]>
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 | #include "OSDMap.h" | |
19 | #include <algorithm> | |
20 | #include "common/config.h" | |
21 | #include "common/Formatter.h" | |
22 | #include "common/TextTable.h" | |
23 | #include "include/ceph_features.h" | |
24 | #include "include/str_map.h" | |
25 | ||
26 | #include "common/code_environment.h" | |
27 | ||
28 | #include "crush/CrushTreeDumper.h" | |
29 | #include "common/Clock.h" | |
30 | ||
31 | #define dout_subsys ceph_subsys_osd | |
32 | ||
33 | MEMPOOL_DEFINE_OBJECT_FACTORY(OSDMap, osdmap, osdmap); | |
34 | MEMPOOL_DEFINE_OBJECT_FACTORY(OSDMap::Incremental, osdmap_inc, osdmap); | |
35 | ||
36 | ||
37 | // ---------------------------------- | |
38 | // osd_info_t | |
39 | ||
40 | void osd_info_t::dump(Formatter *f) const | |
41 | { | |
42 | f->dump_int("last_clean_begin", last_clean_begin); | |
43 | f->dump_int("last_clean_end", last_clean_end); | |
44 | f->dump_int("up_from", up_from); | |
45 | f->dump_int("up_thru", up_thru); | |
46 | f->dump_int("down_at", down_at); | |
47 | f->dump_int("lost_at", lost_at); | |
48 | } | |
49 | ||
50 | void osd_info_t::encode(bufferlist& bl) const | |
51 | { | |
52 | __u8 struct_v = 1; | |
53 | ::encode(struct_v, bl); | |
54 | ::encode(last_clean_begin, bl); | |
55 | ::encode(last_clean_end, bl); | |
56 | ::encode(up_from, bl); | |
57 | ::encode(up_thru, bl); | |
58 | ::encode(down_at, bl); | |
59 | ::encode(lost_at, bl); | |
60 | } | |
61 | ||
62 | void osd_info_t::decode(bufferlist::iterator& bl) | |
63 | { | |
64 | __u8 struct_v; | |
65 | ::decode(struct_v, bl); | |
66 | ::decode(last_clean_begin, bl); | |
67 | ::decode(last_clean_end, bl); | |
68 | ::decode(up_from, bl); | |
69 | ::decode(up_thru, bl); | |
70 | ::decode(down_at, bl); | |
71 | ::decode(lost_at, bl); | |
72 | } | |
73 | ||
74 | void osd_info_t::generate_test_instances(list<osd_info_t*>& o) | |
75 | { | |
76 | o.push_back(new osd_info_t); | |
77 | o.push_back(new osd_info_t); | |
78 | o.back()->last_clean_begin = 1; | |
79 | o.back()->last_clean_end = 2; | |
80 | o.back()->up_from = 30; | |
81 | o.back()->up_thru = 40; | |
82 | o.back()->down_at = 5; | |
83 | o.back()->lost_at = 6; | |
84 | } | |
85 | ||
86 | ostream& operator<<(ostream& out, const osd_info_t& info) | |
87 | { | |
88 | out << "up_from " << info.up_from | |
89 | << " up_thru " << info.up_thru | |
90 | << " down_at " << info.down_at | |
91 | << " last_clean_interval [" << info.last_clean_begin << "," << info.last_clean_end << ")"; | |
92 | if (info.lost_at) | |
93 | out << " lost_at " << info.lost_at; | |
94 | return out; | |
95 | } | |
96 | ||
97 | // ---------------------------------- | |
98 | // osd_xinfo_t | |
99 | ||
100 | void osd_xinfo_t::dump(Formatter *f) const | |
101 | { | |
102 | f->dump_stream("down_stamp") << down_stamp; | |
103 | f->dump_float("laggy_probability", laggy_probability); | |
104 | f->dump_int("laggy_interval", laggy_interval); | |
105 | f->dump_int("features", features); | |
106 | f->dump_unsigned("old_weight", old_weight); | |
107 | } | |
108 | ||
109 | void osd_xinfo_t::encode(bufferlist& bl) const | |
110 | { | |
111 | ENCODE_START(3, 1, bl); | |
112 | ::encode(down_stamp, bl); | |
113 | __u32 lp = laggy_probability * 0xfffffffful; | |
114 | ::encode(lp, bl); | |
115 | ::encode(laggy_interval, bl); | |
116 | ::encode(features, bl); | |
117 | ::encode(old_weight, bl); | |
118 | ENCODE_FINISH(bl); | |
119 | } | |
120 | ||
121 | void osd_xinfo_t::decode(bufferlist::iterator& bl) | |
122 | { | |
123 | DECODE_START(3, bl); | |
124 | ::decode(down_stamp, bl); | |
125 | __u32 lp; | |
126 | ::decode(lp, bl); | |
127 | laggy_probability = (float)lp / (float)0xffffffff; | |
128 | ::decode(laggy_interval, bl); | |
129 | if (struct_v >= 2) | |
130 | ::decode(features, bl); | |
131 | else | |
132 | features = 0; | |
133 | if (struct_v >= 3) | |
134 | ::decode(old_weight, bl); | |
135 | else | |
136 | old_weight = 0; | |
137 | DECODE_FINISH(bl); | |
138 | } | |
139 | ||
140 | void osd_xinfo_t::generate_test_instances(list<osd_xinfo_t*>& o) | |
141 | { | |
142 | o.push_back(new osd_xinfo_t); | |
143 | o.push_back(new osd_xinfo_t); | |
144 | o.back()->down_stamp = utime_t(2, 3); | |
145 | o.back()->laggy_probability = .123; | |
146 | o.back()->laggy_interval = 123456; | |
147 | o.back()->old_weight = 0x7fff; | |
148 | } | |
149 | ||
150 | ostream& operator<<(ostream& out, const osd_xinfo_t& xi) | |
151 | { | |
152 | return out << "down_stamp " << xi.down_stamp | |
153 | << " laggy_probability " << xi.laggy_probability | |
154 | << " laggy_interval " << xi.laggy_interval | |
155 | << " old_weight " << xi.old_weight; | |
156 | } | |
157 | ||
158 | // ---------------------------------- | |
159 | // OSDMap::Incremental | |
160 | ||
161 | int OSDMap::Incremental::get_net_marked_out(const OSDMap *previous) const | |
162 | { | |
163 | int n = 0; | |
164 | for (auto &weight : new_weight) { | |
165 | if (weight.second == CEPH_OSD_OUT && !previous->is_out(weight.first)) | |
166 | n++; // marked out | |
167 | else if (weight.second != CEPH_OSD_OUT && previous->is_out(weight.first)) | |
168 | n--; // marked in | |
169 | } | |
170 | return n; | |
171 | } | |
172 | ||
173 | int OSDMap::Incremental::get_net_marked_down(const OSDMap *previous) const | |
174 | { | |
175 | int n = 0; | |
176 | for (auto &state : new_state) { // | |
177 | if (state.second & CEPH_OSD_UP) { | |
178 | if (previous->is_up(state.first)) | |
179 | n++; // marked down | |
180 | else | |
181 | n--; // marked up | |
182 | } | |
183 | } | |
184 | return n; | |
185 | } | |
186 | ||
187 | int OSDMap::Incremental::identify_osd(uuid_d u) const | |
188 | { | |
189 | for (auto &uuid : new_uuid) | |
190 | if (uuid.second == u) | |
191 | return uuid.first; | |
192 | return -1; | |
193 | } | |
194 | ||
195 | int OSDMap::Incremental::propagate_snaps_to_tiers(CephContext *cct, | |
196 | const OSDMap& osdmap) | |
197 | { | |
198 | assert(epoch == osdmap.get_epoch() + 1); | |
199 | ||
200 | for (auto &new_pool : new_pools) { | |
201 | if (!new_pool.second.tiers.empty()) { | |
202 | pg_pool_t& base = new_pool.second; | |
203 | ||
204 | for (const auto &tier_pool : base.tiers) { | |
205 | const auto &r = new_pools.find(tier_pool); | |
206 | pg_pool_t *tier = 0; | |
207 | if (r == new_pools.end()) { | |
208 | const pg_pool_t *orig = osdmap.get_pg_pool(tier_pool); | |
209 | if (!orig) { | |
210 | lderr(cct) << __func__ << " no pool " << tier_pool << dendl; | |
211 | return -EIO; | |
212 | } | |
213 | tier = get_new_pool(tier_pool, orig); | |
214 | } else { | |
215 | tier = &r->second; | |
216 | } | |
217 | if (tier->tier_of != new_pool.first) { | |
218 | lderr(cct) << __func__ << " " << r->first << " tier_of != " << new_pool.first << dendl; | |
219 | return -EIO; | |
220 | } | |
221 | ||
222 | ldout(cct, 10) << __func__ << " from " << new_pool.first << " to " | |
223 | << tier_pool << dendl; | |
224 | tier->snap_seq = base.snap_seq; | |
225 | tier->snap_epoch = base.snap_epoch; | |
226 | tier->snaps = base.snaps; | |
227 | tier->removed_snaps = base.removed_snaps; | |
228 | } | |
229 | } | |
230 | } | |
231 | return 0; | |
232 | } | |
233 | ||
234 | ||
235 | bool OSDMap::subtree_is_down(int id, set<int> *down_cache) const | |
236 | { | |
237 | if (id >= 0) | |
238 | return is_down(id); | |
239 | ||
240 | if (down_cache && | |
241 | down_cache->count(id)) { | |
242 | return true; | |
243 | } | |
244 | ||
245 | list<int> children; | |
246 | crush->get_children(id, &children); | |
247 | for (const auto &child : children) { | |
248 | if (!subtree_is_down(child, down_cache)) { | |
249 | return false; | |
250 | } | |
251 | } | |
252 | if (down_cache) { | |
253 | down_cache->insert(id); | |
254 | } | |
255 | return true; | |
256 | } | |
257 | ||
258 | bool OSDMap::containing_subtree_is_down(CephContext *cct, int id, int subtree_type, set<int> *down_cache) const | |
259 | { | |
260 | // use a stack-local down_cache if we didn't get one from the | |
261 | // caller. then at least this particular call will avoid duplicated | |
262 | // work. | |
263 | set<int> local_down_cache; | |
264 | if (!down_cache) { | |
265 | down_cache = &local_down_cache; | |
266 | } | |
267 | ||
268 | int current = id; | |
269 | while (true) { | |
270 | int type; | |
271 | if (current >= 0) { | |
272 | type = 0; | |
273 | } else { | |
274 | type = crush->get_bucket_type(current); | |
275 | } | |
276 | assert(type >= 0); | |
277 | ||
278 | if (!subtree_is_down(current, down_cache)) { | |
279 | ldout(cct, 30) << "containing_subtree_is_down(" << id << ") = false" << dendl; | |
280 | return false; | |
281 | } | |
282 | ||
283 | // is this a big enough subtree to be marked as down? | |
284 | if (type >= subtree_type) { | |
285 | ldout(cct, 30) << "containing_subtree_is_down(" << id << ") = true ... " << type << " >= " << subtree_type << dendl; | |
286 | return true; | |
287 | } | |
288 | ||
289 | int r = crush->get_immediate_parent_id(current, ¤t); | |
290 | if (r < 0) { | |
291 | return false; | |
292 | } | |
293 | } | |
294 | } | |
295 | ||
296 | void OSDMap::Incremental::encode_client_old(bufferlist& bl) const | |
297 | { | |
298 | __u16 v = 5; | |
299 | ::encode(v, bl); | |
300 | ::encode(fsid, bl); | |
301 | ::encode(epoch, bl); | |
302 | ::encode(modified, bl); | |
303 | int32_t new_t = new_pool_max; | |
304 | ::encode(new_t, bl); | |
305 | ::encode(new_flags, bl); | |
306 | ::encode(fullmap, bl); | |
307 | ::encode(crush, bl); | |
308 | ||
309 | ::encode(new_max_osd, bl); | |
310 | // for ::encode(new_pools, bl); | |
311 | __u32 n = new_pools.size(); | |
312 | ::encode(n, bl); | |
313 | for (const auto &new_pool : new_pools) { | |
314 | n = new_pool.first; | |
315 | ::encode(n, bl); | |
316 | ::encode(new_pool.second, bl, 0); | |
317 | } | |
318 | // for ::encode(new_pool_names, bl); | |
319 | n = new_pool_names.size(); | |
320 | ::encode(n, bl); | |
321 | ||
322 | for (const auto &new_pool_name : new_pool_names) { | |
323 | n = new_pool_name.first; | |
324 | ::encode(n, bl); | |
325 | ::encode(new_pool_name.second, bl); | |
326 | } | |
327 | // for ::encode(old_pools, bl); | |
328 | n = old_pools.size(); | |
329 | ::encode(n, bl); | |
330 | for (auto &old_pool : old_pools) { | |
331 | n = old_pool; | |
332 | ::encode(n, bl); | |
333 | } | |
334 | ::encode(new_up_client, bl, 0); | |
335 | ::encode(new_state, bl); | |
336 | ::encode(new_weight, bl); | |
337 | // for ::encode(new_pg_temp, bl); | |
338 | n = new_pg_temp.size(); | |
339 | ::encode(n, bl); | |
340 | ||
341 | for (const auto &pg_temp : new_pg_temp) { | |
342 | old_pg_t opg = pg_temp.first.get_old_pg(); | |
343 | ::encode(opg, bl); | |
344 | ::encode(pg_temp.second, bl); | |
345 | } | |
346 | } | |
347 | ||
348 | void OSDMap::Incremental::encode_classic(bufferlist& bl, uint64_t features) const | |
349 | { | |
350 | if ((features & CEPH_FEATURE_PGID64) == 0) { | |
351 | encode_client_old(bl); | |
352 | return; | |
353 | } | |
354 | ||
355 | // base | |
356 | __u16 v = 6; | |
357 | ::encode(v, bl); | |
358 | ::encode(fsid, bl); | |
359 | ::encode(epoch, bl); | |
360 | ::encode(modified, bl); | |
361 | ::encode(new_pool_max, bl); | |
362 | ::encode(new_flags, bl); | |
363 | ::encode(fullmap, bl); | |
364 | ::encode(crush, bl); | |
365 | ||
366 | ::encode(new_max_osd, bl); | |
367 | ::encode(new_pools, bl, features); | |
368 | ::encode(new_pool_names, bl); | |
369 | ::encode(old_pools, bl); | |
370 | ::encode(new_up_client, bl, features); | |
371 | ::encode(new_state, bl); | |
372 | ::encode(new_weight, bl); | |
373 | ::encode(new_pg_temp, bl); | |
374 | ||
375 | // extended | |
376 | __u16 ev = 10; | |
377 | ::encode(ev, bl); | |
378 | ::encode(new_hb_back_up, bl, features); | |
379 | ::encode(new_up_thru, bl); | |
380 | ::encode(new_last_clean_interval, bl); | |
381 | ::encode(new_lost, bl); | |
382 | ::encode(new_blacklist, bl, features); | |
383 | ::encode(old_blacklist, bl, features); | |
384 | ::encode(new_up_cluster, bl, features); | |
385 | ::encode(cluster_snapshot, bl); | |
386 | ::encode(new_uuid, bl); | |
387 | ::encode(new_xinfo, bl); | |
388 | ::encode(new_hb_front_up, bl, features); | |
389 | } | |
390 | ||
391 | void OSDMap::Incremental::encode(bufferlist& bl, uint64_t features) const | |
392 | { | |
393 | if ((features & CEPH_FEATURE_OSDMAP_ENC) == 0) { | |
394 | encode_classic(bl, features); | |
395 | return; | |
396 | } | |
397 | ||
398 | // only a select set of callers should *ever* be encoding new | |
399 | // OSDMaps. others should be passing around the canonical encoded | |
400 | // buffers from on high. select out those callers by passing in an | |
401 | // "impossible" feature bit. | |
402 | assert(features & CEPH_FEATURE_RESERVED); | |
403 | features &= ~CEPH_FEATURE_RESERVED; | |
404 | ||
405 | size_t start_offset = bl.length(); | |
406 | size_t tail_offset; | |
407 | buffer::list::iterator crc_it; | |
408 | ||
409 | // meta-encoding: how we include client-used and osd-specific data | |
410 | ENCODE_START(8, 7, bl); | |
411 | ||
412 | { | |
413 | uint8_t v = 4; | |
414 | if (!HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
415 | v = 3; | |
416 | } | |
417 | ENCODE_START(v, 1, bl); // client-usable data | |
418 | ::encode(fsid, bl); | |
419 | ::encode(epoch, bl); | |
420 | ::encode(modified, bl); | |
421 | ::encode(new_pool_max, bl); | |
422 | ::encode(new_flags, bl); | |
423 | ::encode(fullmap, bl); | |
424 | ::encode(crush, bl); | |
425 | ||
426 | ::encode(new_max_osd, bl); | |
427 | ::encode(new_pools, bl, features); | |
428 | ::encode(new_pool_names, bl); | |
429 | ::encode(old_pools, bl); | |
430 | ::encode(new_up_client, bl, features); | |
431 | ::encode(new_state, bl); | |
432 | ::encode(new_weight, bl); | |
433 | ::encode(new_pg_temp, bl); | |
434 | ::encode(new_primary_temp, bl); | |
435 | ::encode(new_primary_affinity, bl); | |
436 | ::encode(new_erasure_code_profiles, bl); | |
437 | ::encode(old_erasure_code_profiles, bl); | |
438 | if (v >= 4) { | |
439 | ::encode(new_pg_upmap, bl); | |
440 | ::encode(old_pg_upmap, bl); | |
441 | ::encode(new_pg_upmap_items, bl); | |
442 | ::encode(old_pg_upmap_items, bl); | |
443 | } | |
444 | ENCODE_FINISH(bl); // client-usable data | |
445 | } | |
446 | ||
447 | { | |
448 | uint8_t target_v = 5; | |
449 | if (!HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
450 | target_v = 2; | |
451 | } | |
452 | ENCODE_START(target_v, 1, bl); // extended, osd-only data | |
453 | ::encode(new_hb_back_up, bl, features); | |
454 | ::encode(new_up_thru, bl); | |
455 | ::encode(new_last_clean_interval, bl); | |
456 | ::encode(new_lost, bl); | |
457 | ::encode(new_blacklist, bl, features); | |
458 | ::encode(old_blacklist, bl, features); | |
459 | ::encode(new_up_cluster, bl, features); | |
460 | ::encode(cluster_snapshot, bl); | |
461 | ::encode(new_uuid, bl); | |
462 | ::encode(new_xinfo, bl); | |
463 | ::encode(new_hb_front_up, bl, features); | |
464 | ::encode(features, bl); // NOTE: features arg, not the member | |
465 | if (target_v >= 3) { | |
466 | ::encode(new_nearfull_ratio, bl); | |
467 | ::encode(new_full_ratio, bl); | |
468 | ::encode(new_backfillfull_ratio, bl); | |
469 | ::encode(new_require_min_compat_client, bl); | |
470 | } | |
471 | ENCODE_FINISH(bl); // osd-only data | |
472 | } | |
473 | ||
474 | ::encode((uint32_t)0, bl); // dummy inc_crc | |
475 | crc_it = bl.end(); | |
476 | crc_it.advance(-4); | |
477 | tail_offset = bl.length(); | |
478 | ||
479 | ::encode(full_crc, bl); | |
480 | ||
481 | ENCODE_FINISH(bl); // meta-encoding wrapper | |
482 | ||
483 | // fill in crc | |
484 | bufferlist front; | |
