]>
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) 2014 Red Hat | |
7 | * | |
8 | * This is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
10 | * License version 2.1, as published by the Free Software | |
11 | * Foundation. See file COPYING. | |
12 | * | |
13 | */ | |
14 | ||
15 | #ifndef CEPH_OSD_BLUESTORE_H | |
16 | #define CEPH_OSD_BLUESTORE_H | |
17 | ||
18 | #include "acconfig.h" | |
19 | ||
20 | #include <unistd.h> | |
21 | ||
22 | #include <atomic> | |
9f95a23c TL |
23 | #include <chrono> |
24 | #include <ratio> | |
7c673cae FG |
25 | #include <mutex> |
26 | #include <condition_variable> | |
27 | ||
28 | #include <boost/intrusive/list.hpp> | |
29 | #include <boost/intrusive/unordered_set.hpp> | |
30 | #include <boost/intrusive/set.hpp> | |
31 | #include <boost/functional/hash.hpp> | |
32 | #include <boost/dynamic_bitset.hpp> | |
9f95a23c | 33 | #include <boost/circular_buffer.hpp> |
7c673cae | 34 | |
eafe8130 TL |
35 | #include "include/cpp-btree/btree_set.h" |
36 | ||
11fdf7f2 | 37 | #include "include/ceph_assert.h" |
7c673cae | 38 | #include "include/unordered_map.h" |
7c673cae | 39 | #include "include/mempool.h" |
9f95a23c | 40 | #include "include/hash.h" |
11fdf7f2 | 41 | #include "common/bloom_filter.hpp" |
7c673cae | 42 | #include "common/Finisher.h" |
9f95a23c | 43 | #include "common/ceph_mutex.h" |
11fdf7f2 | 44 | #include "common/Throttle.h" |
7c673cae | 45 | #include "common/perf_counters.h" |
91327a77 | 46 | #include "common/PriorityCache.h" |
7c673cae FG |
47 | #include "compressor/Compressor.h" |
48 | #include "os/ObjectStore.h" | |
49 | ||
50 | #include "bluestore_types.h" | |
51 | #include "BlockDevice.h" | |
11fdf7f2 | 52 | #include "BlueFS.h" |
7c673cae FG |
53 | #include "common/EventTrace.h" |
54 | ||
55 | class Allocator; | |
56 | class FreelistManager; | |
11fdf7f2 | 57 | class BlueStoreRepairer; |
7c673cae FG |
58 | |
59 | //#define DEBUG_CACHE | |
60 | //#define DEBUG_DEFERRED | |
61 | ||
31f18b77 FG |
62 | |
63 | ||
64 | // constants for Buffer::optimize() | |
65 | #define MAX_BUFFER_SLOP_RATIO_DEN 8 // so actually 1/N | |
66 | ||
67 | ||
7c673cae FG |
68 | enum { |
69 | l_bluestore_first = 732430, | |
70 | l_bluestore_kv_flush_lat, | |
71 | l_bluestore_kv_commit_lat, | |
11fdf7f2 TL |
72 | l_bluestore_kv_sync_lat, |
73 | l_bluestore_kv_final_lat, | |
7c673cae FG |
74 | l_bluestore_state_prepare_lat, |
75 | l_bluestore_state_aio_wait_lat, | |
76 | l_bluestore_state_io_done_lat, | |
77 | l_bluestore_state_kv_queued_lat, | |
78 | l_bluestore_state_kv_committing_lat, | |
79 | l_bluestore_state_kv_done_lat, | |
80 | l_bluestore_state_deferred_queued_lat, | |
81 | l_bluestore_state_deferred_aio_wait_lat, | |
82 | l_bluestore_state_deferred_cleanup_lat, | |
83 | l_bluestore_state_finishing_lat, | |
84 | l_bluestore_state_done_lat, | |
85 | l_bluestore_throttle_lat, | |
86 | l_bluestore_submit_lat, | |
87 | l_bluestore_commit_lat, | |
88 | l_bluestore_read_lat, | |
89 | l_bluestore_read_onode_meta_lat, | |
90 | l_bluestore_read_wait_aio_lat, | |
91 | l_bluestore_compress_lat, | |
92 | l_bluestore_decompress_lat, | |
93 | l_bluestore_csum_lat, | |
94 | l_bluestore_compress_success_count, | |
95 | l_bluestore_compress_rejected_count, | |
96 | l_bluestore_write_pad_bytes, | |
97 | l_bluestore_deferred_write_ops, | |
98 | l_bluestore_deferred_write_bytes, | |
99 | l_bluestore_write_penalty_read_ops, | |
100 | l_bluestore_allocated, | |
101 | l_bluestore_stored, | |
102 | l_bluestore_compressed, | |
103 | l_bluestore_compressed_allocated, | |
104 | l_bluestore_compressed_original, | |
105 | l_bluestore_onodes, | |
9f95a23c | 106 | l_bluestore_pinned_onodes, |
7c673cae FG |
107 | l_bluestore_onode_hits, |
108 | l_bluestore_onode_misses, | |
109 | l_bluestore_onode_shard_hits, | |
110 | l_bluestore_onode_shard_misses, | |
111 | l_bluestore_extents, | |
112 | l_bluestore_blobs, | |
113 | l_bluestore_buffers, | |
114 | l_bluestore_buffer_bytes, | |
115 | l_bluestore_buffer_hit_bytes, | |
116 | l_bluestore_buffer_miss_bytes, | |
117 | l_bluestore_write_big, | |
118 | l_bluestore_write_big_bytes, | |
119 | l_bluestore_write_big_blobs, | |
120 | l_bluestore_write_small, | |
121 | l_bluestore_write_small_bytes, | |
122 | l_bluestore_write_small_unused, | |
123 | l_bluestore_write_small_deferred, | |
124 | l_bluestore_write_small_pre_read, | |
125 | l_bluestore_write_small_new, | |
126 | l_bluestore_txc, | |
127 | l_bluestore_onode_reshard, | |
128 | l_bluestore_blob_split, | |
129 | l_bluestore_extent_compress, | |
130 | l_bluestore_gc_merged, | |
b32b8144 | 131 | l_bluestore_read_eio, |
f64942e4 | 132 | l_bluestore_reads_with_retries, |
a8e16298 | 133 | l_bluestore_fragmentation, |
11fdf7f2 TL |
134 | l_bluestore_omap_seek_to_first_lat, |
135 | l_bluestore_omap_upper_bound_lat, | |
136 | l_bluestore_omap_lower_bound_lat, | |
137 | l_bluestore_omap_next_lat, | |
494da23a | 138 | l_bluestore_clist_lat, |
7c673cae FG |
139 | l_bluestore_last |
140 | }; | |
141 | ||
11fdf7f2 TL |
142 | #define META_POOL_ID ((uint64_t)-1ull) |
143 | ||
7c673cae | 144 | class BlueStore : public ObjectStore, |
11fdf7f2 | 145 | public BlueFSDeviceExpander, |
7c673cae FG |
146 | public md_config_obs_t { |
147 | // ----------------------------------------------------- | |
148 | // types | |
149 | public: | |
150 | // config observer | |
151 | const char** get_tracked_conf_keys() const override; | |
11fdf7f2 TL |
152 | void handle_conf_change(const ConfigProxy& conf, |
153 | const std::set<std::string> &changed) override; | |
154 | ||
155 | //handler for discard event | |
156 | void handle_discard(interval_set<uint64_t>& to_release); | |
7c673cae FG |
157 | |
158 | void _set_csum(); | |
159 | void _set_compression(); | |
160 | void _set_throttle_params(); | |
31f18b77 | 161 | int _set_cache_sizes(); |
9f95a23c TL |
162 | void _set_max_defer_interval() { |
163 | max_defer_interval = | |
164 | cct->_conf.get_val<double>("bluestore_max_defer_interval"); | |
165 | } | |
7c673cae FG |
166 | |
167 | class TransContext; | |
168 | ||
169 | typedef map<uint64_t, bufferlist> ready_regions_t; | |
170 | ||
eafe8130 | 171 | |
7c673cae FG |
172 | struct BufferSpace; |
173 | struct Collection; | |
174 | typedef boost::intrusive_ptr<Collection> CollectionRef; | |
175 | ||
176 | struct AioContext { | |
177 | virtual void aio_finish(BlueStore *store) = 0; | |
178 | virtual ~AioContext() {} | |
179 | }; | |
180 | ||
181 | /// cached buffer | |
182 | struct Buffer { | |
183 | MEMPOOL_CLASS_HELPERS(); | |
184 | ||
185 | enum { | |
186 | STATE_EMPTY, ///< empty buffer -- used for cache history | |
187 | STATE_CLEAN, ///< clean data that is up to date | |
188 | STATE_WRITING, ///< data that is being written (io not yet complete) | |
189 | }; | |
190 | static const char *get_state_name(int s) { | |
191 | switch (s) { | |
192 | case STATE_EMPTY: return "empty"; | |
193 | case STATE_CLEAN: return "clean"; | |
194 | case STATE_WRITING: return "writing"; | |
195 | default: return "???"; | |
196 | } | |
197 | } | |
198 | enum { | |
199 | FLAG_NOCACHE = 1, ///< trim when done WRITING (do not become CLEAN) | |
200 | // NOTE: fix operator<< when you define a second flag | |
201 | }; | |
202 | static const char *get_flag_name(int s) { | |
203 | switch (s) { | |
204 | case FLAG_NOCACHE: return "nocache"; | |
205 | default: return "???"; | |
206 | } | |
207 | } | |
208 | ||
209 | BufferSpace *space; | |
210 | uint16_t state; ///< STATE_* | |
211 | uint16_t cache_private = 0; ///< opaque (to us) value used by Cache impl | |
212 | uint32_t flags; ///< FLAG_* | |
213 | uint64_t seq; | |
214 | uint32_t offset, length; | |
215 | bufferlist data; | |
216 | ||
217 | boost::intrusive::list_member_hook<> lru_item; | |
218 | boost::intrusive::list_member_hook<> state_item; | |
219 | ||
220 | Buffer(BufferSpace *space, unsigned s, uint64_t q, uint32_t o, uint32_t l, | |
221 | unsigned f = 0) | |
222 | : space(space), state(s), flags(f), seq(q), offset(o), length(l) {} | |
223 | Buffer(BufferSpace *space, unsigned s, uint64_t q, uint32_t o, bufferlist& b, | |
224 | unsigned f = 0) | |
225 | : space(space), state(s), flags(f), seq(q), offset(o), | |
226 | length(b.length()), data(b) {} | |
227 | ||
228 | bool is_empty() const { | |
229 | return state == STATE_EMPTY; | |
230 | } | |
231 | bool is_clean() const { | |
232 | return state == STATE_CLEAN; | |
233 | } | |
234 | bool is_writing() const { | |
235 | return state == STATE_WRITING; | |
236 | } | |
237 | ||
238 | uint32_t end() const { | |
239 | return offset + length; | |
240 | } | |
241 | ||
242 | void truncate(uint32_t newlen) { | |
11fdf7f2 | 243 | ceph_assert(newlen < length); |
7c673cae FG |
244 | if (data.length()) { |
245 | bufferlist t; | |
246 | t.substr_of(data, 0, newlen); | |
247 | data.claim(t); | |
248 | } | |
249 | length = newlen; | |
250 | } | |
31f18b77 FG |
251 | void maybe_rebuild() { |
252 | if (data.length() && | |
253 | (data.get_num_buffers() > 1 || | |
254 | data.front().wasted() > data.length() / MAX_BUFFER_SLOP_RATIO_DEN)) { | |
255 | data.rebuild(); | |
256 | } | |
257 | } | |
7c673cae FG |
258 | |
259 | void dump(Formatter *f) const { | |
260 | f->dump_string("state", get_state_name(state)); | |
261 | f->dump_unsigned("seq", seq); | |
262 | f->dump_unsigned("offset", offset); | |
263 | f->dump_unsigned("length", length); | |
264 | f->dump_unsigned("data_length", data.length()); | |
265 | } | |
266 | }; | |
267 | ||
9f95a23c | 268 | struct BufferCacheShard; |
7c673cae FG |
269 | |
270 | /// map logical extent range (object) onto buffers | |
271 | struct BufferSpace { | |
91327a77 AA |
272 | enum { |
273 | BYPASS_CLEAN_CACHE = 0x1, // bypass clean cache | |
274 | }; | |
275 | ||
7c673cae FG |
276 | typedef boost::intrusive::list< |
277 | Buffer, | |
278 | boost::intrusive::member_hook< | |
279 | Buffer, | |
280 | boost::intrusive::list_member_hook<>, | |
281 | &Buffer::state_item> > state_list_t; | |
282 | ||
f91f0fd5 | 283 | mempool::bluestore_cache_meta::map<uint32_t, std::unique_ptr<Buffer>> |
7c673cae FG |
284 | buffer_map; |
285 | ||
286 | // we use a bare intrusive list here instead of std::map because | |
287 | // it uses less memory and we expect this to be very small (very | |
288 | // few IOs in flight to the same Blob at the same time). | |
289 | state_list_t writing; ///< writing buffers, sorted by seq, ascending | |
290 | ||
291 | ~BufferSpace() { | |
11fdf7f2 TL |
292 | ceph_assert(buffer_map.empty()); |
293 | ceph_assert(writing.empty()); | |
7c673cae FG |
294 | } |
295 | ||
9f95a23c | 296 | void _add_buffer(BufferCacheShard* cache, Buffer *b, int level, Buffer *near) { |
7c673cae FG |
297 | cache->_audit("_add_buffer start"); |
298 | buffer_map[b->offset].reset(b); | |
299 | if (b->is_writing()) { | |
31f18b77 | 300 | b->data.reassign_to_mempool(mempool::mempool_bluestore_writing); |
224ce89b WB |
301 | if (writing.empty() || writing.rbegin()->seq <= b->seq) { |
302 | writing.push_back(*b); | |
303 | } else { | |
304 | auto it = writing.begin(); | |
305 | while (it->seq < b->seq) { | |
306 | ++it; | |
307 | } | |
308 | ||
11fdf7f2 | 309 | ceph_assert(it->seq >= b->seq); |
224ce89b WB |
310 | // note that this will insert b before it |
311 | // hence the order is maintained | |
312 | writing.insert(it, *b); | |
313 | } | |
7c673cae | 314 | } else { |
31f18b77 | 315 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); |
9f95a23c | 316 | cache->_add(b, level, near); |
7c673cae FG |
317 | } |
318 | cache->_audit("_add_buffer end"); | |
319 | } | |
9f95a23c | 320 | void _rm_buffer(BufferCacheShard* cache, Buffer *b) { |
7c673cae FG |
321 | _rm_buffer(cache, buffer_map.find(b->offset)); |
322 | } | |
9f95a23c | 323 | void _rm_buffer(BufferCacheShard* cache, |
31f18b77 | 324 | map<uint32_t, std::unique_ptr<Buffer>>::iterator p) { |
11fdf7f2 | 325 | ceph_assert(p != buffer_map.end()); |
7c673cae FG |
326 | cache->_audit("_rm_buffer start"); |
327 | if (p->second->is_writing()) { | |
328 | writing.erase(writing.iterator_to(*p->second)); | |
329 | } else { | |
9f95a23c | 330 | cache->_rm(p->second.get()); |
7c673cae FG |
331 | } |
332 | buffer_map.erase(p); | |
333 | cache->_audit("_rm_buffer end"); | |
334 | } | |
335 | ||
336 | map<uint32_t,std::unique_ptr<Buffer>>::iterator _data_lower_bound( | |
337 | uint32_t offset) { | |
338 | auto i = buffer_map.lower_bound(offset); | |
339 | if (i != buffer_map.begin()) { | |
340 | --i; | |
341 | if (i->first + i->second->length <= offset) | |
342 | ++i; | |
343 | } | |
344 | return i; | |
345 | } | |
346 | ||
347 | // must be called under protection of the Cache lock | |
9f95a23c | 348 | void _clear(BufferCacheShard* cache); |
7c673cae FG |
349 | |
350 | // return value is the highest cache_private of a trimmed buffer, or 0. | |
9f95a23c | 351 | int discard(BufferCacheShard* cache, uint32_t offset, uint32_t length) { |
11fdf7f2 | 352 | std::lock_guard l(cache->lock); |
9f95a23c TL |
353 | int ret = _discard(cache, offset, length); |
354 | cache->_trim(); | |
355 | return ret; | |
7c673cae | 356 | } |
