |
1228 | bool have_pg_pool(int64_t p) const { |
1229 | return pools.count(p); | |
1230 | } | |
1231 | const pg_pool_t* get_pg_pool(int64_t p) const { | |
1232 | auto i = pools.find(p); | |
1233 | if (i != pools.end()) | |
1234 | return &i->second; | |
1235 | return NULL; | |
1236 | } | |
1237 | unsigned get_pg_size(pg_t pg) const { | |
1238 | auto p = pools.find(pg.pool()); | |
1239 | assert(p != pools.end()); | |
1240 | return p->second.get_size(); | |
1241 | } | |
1242 | int get_pg_type(pg_t pg) const { | |
1243 | auto p = pools.find(pg.pool()); | |
1244 | assert(p != pools.end()); | |
1245 | return p->second.get_type(); | |
1246 | } | |
1247 | ||
1248 | ||
1249 | pg_t raw_pg_to_pg(pg_t pg) const { | |
1250 | auto p = pools.find(pg.pool()); | |
1251 | assert(p != pools.end()); | |
1252 | return p->second.raw_pg_to_pg(pg); | |
1253 | } | |
1254 | ||
1255 | // pg -> acting primary osd | |
1256 | int get_pg_acting_primary(pg_t pg) const { | |
1257 | int primary = -1; | |
1258 | _pg_to_up_acting_osds(pg, nullptr, nullptr, nullptr, &primary); | |
1259 | return primary; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * check whether an spg_t maps to a particular osd | |
1264 | */ | |
1265 | bool is_up_acting_osd_shard(spg_t pg, int osd) const { | |
1266 | vector<int> up, acting; | |
1267 | _pg_to_up_acting_osds(pg.pgid, &up, NULL, &acting, NULL, false); | |
1268 | if (pg.shard == shard_id_t::NO_SHARD) { | |
1269 | if (calc_pg_role(osd, acting, acting.size()) >= 0 || | |
1270 | calc_pg_role(osd, up, up.size()) >= 0) | |
1271 | return true; | |
1272 | } else { | |
1273 | if (pg.shard < (int)acting.size() && acting[pg.shard] == osd) | |
1274 | return true; | |
1275 | if (pg.shard < (int)up.size() && up[pg.shard] == osd) | |
1276 | return true; | |
1277 | } | |
1278 | return false; | |
1279 | } | |
1280 | ||
1281 | ||
1282 | /* what replica # is a given osd? 0 primary, -1 for none. */ | |
1283 | static int calc_pg_rank(int osd, const vector<int>& acting, int nrep=0); | |
1284 | static int calc_pg_role(int osd, const vector<int>& acting, int nrep=0); | |
1285 | static bool primary_changed( | |
1286 | int oldprimary, | |
1287 | const vector<int> &oldacting, | |
1288 | int newprimary, | |
1289 | const vector<int> &newacting); | |
1290 | ||
1291 | /* rank is -1 (stray), 0 (primary), 1,2,3,... (replica) */ | |
1292 | int get_pg_acting_rank(pg_t pg, int osd) const { | |
1293 | vector<int> group; | |
31f18b77 FG |
1294 | pg_to_acting_osds(pg, group); |
1295 | return calc_pg_rank(osd, group, group.size()); | |
7c673cae FG |
1296 | } |
1297 | /* role is -1 (stray), 0 (primary), 1 (replica) */ | |
1298 | int get_pg_acting_role(const pg_t& pg, int osd) const { | |
1299 | vector<int> group; | |
31f18b77 FG |
1300 | pg_to_acting_osds(pg, group); |
1301 | return calc_pg_role(osd, group, group.size()); | |
7c673cae FG |
1302 | } |
1303 | ||
1304 | bool osd_is_valid_op_target(pg_t pg, int osd) const { | |
1305 | int primary; | |
1306 | vector<int> group; | |
31f18b77 | 1307 | pg_to_acting_osds(pg, &group, &primary); |
7c673cae FG |
1308 | if (osd == primary) |
1309 | return true; | |
1310 | if (pg_is_ec(pg)) | |
1311 | return false; | |
1312 | ||
31f18b77 | 1313 | return calc_pg_role(osd, group, group.size()) >= 0; |
7c673cae FG |
1314 | } |
1315 | ||
1316 | int clean_pg_upmaps( | |
1317 | CephContext *cct, | |
1318 | Incremental *pending_inc); | |
1319 | ||
1320 | bool try_pg_upmap( | |
1321 | CephContext *cct, | |
1322 | pg_t pg, ///< pg to potentially remap | |
1323 | const set<int>& overfull, ///< osds we'd want to evacuate | |
1324 | const vector<int>& underfull, ///< osds to move to, in order of preference | |
1325 | vector<int> *orig, | |
1326 | vector<int> *out); ///< resulting alternative mapping | |
1327 | ||
1328 | int calc_pg_upmaps( | |
1329 | CephContext *cct, | |
1330 | float max_deviation, ///< max deviation from target (value < 1.0) | |
1331 | int max_iterations, ///< max iterations to run | |
1332 | const set<int64_t>& pools, ///< [optional] restrict to pool | |
1333 | Incremental *pending_inc | |
1334 | ); | |
1335 | ||
31f18b77 FG |
1336 | int get_osds_by_bucket_name(const string &name, set<int> *osds) const; |
1337 | ||
7c673cae FG |
1338 | /* |
1339 | * handy helpers to build simple maps... | |
1340 | */ | |
1341 | /** | |