485 | front.substr_of(bl, start_offset, crc_it.get_off() - start_offset); | |
486 | inc_crc = front.crc32c(-1); | |
487 | bufferlist tail; | |
488 | tail.substr_of(bl, tail_offset, bl.length() - tail_offset); | |
489 | inc_crc = tail.crc32c(inc_crc); | |
490 | ceph_le32 crc_le; | |
491 | crc_le = inc_crc; | |
492 | crc_it.copy_in(4, (char*)&crc_le); | |
493 | have_crc = true; | |
494 | } | |
495 | ||
496 | void OSDMap::Incremental::decode_classic(bufferlist::iterator &p) | |
497 | { | |
498 | __u32 n, t; | |
499 | // base | |
500 | __u16 v; | |
501 | ::decode(v, p); | |
502 | ::decode(fsid, p); | |
503 | ::decode(epoch, p); | |
504 | ::decode(modified, p); | |
505 | if (v == 4 || v == 5) { | |
506 | ::decode(n, p); | |
507 | new_pool_max = n; | |
508 | } else if (v >= 6) | |
509 | ::decode(new_pool_max, p); | |
510 | ::decode(new_flags, p); | |
511 | ::decode(fullmap, p); | |
512 | ::decode(crush, p); | |
513 | ||
514 | ::decode(new_max_osd, p); | |
515 | if (v < 6) { | |
516 | new_pools.clear(); | |
517 | ::decode(n, p); | |
518 | while (n--) { | |
519 | ::decode(t, p); | |
520 | ::decode(new_pools[t], p); | |
521 | } | |
522 | } else { | |
523 | ::decode(new_pools, p); | |
524 | } | |
525 | if (v == 5) { | |
526 | new_pool_names.clear(); | |
527 | ::decode(n, p); | |
528 | while (n--) { | |
529 | ::decode(t, p); | |
530 | ::decode(new_pool_names[t], p); | |
531 | } | |
532 | } else if (v >= 6) { | |
533 | ::decode(new_pool_names, p); | |
534 | } | |
535 | if (v < 6) { | |
536 | old_pools.clear(); | |
537 | ::decode(n, p); | |
538 | while (n--) { | |
539 | ::decode(t, p); | |
540 | old_pools.insert(t); | |
541 | } | |
542 | } else { | |
543 | ::decode(old_pools, p); | |
544 | } | |
545 | ::decode(new_up_client, p); | |
546 | ::decode(new_state, p); | |
547 | ::decode(new_weight, p); | |
548 | ||
549 | if (v < 6) { | |
550 | new_pg_temp.clear(); | |
551 | ::decode(n, p); | |
552 | while (n--) { | |
553 | old_pg_t opg; | |
554 | ::decode_raw(opg, p); | |
555 | ::decode(new_pg_temp[pg_t(opg)], p); | |
556 | } | |
557 | } else { | |
558 | ::decode(new_pg_temp, p); | |
559 | } | |
560 | ||
561 | // decode short map, too. | |
562 | if (v == 5 && p.end()) | |
563 | return; | |
564 | ||
565 | // extended | |
566 | __u16 ev = 0; | |
567 | if (v >= 5) | |
568 | ::decode(ev, p); | |
569 | ::decode(new_hb_back_up, p); | |
570 | if (v < 5) | |
571 | ::decode(new_pool_names, p); | |
572 | ::decode(new_up_thru, p); | |
573 | ::decode(new_last_clean_interval, p); | |
574 | ::decode(new_lost, p); | |
575 | ::decode(new_blacklist, p); | |
576 | ::decode(old_blacklist, p); | |
577 | if (ev >= 6) | |
578 | ::decode(new_up_cluster, p); | |
579 | if (ev >= 7) | |
580 | ::decode(cluster_snapshot, p); | |
581 | if (ev >= 8) | |
582 | ::decode(new_uuid, p); | |
583 | if (ev >= 9) | |
584 | ::decode(new_xinfo, p); | |
585 | if (ev >= 10) | |
586 | ::decode(new_hb_front_up, p); | |
587 | } | |
588 | ||
589 | void OSDMap::Incremental::decode(bufferlist::iterator& bl) | |
590 | { | |
591 | /** | |
592 | * Older encodings of the Incremental had a single struct_v which | |
593 | * covered the whole encoding, and was prior to our modern | |
594 | * stuff which includes a compatv and a size. So if we see | |
595 | * a struct_v < 7, we must rewind to the beginning and use our | |
596 | * classic decoder. | |
597 | */ | |
598 | size_t start_offset = bl.get_off(); | |
599 | size_t tail_offset = 0; | |
600 | bufferlist crc_front, crc_tail; | |
601 | ||
602 | DECODE_START_LEGACY_COMPAT_LEN(8, 7, 7, bl); // wrapper | |
603 | if (struct_v < 7) { | |
604 | int struct_v_size = sizeof(struct_v); | |
605 | bl.advance(-struct_v_size); | |
606 | decode_classic(bl); | |
607 | encode_features = 0; | |
608 | if (struct_v >= 6) | |
609 | encode_features = CEPH_FEATURE_PGID64; | |
610 | else | |
611 | encode_features = 0; | |
612 | return; | |
613 | } | |
614 | { | |
615 | DECODE_START(4, bl); // client-usable data | |
616 | ::decode(fsid, bl); | |
617 | ::decode(epoch, bl); | |
618 | ::decode(modified, bl); | |
619 | ::decode(new_pool_max, bl); | |
620 | ::decode(new_flags, bl); | |
621 | ::decode(fullmap, bl); | |
622 | ::decode(crush, bl); | |
623 | ||
624 | ::decode(new_max_osd, bl); | |
625 | ::decode(new_pools, bl); | |
626 | ::decode(new_pool_names, bl); | |
627 | ::decode(old_pools, bl); | |
628 | ::decode(new_up_client, bl); | |
629 | ::decode(new_state, bl); | |
630 | ::decode(new_weight, bl); | |
631 | ::decode(new_pg_temp, bl); | |
632 | ::decode(new_primary_temp, bl); | |
633 | if (struct_v >= 2) | |
634 | ::decode(new_primary_affinity, bl); | |
635 | else | |
636 | new_primary_affinity.clear(); | |
637 | if (struct_v >= 3) { | |
638 | ::decode(new_erasure_code_profiles, bl); | |
639 | ::decode(old_erasure_code_profiles, bl); | |
640 | } else { | |
641 | new_erasure_code_profiles.clear(); | |
642 | old_erasure_code_profiles.clear(); | |
643 | } | |
644 | if (struct_v >= 4) { | |
645 | ::decode(new_pg_upmap, bl); | |
646 | ::decode(old_pg_upmap, bl); | |
647 | ::decode(new_pg_upmap_items, bl); | |
648 | ::decode(old_pg_upmap_items, bl); | |
649 | } | |
650 | DECODE_FINISH(bl); // client-usable data | |
651 | } | |
652 | ||
653 | { | |
654 | DECODE_START(5, bl); // extended, osd-only data | |
655 | ::decode(new_hb_back_up, bl); | |
656 | ::decode(new_up_thru, bl); | |
657 | ::decode(new_last_clean_interval, bl); | |
658 | ::decode(new_lost, bl); | |
659 | ::decode(new_blacklist, bl); | |
660 | ::decode(old_blacklist, bl); | |
661 | ::decode(new_up_cluster, bl); | |
662 | ::decode(cluster_snapshot, bl); | |
663 | ::decode(new_uuid, bl); | |
664 | ::decode(new_xinfo, bl); | |
665 | ::decode(new_hb_front_up, bl); | |
666 | if (struct_v >= 2) | |
667 | ::decode(encode_features, bl); | |
668 | else | |
669 | encode_features = CEPH_FEATURE_PGID64 | CEPH_FEATURE_OSDMAP_ENC; | |
670 | if (struct_v >= 3) { | |
671 | ::decode(new_nearfull_ratio, bl); | |
672 | ::decode(new_full_ratio, bl); | |
673 | } else { | |
674 | new_nearfull_ratio = -1; | |
675 | new_full_ratio = -1; | |
676 | } | |
677 | if (struct_v >= 4) { | |
678 | ::decode(new_backfillfull_ratio, bl); | |
679 | } else { | |
680 | new_backfillfull_ratio = -1; | |
681 | } | |
682 | if (struct_v >= 5) | |
683 | ::decode(new_require_min_compat_client, bl); | |
684 | DECODE_FINISH(bl); // osd-only data | |
685 | } | |
686 | ||
687 | if (struct_v >= 8) { | |
688 | have_crc = true; | |
689 | crc_front.substr_of(bl.get_bl(), start_offset, bl.get_off() - start_offset); | |
690 | ::decode(inc_crc, bl); | |
691 | tail_offset = bl.get_off(); | |
692 | ::decode(full_crc, bl); | |
693 | } else { | |
694 | have_crc = false; | |
695 | full_crc = 0; | |
696 | inc_crc = 0; | |
697 | } | |
698 | ||
699 | DECODE_FINISH(bl); // wrapper | |
700 | ||
701 | if (have_crc) { | |
702 | // verify crc | |
703 | uint32_t actual = crc_front.crc32c(-1); | |
704 | if (tail_offset < bl.get_off()) { | |
705 | bufferlist tail; | |
706 | tail.substr_of(bl.get_bl(), tail_offset, bl.get_off() - tail_offset); | |
707 | actual = tail.crc32c(actual); | |
708 | } | |
709 | if (inc_crc != actual) { | |
710 | ostringstream ss; | |
711 | ss << "bad crc, actual " << actual << " != expected " << inc_crc; | |
712 | string s = ss.str(); | |
713 | throw buffer::malformed_input(s.c_str()); | |
714 | } | |
715 | } | |
716 | } | |
717 | ||
718 | void OSDMap::Incremental::dump(Formatter *f) const | |
719 | { | |
720 | f->dump_int("epoch", epoch); | |
721 | f->dump_stream("fsid") << fsid; | |
722 | f->dump_stream("modified") << modified; | |
723 | f->dump_int("new_pool_max", new_pool_max); | |
724 | f->dump_int("new_flags", new_flags); | |
725 | f->dump_float("new_full_ratio", new_full_ratio); | |
726 | f->dump_float("new_nearfull_ratio", new_nearfull_ratio); | |
727 | f->dump_float("new_backfillfull_ratio", new_backfillfull_ratio); | |
728 | f->dump_string("new_require_min_compat_client", new_require_min_compat_client); | |
729 | ||
730 | if (fullmap.length()) { | |
731 | f->open_object_section("full_map"); | |
732 | OSDMap full; | |
733 | bufferlist fbl = fullmap; // kludge around constness. | |
734 | auto p = fbl.begin(); | |
735 | full.decode(p); | |
736 | full.dump(f); | |
737 | f->close_section(); | |
738 | } | |
739 | if (crush.length()) { | |
740 | f->open_object_section("crush"); | |
741 | CrushWrapper c; | |
742 | bufferlist tbl = crush; // kludge around constness. | |
743 | auto p = tbl.begin(); | |
744 | c.decode(p); | |
745 | c.dump(f); | |
746 | f->close_section(); | |
747 | } | |
748 | ||
749 | f->dump_int("new_max_osd", new_max_osd); | |
750 | ||
751 | f->open_array_section("new_pools"); | |
752 | ||
753 | for (const auto &new_pool : new_pools) { | |
754 | f->open_object_section("pool"); | |
755 | f->dump_int("pool", new_pool.first); | |
756 | new_pool.second.dump(f); | |
757 | f->close_section(); | |
758 | } | |
759 | f->close_section(); | |
760 | f->open_array_section("new_pool_names"); | |
761 | ||
762 | for (const auto &new_pool_name : new_pool_names) { | |
763 | f->open_object_section("pool_name"); | |
764 | f->dump_int("pool", new_pool_name.first); | |
765 | f->dump_string("name", new_pool_name.second); | |
766 | f->close_section(); | |
767 | } | |
768 | f->close_section(); | |
769 | f->open_array_section("old_pools"); | |
770 | ||
771 | for (const auto &old_pool : old_pools) | |
772 | f->dump_int("pool", old_pool); | |
773 | f->close_section(); | |
774 | ||
775 | f->open_array_section("new_up_osds"); | |
776 | ||
777 | for (const auto &upclient : new_up_client) { | |
778 | f->open_object_section("osd"); | |
779 | f->dump_int("osd", upclient.first); | |
780 | f->dump_stream("public_addr") << upclient.second; | |
781 | f->dump_stream("cluster_addr") << new_up_cluster.find(upclient.first)->second; | |
782 | f->dump_stream("heartbeat_back_addr") << new_hb_back_up.find(upclient.first)->second; | |
783 | map<int32_t, entity_addr_t>::const_iterator q; | |
784 | if ((q = new_hb_front_up.find(upclient.first)) != new_hb_front_up.end()) | |
785 | f->dump_stream("heartbeat_front_addr") << q->second; | |
786 | f->close_section(); | |
787 | } | |
788 | f->close_section(); | |
789 | ||
790 | f->open_array_section("new_weight"); | |
791 | ||
792 | for (const auto &weight : new_weight) { | |
793 | f->open_object_section("osd"); | |
794 | f->dump_int("osd", weight.first); | |
795 | f->dump_int("weight", weight.second); | |
796 | f->close_section(); | |
797 | } | |
798 | f->close_section(); | |
799 | ||
800 | f->open_array_section("osd_state_xor"); | |
801 | for (const auto &ns : new_state) { | |
802 | f->open_object_section("osd"); | |
803 | f->dump_int("osd", ns.first); | |
804 | set<string> st; | |
805 | calc_state_set(new_state.find(ns.first)->second, st); | |
806 | f->open_array_section("state_xor"); | |
807 | for (auto &state : st) | |
808 | f->dump_string("state", state); | |
809 | f->close_section(); | |
810 | } | |
811 | f->close_section(); | |
812 | ||
813 | f->open_array_section("new_pg_temp"); | |
814 | ||
815 | for (const auto &pg_temp : new_pg_temp) { | |
816 | f->open_object_section("pg"); | |
817 | f->dump_stream("pgid") << pg_temp.first; | |
818 | f->open_array_section("osds"); | |
819 | ||
820 | for (const auto &osd : pg_temp.second) | |
821 | f->dump_int("osd", osd); | |
822 | f->close_section(); | |
823 | f->close_section(); | |
824 | } | |
825 | f->close_section(); | |
826 | ||
827 | f->open_array_section("primary_temp"); | |
828 | ||
829 | for (const auto &primary_temp : new_primary_temp) { | |
830 | f->dump_stream("pgid") << primary_temp.first; | |
831 | f->dump_int("osd", primary_temp.second); | |
832 | } | |
833 | f->close_section(); // primary_temp | |
834 | ||
835 | f->open_array_section("new_pg_upmap"); | |
836 | for (auto& i : new_pg_upmap) { | |
837 | f->open_object_section("mapping"); | |
838 | f->dump_stream("pgid") << i.first; | |
839 | f->open_array_section("osds"); | |
840 | for (auto osd : i.second) { | |
841 | f->dump_int("osd", osd); | |
842 | } | |
843 | f->close_section(); | |
844 | f->close_section(); | |
845 | } | |
846 | f->close_section(); | |
847 | f->open_array_section("old_pg_upmap"); | |
848 | for (auto& i : old_pg_upmap) { | |
849 | f->dump_stream("pgid") << i; | |
850 | } | |
851 | f->close_section(); | |
852 | ||
853 | f->open_array_section("new_pg_upmap_items"); | |
854 | for (auto& i : new_pg_upmap_items) { | |
855 | f->open_object_section("mapping"); | |
856 | f->dump_stream("pgid") << i.first; | |
857 | f->open_array_section("mappings"); | |
858 | for (auto& p : i.second) { | |
859 | f->open_object_section("mapping"); | |
860 | f->dump_int("from", p.first); | |
861 | f->dump_int("to", p.second); | |
862 | f->close_section(); | |
863 | } | |
864 | f->close_section(); | |
865 | f->close_section(); | |
866 | } | |
867 | f->close_section(); | |
868 | f->open_array_section("old_pg_upmap_items"); | |
869 | for (auto& i : old_pg_upmap_items) { | |
870 | f->dump_stream("pgid") << i; | |
871 | } | |
872 | f->close_section(); | |
873 | ||
874 | f->open_array_section("new_up_thru"); | |
875 | ||
876 | for (const auto &up_thru : new_up_thru) { | |
877 | f->open_object_section("osd"); | |
878 | f->dump_int("osd", up_thru.first); | |
879 | f->dump_int("up_thru", up_thru.second); | |
880 | f->close_section(); | |
881 | } | |
882 | f->close_section(); | |
883 | ||
884 | f->open_array_section("new_lost"); | |
885 | ||
886 | for (const auto &lost : new_lost) { | |
887 | f->open_object_section("osd"); | |
888 | f->dump_int("osd", lost.first); | |
889 | f->dump_int("epoch_lost", lost.second); | |
890 | f->close_section(); | |
891 | } | |
892 | f->close_section(); | |
893 | ||
894 | f->open_array_section("new_last_clean_interval"); | |
895 | ||
896 | for (const auto &last_clean_interval : new_last_clean_interval) { | |
897 | f->open_object_section("osd"); | |
898 | f->dump_int("osd", last_clean_interval.first); | |
899 | f->dump_int("first", last_clean_interval.second.first); | |
900 | f->dump_int("last", last_clean_interval.second.second); | |
901 | f->close_section(); | |
902 | } | |
903 | f->close_section(); | |
904 | ||
905 | f->open_array_section("new_blacklist"); | |
906 | for (const auto &blist : new_blacklist) { | |
907 | stringstream ss; | |
908 | ss << blist.first; | |
909 | f->dump_stream(ss.str().c_str()) << blist.second; | |
910 | } | |
911 | f->close_section(); | |
912 | f->open_array_section("old_blacklist"); | |
913 | for (const auto &blist : old_blacklist) | |
914 | f->dump_stream("addr") << blist; | |
915 | f->close_section(); | |
916 | ||
917 | f->open_array_section("new_xinfo"); | |
918 | for (const auto &xinfo : new_xinfo) { | |
919 | f->open_object_section("xinfo"); | |
920 | f->dump_int("osd", xinfo.first); | |
921 | xinfo.second.dump(f); | |
922 | f->close_section(); | |
923 | } | |
924 | f->close_section(); | |
925 | ||
926 | if (cluster_snapshot.size()) | |
927 | f->dump_string("cluster_snapshot", cluster_snapshot); | |
928 | ||
929 | f->open_array_section("new_uuid"); | |
930 | for (const auto &uuid : new_uuid) { | |
931 | f->open_object_section("osd"); | |
932 | f->dump_int("osd", uuid.first); | |
933 | f->dump_stream("uuid") << uuid.second; | |
934 | f->close_section(); | |
935 | } | |
936 | f->close_section(); | |
937 | ||
938 | OSDMap::dump_erasure_code_profiles(new_erasure_code_profiles, f); | |
939 | f->open_array_section("old_erasure_code_profiles"); | |
940 | for (const auto &erasure_code_profile : old_erasure_code_profiles) { | |
941 | f->dump_string("old", erasure_code_profile.c_str()); | |
942 | } | |
943 | f->close_section(); | |
944 | } | |
945 | ||