9f95a23c | 357 | int _discard(BufferCacheShard* cache, uint32_t offset, uint32_t length); |
7c673cae | 358 | |
9f95a23c | 359 | void write(BufferCacheShard* cache, uint64_t seq, uint32_t offset, bufferlist& bl, |
7c673cae | 360 | unsigned flags) { |
11fdf7f2 | 361 | std::lock_guard l(cache->lock); |
7c673cae FG |
362 | Buffer *b = new Buffer(this, Buffer::STATE_WRITING, seq, offset, bl, |
363 | flags); | |
364 | b->cache_private = _discard(cache, offset, bl.length()); | |
365 | _add_buffer(cache, b, (flags & Buffer::FLAG_NOCACHE) ? 0 : 1, nullptr); | |
9f95a23c | 366 | cache->_trim(); |
7c673cae | 367 | } |
9f95a23c TL |
368 | void _finish_write(BufferCacheShard* cache, uint64_t seq); |
369 | void did_read(BufferCacheShard* cache, uint32_t offset, bufferlist& bl) { | |
11fdf7f2 | 370 | std::lock_guard l(cache->lock); |
7c673cae FG |
371 | Buffer *b = new Buffer(this, Buffer::STATE_CLEAN, 0, offset, bl); |
372 | b->cache_private = _discard(cache, offset, bl.length()); | |
373 | _add_buffer(cache, b, 1, nullptr); | |
9f95a23c | 374 | cache->_trim(); |
7c673cae FG |
375 | } |
376 | ||
9f95a23c | 377 | void read(BufferCacheShard* cache, uint32_t offset, uint32_t length, |
7c673cae | 378 | BlueStore::ready_regions_t& res, |
91327a77 AA |
379 | interval_set<uint32_t>& res_intervals, |
380 | int flags = 0); | |
7c673cae | 381 | |
9f95a23c | 382 | void truncate(BufferCacheShard* cache, uint32_t offset) { |
7c673cae FG |
383 | discard(cache, offset, (uint32_t)-1 - offset); |
384 | } | |
385 | ||
9f95a23c | 386 | void split(BufferCacheShard* cache, size_t pos, BufferSpace &r); |
7c673cae | 387 | |
9f95a23c | 388 | void dump(BufferCacheShard* cache, Formatter *f) const { |
11fdf7f2 | 389 | std::lock_guard l(cache->lock); |
7c673cae FG |
390 | f->open_array_section("buffers"); |
391 | for (auto& i : buffer_map) { | |
392 | f->open_object_section("buffer"); | |
11fdf7f2 | 393 | ceph_assert(i.first == i.second->offset); |
7c673cae FG |
394 | i.second->dump(f); |
395 | f->close_section(); | |
396 | } | |
397 | f->close_section(); | |
398 | } | |
399 | }; | |
400 | ||
401 | struct SharedBlobSet; | |
402 | ||
403 | /// in-memory shared blob state (incl cached buffers) | |
404 | struct SharedBlob { | |
405 | MEMPOOL_CLASS_HELPERS(); | |
406 | ||
407 | std::atomic_int nref = {0}; ///< reference count | |
408 | bool loaded = false; | |
409 | ||
410 | CollectionRef coll; | |
411 | union { | |
412 | uint64_t sbid_unloaded; ///< sbid if persistent isn't loaded | |
413 | bluestore_shared_blob_t *persistent; ///< persistent part of the shared blob if any | |
414 | }; | |
415 | BufferSpace bc; ///< buffer cache | |
416 | ||
417 | SharedBlob(Collection *_coll) : coll(_coll), sbid_unloaded(0) { | |
418 | if (get_cache()) { | |
419 | get_cache()->add_blob(); | |
420 | } | |
421 | } | |
422 | SharedBlob(uint64_t i, Collection *_coll); | |
423 | ~SharedBlob(); | |
424 | ||
425 | uint64_t get_sbid() const { | |
426 | return loaded ? persistent->sbid : sbid_unloaded; | |
427 | } | |
428 | ||
429 | friend void intrusive_ptr_add_ref(SharedBlob *b) { b->get(); } | |
430 | friend void intrusive_ptr_release(SharedBlob *b) { b->put(); } | |
431 | ||
9f95a23c | 432 | void dump(Formatter* f) const; |
7c673cae FG |
433 | friend ostream& operator<<(ostream& out, const SharedBlob& sb); |
434 | ||
435 | void get() { | |
436 | ++nref; | |
437 | } | |
438 | void put(); | |
439 | ||
440 | /// get logical references | |
441 | void get_ref(uint64_t offset, uint32_t length); | |
442 | ||
443 | /// put logical references, and get back any released extents | |
444 | void put_ref(uint64_t offset, uint32_t length, | |
11fdf7f2 | 445 | PExtentVector *r, bool *unshare); |
7c673cae | 446 | |
f64942e4 AA |
447 | void finish_write(uint64_t seq); |
448 | ||
7c673cae FG |
449 | friend bool operator==(const SharedBlob &l, const SharedBlob &r) { |
450 | return l.get_sbid() == r.get_sbid(); | |
451 | } | |
9f95a23c | 452 | inline BufferCacheShard* get_cache() { |
7c673cae FG |
453 | return coll ? coll->cache : nullptr; |
454 | } | |
455 | inline SharedBlobSet* get_parent() { | |
456 | return coll ? &(coll->shared_blob_set) : nullptr; | |
457 | } | |
458 | inline bool is_loaded() const { | |
459 | return loaded; | |
460 | } | |
461 | ||
462 | }; | |
463 | typedef boost::intrusive_ptr<SharedBlob> SharedBlobRef; | |
464 | ||
465 | /// a lookup table of SharedBlobs | |
466 | struct SharedBlobSet { | |
11fdf7f2 TL |
467 | /// protect lookup, insertion, removal |
468 | ceph::mutex lock = ceph::make_mutex("BlueStore::SharedBlobSet::lock"); | |
7c673cae FG |
469 | |
470 | // we use a bare pointer because we don't want to affect the ref | |
471 | // count | |
f91f0fd5 | 472 | mempool::bluestore_cache_meta::unordered_map<uint64_t,SharedBlob*> sb_map; |
7c673cae FG |
473 | |
474 | SharedBlobRef lookup(uint64_t sbid) { | |
11fdf7f2 | 475 | std::lock_guard l(lock); |
7c673cae | 476 | auto p = sb_map.find(sbid); |
28e407b8 AA |
477 | if (p == sb_map.end() || |
478 | p->second->nref == 0) { | |
7c673cae FG |
479 | return nullptr; |
480 | } | |
481 | return p->second; | |
482 | } | |
483 | ||
484 | void add(Collection* coll, SharedBlob *sb) { | |
11fdf7f2 | 485 | std::lock_guard l(lock); |
7c673cae FG |
486 | sb_map[sb->get_sbid()] = sb; |
487 | sb->coll = coll; | |
488 | } | |
489 | ||
91327a77 | 490 | bool remove(SharedBlob *sb, bool verify_nref_is_zero=false) { |
11fdf7f2 TL |
491 | std::lock_guard l(lock); |
492 | ceph_assert(sb->get_parent() == this); | |
91327a77 AA |
493 | if (verify_nref_is_zero && sb->nref != 0) { |
494 | return false; | |
495 | } | |
28e407b8 AA |
496 | // only remove if it still points to us |
497 | auto p = sb_map.find(sb->get_sbid()); | |
498 | if (p != sb_map.end() && | |
499 | p->second == sb) { | |
500 | sb_map.erase(p); | |
501 | } | |
91327a77 | 502 | return true; |
3efd9988 FG |
503 | } |
504 | ||
7c673cae | 505 | bool empty() { |
11fdf7f2 | 506 | std::lock_guard l(lock); |
7c673cae FG |
507 | return sb_map.empty(); |
508 | } | |
3efd9988 | 509 | |
11fdf7f2 TL |
510 | template <int LogLevelV> |
511 | void dump(CephContext *cct); | |
7c673cae FG |
512 | }; |
513 | ||
514 | //#define CACHE_BLOB_BL // not sure if this is a win yet or not... :/ | |
515 | ||
516 | /// in-memory blob metadata and associated cached buffers (if any) | |
517 | struct Blob { | |
518 | MEMPOOL_CLASS_HELPERS(); | |
519 | ||
520 | std::atomic_int nref = {0}; ///< reference count | |
521 | int16_t id = -1; ///< id, for spanning blobs only, >= 0 | |
522 | int16_t last_encoded_id = -1; ///< (ephemeral) used during encoding only | |
523 | SharedBlobRef shared_blob; ///< shared blob state (if any) | |
524 | ||
525 | private: | |
526 | mutable bluestore_blob_t blob; ///< decoded blob metadata | |
527 | #ifdef CACHE_BLOB_BL | |
528 | mutable bufferlist blob_bl; ///< cached encoded blob, blob is dirty if empty | |
529 | #endif | |
530 | /// refs from this shard. ephemeral if id<0, persisted if spanning. | |
531 | bluestore_blob_use_tracker_t used_in_blob; | |
532 | ||
533 | public: | |
534 | ||
535 | friend void intrusive_ptr_add_ref(Blob *b) { b->get(); } | |
536 | friend void intrusive_ptr_release(Blob *b) { b->put(); } | |
537 | ||
9f95a23c | 538 | void dump(Formatter* f) const; |
7c673cae FG |
539 | friend ostream& operator<<(ostream& out, const Blob &b); |
540 | ||
541 | const bluestore_blob_use_tracker_t& get_blob_use_tracker() const { | |
542 | return used_in_blob; | |
543 | } | |
544 | bool is_referenced() const { | |
545 | return used_in_blob.is_not_empty(); | |
546 | } | |
547 | uint32_t get_referenced_bytes() const { | |
548 | return used_in_blob.get_referenced_bytes(); | |
549 | } | |
550 | ||
551 | bool is_spanning() const { | |
552 | return id >= 0; | |
553 | } | |
554 | ||
555 | bool can_split() const { | |
11fdf7f2 | 556 | std::lock_guard l(shared_blob->get_cache()->lock); |
7c673cae FG |
557 | // splitting a BufferSpace writing list is too hard; don't try. |
558 | return shared_blob->bc.writing.empty() && | |
559 | used_in_blob.can_split() && | |
560 | get_blob().can_split(); | |
561 | } | |
562 | ||
563 | bool can_split_at(uint32_t blob_offset) const { | |
564 | return used_in_blob.can_split_at(blob_offset) && | |
565 | get_blob().can_split_at(blob_offset); | |
566 | } | |
567 | ||
224ce89b | 568 | bool can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
569 | uint32_t target_blob_size, |
570 | uint32_t b_offset, | |
571 | uint32_t *length0); | |
572 | ||
573 | void dup(Blob& o) { | |
574 | o.shared_blob = shared_blob; | |
575 | o.blob = blob; | |
576 | #ifdef CACHE_BLOB_BL | |
577 | o.blob_bl = blob_bl; | |
578 | #endif | |
579 | } | |
580 | ||
224ce89b | 581 | inline const bluestore_blob_t& get_blob() const { |
7c673cae FG |
582 | return blob; |
583 | } | |
224ce89b | 584 | inline bluestore_blob_t& dirty_blob() { |
7c673cae FG |
585 | #ifdef CACHE_BLOB_BL |
586 | blob_bl.clear(); | |
587 | #endif | |
588 | return blob; | |
589 | } | |
590 | ||
591 | /// discard buffers for unallocated regions | |
592 | void discard_unallocated(Collection *coll); | |
593 | ||
594 | /// get logical references | |
595 | void get_ref(Collection *coll, uint32_t offset, uint32_t length); | |
596 | /// put logical references, and get back any released extents | |
597 | bool put_ref(Collection *coll, uint32_t offset, uint32_t length, | |
598 | PExtentVector *r); | |
599 | ||
600 | /// split the blob | |
601 | void split(Collection *coll, uint32_t blob_offset, Blob *o); | |
602 | ||
603 | void get() { | |
604 | ++nref; | |
605 | } | |
606 | void put() { | |
607 | if (--nref == 0) | |
608 | delete this; | |
609 | } | |
610 | ||
611 | ||
612 | #ifdef CACHE_BLOB_BL | |
613 | void _encode() const { | |
614 | if (blob_bl.length() == 0 ) { | |
11fdf7f2 | 615 | encode(blob, blob_bl); |
7c673cae | 616 | } else { |
11fdf7f2 | 617 | ceph_assert(blob_bl.length()); |
7c673cae FG |
618 | } |
619 | } | |
620 | void bound_encode( | |
621 | size_t& p, | |
622 | bool include_ref_map) const { | |
623 | _encode(); | |
624 | p += blob_bl.length(); | |
625 | if (include_ref_map) { | |
626 | used_in_blob.bound_encode(p); | |
627 | } | |
628 | } | |
629 | void encode( | |
630 | bufferlist::contiguous_appender& p, | |
631 | bool include_ref_map) const { | |
632 | _encode(); | |
633 | p.append(blob_bl); | |
634 | if (include_ref_map) { | |
635 | used_in_blob.encode(p); | |
636 | } | |
637 | } | |
638 | void decode( | |
639 | Collection */*coll*/, | |
11fdf7f2 | 640 | bufferptr::const_iterator& p, |
7c673cae FG |
641 | bool include_ref_map) { |
642 | const char *start = p.get_pos(); | |
643 | denc(blob, p); | |
644 | const char *end = p.get_pos(); | |
645 | blob_bl.clear(); | |
646 | blob_bl.append(start, end - start); | |
647 | if (include_ref_map) { | |
648 | used_in_blob.decode(p); | |
649 | } | |
650 | } | |
651 | #else | |
652 | void bound_encode( | |
653 | size_t& p, | |
654 | uint64_t struct_v, | |
655 | uint64_t sbid, | |
656 | bool include_ref_map) const { | |
657 | denc(blob, p, struct_v); | |
658 | if (blob.is_shared()) { | |
659 | denc(sbid, p); | |
660 | } | |
661 | if (include_ref_map) { | |
662 | used_in_blob.bound_encode(p); | |
663 | } | |
664 | } | |
665 | void encode( | |
666 | bufferlist::contiguous_appender& p, | |
667 | uint64_t struct_v, | |
668 | uint64_t sbid, | |
669 | bool include_ref_map) const { | |
670 | denc(blob, p, struct_v); | |
671 | if (blob.is_shared()) { | |
672 | denc(sbid, p); | |
673 | } | |
674 | if (include_ref_map) { | |
675 | used_in_blob.encode(p); | |
676 | } | |
677 | } | |
678 | void decode( | |
679 | Collection *coll, | |
11fdf7f2 | 680 | bufferptr::const_iterator& p, |
7c673cae FG |
681 | uint64_t struct_v, |
682 | uint64_t* sbid, | |
683 | bool include_ref_map); | |
684 | #endif | |
685 | }; | |
686 | typedef boost::intrusive_ptr<Blob> BlobRef; | |
f91f0fd5 | 687 | typedef mempool::bluestore_cache_meta::map<int,BlobRef> blob_map_t; |
7c673cae FG |
688 | |
689 | /// a logical extent, pointing to (some portion of) a blob | |
690 | typedef boost::intrusive::set_base_hook<boost::intrusive::optimize_size<true> > ExtentBase; //making an alias to avoid build warnings | |
691 | struct Extent : public ExtentBase { | |
692 | MEMPOOL_CLASS_HELPERS(); | |
693 | ||
694 | uint32_t logical_offset = 0; ///< logical offset | |
695 | uint32_t blob_offset = 0; ///< blob offset | |
696 | uint32_t length = 0; ///< length | |
697 | BlobRef blob; ///< the blob with our data | |
698 | ||
699 | /// ctor for lookup only | |
700 | explicit Extent(uint32_t lo) : ExtentBase(), logical_offset(lo) { } | |
701 | /// ctor for delayed initialization (see decode_some()) | |
702 | explicit Extent() : ExtentBase() { | |
703 | } | |
704 | /// ctor for general usage | |
705 | Extent(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) | |
706 | : ExtentBase(), | |
707 | logical_offset(lo), blob_offset(o), length(l) { | |
708 | assign_blob(b); | |
709 | } | |
710 | ~Extent() { | |
711 | if (blob) { | |
712 | blob->shared_blob->get_cache()->rm_extent(); | |
713 | } | |
714 | } | |
715 | ||
9f95a23c TL |
716 | void dump(Formatter* f) const; |
717 | ||