1342 | * Build an OSD map suitable for basic usage. If **num_osd** is >= 0 | |
1343 | * it will be initialized with the specified number of OSDs in a | |
1344 | * single host. If **num_osd** is < 0 the layout of the OSD map will | |
1345 | * be built by reading the content of the configuration file. | |
1346 | * | |
1347 | * @param cct [in] in core ceph context | |
1348 | * @param e [in] initial epoch | |
1349 | * @param fsid [in] id of the cluster | |
1350 | * @param num_osd [in] number of OSDs if >= 0 or read from conf if < 0 | |
1351 | * @return **0** on success, negative errno on error. | |
1352 | */ | |
224ce89b WB |
1353 | private: |
1354 | int build_simple_optioned(CephContext *cct, epoch_t e, uuid_d &fsid, | |
1355 | int num_osd, int pg_bits, int pgp_bits, | |
1356 | bool default_pool); | |
1357 | public: | |
7c673cae | 1358 | int build_simple(CephContext *cct, epoch_t e, uuid_d &fsid, |
224ce89b WB |
1359 | int num_osd) { |
1360 | return build_simple_optioned(cct, e, fsid, num_osd, 0, 0, false); | |
1361 | } | |
1362 | int build_simple_with_pool(CephContext *cct, epoch_t e, uuid_d &fsid, | |
1363 | int num_osd, int pg_bits, int pgp_bits) { | |
1364 | return build_simple_optioned(cct, e, fsid, num_osd, | |
1365 | pg_bits, pgp_bits, true); | |
1366 | } | |
7c673cae FG |
1367 | static int _build_crush_types(CrushWrapper& crush); |
1368 | static int build_simple_crush_map(CephContext *cct, CrushWrapper& crush, | |
1369 | int num_osd, ostream *ss); | |
1370 | static int build_simple_crush_map_from_conf(CephContext *cct, | |
1371 | CrushWrapper& crush, | |
1372 | ostream *ss); | |
31f18b77 FG |
1373 | static int build_simple_crush_rules( |
1374 | CephContext *cct, CrushWrapper& crush, | |
1375 | const string& root, | |
1376 | ostream *ss); | |
7c673cae | 1377 | |
3efd9988 FG |
1378 | bool crush_rule_in_use(int rule_id) const; |
1379 | ||
1380 | int validate_crush_rules(CrushWrapper *crush, ostream *ss) const; | |
7c673cae FG |
1381 | |
1382 | void clear_temp() { | |
1383 | pg_temp->clear(); | |
1384 | primary_temp->clear(); | |
1385 | } | |
1386 | ||
1387 | private: | |
1388 | void print_osd_line(int cur, ostream *out, Formatter *f) const; | |
1389 | public: | |
1390 | void print(ostream& out) const; | |
1391 | void print_pools(ostream& out) const; | |
224ce89b | 1392 | void print_summary(Formatter *f, ostream& out, const string& prefix) const; |
7c673cae | 1393 | void print_oneline_summary(ostream& out) const; |
31f18b77 FG |
1394 | |
1395 | enum { | |
c07f9fc5 FG |
1396 | DUMP_IN = 1, // only 'in' osds |
1397 | DUMP_OUT = 2, // only 'out' osds | |
1398 | DUMP_UP = 4, // only 'up' osds | |
1399 | DUMP_DOWN = 8, // only 'down' osds | |
1400 | DUMP_DESTROYED = 16, // only 'destroyed' osds | |
31f18b77 FG |
1401 | }; |
1402 | void print_tree(Formatter *f, ostream *out, unsigned dump_flags=0) const; | |
7c673cae FG |
1403 | |
1404 | int summarize_mapping_stats( | |
1405 | OSDMap *newmap, | |
1406 | const set<int64_t> *pools, | |
1407 | std::string *out, | |
1408 | Formatter *f) const; | |
1409 | ||
1410 | string get_flag_string() const; | |
1411 | static string get_flag_string(unsigned flags); | |
1412 | static void dump_erasure_code_profiles( | |
1413 | const mempool::osdmap::map<string,map<string,string> > &profiles, | |
1414 | Formatter *f); | |
1415 | void dump(Formatter *f) const; | |
1416 | static void generate_test_instances(list<OSDMap*>& o); | |
1417 | bool check_new_blacklist_entries() const { return new_blacklist_entries; } | |
224ce89b WB |
1418 | |
1419 | void check_health(health_check_map_t *checks) const; | |
35e4c445 FG |
1420 | |
1421 | int parse_osd_id_list(const vector<string>& ls, | |
1422 | set<int> *out, | |
1423 | ostream *ss) const; | |
7c673cae FG |
1424 | }; |
1425 | WRITE_CLASS_ENCODER_FEATURES(OSDMap) | |
1426 | WRITE_CLASS_ENCODER_FEATURES(OSDMap::Incremental) | |
1427 | ||
1428 | typedef ceph::shared_ptr<const OSDMap> OSDMapRef; | |
1429 | ||
1430 | inline ostream& operator<<(ostream& out, const OSDMap& m) { | |
1431 | m.print_oneline_summary(out); | |
1432 | return out; | |
1433 | } | |
1434 | ||
31f18b77 FG |
1435 | class PGStatService; |
1436 | ||
1437 | void print_osd_utilization(const OSDMap& osdmap, | |
1438 | const PGStatService *pgstat, | |
1439 | ostream& out, | |
1440 | Formatter *f, | |
1441 | bool tree); | |
7c673cae FG |
1442 | |
1443 | #endif |