946 | void OSDMap::Incremental::generate_test_instances(list<Incremental*>& o) | |
947 | { | |
948 | o.push_back(new Incremental); | |
949 | } | |
950 | ||
951 | // ---------------------------------- | |
952 | // OSDMap | |
953 | ||
954 | void OSDMap::set_epoch(epoch_t e) | |
955 | { | |
956 | epoch = e; | |
957 | for (auto &pool : pools) | |
958 | pool.second.last_change = e; | |
959 | } | |
960 | ||
961 | bool OSDMap::is_blacklisted(const entity_addr_t& a) const | |
962 | { | |
963 | if (blacklist.empty()) | |
964 | return false; | |
965 | ||
966 | // this specific instance? | |
967 | if (blacklist.count(a)) | |
968 | return true; | |
969 | ||
970 | // is entire ip blacklisted? | |
971 | if (a.is_ip()) { | |
972 | entity_addr_t b = a; | |
973 | b.set_port(0); | |
974 | b.set_nonce(0); | |
975 | if (blacklist.count(b)) { | |
976 | return true; | |
977 | } | |
978 | } | |
979 | ||
980 | return false; | |
981 | } | |
982 | ||
983 | void OSDMap::get_blacklist(list<pair<entity_addr_t,utime_t> > *bl) const | |
984 | { | |
985 | std::copy(blacklist.begin(), blacklist.end(), std::back_inserter(*bl)); | |
986 | } | |
987 | ||
988 | void OSDMap::set_max_osd(int m) | |
989 | { | |
990 | int o = max_osd; | |
991 | max_osd = m; | |
992 | osd_state.resize(m); | |
993 | osd_weight.resize(m); | |
994 | for (; o<max_osd; o++) { | |
995 | osd_state[o] = 0; | |
996 | osd_weight[o] = CEPH_OSD_OUT; | |
997 | } | |
998 | osd_info.resize(m); | |
999 | osd_xinfo.resize(m); | |
1000 | osd_addrs->client_addr.resize(m); | |
1001 | osd_addrs->cluster_addr.resize(m); | |
1002 | osd_addrs->hb_back_addr.resize(m); | |
1003 | osd_addrs->hb_front_addr.resize(m); | |
1004 | osd_uuid->resize(m); | |
1005 | if (osd_primary_affinity) | |
1006 | osd_primary_affinity->resize(m, CEPH_OSD_DEFAULT_PRIMARY_AFFINITY); | |
1007 | ||
1008 | calc_num_osds(); | |
1009 | } | |
1010 | ||
1011 | int OSDMap::calc_num_osds() | |
1012 | { | |
1013 | num_osd = 0; | |
1014 | num_up_osd = 0; | |
1015 | num_in_osd = 0; | |
1016 | for (int i=0; i<max_osd; i++) { | |
1017 | if (osd_state[i] & CEPH_OSD_EXISTS) { | |
1018 | ++num_osd; | |
1019 | if (osd_state[i] & CEPH_OSD_UP) { | |
1020 | ++num_up_osd; | |
1021 | } | |
1022 | if (get_weight(i) != CEPH_OSD_OUT) { | |
1023 | ++num_in_osd; | |
1024 | } | |
1025 | } | |
1026 | } | |
1027 | return num_osd; | |
1028 | } | |
1029 | ||
1030 | void OSDMap::count_full_nearfull_osds(int *full, int *backfill, int *nearfull) const | |
1031 | { | |
1032 | *full = 0; | |
1033 | *backfill = 0; | |
1034 | *nearfull = 0; | |
1035 | for (int i = 0; i < max_osd; ++i) { | |
1036 | if (exists(i) && is_up(i) && is_in(i)) { | |
1037 | if (osd_state[i] & CEPH_OSD_FULL) | |
1038 | ++(*full); | |
1039 | else if (osd_state[i] & CEPH_OSD_BACKFILLFULL) | |
1040 | ++(*backfill); | |
1041 | else if (osd_state[i] & CEPH_OSD_NEARFULL) | |
1042 | ++(*nearfull); | |
1043 | } | |
1044 | } | |
1045 | } | |
1046 | ||
1047 | static bool get_osd_utilization(const ceph::unordered_map<int32_t,osd_stat_t> &osd_stat, | |
1048 | int id, int64_t* kb, int64_t* kb_used, int64_t* kb_avail) { | |
1049 | auto p = osd_stat.find(id); | |
1050 | if (p == osd_stat.end()) | |
1051 | return false; | |
1052 | *kb = p->second.kb; | |
1053 | *kb_used = p->second.kb_used; | |
1054 | *kb_avail = p->second.kb_avail; | |
1055 | return *kb > 0; | |
1056 | } | |
1057 | ||
1058 | void OSDMap::get_full_osd_util(const ceph::unordered_map<int32_t,osd_stat_t> &osd_stat, | |
1059 | map<int, float> *full, map<int, float> *backfill, map<int, float> *nearfull) const | |
1060 | { | |
1061 | full->clear(); | |
1062 | backfill->clear(); | |
1063 | nearfull->clear(); | |
1064 | for (int i = 0; i < max_osd; ++i) { | |
1065 | if (exists(i) && is_up(i) && is_in(i)) { | |
1066 | int64_t kb, kb_used, kb_avail; | |
1067 | if (osd_state[i] & CEPH_OSD_FULL) { | |
1068 | if (get_osd_utilization(osd_stat, i, &kb, &kb_used, &kb_avail)) | |
1069 | full->emplace(i, (float)kb_used / (float)kb); | |
1070 | } else if (osd_state[i] & CEPH_OSD_BACKFILLFULL) { | |
1071 | if (get_osd_utilization(osd_stat, i, &kb, &kb_used, &kb_avail)) | |
1072 | backfill->emplace(i, (float)kb_used / (float)kb); | |
1073 | } else if (osd_state[i] & CEPH_OSD_NEARFULL) { | |
1074 | if (get_osd_utilization(osd_stat, i, &kb, &kb_used, &kb_avail)) | |
1075 | nearfull->emplace(i, (float)kb_used / (float)kb); | |
1076 | } | |
1077 | } | |
1078 | } | |
1079 | } | |
1080 | ||
1081 | void OSDMap::get_all_osds(set<int32_t>& ls) const | |
1082 | { | |
1083 | for (int i=0; i<max_osd; i++) | |
1084 | if (exists(i)) | |
1085 | ls.insert(i); | |
1086 | } | |
1087 | ||
1088 | void OSDMap::get_up_osds(set<int32_t>& ls) const | |
1089 | { | |
1090 | for (int i = 0; i < max_osd; i++) { | |
1091 | if (is_up(i)) | |
1092 | ls.insert(i); | |
1093 | } | |
1094 | } | |
1095 | ||
1096 | void OSDMap::calc_state_set(int state, set<string>& st) | |
1097 | { | |
1098 | unsigned t = state; | |
1099 | for (unsigned s = 1; t; s <<= 1) { | |
1100 | if (t & s) { | |
1101 | t &= ~s; | |
1102 | st.insert(ceph_osd_state_name(s)); | |
1103 | } | |
1104 | } | |
1105 | } | |
1106 | ||
1107 | void OSDMap::adjust_osd_weights(const map<int,double>& weights, Incremental& inc) const | |
1108 | { | |
1109 | float max = 0; | |
1110 | for (const auto &weight : weights) { | |
1111 | if (weight.second > max) | |
1112 | max = weight.second; | |
1113 | } | |
1114 | ||
1115 | for (const auto &weight : weights) { | |
1116 | inc.new_weight[weight.first] = (unsigned)((weight.second / max) * CEPH_OSD_IN); | |
1117 | } | |
1118 | } | |
1119 | ||
1120 | int OSDMap::identify_osd(const entity_addr_t& addr) const | |
1121 | { | |
1122 | for (int i=0; i<max_osd; i++) | |
1123 | if (exists(i) && (get_addr(i) == addr || get_cluster_addr(i) == addr)) | |
1124 | return i; | |
1125 | return -1; | |
1126 | } | |
1127 | ||
1128 | int OSDMap::identify_osd(const uuid_d& u) const | |
1129 | { | |
1130 | for (int i=0; i<max_osd; i++) | |
1131 | if (exists(i) && get_uuid(i) == u) | |
1132 | return i; | |
1133 | return -1; | |
1134 | } | |
1135 | ||
1136 | int OSDMap::identify_osd_on_all_channels(const entity_addr_t& addr) const | |
1137 | { | |
1138 | for (int i=0; i<max_osd; i++) | |
1139 | if (exists(i) && (get_addr(i) == addr || get_cluster_addr(i) == addr || | |
1140 | get_hb_back_addr(i) == addr || get_hb_front_addr(i) == addr)) | |
1141 | return i; | |
1142 | return -1; | |
1143 | } | |
1144 | ||
1145 | int OSDMap::find_osd_on_ip(const entity_addr_t& ip) const | |
1146 | { | |
1147 | for (int i=0; i<max_osd; i++) | |
1148 | if (exists(i) && (get_addr(i).is_same_host(ip) || get_cluster_addr(i).is_same_host(ip))) | |
1149 | return i; | |
1150 | return -1; | |
1151 | } | |
1152 | ||
1153 | ||
1154 | uint64_t OSDMap::get_features(int entity_type, uint64_t *pmask) const | |
1155 | { | |
1156 | uint64_t features = 0; // things we actually have | |
1157 | uint64_t mask = 0; // things we could have | |
1158 | ||
1159 | if (crush->has_nondefault_tunables()) | |
1160 | features |= CEPH_FEATURE_CRUSH_TUNABLES; | |
1161 | if (crush->has_nondefault_tunables2()) | |
1162 | features |= CEPH_FEATURE_CRUSH_TUNABLES2; | |
1163 | if (crush->has_nondefault_tunables3()) | |
1164 | features |= CEPH_FEATURE_CRUSH_TUNABLES3; | |
1165 | if (crush->has_v4_buckets()) | |
1166 | features |= CEPH_FEATURE_CRUSH_V4; | |
1167 | if (crush->has_nondefault_tunables5()) | |
1168 | features |= CEPH_FEATURE_CRUSH_TUNABLES5; | |
1169 | if (crush->has_incompat_chooseargs()) | |
1170 | features |= CEPH_FEATURE_CRUSH_CHOOSEARGS; | |
1171 | mask |= CEPH_FEATURES_CRUSH; | |
1172 | ||
1173 | if (!pg_upmap.empty() || !pg_upmap_items.empty()) | |
1174 | features |= CEPH_FEATUREMASK_OSDMAP_PG_UPMAP; | |
1175 | mask |= CEPH_FEATUREMASK_OSDMAP_PG_UPMAP; | |
1176 | ||
1177 | for (auto &pool: pools) { | |
1178 | if (pool.second.has_flag(pg_pool_t::FLAG_HASHPSPOOL)) { | |
1179 | features |= CEPH_FEATURE_OSDHASHPSPOOL; | |
1180 | } | |
1181 | if (pool.second.is_erasure() && | |
1182 | entity_type != CEPH_ENTITY_TYPE_CLIENT) { // not for clients | |
1183 | features |= CEPH_FEATURE_OSD_ERASURE_CODES; | |
1184 | } | |
1185 | if (!pool.second.tiers.empty() || | |
1186 | pool.second.is_tier()) { | |
1187 | features |= CEPH_FEATURE_OSD_CACHEPOOL; | |
1188 | } | |
1189 | int ruleid = crush->find_rule(pool.second.get_crush_ruleset(), | |
1190 | pool.second.get_type(), | |
1191 | pool.second.get_size()); | |
1192 | if (ruleid >= 0) { | |
1193 | if (crush->is_v2_rule(ruleid)) | |
1194 | features |= CEPH_FEATURE_CRUSH_V2; | |
1195 | if (crush->is_v3_rule(ruleid)) | |
1196 | features |= CEPH_FEATURE_CRUSH_TUNABLES3; | |
1197 | if (crush->is_v5_rule(ruleid)) | |
1198 | features |= CEPH_FEATURE_CRUSH_TUNABLES5; | |
1199 | } | |
1200 | } | |
1201 | if (entity_type == CEPH_ENTITY_TYPE_OSD) { | |
1202 | for (auto &erasure_code_profile : erasure_code_profiles) { | |
1203 | auto& profile = erasure_code_profile.second; | |
1204 | const auto& plugin = profile.find("plugin"); | |
1205 | if (plugin != profile.end()) { | |
1206 | if (plugin->second == "isa" || plugin->second == "lrc") | |
1207 | features |= CEPH_FEATURE_ERASURE_CODE_PLUGINS_V2; | |
1208 | if (plugin->second == "shec") | |
1209 | features |= CEPH_FEATURE_ERASURE_CODE_PLUGINS_V3; | |
1210 | } | |
1211 | } | |
1212 | } | |
1213 | mask |= CEPH_FEATURE_OSDHASHPSPOOL | CEPH_FEATURE_OSD_CACHEPOOL; | |
1214 | if (entity_type != CEPH_ENTITY_TYPE_CLIENT) | |
1215 | mask |= CEPH_FEATURE_OSD_ERASURE_CODES; | |
1216 | ||
1217 | if (osd_primary_affinity) { | |
1218 | for (int i = 0; i < max_osd; ++i) { | |
1219 | if ((*osd_primary_affinity)[i] != CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) { | |
1220 | features |= CEPH_FEATURE_OSD_PRIMARY_AFFINITY; | |
1221 | break; | |
1222 | } | |
1223 | } | |
1224 | } | |
1225 | mask |= CEPH_FEATURE_OSD_PRIMARY_AFFINITY; | |
1226 | ||
1227 | if (entity_type == CEPH_ENTITY_TYPE_OSD) { | |
1228 | const uint64_t jewel_features = CEPH_FEATURE_SERVER_JEWEL; | |
1229 | if (test_flag(CEPH_OSDMAP_REQUIRE_JEWEL)) { | |
1230 | features |= jewel_features; | |
1231 | } | |
1232 | mask |= jewel_features; | |
1233 | ||
1234 | const uint64_t kraken_features = CEPH_FEATUREMASK_SERVER_KRAKEN | |
1235 | | CEPH_FEATURE_MSG_ADDR2; | |
1236 | if (test_flag(CEPH_OSDMAP_REQUIRE_KRAKEN)) { | |
1237 | features |= kraken_features; | |
1238 | } | |
1239 | mask |= kraken_features; | |
1240 | } | |
1241 | ||
1242 | if (pmask) | |
1243 | *pmask = mask; | |
1244 | return features; | |
1245 | } | |
1246 | ||
1247 | pair<string,string> OSDMap::get_min_compat_client() const | |
1248 | { | |
1249 | uint64_t f = get_features(CEPH_ENTITY_TYPE_CLIENT, nullptr); | |
1250 | ||
1251 | if (HAVE_FEATURE(f, OSDMAP_PG_UPMAP) || // v12.0.0-1733-g27d6f43 | |
1252 | HAVE_FEATURE(f, CRUSH_CHOOSEARGS)) { // v12.0.1-2172-gef1ef28 | |
1253 | return make_pair("luminous", "12.2.0"); | |
1254 | } | |
1255 | if (HAVE_FEATURE(f, CRUSH_TUNABLES5)) { // v10.0.0-612-g043a737 | |
1256 | return make_pair("jewel", "10.2.0"); | |
1257 | } | |
1258 | if (HAVE_FEATURE(f, CRUSH_V4)) { // v0.91-678-g325fc56 | |
1259 | return make_pair("hammer", "0.94"); | |
1260 | } | |
1261 | if (HAVE_FEATURE(f, OSD_PRIMARY_AFFINITY) || // v0.76-553-gf825624 | |
1262 | HAVE_FEATURE(f, CRUSH_TUNABLES3) || // v0.76-395-ge20a55d | |
1263 | HAVE_FEATURE(f, OSD_ERASURE_CODES) || // v0.73-498-gbfc86a8 | |
1264 | HAVE_FEATURE(f, OSD_CACHEPOOL)) { // v0.67-401-gb91c1c5 | |
1265 | return make_pair("firefly", "0.80"); | |
1266 | } | |
1267 | if (HAVE_FEATURE(f, CRUSH_TUNABLES2) || // v0.54-684-g0cc47ff | |
1268 | HAVE_FEATURE(f, OSDHASHPSPOOL)) { // v0.57-398-g8cc2b0f | |
1269 | return make_pair("dumpling", "0.67"); | |
1270 | } | |
1271 | if (HAVE_FEATURE(f, CRUSH_TUNABLES)) { // v0.48argonaut-206-g6f381af | |
1272 | return make_pair("argonaut", "0.48argonaut-207"); | |
1273 | } | |
1274 | return make_pair("argonaut", "0.48"); | |
1275 | } | |
1276 | ||
1277 | void OSDMap::_calc_up_osd_features() | |
1278 | { | |
1279 | bool first = true; | |
1280 | cached_up_osd_features = 0; | |
1281 | for (int osd = 0; osd < max_osd; ++osd) { | |
1282 | if (!is_up(osd)) | |
1283 | continue; | |
1284 | const osd_xinfo_t &xi = get_xinfo(osd); | |
1285 | if (first) { | |
1286 | cached_up_osd_features = xi.features; | |
1287 | first = false; | |
1288 | } else { | |
1289 | cached_up_osd_features &= xi.features; | |
1290 | } | |
1291 | } | |
1292 | } | |
1293 | ||
1294 | uint64_t OSDMap::get_up_osd_features() const | |
1295 | { | |
1296 | return cached_up_osd_features; | |
1297 | } | |
1298 | ||
1299 | void OSDMap::dedup(const OSDMap *o, OSDMap *n) | |
1300 | { | |
1301 | if (o->epoch == n->epoch) | |
1302 | return; | |
1303 | ||
1304 | int diff = 0; | |
1305 | ||
1306 | // do addrs match? | |
1307 | if (o->max_osd != n->max_osd) | |
1308 | diff++; | |
1309 | for (int i = 0; i < o->max_osd && i < n->max_osd; i++) { | |
1310 | if ( n->osd_addrs->client_addr[i] && o->osd_addrs->client_addr[i] && | |
1311 | *n->osd_addrs->client_addr[i] == *o->osd_addrs->client_addr[i]) | |
1312 | n->osd_addrs->client_addr[i] = o->osd_addrs->client_addr[i]; | |
1313 | else | |
1314 | diff++; | |
1315 | if ( n->osd_addrs->cluster_addr[i] && o->osd_addrs->cluster_addr[i] && | |
1316 | *n->osd_addrs->cluster_addr[i] == *o->osd_addrs->cluster_addr[i]) | |
1317 | n->osd_addrs->cluster_addr[i] = o->osd_addrs->cluster_addr[i]; | |
1318 | else | |
1319 | diff++; | |
1320 | if ( n->osd_addrs->hb_back_addr[i] && o->osd_addrs->hb_back_addr[i] && | |
1321 | *n->osd_addrs->hb_back_addr[i] == *o->osd_addrs->hb_back_addr[i]) | |
1322 | n->osd_addrs->hb_back_addr[i] = o->osd_addrs->hb_back_addr[i]; | |
1323 | else | |
1324 | diff++; | |
1325 | if ( n->osd_addrs->hb_front_addr[i] && o->osd_addrs->hb_front_addr[i] && | |
1326 | *n->osd_addrs->hb_front_addr[i] == *o->osd_addrs->hb_front_addr[i]) | |
1327 | n->osd_addrs->hb_front_addr[i] = o->osd_addrs->hb_front_addr[i]; | |
1328 | else | |
1329 | diff++; | |
1330 | } | |
1331 | if (diff == 0) { | |
1332 | // zoinks, no differences at all! | |
1333 | n->osd_addrs = o->osd_addrs; | |
1334 | } | |
1335 | ||
1336 | // does crush match? | |
1337 | bufferlist oc, nc; | |
1338 | ::encode(*o->crush, oc, CEPH_FEATURES_SUPPORTED_DEFAULT); | |
1339 | ::encode(*n->crush, nc, CEPH_FEATURES_SUPPORTED_DEFAULT); | |
1340 | if (oc.contents_equal(nc)) { | |
1341 | n->crush = o->crush; | |
1342 | } | |
1343 | ||
1344 | // does pg_temp match? | |
1345 | if (o->pg_temp->size() == n->pg_temp->size()) { | |
1346 | if (*o->pg_temp == *n->pg_temp) | |
1347 | n->pg_temp = o->pg_temp; | |
1348 | } | |
1349 | ||
1350 | // does primary_temp match? | |
1351 | if (o->primary_temp->size() == n->primary_temp->size()) { | |
1352 | if (*o->primary_temp == *n->primary_temp) | |
1353 | n->primary_temp = o->primary_temp; | |
1354 | } | |
1355 | ||
1356 | // do uuids match? | |
1357 | if (o->osd_uuid->size() == n->osd_uuid->size() && | |
1358 | *o->osd_uuid == *n->osd_uuid) | |
1359 | n->osd_uuid = o->osd_uuid; | |
1360 | } | |
1361 | ||
1362 | void OSDMap::clean_temps(CephContext *cct, | |
1363 | const OSDMap& osdmap, Incremental *pending_inc) | |
1364 | { | |
1365 | ldout(cct, 10) << __func__ << dendl; | |
1366 | OSDMap tmpmap; | |
1367 | tmpmap.deepish_copy_from(osdmap); | |
1368 | tmpmap.apply_incremental(*pending_inc); | |
1369 | ||
1370 | for (auto pg : *tmpmap.pg_temp) { | |
1371 | // if pool does not exist, remove any existing pg_temps associated with | |
1372 | // it. we don't care about pg_temps on the pending_inc either; if there | |
1373 | // are new_pg_temp entries on the pending, clear them out just as well. | |