7c673cae | 718 | void assign_blob(const BlobRef& b) { |
11fdf7f2 | 719 | ceph_assert(!blob); |
7c673cae FG |
720 | blob = b; |
721 | blob->shared_blob->get_cache()->add_extent(); | |
722 | } | |
723 | ||
724 | // comparators for intrusive_set | |
725 | friend bool operator<(const Extent &a, const Extent &b) { | |
726 | return a.logical_offset < b.logical_offset; | |
727 | } | |
728 | friend bool operator>(const Extent &a, const Extent &b) { | |
729 | return a.logical_offset > b.logical_offset; | |
730 | } | |
731 | friend bool operator==(const Extent &a, const Extent &b) { | |
732 | return a.logical_offset == b.logical_offset; | |
733 | } | |
734 | ||
735 | uint32_t blob_start() const { | |
736 | return logical_offset - blob_offset; | |
737 | } | |
738 | ||
739 | uint32_t blob_end() const { | |
740 | return blob_start() + blob->get_blob().get_logical_length(); | |
741 | } | |
742 | ||
743 | uint32_t logical_end() const { | |
744 | return logical_offset + length; | |
745 | } | |
746 | ||
747 | // return true if any piece of the blob is out of | |
748 | // the given range [o, o + l]. | |
749 | bool blob_escapes_range(uint32_t o, uint32_t l) const { | |
750 | return blob_start() < o || blob_end() > o + l; | |
751 | } | |
752 | }; | |
753 | typedef boost::intrusive::set<Extent> extent_map_t; | |
754 | ||
755 | ||
756 | friend ostream& operator<<(ostream& out, const Extent& e); | |
757 | ||
758 | struct OldExtent { | |
759 | boost::intrusive::list_member_hook<> old_extent_item; | |
760 | Extent e; | |
761 | PExtentVector r; | |
762 | bool blob_empty; // flag to track the last removed extent that makes blob | |
763 | // empty - required to update compression stat properly | |
764 | OldExtent(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) | |
765 | : e(lo, o, l, b), blob_empty(false) { | |
766 | } | |
767 | static OldExtent* create(CollectionRef c, | |
768 | uint32_t lo, | |
769 | uint32_t o, | |
770 | uint32_t l, | |
771 | BlobRef& b); | |
772 | }; | |
773 | typedef boost::intrusive::list< | |
774 | OldExtent, | |
775 | boost::intrusive::member_hook< | |
776 | OldExtent, | |
777 | boost::intrusive::list_member_hook<>, | |
778 | &OldExtent::old_extent_item> > old_extent_map_t; | |
779 | ||
780 | struct Onode; | |
781 | ||
782 | /// a sharded extent map, mapping offsets to lextents to blobs | |
783 | struct ExtentMap { | |
784 | Onode *onode; | |
785 | extent_map_t extent_map; ///< map of Extents to Blobs | |
786 | blob_map_t spanning_blob_map; ///< blobs that span shards | |
11fdf7f2 | 787 | typedef boost::intrusive_ptr<Onode> OnodeRef; |
7c673cae FG |
788 | |
789 | struct Shard { | |
790 | bluestore_onode_t::shard_info *shard_info = nullptr; | |
791 | unsigned extents = 0; ///< count extents in this shard | |
792 | bool loaded = false; ///< true if shard is loaded | |
793 | bool dirty = false; ///< true if shard is dirty and needs reencoding | |
794 | }; | |
f91f0fd5 | 795 | mempool::bluestore_cache_meta::vector<Shard> shards; ///< shards |
7c673cae FG |
796 | |
797 | bufferlist inline_bl; ///< cached encoded map, if unsharded; empty=>dirty | |
798 | ||
799 | uint32_t needs_reshard_begin = 0; | |
800 | uint32_t needs_reshard_end = 0; | |
801 | ||
11fdf7f2 TL |
802 | void dup(BlueStore* b, TransContext*, CollectionRef&, OnodeRef&, OnodeRef&, |
803 | uint64_t&, uint64_t&, uint64_t&); | |
804 | ||
7c673cae FG |
805 | bool needs_reshard() const { |
806 | return needs_reshard_end > needs_reshard_begin; | |
807 | } | |
808 | void clear_needs_reshard() { | |
809 | needs_reshard_begin = needs_reshard_end = 0; | |
810 | } | |
811 | void request_reshard(uint32_t begin, uint32_t end) { | |
812 | if (begin < needs_reshard_begin) { | |
813 | needs_reshard_begin = begin; | |
814 | } | |
815 | if (end > needs_reshard_end) { | |
816 | needs_reshard_end = end; | |
817 | } | |
818 | } | |
819 | ||
820 | struct DeleteDisposer { | |
821 | void operator()(Extent *e) { delete e; } | |
822 | }; | |
823 | ||
824 | ExtentMap(Onode *o); | |
825 | ~ExtentMap() { | |
826 | extent_map.clear_and_dispose(DeleteDisposer()); | |
827 | } | |
828 | ||
829 | void clear() { | |
830 | extent_map.clear_and_dispose(DeleteDisposer()); | |
831 | shards.clear(); | |
832 | inline_bl.clear(); | |
833 | clear_needs_reshard(); | |
834 | } | |
835 | ||
9f95a23c TL |
836 | void dump(Formatter* f) const; |
837 | ||
7c673cae FG |
838 | bool encode_some(uint32_t offset, uint32_t length, bufferlist& bl, |
839 | unsigned *pn); | |
840 | unsigned decode_some(bufferlist& bl); | |
841 | ||
842 | void bound_encode_spanning_blobs(size_t& p); | |
843 | void encode_spanning_blobs(bufferlist::contiguous_appender& p); | |
11fdf7f2 | 844 | void decode_spanning_blobs(bufferptr::const_iterator& p); |
7c673cae FG |
845 | |
846 | BlobRef get_spanning_blob(int id) { | |
847 | auto p = spanning_blob_map.find(id); | |
11fdf7f2 | 848 | ceph_assert(p != spanning_blob_map.end()); |
7c673cae FG |
849 | return p->second; |
850 | } | |
851 | ||
852 | void update(KeyValueDB::Transaction t, bool force); | |
31f18b77 | 853 | decltype(BlueStore::Blob::id) allocate_spanning_blob_id(); |
7c673cae FG |
854 | void reshard( |
855 | KeyValueDB *db, | |
856 | KeyValueDB::Transaction t); | |
857 | ||
858 | /// initialize Shards from the onode | |
859 | void init_shards(bool loaded, bool dirty); | |
860 | ||
861 | /// return index of shard containing offset | |
862 | /// or -1 if not found | |
863 | int seek_shard(uint32_t offset) { | |
864 | size_t end = shards.size(); | |
865 | size_t mid, left = 0; | |
866 | size_t right = end; // one passed the right end | |
867 | ||
868 | while (left < right) { | |
869 | mid = left + (right - left) / 2; | |
870 | if (offset >= shards[mid].shard_info->offset) { | |
871 | size_t next = mid + 1; | |
872 | if (next >= end || offset < shards[next].shard_info->offset) | |
873 | return mid; | |
874 | //continue to search forwards | |
875 | left = next; | |
876 | } else { | |
877 | //continue to search backwards | |
878 | right = mid; | |
879 | } | |
880 | } | |
881 | ||
882 | return -1; // not found | |
883 | } | |
884 | ||
885 | /// check if a range spans a shard | |
886 | bool spans_shard(uint32_t offset, uint32_t length) { | |
887 | if (shards.empty()) { | |
888 | return false; | |
889 | } | |
890 | int s = seek_shard(offset); | |
11fdf7f2 | 891 | ceph_assert(s >= 0); |
7c673cae FG |
892 | if (s == (int)shards.size() - 1) { |
893 | return false; // last shard | |
894 | } | |
895 | if (offset + length <= shards[s+1].shard_info->offset) { | |
896 | return false; | |
897 | } | |
898 | return true; | |
899 | } | |
900 | ||
901 | /// ensure that a range of the map is loaded | |
902 | void fault_range(KeyValueDB *db, | |
903 | uint32_t offset, uint32_t length); | |
904 | ||
905 | /// ensure a range of the map is marked dirty | |
31f18b77 | 906 | void dirty_range(uint32_t offset, uint32_t length); |
7c673cae | 907 | |
31f18b77 | 908 | /// for seek_lextent test |
7c673cae FG |
909 | extent_map_t::iterator find(uint64_t offset); |
910 | ||
7c673cae FG |
911 | /// seek to the first lextent including or after offset |
912 | extent_map_t::iterator seek_lextent(uint64_t offset); | |
913 | extent_map_t::const_iterator seek_lextent(uint64_t offset) const; | |
914 | ||
915 | /// add a new Extent | |
916 | void add(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) { | |
917 | extent_map.insert(*new Extent(lo, o, l, b)); | |
918 | } | |
919 | ||
920 | /// remove (and delete) an Extent | |
921 | void rm(extent_map_t::iterator p) { | |
922 | extent_map.erase_and_dispose(p, DeleteDisposer()); | |
923 | } | |
924 | ||
925 | bool has_any_lextents(uint64_t offset, uint64_t length); | |
926 | ||
927 | /// consolidate adjacent lextents in extent_map | |
928 | int compress_extent_map(uint64_t offset, uint64_t length); | |
929 | ||
930 | /// punch a logical hole. add lextents to deref to target list. | |
931 | void punch_hole(CollectionRef &c, | |
932 | uint64_t offset, uint64_t length, | |
933 | old_extent_map_t *old_extents); | |
934 | ||
935 | /// put new lextent into lextent_map overwriting existing ones if | |
936 | /// any and update references accordingly | |
937 | Extent *set_lextent(CollectionRef &c, | |
938 | uint64_t logical_offset, | |
939 | uint64_t offset, uint64_t length, | |
940 | BlobRef b, | |
941 | old_extent_map_t *old_extents); | |
942 | ||
943 | /// split a blob (and referring extents) | |
944 | BlobRef split_blob(BlobRef lb, uint32_t blob_offset, uint32_t pos); | |
945 | }; | |
946 | ||
947 | /// Compressed Blob Garbage collector | |
948 | /* | |
949 | The primary idea of the collector is to estimate a difference between | |
950 | allocation units(AU) currently present for compressed blobs and new AUs | |
951 | required to store that data uncompressed. | |
952 | Estimation is performed for protrusive extents within a logical range | |
953 | determined by a concatenation of old_extents collection and specific(current) | |
954 | write request. | |
955 | The root cause for old_extents use is the need to handle blob ref counts | |
956 | properly. Old extents still hold blob refs and hence we need to traverse | |
957 | the collection to determine if blob to be released. | |
958 | Protrusive extents are extents that fit into the blob set in action | |
959 | (ones that are below the logical range from above) but not removed totally | |
960 | due to the current write. | |
961 | E.g. for | |
962 | extent1 <loffs = 100, boffs = 100, len = 100> -> | |
963 | blob1<compressed, len_on_disk=4096, logical_len=8192> | |
964 | extent2 <loffs = 200, boffs = 200, len = 100> -> | |
965 | blob2<raw, len_on_disk=4096, llen=4096> | |
966 | extent3 <loffs = 300, boffs = 300, len = 100> -> | |
967 | blob1<compressed, len_on_disk=4096, llen=8192> | |
968 | extent4 <loffs = 4096, boffs = 0, len = 100> -> | |
969 | blob3<raw, len_on_disk=4096, llen=4096> | |
970 | write(300~100) | |
971 | protrusive extents are within the following ranges <0~300, 400~8192-400> | |
972 | In this case existing AUs that might be removed due to GC (i.e. blob1) | |
973 | use 2x4K bytes. | |
974 | And new AUs expected after GC = 0 since extent1 to be merged into blob2. | |
975 | Hence we should do a collect. | |
976 | */ | |
977 | class GarbageCollector | |
978 | { | |
979 | public: | |
980 | /// return amount of allocation units that might be saved due to GC | |
981 | int64_t estimate( | |
982 | uint64_t offset, | |
983 | uint64_t length, | |
984 | const ExtentMap& extent_map, | |
985 | const old_extent_map_t& old_extents, | |
986 | uint64_t min_alloc_size); | |
987 | ||
988 | /// return a collection of extents to perform GC on | |
eafe8130 | 989 | const interval_set<uint64_t>& get_extents_to_collect() const { |
7c673cae FG |
990 | return extents_to_collect; |
991 | } | |
992 | GarbageCollector(CephContext* _cct) : cct(_cct) {} | |
993 | ||
994 | private: | |
995 | struct BlobInfo { | |
996 | uint64_t referenced_bytes = 0; ///< amount of bytes referenced in blob | |
997 | int64_t expected_allocations = 0; ///< new alloc units required | |
998 | ///< in case of gc fulfilled | |
999 | bool collect_candidate = false; ///< indicate if blob has any extents | |
1000 | ///< eligible for GC. | |
1001 | extent_map_t::const_iterator first_lextent; ///< points to the first | |
1002 | ///< lextent referring to | |
1003 | ///< the blob if any. | |
1004 | ///< collect_candidate flag | |
1005 | ///< determines the validity | |
1006 | extent_map_t::const_iterator last_lextent; ///< points to the last | |
1007 | ///< lextent referring to | |
1008 | ///< the blob if any. | |
1009 | ||
1010 | BlobInfo(uint64_t ref_bytes) : | |
1011 | referenced_bytes(ref_bytes) { | |
1012 | } | |
1013 | }; | |
1014 | CephContext* cct; | |
1015 | map<Blob*, BlobInfo> affected_blobs; ///< compressed blobs and their ref_map | |
1016 | ///< copies that are affected by the | |
1017 | ///< specific write | |
1018 | ||
a8e16298 | 1019 | ///< protrusive extents that should be collected if GC takes place |
eafe8130 | 1020 | interval_set<uint64_t> extents_to_collect; |
7c673cae FG |
1021 | |
1022 | boost::optional<uint64_t > used_alloc_unit; ///< last processed allocation | |
1023 | ///< unit when traversing | |
1024 | ///< protrusive extents. | |
1025 | ///< Other extents mapped to | |
1026 | ///< this AU to be ignored | |
1027 | ///< (except the case where | |
1028 | ///< uncompressed extent follows | |
1029 | ///< compressed one - see below). | |
1030 | BlobInfo* blob_info_counted = nullptr; ///< set if previous allocation unit | |
1031 | ///< caused expected_allocations | |
1032 | ///< counter increment at this blob. | |
1033 | ///< if uncompressed extent follows | |
1034 | ///< a decrement for the | |
1035 | ///< expected_allocations counter | |
1036 | ///< is needed | |
1037 | int64_t expected_allocations = 0; ///< new alloc units required in case | |
1038 | ///< of gc fulfilled | |
1039 | int64_t expected_for_release = 0; ///< alloc units currently used by | |
1040 | ///< compressed blobs that might | |
1041 | ///< gone after GC | |
7c673cae FG |
1042 | |