1374 | if (!osdmap.have_pg_pool(pg.first.pool())) { | |
1375 | ldout(cct, 10) << __func__ << " removing pg_temp " << pg.first | |
1376 | << " for nonexistent pool " << pg.first.pool() << dendl; | |
1377 | pending_inc->new_pg_temp[pg.first].clear(); | |
1378 | continue; | |
1379 | } | |
1380 | // all osds down? | |
1381 | unsigned num_up = 0; | |
1382 | for (auto o : pg.second) { | |
1383 | if (!tmpmap.is_down(o)) { | |
1384 | ++num_up; | |
1385 | break; | |
1386 | } | |
1387 | } | |
1388 | if (num_up == 0) { | |
1389 | ldout(cct, 10) << __func__ << " removing pg_temp " << pg.first | |
1390 | << " with all down osds" << pg.second << dendl; | |
1391 | pending_inc->new_pg_temp[pg.first].clear(); | |
1392 | continue; | |
1393 | } | |
1394 | // redundant pg_temp? | |
1395 | vector<int> raw_up; | |
1396 | int primary; | |
1397 | tmpmap.pg_to_raw_up(pg.first, &raw_up, &primary); | |
1398 | if (vectors_equal(raw_up, pg.second)) { | |
1399 | ldout(cct, 10) << __func__ << " removing pg_temp " << pg.first << " " | |
1400 | << pg.second << " that matches raw_up mapping" << dendl; | |
1401 | if (osdmap.pg_temp->count(pg.first)) | |
1402 | pending_inc->new_pg_temp[pg.first].clear(); | |
1403 | else | |
1404 | pending_inc->new_pg_temp.erase(pg.first); | |
1405 | } | |
1406 | } | |
1407 | ||
1408 | for (auto &pg : *tmpmap.primary_temp) { | |
1409 | // primary down? | |
1410 | if (tmpmap.is_down(pg.second)) { | |
1411 | ldout(cct, 10) << __func__ << " removing primary_temp " << pg.first | |
1412 | << " to down " << pg.second << dendl; | |
1413 | pending_inc->new_primary_temp[pg.first] = -1; | |
1414 | continue; | |
1415 | } | |
1416 | // redundant primary_temp? | |
1417 | vector<int> real_up, templess_up; | |
1418 | int real_primary, templess_primary; | |
1419 | pg_t pgid = pg.first; | |
1420 | tmpmap.pg_to_acting_osds(pgid, &real_up, &real_primary); | |
1421 | tmpmap.pg_to_raw_up(pgid, &templess_up, &templess_primary); | |
1422 | if (real_primary == templess_primary){ | |
1423 | ldout(cct, 10) << __func__ << " removing primary_temp " | |
1424 | << pgid << " -> " << real_primary | |
1425 | << " (unnecessary/redundant)" << dendl; | |
1426 | if (osdmap.primary_temp->count(pgid)) | |
1427 | pending_inc->new_primary_temp[pgid] = -1; | |
1428 | else | |
1429 | pending_inc->new_primary_temp.erase(pgid); | |
1430 | } | |
1431 | } | |
1432 | } | |
1433 | ||
1434 | int OSDMap::apply_incremental(const Incremental &inc) | |
1435 | { | |
1436 | new_blacklist_entries = false; | |
1437 | if (inc.epoch == 1) | |
1438 | fsid = inc.fsid; | |
1439 | else if (inc.fsid != fsid) | |
1440 | return -EINVAL; | |
1441 | ||
1442 | assert(inc.epoch == epoch+1); | |
1443 | ||
1444 | epoch++; | |
1445 | modified = inc.modified; | |
1446 | ||
1447 | // full map? | |
1448 | if (inc.fullmap.length()) { | |
1449 | bufferlist bl(inc.fullmap); | |
1450 | decode(bl); | |
1451 | return 0; | |
1452 | } | |
1453 | ||
1454 | // nope, incremental. | |
1455 | if (inc.new_flags >= 0) | |
1456 | flags = inc.new_flags; | |
1457 | ||
1458 | if (inc.new_max_osd >= 0) | |
1459 | set_max_osd(inc.new_max_osd); | |
1460 | ||
1461 | if (inc.new_pool_max != -1) | |
1462 | pool_max = inc.new_pool_max; | |
1463 | ||
1464 | for (const auto &pool : inc.new_pools) { | |
1465 | pools[pool.first] = pool.second; | |
1466 | pools[pool.first].last_change = epoch; | |
1467 | } | |
1468 | ||
1469 | for (const auto &pname : inc.new_pool_names) { | |
1470 | auto pool_name_entry = pool_name.find(pname.first); | |
1471 | if (pool_name_entry != pool_name.end()) { | |
1472 | name_pool.erase(pool_name_entry->second); | |
1473 | pool_name_entry->second = pname.second; | |
1474 | } else { | |
1475 | pool_name[pname.first] = pname.second; | |
1476 | } | |
1477 | name_pool[pname.second] = pname.first; | |
1478 | } | |
1479 | ||
1480 | for (const auto &pool : inc.old_pools) { | |
1481 | pools.erase(pool); | |
1482 | name_pool.erase(pool_name[pool]); | |
1483 | pool_name.erase(pool); | |
1484 | } | |
1485 | ||
1486 | for (const auto &weight : inc.new_weight) { | |
1487 | set_weight(weight.first, weight.second); | |
1488 | ||
1489 | // if we are marking in, clear the AUTOOUT and NEW bits, and clear | |
1490 | // xinfo old_weight. | |
1491 | if (weight.second) { | |
1492 | osd_state[weight.first] &= ~(CEPH_OSD_AUTOOUT | CEPH_OSD_NEW); | |
1493 | osd_xinfo[weight.first].old_weight = 0; | |
1494 | } | |
1495 | } | |
1496 | ||
1497 | for (const auto &primary_affinity : inc.new_primary_affinity) { | |
1498 | set_primary_affinity(primary_affinity.first, primary_affinity.second); | |
1499 | } | |
1500 | ||
1501 | // erasure_code_profiles | |
1502 | for (const auto &profile : inc.old_erasure_code_profiles) | |
1503 | erasure_code_profiles.erase(profile); | |
1504 | ||
1505 | for (const auto &profile : inc.new_erasure_code_profiles) { | |
1506 | set_erasure_code_profile(profile.first, profile.second); | |
1507 | } | |
1508 | ||
1509 | // up/down | |
1510 | for (const auto &state : inc.new_state) { | |
1511 | const auto osd = state.first; | |
1512 | int s = state.second ? state.second : CEPH_OSD_UP; | |
1513 | if ((osd_state[osd] & CEPH_OSD_UP) && | |
1514 | (s & CEPH_OSD_UP)) { | |
1515 | osd_info[osd].down_at = epoch; | |
1516 | osd_xinfo[osd].down_stamp = modified; | |
1517 | } | |
1518 | if ((osd_state[osd] & CEPH_OSD_EXISTS) && | |
1519 | (s & CEPH_OSD_EXISTS)) { | |
1520 | // osd is destroyed; clear out anything interesting. | |
1521 | (*osd_uuid)[osd] = uuid_d(); | |
1522 | osd_info[osd] = osd_info_t(); | |
1523 | osd_xinfo[osd] = osd_xinfo_t(); | |
1524 | set_primary_affinity(osd, CEPH_OSD_DEFAULT_PRIMARY_AFFINITY); | |
1525 | osd_addrs->client_addr[osd].reset(new entity_addr_t()); | |
1526 | osd_addrs->cluster_addr[osd].reset(new entity_addr_t()); | |
1527 | osd_addrs->hb_front_addr[osd].reset(new entity_addr_t()); | |
1528 | osd_addrs->hb_back_addr[osd].reset(new entity_addr_t()); | |
1529 | osd_state[osd] = 0; | |
1530 | } else { | |
1531 | osd_state[osd] ^= s; | |
1532 | } | |
1533 | } | |
1534 | ||
1535 | for (const auto &client : inc.new_up_client) { | |
1536 | osd_state[client.first] |= CEPH_OSD_EXISTS | CEPH_OSD_UP; | |
1537 | osd_addrs->client_addr[client.first].reset(new entity_addr_t(client.second)); | |
1538 | if (inc.new_hb_back_up.empty()) | |
1539 | osd_addrs->hb_back_addr[client.first].reset(new entity_addr_t(client.second)); //this is a backward-compatibility hack | |
1540 | else | |
1541 | osd_addrs->hb_back_addr[client.first].reset( | |
1542 | new entity_addr_t(inc.new_hb_back_up.find(client.first)->second)); | |
1543 | const auto j = inc.new_hb_front_up.find(client.first); | |
1544 | if (j != inc.new_hb_front_up.end()) | |
1545 | osd_addrs->hb_front_addr[client.first].reset(new entity_addr_t(j->second)); | |
1546 | else | |
1547 | osd_addrs->hb_front_addr[client.first].reset(); | |
1548 | ||
1549 | osd_info[client.first].up_from = epoch; | |
1550 | } | |
1551 | ||
1552 | for (const auto &cluster : inc.new_up_cluster) | |
1553 | osd_addrs->cluster_addr[cluster.first].reset(new entity_addr_t(cluster.second)); | |
1554 | ||
1555 | // info | |
1556 | for (const auto &thru : inc.new_up_thru) | |
1557 | osd_info[thru.first].up_thru = thru.second; | |
1558 | ||
1559 | for (const auto &interval : inc.new_last_clean_interval) { | |
1560 | osd_info[interval.first].last_clean_begin = interval.second.first; | |
1561 | osd_info[interval.first].last_clean_end = interval.second.second; | |
1562 | } | |
1563 | ||
1564 | for (const auto &lost : inc.new_lost) | |
1565 | osd_info[lost.first].lost_at = lost.second; | |
1566 | ||
1567 | // xinfo | |
1568 | for (const auto &xinfo : inc.new_xinfo) | |
1569 | osd_xinfo[xinfo.first] = xinfo.second; | |
1570 | ||
1571 | // uuid | |
1572 | for (const auto &uuid : inc.new_uuid) | |
1573 | (*osd_uuid)[uuid.first] = uuid.second; | |
1574 | ||
1575 | // pg rebuild | |
1576 | for (const auto &pg : inc.new_pg_temp) { | |
1577 | if (pg.second.empty()) | |
1578 | pg_temp->erase(pg.first); | |
1579 | else | |
1580 | (*pg_temp)[pg.first] = pg.second; | |
1581 | } | |
1582 | ||
1583 | for (const auto &pg : inc.new_primary_temp) { | |
1584 | if (pg.second == -1) | |
1585 | primary_temp->erase(pg.first); | |
1586 | else | |
1587 | (*primary_temp)[pg.first] = pg.second; | |
1588 | } | |
1589 | ||
1590 | for (auto& p : inc.new_pg_upmap) { | |
1591 | pg_upmap[p.first] = p.second; | |
1592 | } | |
1593 | for (auto& pg : inc.old_pg_upmap) { | |
1594 | pg_upmap.erase(pg); | |
1595 | } | |
1596 | for (auto& p : inc.new_pg_upmap_items) { | |
1597 | pg_upmap_items[p.first] = p.second; | |
1598 | } | |
1599 | for (auto& pg : inc.old_pg_upmap_items) { | |
1600 | pg_upmap_items.erase(pg); | |
1601 | } | |
1602 | ||
1603 | // blacklist | |
1604 | if (!inc.new_blacklist.empty()) { | |
1605 | blacklist.insert(inc.new_blacklist.begin(),inc.new_blacklist.end()); | |
1606 | new_blacklist_entries = true; | |
1607 | } | |
1608 | for (const auto &addr : inc.old_blacklist) | |
1609 | blacklist.erase(addr); | |
1610 | ||
1611 | // cluster snapshot? | |
1612 | if (inc.cluster_snapshot.length()) { | |
1613 | cluster_snapshot = inc.cluster_snapshot; | |
1614 | cluster_snapshot_epoch = inc.epoch; | |
1615 | } else { | |
1616 | cluster_snapshot.clear(); | |
1617 | cluster_snapshot_epoch = 0; | |
1618 | } | |
1619 | ||
1620 | if (inc.new_nearfull_ratio >= 0) { | |
1621 | nearfull_ratio = inc.new_nearfull_ratio; | |
1622 | } | |
1623 | if (inc.new_backfillfull_ratio >= 0) { | |
1624 | backfillfull_ratio = inc.new_backfillfull_ratio; | |
1625 | } | |
1626 | if (inc.new_full_ratio >= 0) { | |
1627 | full_ratio = inc.new_full_ratio; | |
1628 | } | |
1629 | if (inc.new_require_min_compat_client.length()) { | |
1630 | require_min_compat_client = inc.new_require_min_compat_client; | |
1631 | } | |
1632 | ||
1633 | // do new crush map last (after up/down stuff) | |
1634 | if (inc.crush.length()) { | |
1635 | bufferlist bl(inc.crush); | |
1636 | auto blp = bl.begin(); | |
1637 | crush.reset(new CrushWrapper); | |
1638 | crush->decode(blp); | |
1639 | } | |
1640 | ||
1641 | calc_num_osds(); | |
1642 | _calc_up_osd_features(); | |
1643 | return 0; | |
1644 | } | |
1645 | ||
1646 | // mapping | |
1647 | int OSDMap::map_to_pg( | |
1648 | int64_t poolid, | |
1649 | const string& name, | |
1650 | const string& key, | |
1651 | const string& nspace, | |
1652 | pg_t *pg) const | |
1653 | { | |
1654 | // calculate ps (placement seed) | |
1655 | const pg_pool_t *pool = get_pg_pool(poolid); | |
1656 | if (!pool) | |
1657 | return -ENOENT; | |
1658 | ps_t ps; | |
1659 | if (!key.empty()) | |
1660 | ps = pool->hash_key(key, nspace); | |
1661 | else | |
1662 | ps = pool->hash_key(name, nspace); | |
1663 | *pg = pg_t(ps, poolid); | |
1664 | return 0; | |
1665 | } | |
1666 | ||
1667 | int OSDMap::object_locator_to_pg( | |
1668 | const object_t& oid, const object_locator_t& loc, pg_t &pg) const | |
1669 | { | |
1670 | if (loc.hash >= 0) { | |
1671 | if (!get_pg_pool(loc.get_pool())) { | |
1672 | return -ENOENT; | |
1673 | } | |
1674 | pg = pg_t(loc.hash, loc.get_pool()); | |
1675 | return 0; | |
1676 | } | |
1677 | return map_to_pg(loc.get_pool(), oid.name, loc.key, loc.nspace, &pg); | |
1678 | } | |
1679 | ||
1680 | ceph_object_layout OSDMap::make_object_layout( | |
1681 | object_t oid, int pg_pool, string nspace) const | |
1682 | { | |
1683 | object_locator_t loc(pg_pool, nspace); | |
1684 | ||
1685 | ceph_object_layout ol; | |
1686 | pg_t pgid = object_locator_to_pg(oid, loc); | |
1687 | ol.ol_pgid = pgid.get_old_pg().v; | |
1688 | ol.ol_stripe_unit = 0; | |
1689 | return ol; | |
1690 | } | |
1691 | ||
1692 | void OSDMap::_remove_nonexistent_osds(const pg_pool_t& pool, | |
1693 | vector<int>& osds) const | |
1694 | { | |
1695 | if (pool.can_shift_osds()) { | |
1696 | unsigned removed = 0; | |
1697 | for (unsigned i = 0; i < osds.size(); i++) { | |
1698 | if (!exists(osds[i])) { | |
1699 | removed++; | |
1700 | continue; | |
1701 | } | |
1702 | if (removed) { | |
1703 | osds[i - removed] = osds[i]; | |
1704 | } | |
1705 | } | |
1706 | if (removed) | |
1707 | osds.resize(osds.size() - removed); | |
1708 | } else { | |
1709 | for (auto& osd : osds) { | |
1710 | if (!exists(osd)) | |
1711 | osd = CRUSH_ITEM_NONE; | |
1712 | } | |
1713 | } | |
1714 | } | |
1715 | ||
1716 | int OSDMap::_pg_to_raw_osds( | |
1717 | const pg_pool_t& pool, pg_t pg, | |
1718 | vector<int> *osds, | |
1719 | ps_t *ppps) const | |
1720 | { | |
1721 | // map to osds[] | |
1722 | ps_t pps = pool.raw_pg_to_pps(pg); // placement ps | |
1723 | unsigned size = pool.get_size(); | |
1724 | ||
1725 | // what crush rule? | |
1726 | int ruleno = crush->find_rule(pool.get_crush_ruleset(), pool.get_type(), size); | |
1727 | if (ruleno >= 0) | |
1728 | crush->do_rule(ruleno, pps, *osds, size, osd_weight, pg.pool()); | |
1729 | ||
1730 | _remove_nonexistent_osds(pool, *osds); | |
1731 | ||
1732 | if (ppps) | |
1733 | *ppps = pps; | |
1734 | ||
1735 | return osds->size(); | |
1736 | } | |
1737 | ||
1738 | int OSDMap::_pick_primary(const vector<int>& osds) const | |
1739 | { | |
1740 | for (auto osd : osds) { | |
1741 | if (osd != CRUSH_ITEM_NONE) { | |
1742 | return osd; | |
1743 | } | |
1744 | } | |
1745 | return -1; | |
1746 | } | |
1747 | ||
1748 | void OSDMap::_apply_remap(const pg_pool_t& pi, pg_t raw_pg, vector<int> *raw) const | |
1749 | { | |
1750 | pg_t pg = pi.raw_pg_to_pg(raw_pg); | |
1751 | auto p = pg_upmap.find(pg); | |
1752 | if (p != pg_upmap.end()) { | |
1753 | // make sure targets aren't marked out | |
1754 | for (auto osd : p->second) { | |
1755 | if (osd != CRUSH_ITEM_NONE && osd < max_osd && osd_weight[osd] == 0) { | |
1756 | // reject/ignore the explicit mapping | |
1757 | return; | |
1758 | } | |
1759 | } | |
1760 | *raw = vector<int>(p->second.begin(), p->second.end()); | |
1761 | return; | |
1762 | } | |
1763 | ||
1764 | auto q = pg_upmap_items.find(pg); | |
1765 | if (q != pg_upmap_items.end()) { | |
1766 | // NOTE: this approach does not allow a bidirectional swap, | |
1767 | // e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1]. | |
1768 | for (auto& r : q->second) { | |
1769 | // make sure the replacement value doesn't already appear | |
1770 | bool exists = false; | |
1771 | ssize_t pos = -1; | |
1772 | for (unsigned i = 0; i < raw->size(); ++i) { | |
1773 | int osd = (*raw)[i]; | |
1774 | if (osd == r.second) { | |
1775 | exists = true; | |
1776 | break; | |
1777 | } | |
1778 | // ignore mapping if target is marked out (or invalid osd id) | |
1779 | if (osd == r.first && | |
1780 | pos < 0 && | |
1781 | !(r.second != CRUSH_ITEM_NONE && r.second < max_osd && | |
1782 | osd_weight[r.second] == 0)) { | |
1783 | pos = i; | |
1784 | } | |
1785 | } | |
1786 | if (!exists && pos >= 0) { | |
1787 | (*raw)[pos] = r.second; | |
1788 | return; | |
1789 | } | |
1790 | } | |
1791 | } | |
1792 | } | |
1793 | ||
1794 | // pg -> (up osd list) | |
1795 | void OSDMap::_raw_to_up_osds(const pg_pool_t& pool, const vector<int>& raw, | |
1796 | vector<int> *up) const | |
1797 | { | |
1798 | if (pool.can_shift_osds()) { | |
1799 | // shift left | |
1800 | up->clear(); | |
1801 | up->reserve(raw.size()); | |
1802 | for (unsigned i=0; i<raw.size(); i++) { | |
1803 | if (!exists(raw[i]) || is_down(raw[i])) | |
1804 | continue; | |
1805 | up->push_back(raw[i]); | |
1806 | } | |
1807 | } else { | |
1808 | // set down/dne devices to NONE | |
1809 | up->resize(raw.size()); | |
1810 | for (int i = raw.size() - 1; i >= 0; --i) { | |
1811 | if (!exists(raw[i]) || is_down(raw[i])) { | |
1812 | (*up)[i] = CRUSH_ITEM_NONE; | |
1813 | } else { | |
1814 | (*up)[i] = raw[i]; | |
1815 | } | |
1816 | } | |
1817 | } | |
1818 | } | |
1819 | ||
1820 | void OSDMap::_apply_primary_affinity(ps_t seed, | |
1821 | const pg_pool_t& pool, | |
1822 | vector<int> *osds, | |
1823 | int *primary) const | |
1824 | { | |
1825 | // do we have any non-default primary_affinity values for these osds? | |
1826 | if (!osd_primary_affinity) | |
1827 | return; | |
1828 | ||
1829 | bool any = false; | |
1830 | for (const auto osd : *osds) { | |
1831 | if (osd != CRUSH_ITEM_NONE && | |
1832 | (*osd_primary_affinity)[osd] != CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) { | |
1833 | any = true; | |
1834 | break; | |
1835 | } | |
1836 | } | |
1837 | if (!any) | |
1838 | return; | |
1839 | ||
1840 | // pick the primary. feed both the seed (for the pg) and the osd | |
1841 | // into the hash/rng so that a proportional fraction of an osd's pgs | |
1842 | // get rejected as primary. | |
1843 | int pos = -1; | |
1844 | for (unsigned i = 0; i < osds->size(); ++i) { | |
1845 | int o = (*osds)[i]; | |
1846 | if (o == CRUSH_ITEM_NONE) | |
1847 | continue; | |
1848 | unsigned a = (*osd_primary_affinity)[o]; | |
1849 | if (a < CEPH_OSD_MAX_PRIMARY_AFFINITY && | |
1850 | (crush_hash32_2(CRUSH_HASH_RJENKINS1, | |
1851 | seed, o) >> 16) >= a) { | |
1852 | // we chose not to use this primary. note it anyway as a | |
1853 | // fallback in case we don't pick anyone else, but keep looking. | |
1854 | if (pos < 0) | |
1855 | pos = i; | |
1856 | } else { | |
1857 | pos = i; | |
1858 | break; | |
1859 | } | |
1860 | } | |
1861 | if (pos < 0) | |
1862 | return; | |
1863 | ||
1864 | *primary = (*osds)[pos]; | |
1865 | ||
1866 | if (pool.can_shift_osds() && pos > 0) { | |
1867 | // move the new primary to the front. | |
1868 | for (int i = pos; i > 0; --i) { | |
1869 | (*osds)[i] = (*osds)[i-1]; | |
1870 | } | |
1871 | (*osds)[0] = *primary; | |
1872 | } | |
1873 | } | |
1874 | ||
1875 | void OSDMap::_get_temp_osds(const pg_pool_t& pool, pg_t pg, | |
1876 | vector<int> *temp_pg, int *temp_primary) const | |
1877 | { | |
1878 | pg = pool.raw_pg_to_pg(pg); | |
1879 | const auto p = pg_temp->find(pg); | |
1880 | temp_pg->clear(); | |
1881 | if (p != pg_temp->end()) { | |
1882 | for (unsigned i=0; i<p->second.size(); i++) { | |
1883 | if (!exists(p->second[i]) || is_down(p->second[i])) { | |
1884 | if (pool.can_shift_osds()) { | |
1885 | continue; | |
1886 | } else { | |
1887 | temp_pg->push_back(CRUSH_ITEM_NONE); | |
1888 | } | |
1889 | } else { | |
1890 | temp_pg->push_back(p->second[i]); | |
1891 | } | |
1892 | } | |
1893 | } | |
1894 | const auto &pp = primary_temp->find(pg); | |
1895 | *temp_primary = -1; | |
1896 | if (pp != primary_temp->end()) { | |
1897 | *temp_primary = pp->second; | |
1898 | } else if (!temp_pg->empty()) { // apply pg_temp's primary | |
1899 | for (unsigned i = 0; i < temp_pg->size(); ++i) { | |
1900 | if ((*temp_pg)[i] != CRUSH_ITEM_NONE) { | |
1901 | *temp_primary = (*temp_pg)[i]; | |
1902 | break; | |
1903 | } | |
1904 | } | |
1905 | } | |
1906 | } | |
1907 | ||
1908 | int OSDMap::pg_to_raw_osds(pg_t pg, vector<int> *raw, int *primary) const | |
1909 | { | |
1910 | *primary = -1; | |
1911 | raw->clear(); | |
1912 | const pg_pool_t *pool = get_pg_pool(pg.pool()); | |
1913 | if (!pool) | |
1914 | return 0; | |
1915 | int r = _pg_to_raw_osds(*pool, pg, raw, NULL); | |
1916 | if (primary) | |
1917 | *primary = _pick_primary(*raw); | |
1918 | return r; | |
1919 | } | |
1920 | ||
1921 | void OSDMap::pg_to_raw_up(pg_t pg, vector<int> *up, int *primary) const | |
1922 | { | |
1923 | const pg_pool_t *pool = get_pg_pool(pg.pool()); | |
1924 | if (!pool) { | |
1925 | if (primary) | |
1926 | *primary = -1; | |
1927 | if (up) | |
1928 | up->clear(); | |
1929 | return; | |
1930 | } | |
1931 | vector<int> raw; | |
1932 | ps_t pps; | |
1933 | _pg_to_raw_osds(*pool, pg, &raw, &pps); | |
1934 | _apply_remap(*pool, pg, &raw); | |
1935 | _raw_to_up_osds(*pool, raw, up); | |
1936 | *primary = _pick_primary(raw); | |
1937 | _apply_primary_affinity(pps, *pool, up, primary); | |
1938 | } | |
1939 | ||
1940 | void OSDMap::_pg_to_up_acting_osds( | |
1941 | const pg_t& pg, vector<int> *up, int *up_primary, | |
1942 | vector<int> *acting, int *acting_primary, | |
1943 | bool raw_pg_to_pg) const | |
1944 | { | |
1945 | const pg_pool_t *pool = get_pg_pool(pg.pool()); | |
1946 | if (!pool || | |
1947 | (!raw_pg_to_pg && pg.ps() >= pool->get_pg_num())) { | |
1948 | if (up) | |
1949 | up->clear(); | |
1950 | if (up_primary) | |
1951 | *up_primary = -1; | |
1952 | if (acting) | |
1953 | acting->clear(); | |
1954 | if (acting_primary) | |
1955 | *acting_primary = -1; | |
1956 | return; | |
1957 | } | |
1958 | vector<int> raw; | |
1959 | vector<int> _up; | |
1960 | vector<int> _acting; | |
1961 | int _up_primary; | |
1962 | int _acting_primary; | |
1963 | ps_t pps; | |
1964 | _get_temp_osds(*pool, pg, &_acting, &_acting_primary); | |
1965 | if (_acting.empty() || up || up_primary) { | |
1966 | _pg_to_raw_osds(*pool, pg, &raw, &pps); | |
1967 | _apply_remap(*pool, pg, &raw); | |
1968 | _raw_to_up_osds(*pool, raw, &_up); | |
1969 | _up_primary = _pick_primary(_up); | |
1970 | _apply_primary_affinity(pps, *pool, &_up, &_up_primary); | |
1971 | if (_acting.empty()) { | |
1972 | _acting = _up; | |
1973 | if (_acting_primary == -1) { | |
1974 | _acting_primary = _up_primary; | |
1975 | } | |
1976 | } | |
1977 | ||
1978 | if (up) | |
1979 | up->swap(_up); | |
1980 | if (up_primary) | |
1981 | *up_primary = _up_primary; | |
1982 | } | |
1983 | ||
1984 | if (acting) | |
1985 | acting->swap(_acting); | |
1986 | if (acting_primary) | |
1987 | *acting_primary = _acting_primary; | |
1988 | } | |
1989 | ||
1990 | int OSDMap::calc_pg_rank(int osd, const vector<int>& acting, int nrep) | |
1991 | { | |
1992 | if (!nrep) | |
1993 | nrep = acting.size(); | |
1994 | for (int i=0; i<nrep; i++) | |
1995 | if (acting[i] == osd) | |
1996 | return i; | |
1997 | return -1; | |
1998 | } | |
1999 | ||
2000 | int OSDMap::calc_pg_role(int osd, const vector<int>& acting, int nrep) | |
2001 | { | |
2002 | return calc_pg_rank(osd, acting, nrep); | |
2003 | } | |
2004 | ||
2005 | bool OSDMap::primary_changed( | |
2006 | int oldprimary, | |
2007 | const vector<int> &oldacting, | |
2008 | int newprimary, | |
2009 | const vector<int> &newacting) | |
2010 | { | |
2011 | if (oldacting.empty() && newacting.empty()) | |
2012 | return false; // both still empty | |
2013 | if (oldacting.empty() ^ newacting.empty()) | |
2014 | return true; // was empty, now not, or vice versa | |
2015 | if (oldprimary != newprimary) | |
2016 | return true; // primary changed | |
2017 | if (calc_pg_rank(oldprimary, oldacting) != | |
2018 | calc_pg_rank(newprimary, newacting)) | |
2019 | return true; | |
2020 | return false; // same primary (tho replicas may have changed) | |
2021 | } | |
2022 | ||
2023 | ||
2024 | // serialize, unserialize | |
2025 | void OSDMap::encode_client_old(bufferlist& bl) const | |
2026 | { | |
2027 | __u16 v = 5; | |
2028 | ::encode(v, bl); | |
2029 | ||
2030 | // base | |
2031 | ::encode(fsid, bl); | |
2032 | ::encode(epoch, bl); | |
2033 | ::encode(created, bl); | |
2034 | ::encode(modified, bl); | |
2035 | ||
2036 | // for ::encode(pools, bl); | |
2037 | __u32 n = pools.size(); | |
2038 | ::encode(n, bl); | |
2039 | ||
2040 | for (const auto &pool : pools) { | |
2041 | n = pool.first; | |
2042 | ::encode(n, bl); | |
2043 | ::encode(pool.second, bl, 0); | |
2044 | } | |
2045 | // for ::encode(pool_name, bl); | |
2046 | n = pool_name.size(); | |
2047 | ::encode(n, bl); | |
2048 | for (const auto &pname : pool_name) { | |
2049 | n = pname.first; | |
2050 | ::encode(n, bl); | |
2051 | ::encode(pname.second, bl); | |
2052 | } | |
2053 | // for ::encode(pool_max, bl); | |
2054 | n = pool_max; | |
2055 | ::encode(n, bl); | |
2056 | ||
2057 | ::encode(flags, bl); | |
2058 | ||
2059 | ::encode(max_osd, bl); | |
2060 | ::encode(osd_state, bl); | |
2061 | ::encode(osd_weight, bl); | |
2062 | ::encode(osd_addrs->client_addr, bl, 0); | |
2063 | ||
2064 | // for ::encode(pg_temp, bl); | |
2065 | n = pg_temp->size(); | |
2066 | ::encode(n, bl); | |
2067 | for (const auto pg : *pg_temp) { | |
2068 | old_pg_t opg = pg.first.get_old_pg(); | |
2069 | ::encode(opg, bl); | |
2070 | ::encode(pg.second, bl); | |
2071 | } | |
2072 | ||
2073 | // crush | |
2074 | bufferlist cbl; | |
2075 | crush->encode(cbl, 0 /* legacy (no) features */); | |
2076 | ::encode(cbl, bl); | |
2077 | } | |
2078 | ||
2079 | void OSDMap::encode_classic(bufferlist& bl, uint64_t features) const | |
2080 | { | |
2081 | if ((features & CEPH_FEATURE_PGID64) == 0) { | |
2082 | encode_client_old(bl); | |
2083 | return; | |
2084 | } | |
2085 | ||
2086 | __u16 v = 6; | |
2087 | ::encode(v, bl); | |
2088 | ||
2089 | // base | |
2090 | ::encode(fsid, bl); | |
2091 | ::encode(epoch, bl); | |
2092 | ::encode(created, bl); | |
2093 | ::encode(modified, bl); | |
2094 | ||
2095 | ::encode(pools, bl, features); | |
2096 | ::encode(pool_name, bl); | |
2097 | ::encode(pool_max, bl); | |
2098 | ||
2099 | ::encode(flags, bl); | |
2100 | ||
2101 | ::encode(max_osd, bl); | |
2102 | ::encode(osd_state, bl); | |
2103 | ::encode(osd_weight, bl); | |
2104 | ::encode(osd_addrs->client_addr, bl, features); | |
2105 | ||
2106 | ::encode(*pg_temp, bl); | |
2107 | ||
2108 | // crush | |
2109 | bufferlist cbl; | |
2110 | crush->encode(cbl, 0 /* legacy (no) features */); | |
2111 | ::encode(cbl, bl); | |
2112 | ||
2113 | // extended | |
2114 | __u16 ev = 10; | |
2115 | ::encode(ev, bl); | |
2116 | ::encode(osd_addrs->hb_back_addr, bl, features); | |
2117 | ::encode(osd_info, bl); | |
2118 | ::encode(blacklist, bl, features); | |
2119 | ::encode(osd_addrs->cluster_addr, bl, features); | |
2120 | ::encode(cluster_snapshot_epoch, bl); | |
2121 | ::encode(cluster_snapshot, bl); | |
2122 | ::encode(*osd_uuid, bl); | |
2123 | ::encode(osd_xinfo, bl); | |
2124 | ::encode(osd_addrs->hb_front_addr, bl, features); | |
2125 | } | |
2126 | ||
2127 | void OSDMap::encode(bufferlist& bl, uint64_t features) const | |
2128 | { | |
2129 | if ((features & CEPH_FEATURE_OSDMAP_ENC) == 0) { | |
2130 | encode_classic(bl, features); | |
2131 | return; | |
2132 | } | |
2133 | ||
2134 | // only a select set of callers should *ever* be encoding new | |
2135 | // OSDMaps. others should be passing around the canonical encoded | |
2136 | // buffers from on high. select out those callers by passing in an | |
2137 | // "impossible" feature bit. | |
2138 | assert(features & CEPH_FEATURE_RESERVED); | |
2139 | features &= ~CEPH_FEATURE_RESERVED; | |
2140 | ||
2141 | size_t start_offset = bl.length(); | |
2142 | size_t tail_offset; | |
2143 | buffer::list::iterator crc_it; | |
2144 | ||
2145 | // meta-encoding: how we include client-used and osd-specific data | |
2146 | ENCODE_START(8, 7, bl); | |
2147 | ||
2148 | { | |
2149 | uint8_t v = 4; | |
2150 | if (!HAVE_FEATURE(features, OSDMAP_PG_UPMAP)) { | |
2151 | v = 3; | |
2152 | } | |
2153 | ENCODE_START(v, 1, bl); // client-usable data | |
2154 | // base | |
2155 | ::encode(fsid, bl); | |
2156 | ::encode(epoch, bl); | |
2157 | ::encode(created, bl); | |
2158 | ::encode(modified, bl); | |
2159 | ||
2160 | ::encode(pools, bl, features); | |
2161 | ::encode(pool_name, bl); | |
2162 | ::encode(pool_max, bl); | |
2163 | ||
2164 | ::encode(flags, bl); | |
2165 | ||
2166 | ::encode(max_osd, bl); | |
2167 | ::encode(osd_state, bl); | |
2168 | ::encode(osd_weight, bl); | |
2169 | ::encode(osd_addrs->client_addr, bl, features); | |
2170 | ||
2171 | ::encode(*pg_temp, bl); | |
2172 | ::encode(*primary_temp, bl); | |
2173 | if (osd_primary_affinity) { | |
2174 | ::encode(*osd_primary_affinity, bl); | |
2175 | } else { | |
2176 | vector<__u32> v; | |
2177 | ::encode(v, bl); | |
2178 | } | |
2179 | ||
2180 | // crush | |
2181 | bufferlist cbl; | |
2182 | crush->encode(cbl, features); | |
2183 | ::encode(cbl, bl); | |
2184 | ::encode(erasure_code_profiles, bl); | |
2185 | ||
2186 | if (v >= 4) { | |
2187 | ::encode(pg_upmap, bl); | |
2188 | ::encode(pg_upmap_items, bl); | |
2189 | } else { | |
2190 | assert(pg_upmap.empty()); | |
2191 | assert(pg_upmap_items.empty()); | |
2192 | } | |
2193 | ENCODE_FINISH(bl); // client-usable data | |
2194 | } | |
2195 | ||
2196 | { | |
2197 | uint8_t target_v = 4; | |
2198 | if (!HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
2199 | target_v = 1; | |
2200 | } | |
2201 | ENCODE_START(target_v, 1, bl); // extended, osd-only data | |
2202 | ::encode(osd_addrs->hb_back_addr, bl, features); | |
2203 | ::encode(osd_info, bl); | |
2204 | { | |
2205 | // put this in a sorted, ordered map<> so that we encode in a | |
2206 | // deterministic order. | |
2207 | map<entity_addr_t,utime_t> blacklist_map; | |
2208 | for (const auto &addr : blacklist) | |
2209 | blacklist_map.insert(make_pair(addr.first, addr.second)); | |
2210 | ::encode(blacklist_map, bl, features); | |
2211 | } | |
2212 | ::encode(osd_addrs->cluster_addr, bl, features); | |
2213 | ::encode(cluster_snapshot_epoch, bl); | |
2214 | ::encode(cluster_snapshot, bl); | |
2215 | ::encode(*osd_uuid, bl); | |
2216 | ::encode(osd_xinfo, bl); | |
2217 | ::encode(osd_addrs->hb_front_addr, bl, features); | |
2218 | if (target_v >= 2) { | |
2219 | ::encode(nearfull_ratio, bl); | |
2220 | ::encode(full_ratio, bl); | |
2221 | ::encode(backfillfull_ratio, bl); | |
2222 | ::encode(require_min_compat_client, bl); | |
2223 | } | |
2224 | ENCODE_FINISH(bl); // osd-only data | |
2225 | } | |
2226 | ||
2227 | ::encode((uint32_t)0, bl); // dummy crc | |
2228 | crc_it = bl.end(); | |
2229 | crc_it.advance(-4); | |
2230 | tail_offset = bl.length(); | |
2231 | ||
2232 | ENCODE_FINISH(bl); // meta-encoding wrapper | |
2233 | ||
2234 | // fill in crc | |
2235 | bufferlist front; | |
2236 | front.substr_of(bl, start_offset, crc_it.get_off() - start_offset); | |
2237 | crc = front.crc32c(-1); | |
2238 | if (tail_offset < bl.length()) { | |
2239 | bufferlist tail; | |
2240 | tail.substr_of(bl, tail_offset, bl.length() - tail_offset); | |
2241 | crc = tail.crc32c(crc); | |
2242 | } | |
2243 | ceph_le32 crc_le; | |
2244 | crc_le = crc; | |
2245 | crc_it.copy_in(4, (char*)&crc_le); | |
2246 | crc_defined = true; | |
2247 | } | |
2248 | ||
2249 | void OSDMap::decode(bufferlist& bl) | |
2250 | { | |
2251 | auto p = bl.begin(); | |
2252 | decode(p); | |
2253 | } | |
2254 | ||
2255 | void OSDMap::decode_classic(bufferlist::iterator& p) | |
2256 | { | |
2257 | __u32 n, t; | |
2258 | __u16 v; | |
2259 | ::decode(v, p); | |
2260 | ||
2261 | // base | |
2262 | ::decode(fsid, p); | |
2263 | ::decode(epoch, p); | |
2264 | ::decode(created, p); | |
2265 | ::decode(modified, p); | |
2266 | ||
2267 | if (v < 6) { | |
2268 | if (v < 4) { | |
2269 | int32_t max_pools = 0; | |
2270 | ::decode(max_pools, p); | |
2271 | pool_max = max_pools; | |
2272 | } | |
2273 | pools.clear(); | |
2274 | ::decode(n, p); | |
2275 | while (n--) { | |
2276 | ::decode(t, p); | |
2277 | ::decode(pools[t], p); | |
2278 | } | |
2279 | if (v == 4) { | |
2280 | ::decode(n, p); | |
2281 | pool_max = n; | |
2282 | } else if (v == 5) { | |
2283 | pool_name.clear(); | |
2284 | ::decode(n, p); | |
2285 | while (n--) { | |
2286 | ::decode(t, p); | |
2287 | ::decode(pool_name[t], p); | |
2288 | } | |
2289 | ::decode(n, p); | |
2290 | pool_max = n; | |
2291 | } | |
2292 | } else { | |
2293 | ::decode(pools, p); | |
2294 | ::decode(pool_name, p); | |
2295 | ::decode(pool_max, p); | |
2296 | } | |
2297 | // kludge around some old bug that zeroed out pool_max (#2307) | |