1043 | protected: | |
1044 | void process_protrusive_extents(const BlueStore::ExtentMap& extent_map, | |
1045 | uint64_t start_offset, | |
1046 | uint64_t end_offset, | |
1047 | uint64_t start_touch_offset, | |
1048 | uint64_t end_touch_offset, | |
1049 | uint64_t min_alloc_size); | |
1050 | }; | |
1051 | ||
1052 | struct OnodeSpace; | |
7c673cae FG |
1053 | /// an in-memory object |
1054 | struct Onode { | |
1055 | MEMPOOL_CLASS_HELPERS(); | |
1056 | ||
1057 | std::atomic_int nref; ///< reference count | |
1058 | Collection *c; | |
7c673cae FG |
1059 | ghobject_t oid; |
1060 | ||
1061 | /// key under PREFIX_OBJ where we are stored | |
f91f0fd5 | 1062 | mempool::bluestore_cache_meta::string key; |
7c673cae | 1063 | |
f6b5b4d7 | 1064 | boost::intrusive::list_member_hook<> lru_item; |
7c673cae FG |
1065 | |
1066 | bluestore_onode_t onode; ///< metadata stored as value in kv store | |
1067 | bool exists; ///< true if object logically exists | |
f6b5b4d7 TL |
1068 | bool cached; ///< Onode is logically in the cache |
1069 | /// (it can be pinned and hence physically out | |
1070 | /// of it at the moment though) | |
1071 | bool pinned; ///< Onode is pinned | |
1072 | /// (or should be pinned when cached) | |
7c673cae FG |
1073 | ExtentMap extent_map; |
1074 | ||
1075 | // track txc's that have not been committed to kv store (and whose | |
1076 | // effects cannot be read via the kvdb read methods) | |
1077 | std::atomic<int> flushing_count = {0}; | |
9f95a23c | 1078 | std::atomic<int> waiting_count = {0}; |
11fdf7f2 TL |
1079 | /// protect flush_txns |
1080 | ceph::mutex flush_lock = ceph::make_mutex("BlueStore::Onode::flush_lock"); | |
1081 | ceph::condition_variable flush_cond; ///< wait here for uncommitted txns | |
7c673cae FG |
1082 | |
1083 | Onode(Collection *c, const ghobject_t& o, | |
f91f0fd5 | 1084 | const mempool::bluestore_cache_meta::string& k) |
f6b5b4d7 | 1085 | : nref(0), |
7c673cae FG |
1086 | c(c), |
1087 | oid(o), | |
1088 | key(k), | |
1089 | exists(false), | |
f6b5b4d7 TL |
1090 | cached(false), |
1091 | pinned(false), | |
7c673cae FG |
1092 | extent_map(this) { |
1093 | } | |
eafe8130 | 1094 | Onode(Collection* c, const ghobject_t& o, |
f6b5b4d7 TL |
1095 | const std::string& k) |
1096 | : nref(0), | |
eafe8130 TL |
1097 | c(c), |
1098 | oid(o), | |
1099 | key(k), | |
1100 | exists(false), | |
f6b5b4d7 TL |
1101 | cached(false), |
1102 | pinned(false), | |
eafe8130 TL |
1103 | extent_map(this) { |
1104 | } | |
1105 | Onode(Collection* c, const ghobject_t& o, | |
1106 | const char* k) | |
f6b5b4d7 | 1107 | : nref(0), |
eafe8130 TL |
1108 | c(c), |
1109 | oid(o), | |
1110 | key(k), | |
1111 | exists(false), | |
f6b5b4d7 TL |
1112 | cached(false), |
1113 | pinned(false), | |
eafe8130 TL |
1114 | extent_map(this) { |
1115 | } | |
1116 | ||
1117 | static Onode* decode( | |
1118 | CollectionRef c, | |
1119 | const ghobject_t& oid, | |
1120 | const string& key, | |
1121 | const bufferlist& v); | |
7c673cae | 1122 | |
9f95a23c TL |
1123 | void dump(Formatter* f) const; |
1124 | ||
7c673cae | 1125 | void flush(); |
f6b5b4d7 TL |
1126 | void get(); |
1127 | void put(); | |
1128 | ||
1129 | inline bool put_cache() { | |
1130 | ceph_assert(!cached); | |
1131 | cached = true; | |
1132 | return !pinned; | |
7c673cae | 1133 | } |
f6b5b4d7 TL |
1134 | inline bool pop_cache() { |
1135 | ceph_assert(cached); | |
1136 | cached = false; | |
1137 | return !pinned; | |
7c673cae | 1138 | } |
9f95a23c TL |
1139 | |
1140 | const string& get_omap_prefix(); | |
1141 | void get_omap_header(string *out); | |
1142 | void get_omap_key(const string& key, string *out); | |
1143 | void rewrite_omap_key(const string& old, string *out); | |
1144 | void get_omap_tail(string *out); | |
1145 | void decode_omap_key(const string& key, string *user_key); | |
7c673cae FG |
1146 | }; |
1147 | typedef boost::intrusive_ptr<Onode> OnodeRef; | |
1148 | ||
9f95a23c TL |
1149 | /// A generic Cache Shard |
1150 | struct CacheShard { | |
1151 | CephContext *cct; | |
7c673cae | 1152 | PerfCounters *logger; |
11fdf7f2 TL |
1153 | |
1154 | /// protect lru and other structures | |
1155 | ceph::recursive_mutex lock = { | |
9f95a23c | 1156 | ceph::make_recursive_mutex("BlueStore::CacheShard::lock") }; |
7c673cae | 1157 | |
9f95a23c TL |
1158 | std::atomic<uint64_t> max = {0}; |
1159 | std::atomic<uint64_t> num = {0}; | |
7c673cae | 1160 | |
9f95a23c TL |
1161 | CacheShard(CephContext* cct) : cct(cct), logger(nullptr) {} |
1162 | virtual ~CacheShard() {} | |
7c673cae | 1163 | |
9f95a23c TL |
1164 | void set_max(uint64_t max_) { |
1165 | max = max_; | |
7c673cae FG |
1166 | } |
1167 | ||
9f95a23c TL |
1168 | uint64_t _get_num() { |
1169 | return num; | |
7c673cae | 1170 | } |
7c673cae | 1171 | |
9f95a23c TL |
1172 | virtual void _trim_to(uint64_t new_size) = 0; |
1173 | void _trim() { | |
1174 | if (cct->_conf->objectstore_blackhole) { | |
1175 | // do not trim if we are throwing away IOs a layer down | |
1176 | return; | |
1177 | } | |
1178 | _trim_to(max); | |
1179 | } | |
7c673cae | 1180 | |
9f95a23c | 1181 | void trim() { |
11fdf7f2 | 1182 | std::lock_guard l(lock); |
9f95a23c TL |
1183 | _trim(); |
1184 | } | |
1185 | void flush() { | |
1186 | std::lock_guard l(lock); | |
1187 | // we should not be shutting down after the blackhole is enabled | |
1188 | assert(!cct->_conf->objectstore_blackhole); | |
1189 | _trim_to(0); | |
31f18b77 FG |
1190 | } |
1191 | ||
7c673cae FG |
1192 | #ifdef DEBUG_CACHE |
1193 | virtual void _audit(const char *s) = 0; | |
1194 | #else | |
1195 | void _audit(const char *s) { /* no-op */ } | |
1196 | #endif | |
1197 | }; | |
1198 | ||
9f95a23c TL |
1199 | /// A Generic onode Cache Shard |
1200 | struct OnodeCacheShard : public CacheShard { | |
1201 | std::atomic<uint64_t> num_pinned = {0}; | |
7c673cae | 1202 | |
9f95a23c | 1203 | std::array<std::pair<ghobject_t, mono_clock::time_point>, 64> dumped_onodes; |
f6b5b4d7 TL |
1204 | |
1205 | virtual void _pin(Onode* o) = 0; | |
1206 | virtual void _unpin(Onode* o) = 0; | |
1207 | ||
7c673cae | 1208 | public: |
9f95a23c TL |
1209 | OnodeCacheShard(CephContext* cct) : CacheShard(cct) {} |
1210 | static OnodeCacheShard *create(CephContext* cct, string type, | |
1211 | PerfCounters *logger); | |
f6b5b4d7 TL |
1212 | virtual void _add(Onode* o, int level) = 0; |
1213 | virtual void _rm(Onode* o) = 0; | |
9f95a23c | 1214 | |
f6b5b4d7 | 1215 | void pin(Onode* o, std::function<bool ()> validator) { |
9f95a23c | 1216 | std::lock_guard l(lock); |
f6b5b4d7 TL |
1217 | if (validator()) { |
1218 | _pin(o); | |
1219 | } | |
7c673cae FG |
1220 | } |
1221 | ||
f6b5b4d7 | 1222 | void unpin(Onode* o, std::function<bool()> validator) { |
11fdf7f2 | 1223 | std::lock_guard l(lock); |
f6b5b4d7 TL |
1224 | if (validator()) { |
1225 | _unpin(o); | |
1226 | } | |
7c673cae FG |
1227 | } |
1228 | ||
f6b5b4d7 | 1229 | virtual void move_pinned(OnodeCacheShard *to, Onode *o) = 0; |
9f95a23c TL |
1230 | virtual void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) = 0; |
1231 | bool empty() { | |
1232 | return _get_num() == 0; | |
1233 | } | |
7c673cae FG |
1234 | }; |
1235 | ||
9f95a23c TL |
1236 | /// A Generic buffer Cache Shard |
1237 | struct BufferCacheShard : public CacheShard { | |
1238 | std::atomic<uint64_t> num_extents = {0}; | |
1239 | std::atomic<uint64_t> num_blobs = {0}; | |
1240 | uint64_t buffer_bytes = 0; | |
7c673cae FG |
1241 | |
1242 | public: | |
9f95a23c TL |
1243 | BufferCacheShard(CephContext* cct) : CacheShard(cct) {} |
1244 | static BufferCacheShard *create(CephContext* cct, string type, | |
1245 | PerfCounters *logger); | |
1246 | virtual void _add(Buffer *b, int level, Buffer *near) = 0; | |
1247 | virtual void _rm(Buffer *b) = 0; | |
1248 | virtual void _move(BufferCacheShard *src, Buffer *b) = 0; | |
1249 | virtual void _touch(Buffer *b) = 0; | |
1250 | virtual void _adjust_size(Buffer *b, int64_t delta) = 0; | |
1251 | ||
1252 | uint64_t _get_bytes() { | |
1253 | return buffer_bytes; | |
7c673cae | 1254 | } |
9f95a23c TL |
1255 | |
1256 | void add_extent() { | |
1257 | ++num_extents; | |
7c673cae | 1258 | } |
9f95a23c TL |
1259 | void rm_extent() { |
1260 | --num_extents; | |
7c673cae | 1261 | } |
7c673cae | 1262 | |
9f95a23c TL |
1263 | void add_blob() { |
1264 | ++num_blobs; | |
7c673cae | 1265 | } |
9f95a23c TL |
1266 | void rm_blob() { |
1267 | --num_blobs; | |
7c673cae FG |
1268 | } |
1269 | ||
9f95a23c TL |
1270 | virtual void add_stats(uint64_t *extents, |
1271 | uint64_t *blobs, | |
1272 | uint64_t *buffers, | |
1273 | uint64_t *bytes) = 0; | |
1274 | ||
1275 | bool empty() { | |
1276 | std::lock_guard l(lock); | |
1277 | return _get_bytes() == 0; | |
1278 | } | |
7c673cae FG |
1279 | }; |
1280 | ||
1281 | struct OnodeSpace { | |
9f95a23c | 1282 | OnodeCacheShard *cache; |
7c673cae | 1283 | |
9f95a23c | 1284 | private: |
7c673cae | 1285 | /// forward lookups |
f91f0fd5 | 1286 | mempool::bluestore_cache_meta::unordered_map<ghobject_t,OnodeRef> onode_map; |
7c673cae FG |
1287 | |
1288 | friend class Collection; // for split_cache() | |
1289 | ||
f6b5b4d7 TL |
1290 | friend struct LruOnodeCacheShard; |
1291 | void _remove(const ghobject_t& oid); | |
7c673cae | 1292 | public: |
9f95a23c | 1293 | OnodeSpace(OnodeCacheShard *c) : cache(c) {} |
7c673cae FG |
1294 | ~OnodeSpace() { |
1295 | clear(); | |
1296 | } | |
1297 | ||
f6b5b4d7 | 1298 | OnodeRef add(const ghobject_t& oid, OnodeRef& o); |
7c673cae | 1299 | OnodeRef lookup(const ghobject_t& o); |
7c673cae FG |
1300 | void rename(OnodeRef& o, const ghobject_t& old_oid, |
1301 | const ghobject_t& new_oid, | |
f91f0fd5 | 1302 | const mempool::bluestore_cache_meta::string& new_okey); |
7c673cae FG |
1303 | void clear(); |
1304 | bool empty(); | |
1305 | ||
11fdf7f2 TL |
1306 | template <int LogLevelV> |
1307 | void dump(CephContext *cct); | |
3efd9988 | 1308 | |
7c673cae FG |
1309 | /// return true if f true for any item |
1310 | bool map_any(std::function<bool(OnodeRef)> f); | |
1311 | }; | |
1312 | ||
11fdf7f2 | 1313 | class OpSequencer; |
9f95a23c | 1314 | using OpSequencerRef = ceph::ref_t<OpSequencer>; |
11fdf7f2 | 1315 | |
7c673cae FG |
1316 | struct Collection : public CollectionImpl { |
1317 | BlueStore *store; | |
11fdf7f2 | 1318 | OpSequencerRef osr; |
9f95a23c | 1319 | BufferCacheShard *cache; ///< our cache shard |
7c673cae | 1320 | bluestore_cnode_t cnode; |
9f95a23c TL |
1321 | ceph::shared_mutex lock = |
1322 | ceph::make_shared_mutex("BlueStore::Collection::lock", true, false); | |
7c673cae FG |
1323 | |
1324 | bool exists; | |
1325 | ||
1326 | SharedBlobSet shared_blob_set; ///< open SharedBlobs | |
1327 | ||
1328 | // cache onodes on a per-collection basis to avoid lock | |
1329 | // contention. | |
1330 | OnodeSpace onode_map; | |
1331 | ||
1332 | //pool options | |
1333 | pool_opts_t pool_opts; | |
11fdf7f2 | 1334 | ContextQueue *commit_queue; |
7c673cae | 1335 | |
f6b5b4d7 TL |
1336 | OnodeCacheShard* get_onode_cache() const { |
1337 | return onode_map.cache; | |
1338 | } | |
9f95a23c | 1339 | OnodeRef get_onode(const ghobject_t& oid, bool create, bool is_createop=false); |
7c673cae FG |
1340 | |
1341 | // the terminology is confusing here, sorry! | |
1342 | // | |
1343 | // blob_t shared_blob_t | |
1344 | // !shared unused -> open | |
1345 | // shared !loaded -> open + shared | |
1346 | // shared loaded -> open + shared + loaded | |
1347 | // | |
1348 | // i.e., | |
1349 | // open = SharedBlob is instantiated | |
1350 | // shared = blob_t shared flag is set; SharedBlob is hashed. | |
1351 | // loaded = SharedBlob::shared_blob_t is loaded from kv store | |
1352 | void open_shared_blob(uint64_t sbid, BlobRef b); | |
1353 | void load_shared_blob(SharedBlobRef sb); | |
1354 | void make_blob_shared(uint64_t sbid, BlobRef b); | |
31f18b77 | 1355 | uint64_t make_blob_unshared(SharedBlob *sb); |
7c673cae FG |
1356 | |
1357 | BlobRef new_blob() { | |
1358 | BlobRef b = new Blob(); | |
1359 | b->shared_blob = new SharedBlob(this); | |
1360 | return b; | |
1361 | } | |
1362 | ||
7c673cae FG |
1363 | bool contains(const ghobject_t& oid) { |
1364 | if (cid.is_meta()) | |
1365 | return oid.hobj.pool == -1; | |
1366 | spg_t spgid; | |
1367 | if (cid.is_pg(&spgid)) | |
1368 | return | |
1369 | spgid.pgid.contains(cnode.bits, oid) && | |
1370 | oid.shard_id == spgid.shard; | |
1371 | return false; | |
1372 | } | |
1373 | ||
9f95a23c TL |
1374 | int64_t pool() const { |
1375 | return cid.pool(); | |
1376 | } | |
1377 | ||
7c673cae | 1378 | void split_cache(Collection *dest); |
7c673cae | 1379 | |
11fdf7f2 TL |
1380 | bool flush_commit(Context *c) override; |
1381 | void flush() override; | |
1382 | void flush_all_but_last(); | |
1383 | ||
9f95a23c | 1384 | Collection(BlueStore *ns, OnodeCacheShard *oc, BufferCacheShard *bc, coll_t c); |
7c673cae FG |
1385 | }; |
1386 | ||
1387 | class OmapIteratorImpl : public ObjectMap::ObjectMapIteratorImpl { | |
1388 | CollectionRef c; | |
1389 | OnodeRef o; | |
1390 | KeyValueDB::Iterator it; | |
1391 | string head, tail; | |
11fdf7f2 TL |
1392 | |
1393 | string _stringify() const; | |