2298 | if (pools.size() && pool_max < pools.rbegin()->first) { | |
2299 | pool_max = pools.rbegin()->first; | |
2300 | } | |
2301 | ||
2302 | ::decode(flags, p); | |
2303 | ||
2304 | ::decode(max_osd, p); | |
2305 | ::decode(osd_state, p); | |
2306 | ::decode(osd_weight, p); | |
2307 | ::decode(osd_addrs->client_addr, p); | |
2308 | if (v <= 5) { | |
2309 | pg_temp->clear(); | |
2310 | ::decode(n, p); | |
2311 | while (n--) { | |
2312 | old_pg_t opg; | |
2313 | ::decode_raw(opg, p); | |
2314 | ::decode((*pg_temp)[pg_t(opg)], p); | |
2315 | } | |
2316 | } else { | |
2317 | ::decode(*pg_temp, p); | |
2318 | } | |
2319 | ||
2320 | // crush | |
2321 | bufferlist cbl; | |
2322 | ::decode(cbl, p); | |
2323 | auto cblp = cbl.begin(); | |
2324 | crush->decode(cblp); | |
2325 | ||
2326 | // extended | |
2327 | __u16 ev = 0; | |
2328 | if (v >= 5) | |
2329 | ::decode(ev, p); | |
2330 | ::decode(osd_addrs->hb_back_addr, p); | |
2331 | ::decode(osd_info, p); | |
2332 | if (v < 5) | |
2333 | ::decode(pool_name, p); | |
2334 | ||
2335 | ::decode(blacklist, p); | |
2336 | if (ev >= 6) | |
2337 | ::decode(osd_addrs->cluster_addr, p); | |
2338 | else | |
2339 | osd_addrs->cluster_addr.resize(osd_addrs->client_addr.size()); | |
2340 | ||
2341 | if (ev >= 7) { | |
2342 | ::decode(cluster_snapshot_epoch, p); | |
2343 | ::decode(cluster_snapshot, p); | |
2344 | } | |
2345 | ||
2346 | if (ev >= 8) { | |
2347 | ::decode(*osd_uuid, p); | |
2348 | } else { | |
2349 | osd_uuid->resize(max_osd); | |
2350 | } | |
2351 | if (ev >= 9) | |
2352 | ::decode(osd_xinfo, p); | |
2353 | else | |
2354 | osd_xinfo.resize(max_osd); | |
2355 | ||
2356 | if (ev >= 10) | |
2357 | ::decode(osd_addrs->hb_front_addr, p); | |
2358 | else | |
2359 | osd_addrs->hb_front_addr.resize(osd_addrs->hb_back_addr.size()); | |
2360 | ||
2361 | osd_primary_affinity.reset(); | |
2362 | ||
2363 | post_decode(); | |
2364 | } | |
2365 | ||
2366 | void OSDMap::decode(bufferlist::iterator& bl) | |
2367 | { | |
2368 | /** | |
2369 | * Older encodings of the OSDMap had a single struct_v which | |
2370 | * covered the whole encoding, and was prior to our modern | |
2371 | * stuff which includes a compatv and a size. So if we see | |
2372 | * a struct_v < 7, we must rewind to the beginning and use our | |
2373 | * classic decoder. | |
2374 | */ | |
2375 | size_t start_offset = bl.get_off(); | |
2376 | size_t tail_offset = 0; | |
2377 | bufferlist crc_front, crc_tail; | |
2378 | ||
2379 | DECODE_START_LEGACY_COMPAT_LEN(8, 7, 7, bl); // wrapper | |
2380 | if (struct_v < 7) { | |
2381 | int struct_v_size = sizeof(struct_v); | |
2382 | bl.advance(-struct_v_size); | |
2383 | decode_classic(bl); | |
2384 | return; | |
2385 | } | |
2386 | /** | |
2387 | * Since we made it past that hurdle, we can use our normal paths. | |
2388 | */ | |
2389 | { | |
2390 | DECODE_START(4, bl); // client-usable data | |
2391 | // base | |
2392 | ::decode(fsid, bl); | |
2393 | ::decode(epoch, bl); | |
2394 | ::decode(created, bl); | |
2395 | ::decode(modified, bl); | |
2396 | ||
2397 | ::decode(pools, bl); | |
2398 | ::decode(pool_name, bl); | |
2399 | ::decode(pool_max, bl); | |
2400 | ||
2401 | ::decode(flags, bl); | |
2402 | ||
2403 | ::decode(max_osd, bl); | |
2404 | ::decode(osd_state, bl); | |
2405 | ::decode(osd_weight, bl); | |
2406 | ::decode(osd_addrs->client_addr, bl); | |
2407 | ||
2408 | ::decode(*pg_temp, bl); | |
2409 | ::decode(*primary_temp, bl); | |
2410 | if (struct_v >= 2) { | |
2411 | osd_primary_affinity.reset(new mempool::osdmap::vector<__u32>); | |
2412 | ::decode(*osd_primary_affinity, bl); | |
2413 | if (osd_primary_affinity->empty()) | |
2414 | osd_primary_affinity.reset(); | |
2415 | } else { | |
2416 | osd_primary_affinity.reset(); | |
2417 | } | |
2418 | ||
2419 | // crush | |
2420 | bufferlist cbl; | |
2421 | ::decode(cbl, bl); | |
2422 | auto cblp = cbl.begin(); | |
2423 | crush->decode(cblp); | |
2424 | if (struct_v >= 3) { | |
2425 | ::decode(erasure_code_profiles, bl); | |
2426 | } else { | |
2427 | erasure_code_profiles.clear(); | |
2428 | } | |
2429 | if (struct_v >= 4) { | |
2430 | ::decode(pg_upmap, bl); | |
2431 | ::decode(pg_upmap_items, bl); | |
2432 | } else { | |
2433 | pg_upmap.clear(); | |
2434 | pg_upmap_items.clear(); | |
2435 | } | |
2436 | DECODE_FINISH(bl); // client-usable data | |
2437 | } | |
2438 | ||
2439 | { | |
2440 | DECODE_START(4, bl); // extended, osd-only data | |
2441 | ::decode(osd_addrs->hb_back_addr, bl); | |
2442 | ::decode(osd_info, bl); | |
2443 | ::decode(blacklist, bl); | |
2444 | ::decode(osd_addrs->cluster_addr, bl); | |
2445 | ::decode(cluster_snapshot_epoch, bl); | |
2446 | ::decode(cluster_snapshot, bl); | |
2447 | ::decode(*osd_uuid, bl); | |
2448 | ::decode(osd_xinfo, bl); | |
2449 | ::decode(osd_addrs->hb_front_addr, bl); | |
2450 | if (struct_v >= 2) { | |
2451 | ::decode(nearfull_ratio, bl); | |
2452 | ::decode(full_ratio, bl); | |
2453 | } else { | |
2454 | nearfull_ratio = 0; | |
2455 | full_ratio = 0; | |
2456 | } | |
2457 | if (struct_v >= 3) { | |
2458 | ::decode(backfillfull_ratio, bl); | |
2459 | } else { | |
2460 | backfillfull_ratio = 0; | |
2461 | } | |
2462 | if (struct_v >= 4) | |
2463 | ::decode(require_min_compat_client, bl); | |
2464 | DECODE_FINISH(bl); // osd-only data | |
2465 | } | |
2466 | ||
2467 | if (struct_v >= 8) { | |
2468 | crc_front.substr_of(bl.get_bl(), start_offset, bl.get_off() - start_offset); | |
2469 | ::decode(crc, bl); | |
2470 | tail_offset = bl.get_off(); | |
2471 | crc_defined = true; | |
2472 | } else { | |
2473 | crc_defined = false; | |
2474 | crc = 0; | |
2475 | } | |
2476 | ||
2477 | DECODE_FINISH(bl); // wrapper | |
2478 | ||
2479 | if (tail_offset) { | |
2480 | // verify crc | |
2481 | uint32_t actual = crc_front.crc32c(-1); | |
2482 | if (tail_offset < bl.get_off()) { | |
2483 | bufferlist tail; | |
2484 | tail.substr_of(bl.get_bl(), tail_offset, bl.get_off() - tail_offset); | |
2485 | actual = tail.crc32c(actual); | |
2486 | } | |
2487 | if (crc != actual) { | |
2488 | ostringstream ss; | |
2489 | ss << "bad crc, actual " << actual << " != expected " << crc; | |
2490 | string s = ss.str(); | |
2491 | throw buffer::malformed_input(s.c_str()); | |
2492 | } | |
2493 | } | |
2494 | ||
2495 | post_decode(); | |
2496 | } | |
2497 | ||
2498 | void OSDMap::post_decode() | |
2499 | { | |
2500 | // index pool names | |
2501 | name_pool.clear(); | |
2502 | for (const auto &pname : pool_name) { | |
2503 | name_pool[pname.second] = pname.first; | |
2504 | } | |
2505 | ||
2506 | calc_num_osds(); | |
2507 | _calc_up_osd_features(); | |
2508 | } | |
2509 | ||
2510 | void OSDMap::dump_erasure_code_profiles( | |
2511 | const mempool::osdmap::map<string,map<string,string>>& profiles, | |
2512 | Formatter *f) | |
2513 | { | |
2514 | f->open_object_section("erasure_code_profiles"); | |
2515 | for (const auto &profile : profiles) { | |
2516 | f->open_object_section(profile.first.c_str()); | |
2517 | for (const auto &profm : profile.second) { | |
2518 | f->dump_string(profm.first.c_str(), profm.second.c_str()); | |
2519 | } | |
2520 | f->close_section(); | |
2521 | } | |
2522 | f->close_section(); | |
2523 | } | |
2524 | ||
2525 | void OSDMap::dump(Formatter *f) const | |
2526 | { | |
2527 | f->dump_int("epoch", get_epoch()); | |
2528 | f->dump_stream("fsid") << get_fsid(); | |
2529 | f->dump_stream("created") << get_created(); | |
2530 | f->dump_stream("modified") << get_modified(); | |
2531 | f->dump_string("flags", get_flag_string()); | |
2532 | f->dump_float("full_ratio", full_ratio); | |
2533 | f->dump_float("backfillfull_ratio", backfillfull_ratio); | |
2534 | f->dump_float("nearfull_ratio", nearfull_ratio); | |
2535 | f->dump_string("cluster_snapshot", get_cluster_snapshot()); | |
2536 | f->dump_int("pool_max", get_pool_max()); | |
2537 | f->dump_int("max_osd", get_max_osd()); | |
2538 | f->dump_string("require_min_compat_client", require_min_compat_client); | |
2539 | auto mv = get_min_compat_client(); | |
2540 | f->dump_string("min_compat_client", mv.first); | |
2541 | f->dump_string("min_compat_client_version", mv.second); | |
2542 | ||
2543 | f->open_array_section("pools"); | |
2544 | for (const auto &pool : pools) { | |
2545 | std::string name("<unknown>"); | |
2546 | const auto &pni = pool_name.find(pool.first); | |
2547 | if (pni != pool_name.end()) | |
2548 | name = pni->second; | |
2549 | f->open_object_section("pool"); | |
2550 | f->dump_int("pool", pool.first); | |
2551 | f->dump_string("pool_name", name); | |
2552 | pool.second.dump(f); | |
2553 | f->close_section(); | |
2554 | } | |
2555 | f->close_section(); | |
2556 | ||
2557 | f->open_array_section("osds"); | |
2558 | for (int i=0; i<get_max_osd(); i++) | |
2559 | if (exists(i)) { | |
2560 | f->open_object_section("osd_info"); | |
2561 | f->dump_int("osd", i); | |
2562 | f->dump_stream("uuid") << get_uuid(i); | |
2563 | f->dump_int("up", is_up(i)); | |
2564 | f->dump_int("in", is_in(i)); | |
2565 | f->dump_float("weight", get_weightf(i)); | |
2566 | f->dump_float("primary_affinity", get_primary_affinityf(i)); | |
2567 | get_info(i).dump(f); | |
2568 | f->dump_stream("public_addr") << get_addr(i); | |
2569 | f->dump_stream("cluster_addr") << get_cluster_addr(i); | |
2570 | f->dump_stream("heartbeat_back_addr") << get_hb_back_addr(i); | |
2571 | f->dump_stream("heartbeat_front_addr") << get_hb_front_addr(i); | |
2572 | ||
2573 | set<string> st; | |
2574 | get_state(i, st); | |
2575 | f->open_array_section("state"); | |
2576 | for (const auto &state : st) | |
2577 | f->dump_string("state", state); | |
2578 | f->close_section(); | |
2579 | ||
2580 | f->close_section(); | |
2581 | } | |
2582 | f->close_section(); | |
2583 | ||
2584 | f->open_array_section("osd_xinfo"); | |
2585 | for (int i=0; i<get_max_osd(); i++) { | |
2586 | if (exists(i)) { | |
2587 | f->open_object_section("xinfo"); | |
2588 | f->dump_int("osd", i); | |
2589 | osd_xinfo[i].dump(f); | |
2590 | f->close_section(); | |
2591 | } | |
2592 | } | |
2593 | f->close_section(); | |
2594 | ||
2595 | f->open_array_section("pg_upmap"); | |
2596 | for (auto& p : pg_upmap) { | |
2597 | f->open_object_section("mapping"); | |
2598 | f->dump_stream("pgid") << p.first; | |
2599 | f->open_array_section("osds"); | |
2600 | for (auto q : p.second) { | |
2601 | f->dump_int("osd", q); | |
2602 | } | |
2603 | f->close_section(); | |
2604 | f->close_section(); | |
2605 | } | |
2606 | f->close_section(); | |
2607 | f->open_array_section("pg_upmap_items"); | |
2608 | for (auto& p : pg_upmap_items) { | |
2609 | f->open_object_section("mapping"); | |
2610 | f->dump_stream("pgid") << p.first; | |
2611 | f->open_array_section("mappings"); | |
2612 | for (auto& q : p.second) { | |
2613 | f->open_object_section("mapping"); | |
2614 | f->dump_int("from", q.first); | |
2615 | f->dump_int("to", q.second); | |
2616 | f->close_section(); | |
2617 | } | |
2618 | f->close_section(); | |
2619 | f->close_section(); | |
2620 | } | |
2621 | f->close_section(); | |
2622 | f->open_array_section("pg_temp"); | |
2623 | for (const auto &pg : *pg_temp) { | |
2624 | f->open_object_section("osds"); | |
2625 | f->dump_stream("pgid") << pg.first; | |
2626 | f->open_array_section("osds"); | |
2627 | for (const auto osd : pg.second) | |
2628 | f->dump_int("osd", osd); | |
2629 | f->close_section(); | |
2630 | f->close_section(); | |
2631 | } | |
2632 | f->close_section(); | |
2633 | ||
2634 | f->open_array_section("primary_temp"); | |
2635 | for (const auto &pg : *primary_temp) { | |
2636 | f->dump_stream("pgid") << pg.first; | |
2637 | f->dump_int("osd", pg.second); | |
2638 | } | |
2639 | f->close_section(); // primary_temp | |
2640 | ||
2641 | f->open_object_section("blacklist"); | |
2642 | for (const auto &addr : blacklist) { | |
2643 | stringstream ss; | |
2644 | ss << addr.first; | |
2645 | f->dump_stream(ss.str().c_str()) << addr.second; | |
2646 | } | |
2647 | f->close_section(); | |
2648 | ||
2649 | dump_erasure_code_profiles(erasure_code_profiles, f); | |
2650 | } | |
2651 | ||
2652 | void OSDMap::generate_test_instances(list<OSDMap*>& o) | |
2653 | { | |
2654 | o.push_back(new OSDMap); | |
2655 | ||
2656 | CephContext *cct = new CephContext(CODE_ENVIRONMENT_UTILITY); | |
2657 | o.push_back(new OSDMap); | |
2658 | uuid_d fsid; | |
2659 | o.back()->build_simple(cct, 1, fsid, 16, 7, 8); | |
2660 | o.back()->created = o.back()->modified = utime_t(1, 2); // fix timestamp | |
2661 | o.back()->blacklist[entity_addr_t()] = utime_t(5, 6); | |
2662 | cct->put(); | |
2663 | } | |
2664 | ||
2665 | string OSDMap::get_flag_string(unsigned f) | |
2666 | { | |
2667 | string s; | |
2668 | if ( f& CEPH_OSDMAP_NEARFULL) | |
2669 | s += ",nearfull"; | |
2670 | if (f & CEPH_OSDMAP_FULL) | |
2671 | s += ",full"; | |
2672 | if (f & CEPH_OSDMAP_PAUSERD) | |
2673 | s += ",pauserd"; | |
2674 | if (f & CEPH_OSDMAP_PAUSEWR) | |
2675 | s += ",pausewr"; | |
2676 | if (f & CEPH_OSDMAP_PAUSEREC) | |
2677 | s += ",pauserec"; | |
2678 | if (f & CEPH_OSDMAP_NOUP) | |
2679 | s += ",noup"; | |
2680 | if (f & CEPH_OSDMAP_NODOWN) | |
2681 | s += ",nodown"; | |
2682 | if (f & CEPH_OSDMAP_NOOUT) | |
2683 | s += ",noout"; | |
2684 | if (f & CEPH_OSDMAP_NOIN) | |
2685 | s += ",noin"; | |
2686 | if (f & CEPH_OSDMAP_NOBACKFILL) | |
2687 | s += ",nobackfill"; | |
2688 | if (f & CEPH_OSDMAP_NOREBALANCE) | |
2689 | s += ",norebalance"; | |
2690 | if (f & CEPH_OSDMAP_NORECOVER) | |
2691 | s += ",norecover"; | |
2692 | if (f & CEPH_OSDMAP_NOSCRUB) | |
2693 | s += ",noscrub"; | |
2694 | if (f & CEPH_OSDMAP_NODEEP_SCRUB) | |
2695 | s += ",nodeep-scrub"; | |
2696 | if (f & CEPH_OSDMAP_NOTIERAGENT) | |
2697 | s += ",notieragent"; | |
2698 | if (f & CEPH_OSDMAP_SORTBITWISE) | |
2699 | s += ",sortbitwise"; | |
2700 | if (f & CEPH_OSDMAP_REQUIRE_JEWEL) | |
2701 | s += ",require_jewel_osds"; | |
2702 | if (f & CEPH_OSDMAP_REQUIRE_KRAKEN) | |
2703 | s += ",require_kraken_osds"; | |
2704 | if (f & CEPH_OSDMAP_REQUIRE_LUMINOUS) | |
2705 | s += ",require_luminous_osds"; | |
2706 | if (s.length()) | |
2707 | s.erase(0, 1); | |
2708 | return s; | |
2709 | } | |
2710 | ||
2711 | string OSDMap::get_flag_string() const | |
2712 | { | |
2713 | return get_flag_string(flags); | |
2714 | } | |
2715 | ||
2716 | struct qi { | |
2717 | int item; | |
2718 | int depth; | |
2719 | float weight; | |
2720 | qi() : item(0), depth(0), weight(0) {} | |
2721 | qi(int i, int d, float w) : item(i), depth(d), weight(w) {} | |
2722 | }; | |
2723 | ||
2724 | void OSDMap::print_pools(ostream& out) const | |
2725 | { | |
2726 | for (const auto &pool : pools) { | |
2727 | std::string name("<unknown>"); | |
2728 | const auto &pni = pool_name.find(pool.first); | |
2729 | if (pni != pool_name.end()) | |
2730 | name = pni->second; | |
2731 | out << "pool " << pool.first | |
2732 | << " '" << name | |
2733 | << "' " << pool.second << "\n"; | |
2734 | ||
2735 | for (const auto &snap : pool.second.snaps) | |
2736 | out << "\tsnap " << snap.second.snapid << " '" << snap.second.name << "' " << snap.second.stamp << "\n"; | |
2737 | ||
2738 | if (!pool.second.removed_snaps.empty()) | |
2739 | out << "\tremoved_snaps " << pool.second.removed_snaps << "\n"; | |
2740 | } | |
2741 | out << std::endl; | |
2742 | } | |
2743 | ||
2744 | void OSDMap::print(ostream& out) const | |
2745 | { | |
2746 | out << "epoch " << get_epoch() << "\n" | |
2747 | << "fsid " << get_fsid() << "\n" | |
2748 | << "created " << get_created() << "\n" | |
2749 | << "modified " << get_modified() << "\n"; | |
2750 | ||
2751 | out << "flags " << get_flag_string() << "\n"; | |
2752 | out << "full_ratio " << full_ratio << "\n"; | |
2753 | out << "backfillfull_ratio " << backfillfull_ratio << "\n"; | |
2754 | out << "nearfull_ratio " << nearfull_ratio << "\n"; | |
2755 | if (require_min_compat_client.length()) { | |
2756 | out << "require_min_compat_client " << require_min_compat_client << "\n"; | |
2757 | } | |
2758 | auto mv = get_min_compat_client(); | |
2759 | out << "min_compat_client " << mv.first << " " << mv.second << "\n"; | |
2760 | if (get_cluster_snapshot().length()) | |
2761 | out << "cluster_snapshot " << get_cluster_snapshot() << "\n"; | |
2762 | out << "\n"; | |
2763 | ||
2764 | print_pools(out); | |
2765 | ||
2766 | out << "max_osd " << get_max_osd() << "\n"; | |
2767 | for (int i=0; i<get_max_osd(); i++) { | |
2768 | if (exists(i)) { | |
2769 | out << "osd." << i; | |
2770 | out << (is_up(i) ? " up ":" down"); | |
2771 | out << (is_in(i) ? " in ":" out"); | |
2772 | out << " weight " << get_weightf(i); | |
2773 | if (get_primary_affinity(i) != CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) | |
2774 | out << " primary_affinity " << get_primary_affinityf(i); | |
2775 | const osd_info_t& info(get_info(i)); | |
2776 | out << " " << info; | |
2777 | out << " " << get_addr(i) << " " << get_cluster_addr(i) << " " << get_hb_back_addr(i) | |
2778 | << " " << get_hb_front_addr(i); | |
2779 | set<string> st; | |
2780 | get_state(i, st); | |
2781 | out << " " << st; | |
2782 | if (!get_uuid(i).is_zero()) | |
2783 | out << " " << get_uuid(i); | |
2784 | out << "\n"; | |
2785 | } | |
2786 | } | |
2787 | out << std::endl; | |
2788 | ||
2789 | for (auto& p : pg_upmap) { | |
2790 | out << "pg_upmap " << p.first << " " << p.second << "\n"; | |
2791 | } | |
2792 | for (auto& p : pg_upmap_items) { | |
2793 | out << "pg_upmap_items " << p.first << " " << p.second << "\n"; | |
2794 | } | |
2795 | ||
2796 | for (const auto pg : *pg_temp) | |
2797 | out << "pg_temp " << pg.first << " " << pg.second << "\n"; | |
2798 | ||
2799 | for (const auto pg : *primary_temp) | |
2800 | out << "primary_temp " << pg.first << " " << pg.second << "\n"; | |
2801 | ||
2802 | for (const auto &addr : blacklist) | |
2803 | out << "blacklist " << addr.first << " expires " << addr.second << "\n"; | |
2804 | ||
2805 | // ignore pg_swap_primary | |
2806 | } | |
2807 | ||
2808 | class OSDTreePlainDumper : public CrushTreeDumper::Dumper<TextTable> { | |
2809 | public: | |
2810 | typedef CrushTreeDumper::Dumper<TextTable> Parent; | |
2811 | OSDTreePlainDumper(const CrushWrapper *crush, const OSDMap *osdmap_) | |
2812 | : Parent(crush), osdmap(osdmap_) {} | |
2813 | ||
2814 | void dump(TextTable *tbl) { | |
2815 | tbl->define_column("ID", TextTable::LEFT, TextTable::RIGHT); | |
2816 | tbl->define_column("WEIGHT", TextTable::LEFT, TextTable::RIGHT); | |
2817 | tbl->define_column("TYPE NAME", TextTable::LEFT, TextTable::LEFT); | |
2818 | tbl->define_column("UP/DOWN", TextTable::LEFT, TextTable::RIGHT); | |
2819 | tbl->define_column("REWEIGHT", TextTable::LEFT, TextTable::RIGHT); | |
2820 | tbl->define_column("PRIMARY-AFFINITY", TextTable::LEFT, TextTable::RIGHT); | |
2821 | ||
2822 | Parent::dump(tbl); | |
2823 | ||
2824 | for (int i = 0; i < osdmap->get_max_osd(); i++) { | |
2825 | if (osdmap->exists(i) && !is_touched(i)) | |
2826 | dump_item(CrushTreeDumper::Item(i, 0, 0), tbl); | |
2827 | } | |
2828 | } | |
2829 | ||
2830 | protected: | |
2831 | void dump_item(const CrushTreeDumper::Item &qi, TextTable *tbl) override { | |
2832 | ||
2833 | *tbl << qi.id | |
2834 | << weightf_t(qi.weight); | |
2835 | ||
2836 | ostringstream name; | |
2837 | for (int k = 0; k < qi.depth; k++) | |
2838 | name << " "; | |
2839 | if (qi.is_bucket()) { | |
2840 | name << crush->get_type_name(crush->get_bucket_type(qi.id)) << " " | |
2841 | << crush->get_item_name(qi.id); | |
2842 | } else { | |
2843 | name << "osd." << qi.id; | |
2844 | } | |
2845 | *tbl << name.str(); | |
2846 | ||
2847 | if (!qi.is_bucket()) { | |
2848 | if (!osdmap->exists(qi.id)) { | |
2849 | *tbl << "DNE" | |
2850 | << 0; | |
2851 | } else { | |
2852 | *tbl << (osdmap->is_up(qi.id) ? "up" : "down") | |
2853 | << weightf_t(osdmap->get_weightf(qi.id)) | |
2854 | << weightf_t(osdmap->get_primary_affinityf(qi.id)); | |
2855 | } | |
2856 | } | |
2857 | *tbl << TextTable::endrow; | |
2858 | } | |
2859 | ||
2860 | private: | |
2861 | const OSDMap *osdmap; | |
2862 | }; | |
2863 | ||
2864 | class OSDTreeFormattingDumper : public CrushTreeDumper::FormattingDumper { | |
2865 | public: | |
2866 | typedef CrushTreeDumper::FormattingDumper Parent; | |
2867 | ||
2868 | OSDTreeFormattingDumper(const CrushWrapper *crush, const OSDMap *osdmap_) | |
2869 | : Parent(crush), osdmap(osdmap_) {} | |
2870 | ||
2871 | void dump(Formatter *f) { | |
2872 | f->open_array_section("nodes"); | |
2873 | Parent::dump(f); | |
2874 | f->close_section(); | |
2875 | f->open_array_section("stray"); | |
2876 | for (int i = 0; i < osdmap->get_max_osd(); i++) { | |
2877 | if (osdmap->exists(i) && !is_touched(i)) | |
2878 | dump_item(CrushTreeDumper::Item(i, 0, 0), f); | |
2879 | } | |
2880 | f->close_section(); | |
2881 | } | |
2882 | ||
2883 | protected: | |
2884 | void dump_item_fields(const CrushTreeDumper::Item &qi, Formatter *f) override { | |
2885 | Parent::dump_item_fields(qi, f); | |
2886 | if (!qi.is_bucket()) | |
2887 | { | |
2888 | f->dump_unsigned("exists", (int)osdmap->exists(qi.id)); | |
2889 | f->dump_string("status", osdmap->is_up(qi.id) ? "up" : "down"); | |
2890 | f->dump_float("reweight", osdmap->get_weightf(qi.id)); | |
2891 | f->dump_float("primary_affinity", osdmap->get_primary_affinityf(qi.id)); | |
2892 | } | |
2893 | } | |
2894 | ||
2895 | private: | |
2896 | const OSDMap *osdmap; | |
2897 | }; | |
2898 | ||
2899 | void OSDMap::print_tree(Formatter *f, ostream *out) const | |
2900 | { | |
2901 | if (f) | |
2902 | OSDTreeFormattingDumper(crush.get(), this).dump(f); | |
2903 | else { | |
2904 | assert(out); | |
2905 | TextTable tbl; | |
2906 | OSDTreePlainDumper(crush.get(), this).dump(&tbl); | |
2907 | *out << tbl; | |
2908 | } | |
2909 | } | |
2910 | ||
2911 | void OSDMap::print_summary(Formatter *f, ostream& out) const | |
2912 | { | |
2913 | if (f) { | |
2914 | f->open_object_section("osdmap"); | |
2915 | f->dump_int("epoch", get_epoch()); | |
2916 | f->dump_int("num_osds", get_num_osds()); | |
2917 | f->dump_int("num_up_osds", get_num_up_osds()); | |
2918 | f->dump_int("num_in_osds", get_num_in_osds()); | |
2919 | f->dump_bool("full", test_flag(CEPH_OSDMAP_FULL) ? true : false); | |
2920 | f->dump_bool("nearfull", test_flag(CEPH_OSDMAP_NEARFULL) ? true : false); | |
2921 | f->dump_unsigned("num_remapped_pgs", get_num_pg_temp()); | |
2922 | f->close_section(); | |
2923 | } else { | |
2924 | out << " osdmap e" << get_epoch() << ": " | |
2925 | << get_num_osds() << " osds: " | |
2926 | << get_num_up_osds() << " up, " | |
2927 | << get_num_in_osds() << " in"; | |
2928 | if (get_num_pg_temp()) | |
2929 | out << "; " << get_num_pg_temp() << " remapped pgs"; | |
2930 | out << "\n"; | |
2931 | uint64_t important_flags = flags & ~CEPH_OSDMAP_SEMIHIDDEN_FLAGS; | |
2932 | if (important_flags) | |
2933 | out << " flags " << get_flag_string(important_flags) << "\n"; | |
2934 | } | |
2935 | } | |
2936 | ||
2937 | void OSDMap::print_oneline_summary(ostream& out) const | |
2938 | { | |
2939 | out << "e" << get_epoch() << ": " | |
2940 | << get_num_osds() << " osds: " | |
2941 | << get_num_up_osds() << " up, " | |
2942 | << get_num_in_osds() << " in"; | |
2943 | if (test_flag(CEPH_OSDMAP_FULL)) | |
2944 | out << " full"; | |
2945 | else if (test_flag(CEPH_OSDMAP_NEARFULL)) | |
2946 | out << " nearfull"; | |
2947 | } | |
2948 | ||
2949 | bool OSDMap::crush_ruleset_in_use(int ruleset) const | |
2950 | { | |
2951 | for (const auto &pool : pools) { | |
2952 | if (pool.second.crush_ruleset == ruleset) | |
2953 | return true; | |
2954 | } | |
2955 | return false; | |
2956 | } | |
2957 | ||
2958 | int OSDMap::build_simple(CephContext *cct, epoch_t e, uuid_d &fsid, | |
2959 | int nosd, int pg_bits, int pgp_bits) | |
2960 | { | |
2961 | ldout(cct, 10) << "build_simple on " << num_osd | |
2962 | << " osds with " << pg_bits << " pg bits per osd, " | |
2963 | << dendl; | |
2964 | epoch = e; | |
2965 | set_fsid(fsid); | |
2966 | created = modified = ceph_clock_now(); | |
2967 | ||
2968 | if (nosd >= 0) { | |
2969 | set_max_osd(nosd); | |
2970 | } else { | |
2971 | // count osds | |
2972 | int maxosd = 0; | |
2973 | const md_config_t *conf = cct->_conf; | |
2974 | vector<string> sections; | |
2975 | conf->get_all_sections(sections); | |
2976 | ||
2977 | for (auto §ion : sections) { | |
2978 | if (section.find("osd.") != 0) | |
2979 | continue; | |
2980 | ||
2981 | const char *begin = section.c_str() + 4; | |
2982 | char *end = (char*)begin; | |
2983 | int o = strtol(begin, &end, 10); | |
2984 | if (*end != '\0') | |
2985 | continue; | |
2986 | ||
2987 | if (o > cct->_conf->mon_max_osd) { | |
2988 | lderr(cct) << "[osd." << o << "] in config has id > mon_max_osd " << cct->_conf->mon_max_osd << dendl; | |
2989 | return -ERANGE; | |
2990 | } | |
2991 | ||
2992 | if (o > maxosd) | |
2993 | maxosd = o; | |
2994 | } | |
2995 | ||
2996 | set_max_osd(maxosd + 1); | |
2997 | } | |
2998 | ||
2999 | // pgp_num <= pg_num | |
3000 | if (pgp_bits > pg_bits) | |
3001 | pgp_bits = pg_bits; | |
3002 | ||
3003 | vector<string> pool_names; | |
3004 | pool_names.push_back("rbd"); | |
3005 | ||
3006 | stringstream ss; | |
3007 | int r; | |
3008 | if (nosd >= 0) | |
3009 | r = build_simple_crush_map(cct, *crush, nosd, &ss); | |
3010 | else | |
3011 | r = build_simple_crush_map_from_conf(cct, *crush, &ss); | |
3012 | assert(r == 0); | |
3013 | ||
3014 | int poolbase = get_max_osd() ? get_max_osd() : 1; | |
3015 | ||
3016 | int const default_replicated_ruleset = crush->get_osd_pool_default_crush_replicated_ruleset(cct); | |
3017 | assert(default_replicated_ruleset >= 0); | |
3018 | ||
3019 | for (auto &plname : pool_names) { | |
3020 | int64_t pool = ++pool_max; | |
3021 | pools[pool].type = pg_pool_t::TYPE_REPLICATED; | |
3022 | pools[pool].flags = cct->_conf->osd_pool_default_flags; | |
3023 | if (cct->_conf->osd_pool_default_flag_hashpspool) | |
3024 | pools[pool].set_flag(pg_pool_t::FLAG_HASHPSPOOL); | |
3025 | if (cct->_conf->osd_pool_default_flag_nodelete) | |
3026 | pools[pool].set_flag(pg_pool_t::FLAG_NODELETE); | |
3027 | if (cct->_conf->osd_pool_default_flag_nopgchange) | |
3028 | pools[pool].set_flag(pg_pool_t::FLAG_NOPGCHANGE); | |
3029 | if (cct->_conf->osd_pool_default_flag_nosizechange) | |
3030 | pools[pool].set_flag(pg_pool_t::FLAG_NOSIZECHANGE); | |
3031 | pools[pool].size = cct->_conf->osd_pool_default_size; | |
3032 | pools[pool].min_size = cct->_conf->get_osd_pool_default_min_size(); | |
3033 | pools[pool].crush_ruleset = default_replicated_ruleset; | |
3034 | pools[pool].object_hash = CEPH_STR_HASH_RJENKINS; | |
3035 | pools[pool].set_pg_num(poolbase << pg_bits); | |
3036 | pools[pool].set_pgp_num(poolbase << pgp_bits); | |
3037 | pools[pool].last_change = epoch; | |
3038 | pool_name[pool] = plname; | |
3039 | name_pool[plname] = pool; | |
3040 | } | |
3041 | ||
3042 | for (int i=0; i<get_max_osd(); i++) { | |
3043 | set_state(i, 0); | |
3044 | set_weight(i, CEPH_OSD_OUT); | |
3045 | } | |
3046 | ||
3047 | map<string,string> profile_map; | |
3048 | r = get_erasure_code_profile_default(cct, profile_map, &ss); | |
3049 | if (r < 0) { | |
3050 | lderr(cct) << ss.str() << dendl; | |
3051 | return r; | |
3052 | } | |
3053 | set_erasure_code_profile("default", profile_map); | |
3054 | return 0; | |
3055 | } | |
3056 | ||
3057 | int OSDMap::get_erasure_code_profile_default(CephContext *cct, | |
3058 | map<string,string> &profile_map, | |
3059 | ostream *ss) | |
3060 | { | |
3061 | int r = get_json_str_map(cct->_conf->osd_pool_default_erasure_code_profile, | |
3062 | *ss, | |
3063 | &profile_map); | |
3064 | return r; | |
3065 | } | |
3066 | ||
3067 | int OSDMap::_build_crush_types(CrushWrapper& crush) | |
3068 | { | |
3069 | crush.set_type_name(0, "osd"); | |
3070 | crush.set_type_name(1, "host"); | |
3071 | crush.set_type_name(2, "chassis"); | |
3072 | crush.set_type_name(3, "rack"); | |
3073 | crush.set_type_name(4, "row"); | |
3074 | crush.set_type_name(5, "pdu"); | |
3075 | crush.set_type_name(6, "pod"); | |
3076 | crush.set_type_name(7, "room"); | |
3077 | crush.set_type_name(8, "datacenter"); | |
3078 | crush.set_type_name(9, "region"); | |
3079 | crush.set_type_name(10, "root"); | |
3080 | return 10; | |
3081 | } | |
3082 | ||
3083 | int OSDMap::build_simple_crush_map(CephContext *cct, CrushWrapper& crush, | |
3084 | int nosd, ostream *ss) | |
3085 | { | |
3086 | crush.create(); | |
3087 | ||
3088 | // root | |
3089 | int root_type = _build_crush_types(crush); | |
3090 | int rootid; | |
3091 | int r = crush.add_bucket(0, 0, CRUSH_HASH_DEFAULT, | |
3092 | root_type, 0, NULL, NULL, &rootid); | |
3093 | assert(r == 0); | |
3094 | crush.set_item_name(rootid, "default"); | |
3095 | ||
3096 | for (int o=0; o<nosd; o++) { | |
3097 | map<string,string> loc; | |
3098 | loc["host"] = "localhost"; | |
3099 | loc["rack"] = "localrack"; | |
3100 | loc["root"] = "default"; | |
3101 | ldout(cct, 10) << " adding osd." << o << " at " << loc << dendl; | |
3102 | char name[32]; | |
3103 | snprintf(name, sizeof(name), "osd.%d", o); | |
3104 | crush.insert_item(cct, o, 1.0, name, loc); | |
3105 | } | |
3106 | ||
3107 | build_simple_crush_rulesets(cct, crush, "default", ss); | |
3108 | ||
3109 | crush.finalize(); | |
3110 | ||
3111 | return 0; | |
3112 | } | |
3113 | ||
3114 | int OSDMap::build_simple_crush_map_from_conf(CephContext *cct, | |
3115 | CrushWrapper& crush, | |
3116 | ostream *ss) | |
3117 | { | |
3118 | const md_config_t *conf = cct->_conf; | |
3119 | ||
3120 | crush.create(); | |
3121 | ||
3122 | // root | |
3123 | int root_type = _build_crush_types(crush); | |
3124 | int rootid; | |
3125 | int r = crush.add_bucket(0, 0, | |
3126 | CRUSH_HASH_DEFAULT, | |
3127 | root_type, 0, NULL, NULL, &rootid); | |
3128 | assert(r == 0); | |
3129 | crush.set_item_name(rootid, "default"); | |
3130 | ||
3131 | // add osds | |
3132 | vector<string> sections; | |
3133 | conf->get_all_sections(sections); | |
3134 | ||
3135 | for (auto §ion : sections) { | |
3136 | if (section.find("osd.") != 0) | |
3137 | continue; | |
3138 | ||
3139 | const char *begin = section.c_str() + 4; | |
3140 | char *end = (char*)begin; | |
3141 | int o = strtol(begin, &end, 10); | |
3142 | if (*end != '\0') | |
3143 | continue; | |
3144 | ||
3145 | string host, rack, row, room, dc, pool; | |
3146 | vector<string> sectiontmp; | |
3147 | sectiontmp.push_back("osd"); | |
3148 | sectiontmp.push_back(section); | |
3149 | conf->get_val_from_conf_file(sectiontmp, "host", host, false); | |
3150 | conf->get_val_from_conf_file(sectiontmp, "rack", rack, false); | |
3151 | conf->get_val_from_conf_file(sectiontmp, "row", row, false); | |
3152 | conf->get_val_from_conf_file(sectiontmp, "room", room, false); | |
3153 | conf->get_val_from_conf_file(sectiontmp, "datacenter", dc, false); | |
3154 | conf->get_val_from_conf_file(sectiontmp, "root", pool, false); | |
3155 | ||
3156 | if (host.length() == 0) | |
3157 | host = "unknownhost"; | |
3158 | if (rack.length() == 0) | |
3159 | rack = "unknownrack"; | |
3160 | ||
3161 | map<string,string> loc; | |
3162 | loc["host"] = host; | |
3163 | loc["rack"] = rack; | |
3164 | if (row.size()) | |
3165 | loc["row"] = row; | |
3166 | if (room.size()) | |
3167 | loc["room"] = room; | |
3168 | if (dc.size()) | |
3169 | loc["datacenter"] = dc; | |
3170 | loc["root"] = "default"; | |