1394 | ||
7c673cae FG |
1395 | public: |
1396 | OmapIteratorImpl(CollectionRef c, OnodeRef o, KeyValueDB::Iterator it); | |
1397 | int seek_to_first() override; | |
1398 | int upper_bound(const string &after) override; | |
1399 | int lower_bound(const string &to) override; | |
1400 | bool valid() override; | |
11fdf7f2 | 1401 | int next() override; |
7c673cae FG |
1402 | string key() override; |
1403 | bufferlist value() override; | |
9f95a23c TL |
1404 | std::string tail_key() { |
1405 | return tail; | |
1406 | } | |
1407 | ||
7c673cae FG |
1408 | int status() override { |
1409 | return 0; | |
1410 | } | |
1411 | }; | |
1412 | ||
31f18b77 FG |
1413 | struct volatile_statfs{ |
1414 | enum { | |
1415 | STATFS_ALLOCATED = 0, | |
1416 | STATFS_STORED, | |
1417 | STATFS_COMPRESSED_ORIGINAL, | |
1418 | STATFS_COMPRESSED, | |
1419 | STATFS_COMPRESSED_ALLOCATED, | |
1420 | STATFS_LAST | |
1421 | }; | |
1422 | int64_t values[STATFS_LAST]; | |
1423 | volatile_statfs() { | |
1424 | memset(this, 0, sizeof(volatile_statfs)); | |
1425 | } | |
1426 | void reset() { | |
1427 | *this = volatile_statfs(); | |
1428 | } | |
11fdf7f2 TL |
1429 | void publish(store_statfs_t* buf) const { |
1430 | buf->allocated = allocated(); | |
1431 | buf->data_stored = stored(); | |
1432 | buf->data_compressed = compressed(); | |
1433 | buf->data_compressed_original = compressed_original(); | |
1434 | buf->data_compressed_allocated = compressed_allocated(); | |
1435 | } | |
1436 | ||
31f18b77 FG |
1437 | volatile_statfs& operator+=(const volatile_statfs& other) { |
1438 | for (size_t i = 0; i < STATFS_LAST; ++i) { | |
1439 | values[i] += other.values[i]; | |
1440 | } | |
1441 | return *this; | |
1442 | } | |
1443 | int64_t& allocated() { | |
1444 | return values[STATFS_ALLOCATED]; | |
1445 | } | |
1446 | int64_t& stored() { | |
1447 | return values[STATFS_STORED]; | |
1448 | } | |
1449 | int64_t& compressed_original() { | |
1450 | return values[STATFS_COMPRESSED_ORIGINAL]; | |
1451 | } | |
1452 | int64_t& compressed() { | |
1453 | return values[STATFS_COMPRESSED]; | |
1454 | } | |
1455 | int64_t& compressed_allocated() { | |
1456 | return values[STATFS_COMPRESSED_ALLOCATED]; | |
1457 | } | |
11fdf7f2 TL |
1458 | int64_t allocated() const { |
1459 | return values[STATFS_ALLOCATED]; | |
1460 | } | |
1461 | int64_t stored() const { | |
1462 | return values[STATFS_STORED]; | |
1463 | } | |
1464 | int64_t compressed_original() const { | |
1465 | return values[STATFS_COMPRESSED_ORIGINAL]; | |
1466 | } | |
1467 | int64_t compressed() const { | |
1468 | return values[STATFS_COMPRESSED]; | |
1469 | } | |
1470 | int64_t compressed_allocated() const { | |
1471 | return values[STATFS_COMPRESSED_ALLOCATED]; | |
1472 | } | |
1473 | volatile_statfs& operator=(const store_statfs_t& st) { | |
1474 | values[STATFS_ALLOCATED] = st.allocated; | |
1475 | values[STATFS_STORED] = st.data_stored; | |
1476 | values[STATFS_COMPRESSED_ORIGINAL] = st.data_compressed_original; | |
1477 | values[STATFS_COMPRESSED] = st.data_compressed; | |
1478 | values[STATFS_COMPRESSED_ALLOCATED] = st.data_compressed_allocated; | |
1479 | return *this; | |
1480 | } | |
31f18b77 FG |
1481 | bool is_empty() { |
1482 | return values[STATFS_ALLOCATED] == 0 && | |
1483 | values[STATFS_STORED] == 0 && | |
1484 | values[STATFS_COMPRESSED] == 0 && | |
1485 | values[STATFS_COMPRESSED_ORIGINAL] == 0 && | |
1486 | values[STATFS_COMPRESSED_ALLOCATED] == 0; | |
1487 | } | |
11fdf7f2 TL |
1488 | void decode(bufferlist::const_iterator& it) { |
1489 | using ceph::decode; | |
31f18b77 | 1490 | for (size_t i = 0; i < STATFS_LAST; i++) { |
11fdf7f2 | 1491 | decode(values[i], it); |
31f18b77 FG |
1492 | } |
1493 | } | |
1494 | ||
1495 | void encode(bufferlist& bl) { | |
11fdf7f2 | 1496 | using ceph::encode; |
31f18b77 | 1497 | for (size_t i = 0; i < STATFS_LAST; i++) { |
11fdf7f2 | 1498 | encode(values[i], bl); |
31f18b77 FG |
1499 | } |
1500 | } | |
1501 | }; | |
1502 | ||
11fdf7f2 | 1503 | struct TransContext final : public AioContext { |
31f18b77 FG |
1504 | MEMPOOL_CLASS_HELPERS(); |
1505 | ||
7c673cae FG |
1506 | typedef enum { |
1507 | STATE_PREPARE, | |
1508 | STATE_AIO_WAIT, | |
1509 | STATE_IO_DONE, | |
1510 | STATE_KV_QUEUED, // queued for kv_sync_thread submission | |
1511 | STATE_KV_SUBMITTED, // submitted to kv; not yet synced | |
1512 | STATE_KV_DONE, | |
1513 | STATE_DEFERRED_QUEUED, // in deferred_queue (pending or running) | |
1514 | STATE_DEFERRED_CLEANUP, // remove deferred kv record | |
1515 | STATE_DEFERRED_DONE, | |
1516 | STATE_FINISHING, | |
1517 | STATE_DONE, | |
1518 | } state_t; | |
1519 | ||
1520 | state_t state = STATE_PREPARE; | |
1521 | ||
1522 | const char *get_state_name() { | |
1523 | switch (state) { | |
1524 | case STATE_PREPARE: return "prepare"; | |
1525 | case STATE_AIO_WAIT: return "aio_wait"; | |
1526 | case STATE_IO_DONE: return "io_done"; | |
1527 | case STATE_KV_QUEUED: return "kv_queued"; | |
1528 | case STATE_KV_SUBMITTED: return "kv_submitted"; | |
1529 | case STATE_KV_DONE: return "kv_done"; | |
1530 | case STATE_DEFERRED_QUEUED: return "deferred_queued"; | |
1531 | case STATE_DEFERRED_CLEANUP: return "deferred_cleanup"; | |
1532 | case STATE_DEFERRED_DONE: return "deferred_done"; | |
1533 | case STATE_FINISHING: return "finishing"; | |
1534 | case STATE_DONE: return "done"; | |
1535 | } | |
1536 | return "???"; | |
1537 | } | |
1538 | ||
9f95a23c | 1539 | #if defined(WITH_LTTNG) |
7c673cae FG |
1540 | const char *get_state_latency_name(int state) { |
1541 | switch (state) { | |
1542 | case l_bluestore_state_prepare_lat: return "prepare"; | |
1543 | case l_bluestore_state_aio_wait_lat: return "aio_wait"; | |
1544 | case l_bluestore_state_io_done_lat: return "io_done"; | |
1545 | case l_bluestore_state_kv_queued_lat: return "kv_queued"; | |
1546 | case l_bluestore_state_kv_committing_lat: return "kv_committing"; | |
1547 | case l_bluestore_state_kv_done_lat: return "kv_done"; | |
1548 | case l_bluestore_state_deferred_queued_lat: return "deferred_queued"; | |
1549 | case l_bluestore_state_deferred_cleanup_lat: return "deferred_cleanup"; | |
1550 | case l_bluestore_state_finishing_lat: return "finishing"; | |
1551 | case l_bluestore_state_done_lat: return "done"; | |
1552 | } | |
1553 | return "???"; | |
1554 | } | |
1555 | #endif | |
1556 | ||
11fdf7f2 TL |
1557 | CollectionRef ch; |
1558 | OpSequencerRef osr; // this should be ch->osr | |
7c673cae FG |
1559 | boost::intrusive::list_member_hook<> sequencer_item; |
1560 | ||
9f95a23c | 1561 | uint64_t bytes = 0, ios = 0, cost = 0; |
7c673cae FG |
1562 | |
1563 | set<OnodeRef> onodes; ///< these need to be updated/written | |
1564 | set<OnodeRef> modified_objects; ///< objects we modified (and need a ref) | |
1565 | set<SharedBlobRef> shared_blobs; ///< these need to be updated/written | |
1566 | set<SharedBlobRef> shared_blobs_written; ///< update these on io completion | |
1567 | ||
1568 | KeyValueDB::Transaction t; ///< then we will commit this | |
7c673cae FG |
1569 | list<Context*> oncommits; ///< more commit completions |
1570 | list<CollectionRef> removed_collections; ///< colls we removed | |
1571 | ||
1572 | boost::intrusive::list_member_hook<> deferred_queue_item; | |
1573 | bluestore_deferred_transaction_t *deferred_txn = nullptr; ///< if any | |
1574 | ||
1575 | interval_set<uint64_t> allocated, released; | |
11fdf7f2 TL |
1576 | volatile_statfs statfs_delta; ///< overall store statistics delta |
1577 | uint64_t osd_pool_id = META_POOL_ID; ///< osd pool id we're operating on | |
1578 | ||
7c673cae FG |
1579 | IOContext ioc; |
1580 | bool had_ios = false; ///< true if we submitted IOs before our kv txn | |
1581 | ||
7c673cae | 1582 | uint64_t seq = 0; |
9f95a23c TL |
1583 | mono_clock::time_point start; |
1584 | mono_clock::time_point last_stamp; | |
7c673cae FG |
1585 | |
1586 | uint64_t last_nid = 0; ///< if non-zero, highest new nid we allocated | |
1587 | uint64_t last_blobid = 0; ///< if non-zero, highest new blobid we allocated | |
1588 | ||
9f95a23c TL |
1589 | #if defined(WITH_LTTNG) |
1590 | bool tracing = false; | |
1591 | #endif | |
1592 | ||
11fdf7f2 TL |
1593 | explicit TransContext(CephContext* cct, Collection *c, OpSequencer *o, |
1594 | list<Context*> *on_commits) | |
1595 | : ch(c), | |
1596 | osr(o), | |
7c673cae | 1597 | ioc(cct, this), |
9f95a23c | 1598 | start(mono_clock::now()) { |
7c673cae | 1599 | last_stamp = start; |
11fdf7f2 TL |
1600 | if (on_commits) { |
1601 | oncommits.swap(*on_commits); | |
1602 | } | |
7c673cae FG |
1603 | } |
1604 | ~TransContext() { | |
1605 | delete deferred_txn; | |
1606 | } | |
1607 | ||
1608 | void write_onode(OnodeRef &o) { | |
1609 | onodes.insert(o); | |
1610 | } | |
1611 | void write_shared_blob(SharedBlobRef &sb) { | |
1612 | shared_blobs.insert(sb); | |
1613 | } | |
31f18b77 FG |
1614 | void unshare_blob(SharedBlob *sb) { |
1615 | shared_blobs.erase(sb); | |
1616 | } | |
1617 | ||
7c673cae FG |
1618 | /// note we logically modified object (when onode itself is unmodified) |
1619 | void note_modified_object(OnodeRef &o) { | |
1620 | // onode itself isn't written, though | |
1621 | modified_objects.insert(o); | |
1622 | } | |
a8e16298 | 1623 | void note_removed_object(OnodeRef& o) { |
7c673cae | 1624 | onodes.erase(o); |
a8e16298 | 1625 | modified_objects.insert(o); |
7c673cae FG |
1626 | } |
1627 | ||
1628 | void aio_finish(BlueStore *store) override { | |
1629 | store->txc_aio_finish(this); | |
1630 | } | |
1631 | }; | |
1632 | ||
9f95a23c TL |
1633 | class BlueStoreThrottle { |
1634 | #if defined(WITH_LTTNG) | |
1635 | const std::chrono::time_point<mono_clock> time_base = mono_clock::now(); | |
1636 | ||
1637 | // Time of last chosen io (microseconds) | |
1638 | std::atomic<uint64_t> previous_emitted_tp_time_mono_mcs = {0}; | |
1639 | std::atomic<uint64_t> ios_started_since_last_traced = {0}; | |
1640 | std::atomic<uint64_t> ios_completed_since_last_traced = {0}; | |
1641 | ||
1642 | std::atomic_uint pending_kv_ios = {0}; | |
1643 | std::atomic_uint pending_deferred_ios = {0}; | |
1644 | ||
1645 | // Min period between trace points (microseconds) | |
1646 | std::atomic<uint64_t> trace_period_mcs = {0}; | |
1647 | ||
1648 | bool should_trace( | |
1649 | uint64_t *started, | |
1650 | uint64_t *completed) { | |
1651 | uint64_t min_period_mcs = trace_period_mcs.load( | |
1652 | std::memory_order_relaxed); | |
1653 | ||
1654 | if (min_period_mcs == 0) { | |
1655 | *started = 1; | |
1656 | *completed = ios_completed_since_last_traced.exchange(0); | |
1657 | return true; | |
1658 | } else { | |
1659 | ios_started_since_last_traced++; | |
1660 | auto now_mcs = ceph::to_microseconds<uint64_t>( | |
1661 | mono_clock::now() - time_base); | |
1662 | uint64_t previous_mcs = previous_emitted_tp_time_mono_mcs; | |
1663 | uint64_t period_mcs = now_mcs - previous_mcs; | |
1664 | if (period_mcs > min_period_mcs) { | |
1665 | if (previous_emitted_tp_time_mono_mcs.compare_exchange_strong( | |
1666 | previous_mcs, now_mcs)) { | |
1667 | // This would be racy at a sufficiently extreme trace rate, but isn't | |
1668 | // worth the overhead of doing it more carefully. | |
1669 | *started = ios_started_since_last_traced.exchange(0); | |
1670 | *completed = ios_completed_since_last_traced.exchange(0); | |
1671 | return true; | |
1672 | } | |
1673 | } | |
1674 | return false; | |
1675 | } | |
1676 | } | |
1677 | #endif | |
1678 | ||
1679 | #if defined(WITH_LTTNG) | |
1680 | void emit_initial_tracepoint( | |
1681 | KeyValueDB &db, | |
1682 | TransContext &txc, | |
1683 | mono_clock::time_point); | |
1684 | #else | |
1685 | void emit_initial_tracepoint( | |
1686 | KeyValueDB &db, | |
1687 | TransContext &txc, | |
1688 | mono_clock::time_point) {} | |
1689 | #endif | |
1690 | ||
1691 | Throttle throttle_bytes; ///< submit to commit | |
1692 | Throttle throttle_deferred_bytes; ///< submit to deferred complete | |
1693 | ||
1694 | public: | |
1695 | BlueStoreThrottle(CephContext *cct) : | |
1696 | throttle_bytes(cct, "bluestore_throttle_bytes", 0), | |
1697 | throttle_deferred_bytes(cct, "bluestore_throttle_deferred_bytes", 0) | |
1698 | { | |
1699 | reset_throttle(cct->_conf); | |
1700 | } | |
1701 | ||
1702 | #if defined(WITH_LTTNG) | |
1703 | void complete_kv(TransContext &txc); | |
1704 | void complete(TransContext &txc); | |
1705 | #else | |
1706 | void complete_kv(TransContext &txc) {} | |
1707 | void complete(TransContext &txc) {} | |
1708 | #endif | |
1709 | ||
1710 | mono_clock::duration log_state_latency( | |
1711 | TransContext &txc, PerfCounters *logger, int state); | |
1712 | bool try_start_transaction( | |
1713 | KeyValueDB &db, | |
1714 | TransContext &txc, | |
1715 | mono_clock::time_point); | |
1716 | void finish_start_transaction( | |
1717 | KeyValueDB &db, | |
1718 | TransContext &txc, | |
1719 | mono_clock::time_point); | |
1720 | void release_kv_throttle(uint64_t cost) { | |
1721 | throttle_bytes.put(cost); | |
1722 | } | |
1723 | void release_deferred_throttle(uint64_t cost) { | |
1724 | throttle_deferred_bytes.put(cost); | |
1725 | } | |
1726 | bool should_submit_deferred() { | |
1727 | return throttle_deferred_bytes.past_midpoint(); | |
1728 | } | |
1729 | void reset_throttle(const ConfigProxy &conf) { | |
1730 | throttle_bytes.reset_max(conf->bluestore_throttle_bytes); | |
1731 | throttle_deferred_bytes.reset_max( | |
1732 | conf->bluestore_throttle_bytes + | |
1733 | conf->bluestore_throttle_deferred_bytes); | |
1734 | #if defined(WITH_LTTNG) | |
1735 | double rate = conf.get_val<double>("bluestore_throttle_trace_rate"); | |
1736 | trace_period_mcs = rate > 0 ? floor((1/rate) * 1000000.0) : 0; | |
1737 | #endif | |
1738 | } | |
1739 | } throttle; | |
1740 | ||
7c673cae FG |
1741 | typedef boost::intrusive::list< |
1742 | TransContext, | |
1743 | boost::intrusive::member_hook< | |
1744 | TransContext, | |
1745 | boost::intrusive::list_member_hook<>, | |
1746 | &TransContext::deferred_queue_item> > deferred_queue_t; | |
1747 | ||
11fdf7f2 | 1748 | struct DeferredBatch final : public AioContext { |
7c673cae FG |
1749 | OpSequencer *osr; |
1750 | struct deferred_io { | |
1751 | bufferlist bl; ///< data | |
1752 | uint64_t seq; ///< deferred transaction seq | |
1753 | }; | |
1754 | map<uint64_t,deferred_io> iomap; ///< map of ios in this batch | |
1755 | deferred_queue_t txcs; ///< txcs in this batch | |
1756 | IOContext ioc; ///< our aios | |
1757 | /// bytes of pending io for each deferred seq (may be 0) | |
1758 | map<uint64_t,int> seq_bytes; | |
1759 | ||
1760 | void _discard(CephContext *cct, uint64_t offset, uint64_t length); | |
1761 | void _audit(CephContext *cct); | |
1762 | ||
1763 | DeferredBatch(CephContext *cct, OpSequencer *osr) | |
1764 | : osr(osr), ioc(cct, this) {} | |
1765 | ||
1766 | /// prepare a write | |
1767 | void prepare_write(CephContext *cct, | |
1768 | uint64_t seq, uint64_t offset, uint64_t length, | |
1769 | bufferlist::const_iterator& p); | |
1770 | ||
1771 | void aio_finish(BlueStore *store) override { | |
1772 | store->_deferred_aio_finish(osr); | |
1773 | } | |
1774 | }; | |
1775 | ||
11fdf7f2 | 1776 | class OpSequencer : public RefCountedObject { |
7c673cae | 1777 | public: |
11fdf7f2 TL |
1778 | ceph::mutex qlock = ceph::make_mutex("BlueStore::OpSequencer::qlock"); |
1779 | ceph::condition_variable qcond; | |
7c673cae FG |
1780 | typedef boost::intrusive::list< |
1781 | TransContext, | |
1782 | boost::intrusive::member_hook< | |
1783 | TransContext, | |
1784 | boost::intrusive::list_member_hook<>, | |
1785 | &TransContext::sequencer_item> > q_list_t; | |
1786 | q_list_t q; ///< transactions | |
1787 | ||
1788 | boost::intrusive::list_member_hook<> deferred_osr_queue_item; | |
1789 | ||
1790 | DeferredBatch *deferred_running = nullptr; | |
1791 | DeferredBatch *deferred_pending = nullptr; | |
1792 | ||
7c673cae | 1793 | BlueStore *store; |
11fdf7f2 | 1794 | coll_t cid; |
7c673cae FG |
1795 | |
1796 | uint64_t last_seq = 0; | |
1797 | ||
1798 | std::atomic_int txc_with_unstable_io = {0}; ///< num txcs with unstable io | |
1799 | ||
1800 | std::atomic_int kv_committing_serially = {0}; | |
1801 | ||
1802 | std::atomic_int kv_submitted_waiters = {0}; | |
1803 | ||
11fdf7f2 | 1804 | std::atomic_bool zombie = {false}; ///< in zombie_osr set (collection going away) |
7c673cae | 1805 | |
9f95a23c TL |
1806 | const uint32_t sequencer_id; |
1807 | ||
1808 | uint32_t get_sequencer_id() const { | |
1809 | return sequencer_id; | |
7c673cae FG |
1810 | } |
1811 | ||
1812 | void queue_new(TransContext *txc) { | |
11fdf7f2 | 1813 | std::lock_guard l(qlock); |
7c673cae FG |
1814 | txc->seq = ++last_seq; |
1815 | q.push_back(*txc); | |
1816 | } | |
1817 | ||
1818 | void drain() { | |
11fdf7f2 | 1819 | std::unique_lock l(qlock); |
7c673cae FG |
1820 | while (!q.empty()) |
1821 | qcond.wait(l); | |
1822 | } | |
1823 | ||
1824 | void drain_preceding(TransContext *txc) { | |
11fdf7f2 | 1825 | std::unique_lock l(qlock); |
9f95a23c | 1826 | while (&q.front() != txc) |
7c673cae FG |
1827 | qcond.wait(l); |
1828 | } | |
1829 | ||
1830 | bool _is_all_kv_submitted() { | |
11fdf7f2 TL |
1831 | // caller must hold qlock & q.empty() must not empty |
1832 | ceph_assert(!q.empty()); | |
7c673cae FG |
1833 | TransContext *txc = &q.back(); |
1834 | if (txc->state >= TransContext::STATE_KV_SUBMITTED) { | |
1835 | return true; | |
1836 | } | |
1837 | return false; | |
1838 | } | |
1839 | ||
11fdf7f2 TL |
1840 | void flush() { |
1841 | std::unique_lock l(qlock); | |
1842 | while (true) { | |
1843 | // set flag before the check because the condition | |
1844 | // may become true outside qlock, and we need to make | |
1845 | // sure those threads see waiters and signal qcond. | |
1846 | ++kv_submitted_waiters; | |
1847 | if (q.empty() || _is_all_kv_submitted()) { | |
1848 | --kv_submitted_waiters; | |
1849 | return; | |
1850 | } | |
1851 | qcond.wait(l); | |
1852 | --kv_submitted_waiters; | |
1853 | } | |
1854 | } | |
1855 | ||
1856 | void flush_all_but_last() { | |
1857 | std::unique_lock l(qlock); | |
1858 | assert (q.size() >= 1); | |
7c673cae FG |
1859 | while (true) { |
1860 | // set flag before the check because the condition | |
1861 | // may become true outside qlock, and we need to make | |
1862 | // sure those threads see waiters and signal qcond. | |
1863 | ++kv_submitted_waiters; | |
11fdf7f2 TL |
1864 | if (q.size() <= 1) { |
1865 | --kv_submitted_waiters; | |
7c673cae | 1866 | return; |
11fdf7f2 TL |
1867 | } else { |
1868 | auto it = q.rbegin(); | |
1869 | it++; | |
1870 | if (it->state >= TransContext::STATE_KV_SUBMITTED) { | |
eafe8130 | 1871 | --kv_submitted_waiters; |
11fdf7f2 TL |
1872 | return; |
1873 | } | |
7c673cae FG |
1874 | } |
1875 | qcond.wait(l); | |
1876 | --kv_submitted_waiters; | |
1877 | } | |
1878 | } | |
1879 | ||
11fdf7f2 TL |
1880 | bool flush_commit(Context *c) { |
1881 | std::lock_guard l(qlock); | |
7c673cae FG |
1882 | if (q.empty()) { |
1883 | return true; | |
1884 | } | |
1885 | TransContext *txc = &q.back(); | |
1886 | if (txc->state >= TransContext::STATE_KV_DONE) { | |
1887 | return true; | |
1888 | } | |
1889 | txc->oncommits.push_back(c); | |
1890 | return false; | |
1891 | } | |
9f95a23c TL |
1892 | private: |
1893 | FRIEND_MAKE_REF(OpSequencer); | |
1894 | OpSequencer(BlueStore *store, uint32_t sequencer_id, const coll_t& c) | |
1895 | : RefCountedObject(store->cct), | |
1896 | store(store), cid(c), sequencer_id(sequencer_id) { | |
1897 | } | |
1898 | ~OpSequencer() { | |
1899 | ceph_assert(q.empty()); | |
1900 | } | |
7c673cae FG |
1901 | }; |
1902 | ||
1903 | typedef boost::intrusive::list< | |
1904 | OpSequencer, | |
1905 | boost::intrusive::member_hook< | |
1906 | OpSequencer, | |
1907 | boost::intrusive::list_member_hook<>, | |
1908 | &OpSequencer::deferred_osr_queue_item> > deferred_osr_queue_t; | |
1909 | ||
1910 | struct KVSyncThread : public Thread { | |
1911 | BlueStore *store; | |
1912 | explicit KVSyncThread(BlueStore *s) : store(s) {} | |
1913 | void *entry() override { | |
1914 | store->_kv_sync_thread(); | |
1915 | return NULL; | |
1916 | } | |
1917 | }; | |
31f18b77 FG |
1918 | struct KVFinalizeThread : public Thread { |
1919 | BlueStore *store; | |
1920 | explicit KVFinalizeThread(BlueStore *s) : store(s) {} | |
1921 | void *entry() { | |
1922 | store->_kv_finalize_thread(); | |
1923 | return NULL; | |
1924 | } | |
1925 | }; | |
7c673cae FG |
1926 | |
1927 | struct DBHistogram { | |
1928 | struct value_dist { | |
1929 | uint64_t count; | |
1930 | uint32_t max_len; | |
1931 | }; | |
1932 | ||
1933 | struct key_dist { | |
1934 | uint64_t count; | |
1935 | uint32_t max_len; | |
1936 | map<int, struct value_dist> val_map; ///< slab id to count, max length of value and key | |
1937 | }; | |
1938 | ||
1939 | map<string, map<int, struct key_dist> > key_hist; | |
1940 | map<int, uint64_t> value_hist; | |
1941 | int get_key_slab(size_t sz); | |
1942 | string get_key_slab_to_range(int slab); | |
1943 | int get_value_slab(size_t sz); | |
1944 | string get_value_slab_to_range(int slab); | |
1945 | void update_hist_entry(map<string, map<int, struct key_dist> > &key_hist, | |
1946 | const string &prefix, size_t key_size, size_t value_size); | |
1947 | void dump(Formatter *f); | |
1948 | }; | |
1949 | ||
1950 | // -------------------------------------------------------- | |
1951 | // members | |
1952 | private: | |
1953 | BlueFS *bluefs = nullptr; | |
9f95a23c | 1954 | bluefs_layout_t bluefs_layout; |
11fdf7f2 TL |
1955 | mono_time bluefs_last_balance; |
1956 | utime_t next_dump_on_bluefs_alloc_failure; | |
7c673cae FG |
1957 | |
1958 | KeyValueDB *db = nullptr; | |
1959 | BlockDevice *bdev = nullptr; | |
1960 | std::string freelist_type; | |
1961 | FreelistManager *fm = nullptr; | |
1962 | Allocator *alloc = nullptr; | |
1963 | uuid_d fsid; | |
1964 | int path_fd = -1; ///< open handle to $path | |
1965 | int fsid_fd = -1; ///< open handle (locked) to $path/fsid | |
1966 | bool mounted = false; | |
1967 | ||
9f95a23c | 1968 | ceph::shared_mutex coll_lock = ceph::make_shared_mutex("BlueStore::coll_lock"); ///< rwlock to protect coll_map |
31f18b77 | 1969 | mempool::bluestore_cache_other::unordered_map<coll_t, CollectionRef> coll_map; |
eafe8130 | 1970 | bool collections_had_errors = false; |
11fdf7f2 | 1971 | map<coll_t,CollectionRef> new_coll_map; |
7c673cae | 1972 | |
9f95a23c TL |
1973 | vector<OnodeCacheShard*> onode_cache_shards; |
1974 | vector<BufferCacheShard*> buffer_cache_shards; | |
7c673cae | 1975 | |
11fdf7f2 TL |
1976 | /// protect zombie_osr_set |
1977 | ceph::mutex zombie_osr_lock = ceph::make_mutex("BlueStore::zombie_osr_lock"); | |
9f95a23c | 1978 | uint32_t next_sequencer_id = 0; |
11fdf7f2 | 1979 | std::map<coll_t,OpSequencerRef> zombie_osr_set; ///< set of OpSequencers for deleted collections |
7c673cae FG |
1980 | |
1981 | std::atomic<uint64_t> nid_last = {0}; | |
1982 | std::atomic<uint64_t> nid_max = {0}; | |
1983 | std::atomic<uint64_t> blobid_last = {0}; | |
1984 | std::atomic<uint64_t> blobid_max = {0}; | |
1985 | ||
7c673cae FG |
1986 | interval_set<uint64_t> bluefs_extents; ///< block extents owned by bluefs |
1987 | interval_set<uint64_t> bluefs_extents_reclaiming; ///< currently reclaiming | |
1988 | ||
11fdf7f2 | 1989 | ceph::mutex deferred_lock = ceph::make_mutex("BlueStore::deferred_lock"); |
7c673cae FG |
1990 | std::atomic<uint64_t> deferred_seq = {0}; |
1991 | deferred_osr_queue_t deferred_queue; ///< osr's with deferred io pending | |
1992 | int deferred_queue_size = 0; ///< num txc's queued across all osrs | |
1993 | atomic_int deferred_aggressive = {0}; ///< aggressive wakeup of kv thread | |
9f95a23c TL |
1994 | Finisher finisher; |
1995 | utime_t deferred_last_submitted = utime_t(); | |
7c673cae FG |
1996 | |
1997 | KVSyncThread kv_sync_thread; | |
11fdf7f2 TL |
1998 | ceph::mutex kv_lock = ceph::make_mutex("BlueStore::kv_lock"); |
1999 | ceph::condition_variable kv_cond; | |
3efd9988 | 2000 | bool _kv_only = false; |
31f18b77 | 2001 | bool kv_sync_started = false; |
7c673cae | 2002 | bool kv_stop = false; |
31f18b77 FG |
2003 | bool kv_finalize_started = false; |
2004 | bool kv_finalize_stop = false; | |
7c673cae FG |
2005 | deque<TransContext*> kv_queue; ///< ready, already submitted |
2006 | deque<TransContext*> kv_queue_unsubmitted; ///< ready, need submit by kv thread | |
2007 | deque<TransContext*> kv_committing; ///< currently syncing | |
2008 | deque<DeferredBatch*> deferred_done_queue; ///< deferred ios done | |
9f95a23c | 2009 | bool kv_sync_in_progress = false; |
7c673cae | 2010 | |
31f18b77 | 2011 | KVFinalizeThread kv_finalize_thread; |
11fdf7f2 TL |
2012 | ceph::mutex kv_finalize_lock = ceph::make_mutex("BlueStore::kv_finalize_lock"); |
2013 | ceph::condition_variable kv_finalize_cond; | |
31f18b77 FG |
2014 | deque<TransContext*> kv_committing_to_finalize; ///< pending finalization |
2015 | deque<DeferredBatch*> deferred_stable_to_finalize; ///< pending finalization | |
9f95a23c | 2016 | bool kv_finalize_in_progress = false; |
31f18b77 | 2017 | |
7c673cae FG |
2018 | PerfCounters *logger = nullptr; |
2019 | ||
7c673cae FG |
2020 | list<CollectionRef> removed_collections; |
2021 | ||
9f95a23c TL |
2022 | ceph::shared_mutex debug_read_error_lock = |
2023 | ceph::make_shared_mutex("BlueStore::debug_read_error_lock"); | |
7c673cae FG |
2024 | set<ghobject_t> debug_data_error_objects; |
2025 | set<ghobject_t> debug_mdata_error_objects; | |
2026 | ||
2027 | std::atomic<int> csum_type = {Checksummer::CSUM_CRC32C}; | |
2028 | ||
2029 | uint64_t block_size = 0; ///< block size of block device (power of 2) | |
2030 | uint64_t block_mask = 0; ///< mask to get just the block offset | |
2031 | size_t block_size_order = 0; ///< bits to shift to get block size | |
2032 | ||
9f95a23c | 2033 | uint64_t min_alloc_size; ///< minimum allocation unit (power of 2) |
7c673cae | 2034 | ///< bits for min_alloc_size |
224ce89b | 2035 | uint8_t min_alloc_size_order = 0; |
7c673cae FG |
2036 | static_assert(std::numeric_limits<uint8_t>::max() > |
2037 | std::numeric_limits<decltype(min_alloc_size)>::digits, | |
2038 | "not enough bits for min_alloc_size"); | |
2039 | ||
9f95a23c TL |