3171 | ||
3172 | ldout(cct, 5) << " adding osd." << o << " at " << loc << dendl; | |
3173 | crush.insert_item(cct, o, 1.0, section, loc); | |
3174 | } | |
3175 | ||
3176 | build_simple_crush_rulesets(cct, crush, "default", ss); | |
3177 | ||
3178 | crush.finalize(); | |
3179 | ||
3180 | return 0; | |
3181 | } | |
3182 | ||
3183 | ||
3184 | int OSDMap::build_simple_crush_rulesets(CephContext *cct, | |
3185 | CrushWrapper& crush, | |
3186 | const string& root, | |
3187 | ostream *ss) | |
3188 | { | |
3189 | int crush_ruleset = | |
3190 | crush._get_osd_pool_default_crush_replicated_ruleset(cct, true); | |
3191 | string failure_domain = | |
3192 | crush.get_type_name(cct->_conf->osd_crush_chooseleaf_type); | |
3193 | ||
3194 | if (crush_ruleset == CEPH_DEFAULT_CRUSH_REPLICATED_RULESET) | |
3195 | crush_ruleset = -1; // create ruleset 0 by default | |
3196 | ||
3197 | int r; | |
3198 | r = crush.add_simple_ruleset_at("replicated_ruleset", root, failure_domain, | |
3199 | "firstn", pg_pool_t::TYPE_REPLICATED, | |
3200 | crush_ruleset, ss); | |
3201 | if (r < 0) | |
3202 | return r; | |
3203 | // do not add an erasure rule by default or else we will implicitly | |
3204 | // require the crush_v2 feature of clients | |
3205 | return 0; | |
3206 | } | |
3207 | ||
3208 | int OSDMap::summarize_mapping_stats( | |
3209 | OSDMap *newmap, | |
3210 | const set<int64_t> *pools, | |
3211 | std::string *out, | |
3212 | Formatter *f) const | |
3213 | { | |
3214 | set<int64_t> ls; | |
3215 | if (pools) { | |
3216 | ls = *pools; | |
3217 | } else { | |
3218 | for (auto &p : get_pools()) | |
3219 | ls.insert(p.first); | |
3220 | } | |
3221 | ||
3222 | unsigned total_pg = 0; | |
3223 | unsigned moved_pg = 0; | |
3224 | vector<unsigned> base_by_osd(get_max_osd(), 0); | |
3225 | vector<unsigned> new_by_osd(get_max_osd(), 0); | |
3226 | for (int64_t pool_id : ls) { | |
3227 | const pg_pool_t *pi = get_pg_pool(pool_id); | |
3228 | vector<int> up, up2, acting; | |
3229 | int up_primary, acting_primary; | |
3230 | for (unsigned ps = 0; ps < pi->get_pg_num(); ++ps) { | |
3231 | pg_t pgid(ps, pool_id, -1); | |
3232 | total_pg += pi->get_size(); | |
3233 | pg_to_up_acting_osds(pgid, &up, &up_primary, | |
3234 | &acting, &acting_primary); | |
3235 | for (int osd : up) { | |
3236 | if (osd >= 0 && osd < get_max_osd()) | |
3237 | ++base_by_osd[osd]; | |
3238 | } | |
3239 | if (newmap) { | |
3240 | newmap->pg_to_up_acting_osds(pgid, &up2, &up_primary, | |
3241 | &acting, &acting_primary); | |
3242 | for (int osd : up2) { | |
3243 | if (osd >= 0 && osd < get_max_osd()) | |
3244 | ++new_by_osd[osd]; | |
3245 | } | |
3246 | if (pi->type == pg_pool_t::TYPE_ERASURE) { | |
3247 | for (unsigned i=0; i<up.size(); ++i) { | |
3248 | if (up[i] != up2[i]) { | |
3249 | ++moved_pg; | |
3250 | } | |
3251 | } | |
3252 | } else if (pi->type == pg_pool_t::TYPE_REPLICATED) { | |
3253 | for (int osd : up) { | |
3254 | if (std::find(up2.begin(), up2.end(), osd) == up2.end()) { | |
3255 | ++moved_pg; | |
3256 | } | |
3257 | } | |
3258 | } else { | |
3259 | assert(0 == "unhandled pool type"); | |
3260 | } | |
3261 | } | |
3262 | } | |
3263 | } | |
3264 | ||
3265 | unsigned num_up_in = 0; | |
3266 | for (int osd = 0; osd < get_max_osd(); ++osd) { | |
3267 | if (is_up(osd) && is_in(osd)) | |
3268 | ++num_up_in; | |
3269 | } | |
3270 | if (!num_up_in) { | |
3271 | return -EINVAL; | |
3272 | } | |
3273 | ||
3274 | float avg_pg = (float)total_pg / (float)num_up_in; | |
3275 | float base_stddev = 0, new_stddev = 0; | |
3276 | int min = -1, max = -1; | |
3277 | unsigned min_base_pg = 0, max_base_pg = 0; | |
3278 | unsigned min_new_pg = 0, max_new_pg = 0; | |
3279 | for (int osd = 0; osd < get_max_osd(); ++osd) { | |
3280 | if (is_up(osd) && is_in(osd)) { | |
3281 | float base_diff = (float)base_by_osd[osd] - avg_pg; | |
3282 | base_stddev += base_diff * base_diff; | |
3283 | float new_diff = (float)new_by_osd[osd] - avg_pg; | |
3284 | new_stddev += new_diff * new_diff; | |
3285 | if (min < 0 || base_by_osd[osd] < min_base_pg) { | |
3286 | min = osd; | |
3287 | min_base_pg = base_by_osd[osd]; | |
3288 | min_new_pg = new_by_osd[osd]; | |
3289 | } | |
3290 | if (max < 0 || base_by_osd[osd] > max_base_pg) { | |
3291 | max = osd; | |
3292 | max_base_pg = base_by_osd[osd]; | |
3293 | max_new_pg = new_by_osd[osd]; | |
3294 | } | |
3295 | } | |
3296 | } | |
3297 | base_stddev = sqrt(base_stddev / num_up_in); | |
3298 | new_stddev = sqrt(new_stddev / num_up_in); | |
3299 | ||
3300 | float edev = sqrt(avg_pg * (1.0 - (1.0 / (double)num_up_in))); | |
3301 | ||
3302 | ostringstream ss; | |
3303 | if (f) | |
3304 | f->open_object_section("utilization"); | |
3305 | if (newmap) { | |
3306 | if (f) { | |
3307 | f->dump_unsigned("moved_pgs", moved_pg); | |
3308 | f->dump_unsigned("total_pgs", total_pg); | |
3309 | } else { | |
3310 | float percent = 0; | |
3311 | if (total_pg) | |
3312 | percent = (float)moved_pg * 100.0 / (float)total_pg; | |
3313 | ss << "moved " << moved_pg << " / " << total_pg | |
3314 | << " (" << percent << "%)\n"; | |
3315 | } | |
3316 | } | |
3317 | if (f) { | |
3318 | f->dump_float("avg_pgs", avg_pg); | |
3319 | f->dump_float("std_dev", base_stddev); | |
3320 | f->dump_float("expected_baseline_std_dev", edev); | |
3321 | if (newmap) | |
3322 | f->dump_float("new_std_dev", new_stddev); | |
3323 | } else { | |
3324 | ss << "avg " << avg_pg << "\n"; | |
3325 | ss << "stddev " << base_stddev; | |
3326 | if (newmap) | |
3327 | ss << " -> " << new_stddev; | |
3328 | ss << " (expected baseline " << edev << ")\n"; | |
3329 | } | |
3330 | if (min >= 0) { | |
3331 | if (f) { | |
3332 | f->dump_unsigned("min_osd", min); | |
3333 | f->dump_unsigned("min_osd_pgs", min_base_pg); | |
3334 | if (newmap) | |
3335 | f->dump_unsigned("new_min_osd_pgs", min_new_pg); | |
3336 | } else { | |
3337 | ss << "min osd." << min << " with " << min_base_pg; | |
3338 | if (newmap) | |
3339 | ss << " -> " << min_new_pg; | |
3340 | ss << " pgs (" << (float)min_base_pg / avg_pg; | |
3341 | if (newmap) | |
3342 | ss << " -> " << (float)min_new_pg / avg_pg; | |
3343 | ss << " * mean)\n"; | |
3344 | } | |
3345 | } | |
3346 | if (max >= 0) { | |
3347 | if (f) { | |
3348 | f->dump_unsigned("max_osd", max); | |
3349 | f->dump_unsigned("max_osd_pgs", max_base_pg); | |
3350 | if (newmap) | |
3351 | f->dump_unsigned("new_max_osd_pgs", max_new_pg); | |
3352 | } else { | |
3353 | ss << "max osd." << max << " with " << max_base_pg; | |
3354 | if (newmap) | |
3355 | ss << " -> " << max_new_pg; | |
3356 | ss << " pgs (" << (float)max_base_pg / avg_pg; | |
3357 | if (newmap) | |
3358 | ss << " -> " << (float)max_new_pg / avg_pg; | |
3359 | ss << " * mean)\n"; | |
3360 | } | |
3361 | } | |
3362 | if (f) | |
3363 | f->close_section(); | |
3364 | if (out) | |
3365 | *out = ss.str(); | |
3366 | return 0; | |
3367 | } | |
3368 | ||
3369 | ||
3370 | int OSDMap::clean_pg_upmaps( | |
3371 | CephContext *cct, | |
3372 | Incremental *pending_inc) | |
3373 | { | |
3374 | ldout(cct, 10) << __func__ << dendl; | |
3375 | int changed = 0; | |
3376 | for (auto& p : pg_upmap) { | |
3377 | vector<int> raw; | |
3378 | int primary; | |
3379 | pg_to_raw_osds(p.first, &raw, &primary); | |
3380 | if (vectors_equal(raw, p.second)) { | |
3381 | ldout(cct, 10) << " removing redundant pg_upmap " << p.first << " " | |
3382 | << p.second << dendl; | |
3383 | pending_inc->old_pg_upmap.insert(p.first); | |
3384 | ++changed; | |
3385 | } | |
3386 | } | |
3387 | for (auto& p : pg_upmap_items) { | |
3388 | vector<int> raw; | |
3389 | int primary; | |
3390 | pg_to_raw_osds(p.first, &raw, &primary); | |
3391 | mempool::osdmap::vector<pair<int,int>> newmap; | |
3392 | for (auto& q : p.second) { | |
3393 | if (std::find(raw.begin(), raw.end(), q.first) != raw.end()) { | |
3394 | newmap.push_back(q); | |
3395 | } | |
3396 | } | |
3397 | if (newmap.empty()) { | |
3398 | ldout(cct, 10) << " removing no-op pg_upmap_items " << p.first << " " | |
3399 | << p.second << dendl; | |
3400 | pending_inc->old_pg_upmap_items.insert(p.first); | |
3401 | ++changed; | |
3402 | } else if (newmap != p.second) { | |
3403 | ldout(cct, 10) << " simplifying partially no-op pg_upmap_items " | |
3404 | << p.first << " " << p.second << " -> " << newmap << dendl; | |
3405 | pending_inc->new_pg_upmap_items[p.first] = newmap; | |
3406 | ++changed; | |
3407 | } | |
3408 | } | |
3409 | return changed; | |
3410 | } | |
3411 | ||
3412 | bool OSDMap::try_pg_upmap( | |
3413 | CephContext *cct, | |
3414 | pg_t pg, ///< pg to potentially remap | |
3415 | const set<int>& overfull, ///< osds we'd want to evacuate | |
3416 | const vector<int>& underfull, ///< osds to move to, in order of preference | |
3417 | vector<int> *orig, | |
3418 | vector<int> *out) ///< resulting alternative mapping | |
3419 | { | |
3420 | const pg_pool_t *pool = get_pg_pool(pg.pool()); | |
3421 | if (!pool) | |
3422 | return false; | |
3423 | int rule = crush->find_rule(pool->get_crush_ruleset(), pool->get_type(), | |
3424 | pool->get_size()); | |
3425 | if (rule < 0) | |
3426 | return false; | |
3427 | ||
3428 | // get original mapping | |
3429 | _pg_to_raw_osds(*pool, pg, orig, NULL); | |
3430 | ||
3431 | // make sure there is something there to remap | |
3432 | bool any = false; | |
3433 | for (auto osd : *orig) { | |
3434 | if (overfull.count(osd)) { | |
3435 | any = true; | |
3436 | break; | |
3437 | } | |
3438 | } | |
3439 | if (!any) { | |
3440 | return false; | |
3441 | } | |
3442 | ||
3443 | int r = crush->try_remap_rule( | |
3444 | cct, | |
3445 | rule, | |
3446 | pool->get_size(), | |
3447 | overfull, underfull, | |
3448 | *orig, | |
3449 | out); | |
3450 | if (r < 0) | |
3451 | return false; | |
3452 | if (*out == *orig) | |
3453 | return false; | |
3454 | return true; | |
3455 | } | |
3456 | ||
3457 | int OSDMap::calc_pg_upmaps( | |
3458 | CephContext *cct, | |
3459 | float max_deviation, | |
3460 | int max, | |
3461 | const set<int64_t>& only_pools, | |
3462 | OSDMap::Incremental *pending_inc) | |
3463 | { | |
3464 | OSDMap tmp; | |
3465 | tmp.deepish_copy_from(*this); | |
3466 | int num_changed = 0; | |
3467 | while (true) { | |
3468 | map<int,set<pg_t>> pgs_by_osd; | |
3469 | int total_pgs = 0; | |
3470 | for (auto& i : pools) { | |
3471 | if (!only_pools.empty() && !only_pools.count(i.first)) | |
3472 | continue; | |
3473 | for (unsigned ps = 0; ps < i.second.get_pg_num(); ++ps) { | |
3474 | pg_t pg(ps, i.first); | |
3475 | vector<int> up; | |
3476 | tmp.pg_to_up_acting_osds(pg, &up, nullptr, nullptr, nullptr); | |
3477 | for (auto osd : up) { | |
3478 | if (osd != CRUSH_ITEM_NONE) | |
3479 | pgs_by_osd[osd].insert(pg); | |
3480 | } | |
3481 | } | |
3482 | total_pgs += i.second.get_size() * i.second.get_pg_num(); | |
3483 | } | |
3484 | float osd_weight_total = 0; | |
3485 | map<int,float> osd_weight; | |
3486 | for (auto& i : pgs_by_osd) { | |
3487 | float w = crush->get_item_weightf(i.first); | |
3488 | osd_weight[i.first] = w; | |
3489 | osd_weight_total += w; | |
3490 | ldout(cct, 20) << " osd." << i.first << " weight " << w | |
3491 | << " pgs " << i.second.size() << dendl; | |
3492 | } | |
3493 | ||
3494 | // NOTE: we assume we touch all osds with CRUSH! | |
3495 | float pgs_per_weight = total_pgs / osd_weight_total; | |
3496 | ldout(cct, 10) << " osd_weight_total " << osd_weight_total << dendl; | |
3497 | ldout(cct, 10) << " pgs_per_weight " << pgs_per_weight << dendl; | |
3498 | ||
3499 | // osd deviation | |
3500 | map<int,float> osd_deviation; // osd, deviation(pgs) | |
3501 | multimap<float,int> deviation_osd; // deviation(pgs), osd | |
3502 | set<int> overfull; | |
3503 | for (auto& i : pgs_by_osd) { | |
3504 | float target = osd_weight[i.first] * pgs_per_weight; | |
3505 | float deviation = (float)i.second.size() - target; | |
3506 | ldout(cct, 20) << " osd." << i.first | |
3507 | << "\tpgs " << i.second.size() | |
3508 | << "\ttarget " << target | |
3509 | << "\tdeviation " << deviation | |
3510 | << dendl; | |
3511 | osd_deviation[i.first] = deviation; | |
3512 | deviation_osd.insert(make_pair(deviation, i.first)); | |
3513 | if (deviation > 0) | |
3514 | overfull.insert(i.first); | |
3515 | } | |
3516 | ||
3517 | // build underfull, sorted from least-full to most-average | |
3518 | vector<int> underfull; | |
3519 | for (auto i = deviation_osd.begin(); | |
3520 | i != deviation_osd.end(); | |
3521 | ++i) { | |
3522 | if (i->first >= -.999) | |
3523 | break; | |
3524 | underfull.push_back(i->second); | |
3525 | } | |
3526 | ldout(cct, 10) << " overfull " << overfull | |
3527 | << " underfull " << underfull << dendl; | |
3528 | if (overfull.empty() || underfull.empty()) | |
3529 | break; | |
3530 | ||
3531 | // pick fullest | |
3532 | bool restart = false; | |
3533 | for (auto p = deviation_osd.rbegin(); p != deviation_osd.rend(); ++p) { | |
3534 | int osd = p->second; | |
3535 | float target = osd_weight[osd] * pgs_per_weight; | |
3536 | float deviation = deviation_osd.rbegin()->first; | |
3537 | if (deviation/target < max_deviation) { | |
3538 | ldout(cct, 10) << " osd." << osd | |
3539 | << " target " << target | |
3540 | << " deviation " << deviation | |
3541 | << " -> " << deviation/target | |
3542 | << " < max " << max_deviation << dendl; | |
3543 | break; | |
3544 | } | |
3545 | int num_to_move = deviation; | |
3546 | ldout(cct, 10) << " osd." << osd << " move " << num_to_move << dendl; | |
3547 | if (num_to_move < 1) | |
3548 | break; | |
3549 | ||
3550 | set<pg_t>& pgs = pgs_by_osd[osd]; | |
3551 | ||
3552 | // look for remaps we can un-remap | |
3553 | for (auto pg : pgs) { | |
3554 | auto p = tmp.pg_upmap_items.find(pg); | |
3555 | if (p != tmp.pg_upmap_items.end()) { | |
3556 | for (auto q : p->second) { | |
3557 | if (q.second == osd) { | |
3558 | ldout(cct, 10) << " dropping pg_upmap_items " << pg | |
3559 | << " " << p->second << dendl; | |
3560 | tmp.pg_upmap_items.erase(p); | |
3561 | pending_inc->old_pg_upmap_items.insert(pg); | |
3562 | ++num_changed; | |
3563 | restart = true; | |
3564 | } | |
3565 | } | |
3566 | } | |
3567 | if (restart) | |
3568 | break; | |
3569 | } // pg loop | |
3570 | if (restart) | |
3571 | break; | |
3572 | ||
3573 | for (auto pg : pgs) { | |
3574 | if (tmp.pg_upmap.count(pg) || | |
3575 | tmp.pg_upmap_items.count(pg)) { | |
3576 | ldout(cct, 20) << " already remapped " << pg << dendl; | |
3577 | continue; | |
3578 | } | |
3579 | ldout(cct, 10) << " trying " << pg << dendl; | |
3580 | vector<int> orig, out; | |
3581 | if (!try_pg_upmap(cct, pg, overfull, underfull, &orig, &out)) { | |
3582 | continue; | |
3583 | } | |
3584 | ldout(cct, 10) << " " << pg << " " << orig << " -> " << out << dendl; | |
3585 | if (orig.size() != out.size()) { | |
3586 | continue; | |
3587 | } | |
3588 | assert(orig != out); | |
3589 | auto& rmi = tmp.pg_upmap_items[pg]; | |
3590 | for (unsigned i = 0; i < out.size(); ++i) { | |
3591 | if (orig[i] != out[i]) { | |
3592 | rmi.push_back(make_pair(orig[i], out[i])); | |
3593 | } | |
3594 | } | |
3595 | pending_inc->new_pg_upmap_items[pg] = rmi; | |
3596 | ldout(cct, 10) << " " << pg << " pg_upmap_items " << rmi << dendl; | |
3597 | restart = true; | |
3598 | ++num_changed; | |
3599 | break; | |
3600 | } // pg loop | |
3601 | if (restart) | |
3602 | break; | |
3603 | } // osd loop | |
3604 | ||
3605 | if (!restart) { | |
3606 | ldout(cct, 10) << " failed to find any changes to make" << dendl; | |
3607 | break; | |
3608 | } | |
3609 | if (--max == 0) { | |
3610 | ldout(cct, 10) << " hit max iterations, stopping" << dendl; | |
3611 | break; | |
3612 | } | |
3613 | } | |
3614 | return num_changed; | |
3615 | } |