2040 | bool per_pool_omap = false; |
2041 | ||
7c673cae FG |
2042 | ///< maximum allocation unit (power of 2) |
2043 | std::atomic<uint64_t> max_alloc_size = {0}; | |
2044 | ||
224ce89b WB |
2045 | ///< number threshold for forced deferred writes |
2046 | std::atomic<int> deferred_batch_ops = {0}; | |
2047 | ||
2048 | ///< size threshold for forced deferred writes | |
2049 | std::atomic<uint64_t> prefer_deferred_size = {0}; | |
2050 | ||
7c673cae FG |
2051 | ///< approx cost per io, in bytes |
2052 | std::atomic<uint64_t> throttle_cost_per_io = {0}; | |
2053 | ||
224ce89b WB |
2054 | std::atomic<Compressor::CompressionMode> comp_mode = |
2055 | {Compressor::COMP_NONE}; ///< compression mode | |
7c673cae FG |
2056 | CompressorRef compressor; |
2057 | std::atomic<uint64_t> comp_min_blob_size = {0}; | |
2058 | std::atomic<uint64_t> comp_max_blob_size = {0}; | |
2059 | ||
2060 | std::atomic<uint64_t> max_blob_size = {0}; ///< maximum blob size | |
2061 | ||
31f18b77 FG |
2062 | uint64_t kv_ios = 0; |
2063 | uint64_t kv_throttle_costs = 0; | |
2064 | ||
7c673cae | 2065 | // cache trim control |
91327a77 AA |
2066 | uint64_t cache_size = 0; ///< total cache size |
2067 | double cache_meta_ratio = 0; ///< cache ratio dedicated to metadata | |
2068 | double cache_kv_ratio = 0; ///< cache ratio dedicated to kv (e.g., rocksdb) | |
2069 | double cache_data_ratio = 0; ///< cache ratio dedicated to object data | |
2070 | bool cache_autotune = false; ///< cache autotune setting | |
91327a77 AA |
2071 | double cache_autotune_interval = 0; ///< time to wait between cache rebalancing |
2072 | uint64_t osd_memory_target = 0; ///< OSD memory target when autotuning cache | |
2073 | uint64_t osd_memory_base = 0; ///< OSD base memory when autotuning cache | |
2074 | double osd_memory_expected_fragmentation = 0; ///< expected memory fragmentation | |
11fdf7f2 | 2075 | uint64_t osd_memory_cache_min = 0; ///< Min memory to assign when autotuning cache |
91327a77 | 2076 | double osd_memory_cache_resize_interval = 0; ///< Time to wait between cache resizing |
9f95a23c | 2077 | double max_defer_interval = 0; ///< Time to wait between last deferred submit |
92f5a8d4 | 2078 | std::atomic<uint32_t> config_changed = {0}; ///< Counter to determine if there is a configuration change. |
11fdf7f2 TL |
2079 | |
2080 | typedef map<uint64_t, volatile_statfs> osd_pools_map; | |
2081 | ||
2082 | ceph::mutex vstatfs_lock = ceph::make_mutex("BlueStore::vstatfs_lock"); | |
31f18b77 | 2083 | volatile_statfs vstatfs; |
11fdf7f2 TL |
2084 | osd_pools_map osd_pools; // protected by vstatfs_lock as well |
2085 | ||
2086 | bool per_pool_stat_collection = true; | |
7c673cae FG |
2087 | |
2088 | struct MempoolThread : public Thread { | |
91327a77 | 2089 | public: |
7c673cae | 2090 | BlueStore *store; |
91327a77 | 2091 | |
11fdf7f2 TL |
2092 | ceph::condition_variable cond; |
2093 | ceph::mutex lock = ceph::make_mutex("BlueStore::MempoolThread::lock"); | |
7c673cae | 2094 | bool stop = false; |
11fdf7f2 | 2095 | std::shared_ptr<PriorityCache::PriCache> binned_kv_cache = nullptr; |
eafe8130 | 2096 | std::shared_ptr<PriorityCache::Manager> pcm = nullptr; |
91327a77 AA |
2097 | |
2098 | struct MempoolCache : public PriorityCache::PriCache { | |
2099 | BlueStore *store; | |
11fdf7f2 TL |
2100 | int64_t cache_bytes[PriorityCache::Priority::LAST+1] = {0}; |
2101 | int64_t committed_bytes = 0; | |
91327a77 AA |
2102 | double cache_ratio = 0; |
2103 | ||
2104 | MempoolCache(BlueStore *s) : store(s) {}; | |
2105 | ||
2106 | virtual uint64_t _get_used_bytes() const = 0; | |
2107 | ||
2108 | virtual int64_t request_cache_bytes( | |
11fdf7f2 | 2109 | PriorityCache::Priority pri, uint64_t total_cache) const { |
91327a77 AA |
2110 | int64_t assigned = get_cache_bytes(pri); |
2111 | ||
2112 | switch (pri) { | |
eafe8130 TL |
2113 | // All cache items are currently shoved into the PRI1 priority |
2114 | case PriorityCache::Priority::PRI1: | |
91327a77 | 2115 | { |
11fdf7f2 | 2116 | int64_t request = _get_used_bytes(); |
91327a77 AA |
2117 | return(request > assigned) ? request - assigned : 0; |
2118 | } | |
2119 | default: | |
2120 | break; | |
2121 | } | |
2122 | return -EOPNOTSUPP; | |
2123 | } | |
2124 | ||
2125 | virtual int64_t get_cache_bytes(PriorityCache::Priority pri) const { | |
2126 | return cache_bytes[pri]; | |
2127 | } | |
2128 | virtual int64_t get_cache_bytes() const { | |
2129 | int64_t total = 0; | |
2130 | ||
2131 | for (int i = 0; i < PriorityCache::Priority::LAST + 1; i++) { | |
2132 | PriorityCache::Priority pri = static_cast<PriorityCache::Priority>(i); | |
2133 | total += get_cache_bytes(pri); | |
2134 | } | |
2135 | return total; | |
2136 | } | |
2137 | virtual void set_cache_bytes(PriorityCache::Priority pri, int64_t bytes) { | |
2138 | cache_bytes[pri] = bytes; | |
2139 | } | |
2140 | virtual void add_cache_bytes(PriorityCache::Priority pri, int64_t bytes) { | |
2141 | cache_bytes[pri] += bytes; | |
2142 | } | |
11fdf7f2 TL |
2143 | virtual int64_t commit_cache_size(uint64_t total_cache) { |
2144 | committed_bytes = PriorityCache::get_chunk( | |
2145 | get_cache_bytes(), total_cache); | |
2146 | return committed_bytes; | |
2147 | } | |
2148 | virtual int64_t get_committed_size() const { | |
2149 | return committed_bytes; | |
91327a77 AA |
2150 | } |
2151 | virtual double get_cache_ratio() const { | |
2152 | return cache_ratio; | |
2153 | } | |
2154 | virtual void set_cache_ratio(double ratio) { | |
2155 | cache_ratio = ratio; | |
2156 | } | |
2157 | virtual string get_cache_name() const = 0; | |
2158 | }; | |
2159 | ||
2160 | struct MetaCache : public MempoolCache { | |
2161 | MetaCache(BlueStore *s) : MempoolCache(s) {}; | |
2162 | ||
2163 | virtual uint64_t _get_used_bytes() const { | |
f91f0fd5 TL |
2164 | return mempool::bluestore_Buffer::allocated_bytes() + |
2165 | mempool::bluestore_Blob::allocated_bytes() + | |
2166 | mempool::bluestore_Extent::allocated_bytes() + | |
2167 | mempool::bluestore_cache_meta::allocated_bytes() + | |
2168 | mempool::bluestore_cache_other::allocated_bytes() + | |
2169 | mempool::bluestore_cache_onode::allocated_bytes() + | |
2170 | mempool::bluestore_SharedBlob::allocated_bytes() + | |
2171 | mempool::bluestore_inline_bl::allocated_bytes(); | |
91327a77 AA |
2172 | } |
2173 | ||
2174 | virtual string get_cache_name() const { | |
2175 | return "BlueStore Meta Cache"; | |
2176 | } | |
2177 | ||
2178 | uint64_t _get_num_onodes() const { | |
2179 | uint64_t onode_num = | |
2180 | mempool::bluestore_cache_onode::allocated_items(); | |
2181 | return (2 > onode_num) ? 2 : onode_num; | |
2182 | } | |
2183 | ||
2184 | double get_bytes_per_onode() const { | |
2185 | return (double)_get_used_bytes() / (double)_get_num_onodes(); | |
2186 | } | |
11fdf7f2 TL |
2187 | }; |
2188 | std::shared_ptr<MetaCache> meta_cache; | |
91327a77 AA |
2189 | |
2190 | struct DataCache : public MempoolCache { | |
2191 | DataCache(BlueStore *s) : MempoolCache(s) {}; | |
2192 | ||
2193 | virtual uint64_t _get_used_bytes() const { | |
2194 | uint64_t bytes = 0; | |
9f95a23c TL |
2195 | for (auto i : store->buffer_cache_shards) { |
2196 | bytes += i->_get_bytes(); | |
91327a77 AA |
2197 | } |
2198 | return bytes; | |
2199 | } | |
2200 | virtual string get_cache_name() const { | |
2201 | return "BlueStore Data Cache"; | |
2202 | } | |
11fdf7f2 TL |
2203 | }; |
2204 | std::shared_ptr<DataCache> data_cache; | |
91327a77 | 2205 | |
7c673cae FG |
2206 | public: |
2207 | explicit MempoolThread(BlueStore *s) | |
2208 | : store(s), | |
11fdf7f2 TL |
2209 | meta_cache(new MetaCache(s)), |
2210 | data_cache(new DataCache(s)) {} | |
91327a77 | 2211 | |
7c673cae FG |
2212 | void *entry() override; |
2213 | void init() { | |
11fdf7f2 | 2214 | ceph_assert(stop == false); |
7c673cae FG |
2215 | create("bstore_mempool"); |
2216 | } | |
2217 | void shutdown() { | |
11fdf7f2 | 2218 | lock.lock(); |
7c673cae | 2219 | stop = true; |
11fdf7f2 TL |
2220 | cond.notify_all(); |
2221 | lock.unlock(); | |
7c673cae FG |
2222 | join(); |
2223 | } | |
91327a77 AA |
2224 | |
2225 | private: | |
2226 | void _adjust_cache_settings(); | |
92f5a8d4 | 2227 | void _update_cache_settings(); |
9f95a23c | 2228 | void _resize_shards(bool interval_stats); |
7c673cae FG |
2229 | } mempool_thread; |
2230 | ||
2231 | // -------------------------------------------------------- | |
2232 | // private methods | |
2233 | ||
2234 | void _init_logger(); | |
2235 | void _shutdown_logger(); | |
2236 | int _reload_logger(); | |
2237 | ||
2238 | int _open_path(); | |
2239 | void _close_path(); | |
2240 | int _open_fsid(bool create); | |
2241 | int _lock_fsid(); | |
2242 | int _read_fsid(uuid_d *f); | |
2243 | int _write_fsid(); | |
2244 | void _close_fsid(); | |
2245 | void _set_alloc_sizes(); | |
2246 | void _set_blob_size(); | |
1adf2230 | 2247 | void _set_finisher_num(); |
9f95a23c | 2248 | void _set_per_pool_omap(); |
92f5a8d4 | 2249 | void _update_osd_memory_options(); |
7c673cae FG |
2250 | |
2251 | int _open_bdev(bool create); | |
11fdf7f2 TL |
2252 | // Verifies if disk space is enough for reserved + min bluefs |
2253 | // and alters the latter if needed. | |
2254 | // Depends on min_alloc_size hence should be called after | |
2255 | // its initialization (and outside of _open_bdev) | |
2256 | void _validate_bdev(); | |
7c673cae | 2257 | void _close_bdev(); |
11fdf7f2 TL |
2258 | |
2259 | int _minimal_open_bluefs(bool create); | |
2260 | void _minimal_close_bluefs(); | |
2261 | int _open_bluefs(bool create); | |
1911f103 | 2262 | void _close_bluefs(bool cold_close); |
11fdf7f2 TL |
2263 | |
2264 | // Limited (u)mount intended for BlueFS operations only | |
2265 | int _mount_for_bluefs(); | |
2266 | void _umount_for_bluefs(); | |
2267 | ||
2268 | ||
2269 | int _is_bluefs(bool create, bool* ret); | |
2270 | /* | |
2271 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
2272 | * in the proper order | |
2273 | */ | |
2274 | int _open_db_and_around(bool read_only); | |
1911f103 | 2275 | void _close_db_and_around(bool read_only); |
11fdf7f2 TL |
2276 | |
2277 | // updates legacy bluefs related recs in DB to a state valid for | |
2278 | // downgrades from nautilus. | |
2279 | void _sync_bluefs_and_fm(); | |
2280 | ||
2281 | /* | |
2282 | * @warning to_repair_db means that we open this db to repair it, will not | |
2283 | * hold the rocksdb's file lock. | |
2284 | */ | |
2285 | int _open_db(bool create, | |
2286 | bool to_repair_db=false, | |
2287 | bool read_only = false); | |
1911f103 TL |
2288 | void _close_db(bool read_only); |
2289 | int _open_fm(KeyValueDB::Transaction t, bool read_only); | |
7c673cae | 2290 | void _close_fm(); |
1911f103 TL |
2291 | int _write_out_fm_meta(uint64_t target_size, |
2292 | bool update_root_size = false, | |
2293 | bluestore_bdev_label_t* res_label = nullptr); | |
7c673cae FG |
2294 | int _open_alloc(); |
2295 | void _close_alloc(); | |
eafe8130 TL |
2296 | int _open_collections(); |
2297 | void _fsck_collections(int64_t* errors); | |
7c673cae FG |
2298 | void _close_collections(); |
2299 | ||
2300 | int _setup_block_symlink_or_file(string name, string path, uint64_t size, | |
2301 | bool create); | |
2302 | ||
7c673cae | 2303 | public: |
9f95a23c TL |
2304 | utime_t get_deferred_last_submitted() { |
2305 | std::lock_guard l(deferred_lock); | |
2306 | return deferred_last_submitted; | |
2307 | } | |
2308 | ||
3efd9988 FG |
2309 | static int _write_bdev_label(CephContext* cct, |
2310 | string path, bluestore_bdev_label_t label); | |
7c673cae FG |
2311 | static int _read_bdev_label(CephContext* cct, string path, |
2312 | bluestore_bdev_label_t *label); | |
2313 | private: | |
2314 | int _check_or_set_bdev_label(string path, uint64_t size, string desc, | |
2315 | bool create); | |
2316 | ||
2317 | int _open_super_meta(); | |
2318 | ||
224ce89b | 2319 | void _open_statfs(); |
11fdf7f2 | 2320 | void _get_statfs_overall(struct store_statfs_t *buf); |
31f18b77 | 2321 | |
11fdf7f2 TL |
2322 | void _dump_alloc_on_failure(); |
2323 | ||
2324 | int64_t _get_bluefs_size_delta(uint64_t bluefs_free, uint64_t bluefs_total); | |
2325 | int _balance_bluefs_freespace(); | |
7c673cae FG |
2326 | |
2327 | CollectionRef _get_collection(const coll_t& cid); | |
2328 | void _queue_reap_collection(CollectionRef& c); | |
2329 | void _reap_collections(); | |
2330 | void _update_cache_logger(); | |
2331 | ||
2332 | void _assign_nid(TransContext *txc, OnodeRef o); | |
2333 | uint64_t _assign_blobid(TransContext *txc); | |
2334 | ||
81eedcae TL |
2335 | template <int LogLevelV> |
2336 | friend void _dump_onode(CephContext *cct, const Onode& o); | |
2337 | template <int LogLevelV> | |
2338 | friend void _dump_extent_map(CephContext *cct, const ExtentMap& em); | |
2339 | template <int LogLevelV> | |
2340 | friend void _dump_transaction(CephContext *cct, Transaction *t); | |
7c673cae | 2341 | |
11fdf7f2 TL |
2342 | TransContext *_txc_create(Collection *c, OpSequencer *osr, |
2343 | list<Context*> *on_commits); | |
7c673cae FG |
2344 | void _txc_update_store_statfs(TransContext *txc); |
2345 | void _txc_add_transaction(TransContext *txc, Transaction *t); | |
2346 | void _txc_calc_cost(TransContext *txc); | |
2347 | void _txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t); | |
2348 | void _txc_state_proc(TransContext *txc); | |
2349 | void _txc_aio_submit(TransContext *txc); | |
2350 | public: | |
2351 | void txc_aio_finish(void *p) { | |
2352 | _txc_state_proc(static_cast<TransContext*>(p)); | |
2353 | } | |
2354 | private: | |
2355 | void _txc_finish_io(TransContext *txc); | |
2356 | void _txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t); | |
9f95a23c | 2357 | void _txc_apply_kv(TransContext *txc, bool sync_submit_transaction); |
7c673cae FG |
2358 | void _txc_committed_kv(TransContext *txc); |
2359 | void _txc_finish(TransContext *txc); | |
2360 | void _txc_release_alloc(TransContext *txc); | |
2361 | ||
11fdf7f2 TL |
2362 | void _osr_attach(Collection *c); |
2363 | void _osr_register_zombie(OpSequencer *osr); | |
2364 | void _osr_drain(OpSequencer *osr); | |
7c673cae FG |
2365 | void _osr_drain_preceding(TransContext *txc); |
2366 | void _osr_drain_all(); | |
7c673cae | 2367 | |
31f18b77 FG |
2368 | void _kv_start(); |
2369 | void _kv_stop(); | |
7c673cae | 2370 | void _kv_sync_thread(); |
31f18b77 | 2371 | void _kv_finalize_thread(); |
7c673cae | 2372 | |
9f95a23c | 2373 | bluestore_deferred_op_t *_get_deferred_op(TransContext *txc); |
7c673cae | 2374 | void _deferred_queue(TransContext *txc); |
3efd9988 | 2375 | public: |
224ce89b | 2376 | void deferred_try_submit(); |
3efd9988 | 2377 | private: |
224ce89b | 2378 | void _deferred_submit_unlock(OpSequencer *osr); |
7c673cae FG |
2379 | void _deferred_aio_finish(OpSequencer *osr); |
2380 | int _deferred_replay(); | |
2381 | ||
2382 | public: | |
2383 | using mempool_dynamic_bitset = | |
2384 | boost::dynamic_bitset<uint64_t, | |
2385 | mempool::bluestore_fsck::pool_allocator<uint64_t>>; | |
eafe8130 TL |
2386 | using per_pool_statfs = |
2387 | mempool::bluestore_fsck::map<uint64_t, store_statfs_t>; | |
2388 | ||
2389 | enum FSCKDepth { | |
2390 | FSCK_REGULAR, | |
2391 | FSCK_DEEP, | |
2392 | FSCK_SHALLOW | |
2393 | }; | |
9f95a23c TL |
2394 | enum { |
2395 | MAX_FSCK_ERROR_LINES = 100, | |
2396 | }; | |
7c673cae FG |
2397 | |
2398 | private: | |
2399 | int _fsck_check_extents( | |
11fdf7f2 | 2400 | const coll_t& cid, |
7c673cae FG |
2401 | const ghobject_t& oid, |
2402 | const PExtentVector& extents, | |
2403 | bool compressed, | |
2404 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 2405 | uint64_t granularity, |
11fdf7f2 | 2406 | BlueStoreRepairer* repairer, |
eafe8130 TL |
2407 | store_statfs_t& expected_statfs, |
2408 | FSCKDepth depth); | |
7c673cae | 2409 | |
11fdf7f2 TL |
2410 | void _fsck_check_pool_statfs( |
2411 | per_pool_statfs& expected_pool_statfs, | |
eafe8130 TL |
2412 | int64_t& errors, |
2413 | int64_t &warnings, | |
11fdf7f2 TL |
2414 | BlueStoreRepairer* repairer); |
2415 | ||
eafe8130 TL |
2416 | int _fsck(FSCKDepth depth, bool repair); |
2417 | int _fsck_on_open(BlueStore::FSCKDepth depth, bool repair); | |
2418 | ||
7c673cae FG |
2419 | void _buffer_cache_write( |
2420 | TransContext *txc, | |
2421 | BlobRef b, | |
2422 | uint64_t offset, | |
2423 | bufferlist& bl, | |
2424 | unsigned flags) { | |
2425 | b->shared_blob->bc.write(b->shared_blob->get_cache(), txc->seq, offset, bl, | |
2426 | flags); | |
2427 | txc->shared_blobs_written.insert(b->shared_blob); | |
2428 | } | |
2429 | ||
2430 | int _collection_list( | |
2431 | Collection *c, const ghobject_t& start, const ghobject_t& end, | |
f91f0fd5 | 2432 | int max, bool legacy, vector<ghobject_t> *ls, ghobject_t *next); |
7c673cae FG |
2433 | |
2434 | template <typename T, typename F> | |
2435 | T select_option(const std::string& opt_name, T val1, F f) { | |
2436 | //NB: opt_name reserved for future use | |
2437 | boost::optional<T> val2 = f(); | |
2438 | if (val2) { | |
2439 | return *val2; | |
2440 | } | |
2441 | return val1; | |
2442 | } | |
2443 | ||
2444 | void _apply_padding(uint64_t head_pad, | |
2445 | uint64_t tail_pad, | |
7c673cae FG |
2446 | bufferlist& padded); |
2447 | ||
11fdf7f2 TL |
2448 | void _record_onode(OnodeRef &o, KeyValueDB::Transaction &txn); |
2449 | ||
7c673cae FG |
2450 | // -- ondisk version --- |
2451 | public: | |
1911f103 | 2452 | const int32_t latest_ondisk_format = 4; ///< our version |
7c673cae | 2453 | const int32_t min_readable_ondisk_format = 1; ///< what we can read |
9f95a23c | 2454 | const int32_t min_compat_ondisk_format = 3; ///< who can read us |
7c673cae FG |
2455 | |
2456 | private: | |
2457 | int32_t ondisk_format = 0; ///< value detected on mount | |
2458 | ||
2459 | int _upgrade_super(); ///< upgrade (called during open_super) | |
11fdf7f2 | 2460 | uint64_t _get_ondisk_reserved() const; |
7c673cae FG |
2461 | void _prepare_ondisk_format_super(KeyValueDB::Transaction& t); |
2462 | ||
2463 | // --- public interface --- | |
2464 | public: | |
2465 | BlueStore(CephContext *cct, const string& path); | |
2466 | BlueStore(CephContext *cct, const string& path, uint64_t min_alloc_size); // Ctor for UT only | |
2467 | ~BlueStore() override; | |
2468 | ||
2469 | string get_type() override { | |
2470 | return "bluestore"; | |
2471 | } | |
2472 | ||
2473 | bool needs_journal() override { return false; }; | |
2474 | bool wants_journal() override { return false; }; | |
2475 | bool allows_journal() override { return false; }; | |
2476 | ||
9f95a23c TL |
2477 | uint64_t get_min_alloc_size() const override { |
2478 | return min_alloc_size; | |
2479 | } | |
2480 | ||
11fdf7f2 TL |
2481 | int get_devices(set<string> *ls) override; |
2482 | ||
31f18b77 | 2483 | bool is_rotational() override; |
d2e6a577 | 2484 | bool is_journal_rotational() override; |
31f18b77 | 2485 | |
224ce89b WB |
2486 | string get_default_device_class() override { |
2487 | string device_class; | |
2488 | map<string, string> metadata; | |
2489 | collect_metadata(&metadata); | |
2490 | auto it = metadata.find("bluestore_bdev_type"); | |
2491 | if (it != metadata.end()) { | |
2492 | device_class = it->second; | |
2493 | } | |
2494 | return device_class; | |
2495 | } | |
2496 | ||
11fdf7f2 TL |
2497 | int get_numa_node( |
2498 | int *numa_node, | |
2499 | set<int> *nodes, | |
2500 | set<string> *failed) override; | |
2501 | ||
7c673cae FG |
2502 | static int get_block_device_fsid(CephContext* cct, const string& path, |
2503 | uuid_d *fsid); | |
2504 | ||
2505 | bool test_mount_in_use() override; | |
2506 | ||
2507 | private: | |
11fdf7f2 | 2508 | int _mount(bool kv_only, bool open_db=true); |
7c673cae FG |
2509 | public: |
2510 | int mount() override { | |
2511 | return _mount(false); | |
2512 | } | |
2513 | int umount() override; | |
2514 | ||
11fdf7f2 TL |
2515 | int start_kv_only(KeyValueDB **pdb, bool open_db=true) { |
2516 | int r = _mount(true, open_db); | |
7c673cae FG |
2517 | if (r < 0) |
2518 | return r; | |
2519 | *pdb = db; | |
2520 | return 0; | |
2521 | } | |
2522 | ||
3efd9988 FG |
2523 | int write_meta(const std::string& key, const std::string& value) override; |
2524 | int read_meta(const std::string& key, std::string *value) override; | |
2525 | ||
eafe8130 TL |
2526 | int cold_open(); |
2527 | int cold_close(); | |
3efd9988 FG |
2528 | |
2529 | int fsck(bool deep) override { | |
eafe8130 | 2530 | return _fsck(deep ? FSCK_DEEP : FSCK_REGULAR, false); |
3efd9988 FG |
2531 | } |
2532 | int repair(bool deep) override { | |
eafe8130 TL |
2533 | return _fsck(deep ? FSCK_DEEP : FSCK_REGULAR, true); |
2534 | } | |
2535 | int quick_fix() override { | |
2536 | return _fsck(FSCK_SHALLOW, true); | |
3efd9988 | 2537 | } |
7c673cae FG |
2538 | |
2539 | void set_cache_shards(unsigned num) override; | |
11fdf7f2 TL |
2540 | void dump_cache_stats(Formatter *f) override { |
2541 | int onode_count = 0, buffers_bytes = 0; | |
9f95a23c TL |
2542 | for (auto i: onode_cache_shards) { |
2543 | onode_count += i->_get_num(); | |
2544 | } | |
2545 | for (auto i: buffer_cache_shards) { | |
2546 | buffers_bytes += i->_get_bytes(); | |
11fdf7f2 TL |
2547 | } |
2548 | f->dump_int("bluestore_onode", onode_count); | |
2549 | f->dump_int("bluestore_buffers", buffers_bytes); | |
2550 | } | |
2551 | void dump_cache_stats(ostream& ss) override { | |
2552 | int onode_count = 0, buffers_bytes = 0; | |
9f95a23c TL |
2553 | for (auto i: onode_cache_shards) { |
2554 | onode_count += i->_get_num(); | |
2555 | } | |
2556 | for (auto i: buffer_cache_shards) { | |
2557 | buffers_bytes += i->_get_bytes(); | |
11fdf7f2 TL |
2558 | } |
2559 | ss << "bluestore_onode: " << onode_count; | |
2560 | ss << "bluestore_buffers: " << buffers_bytes; | |
2561 | } | |
7c673cae FG |
2562 | |
2563 | int validate_hobject_key(const hobject_t &obj) const override { | |
2564 | return 0; | |
2565 | } | |
2566 | unsigned get_max_attr_name_length() override { | |
2567 | return 256; // arbitrary; there is no real limit internally | |
2568 | } | |
2569 | ||
2570 | int mkfs() override; | |
2571 | int mkjournal() override { | |
2572 | return 0; | |
2573 | } | |
2574 | ||
2575 | void get_db_statistics(Formatter *f) override; | |
2576 | void generate_db_histogram(Formatter *f) override; | |
f6b5b4d7 | 2577 | void _shutdown_cache(); |
11fdf7f2 | 2578 | int flush_cache(ostream *os = NULL) override; |
7c673cae FG |
2579 | void dump_perf_counters(Formatter *f) override { |
2580 | f->open_object_section("perf_counters"); | |
2581 | logger->dump_formatted(f, false); | |
2582 | f->close_section(); | |
2583 | } | |
2584 | ||
11fdf7f2 TL |
2585 | int add_new_bluefs_device(int id, const string& path); |
2586 | int migrate_to_existing_bluefs_device(const set<int>& devs_source, | |
2587 | int id); | |
2588 | int migrate_to_new_bluefs_device(const set<int>& devs_source, | |
2589 | int id, | |
2590 | const string& path); | |
2591 | int expand_devices(ostream& out); | |
2592 | string get_device_path(unsigned id); | |
7c673cae | 2593 | |
1911f103 TL |
2594 | int dump_bluefs_sizes(ostream& out); |
2595 | ||
7c673cae | 2596 | public: |
11fdf7f2 TL |
2597 | int statfs(struct store_statfs_t *buf, |
2598 | osd_alert_list_t* alerts = nullptr) override; | |
9f95a23c TL |
2599 | int pool_statfs(uint64_t pool_id, struct store_statfs_t *buf, |
2600 | bool *per_pool_omap) override; | |
7c673cae FG |
2601 | |
2602 | void collect_metadata(map<string,string> *pm) override; | |
2603 | ||
7c673cae FG |
2604 | bool exists(CollectionHandle &c, const ghobject_t& oid) override; |
2605 | int set_collection_opts( | |
11fdf7f2 | 2606 | CollectionHandle& c, |
7c673cae | 2607 | const pool_opts_t& opts) override; |
7c673cae FG |
2608 | int stat( |
2609 | CollectionHandle &c, | |
2610 | const ghobject_t& oid, | |
2611 | struct stat *st, | |
2612 | bool allow_eio = false) override; | |
7c673cae FG |
2613 | int read( |
2614 | CollectionHandle &c, | |
2615 | const ghobject_t& oid, | |
2616 | uint64_t offset, | |
2617 | size_t len, | |
2618 | bufferlist& bl, | |
224ce89b | 2619 | uint32_t op_flags = 0) override; |
9f95a23c TL |
2620 | |
2621 | private: | |
2622 | ||
2623 | // -------------------------------------------------------- | |
2624 | // intermediate data structures used while reading | |
2625 | struct region_t { | |
2626 | uint64_t logical_offset; | |
2627 | uint64_t blob_xoffset; //region offset within the blob | |
2628 | uint64_t length; | |
2629 | ||
2630 | // used later in read process | |
2631 | uint64_t front = 0; | |
2632 | ||
2633 | region_t(uint64_t offset, uint64_t b_offs, uint64_t len, uint64_t front = 0) | |
2634 | : logical_offset(offset), | |
2635 | blob_xoffset(b_offs), | |
2636 | length(len), | |
2637 | front(front){} | |
2638 | region_t(const region_t& from) | |
2639 | : logical_offset(from.logical_offset), | |
2640 | blob_xoffset(from.blob_xoffset), | |
2641 | length(from.length), | |
2642 | front(from.front){} | |
2643 | ||
2644 | friend ostream& operator<<(ostream& out, const region_t& r) { | |
2645 | return out << "0x" << std::hex << r.logical_offset << ":" | |
2646 | << r.blob_xoffset << "~" << r.length << std::dec; | |
2647 | } | |
2648 | }; | |
2649 | ||
2650 | // merged blob read request | |
2651 | struct read_req_t { | |
2652 | uint64_t r_off = 0; | |
2653 | uint64_t r_len = 0; | |
2654 | bufferlist bl; | |
2655 | std::list<region_t> regs; // original read regions | |
2656 | ||
2657 | read_req_t(uint64_t off, uint64_t len) : r_off(off), r_len(len) {} | |
2658 | ||
2659 | friend ostream& operator<<(ostream& out, const read_req_t& r) { | |
2660 | out << "{<0x" << std::hex << r.r_off << ", 0x" << r.r_len << "> : ["; | |
2661 | for (const auto& reg : r.regs) | |
2662 | out << reg; | |
2663 | return out << "]}" << std::dec; | |
2664 | } | |
2665 | }; | |
2666 | ||
2667 | typedef list<read_req_t> regions2read_t; | |
2668 | typedef map<BlueStore::BlobRef, regions2read_t> blobs2read_t; | |
2669 | ||
2670 | void _read_cache( | |
2671 | OnodeRef o, | |
2672 | uint64_t offset, | |
2673 | size_t length, | |
2674 | int read_cache_policy, | |
2675 | ready_regions_t& ready_regions, | |
2676 | blobs2read_t& blobs2read); | |
2677 | ||
2678 | ||
2679 | int _prepare_read_ioc( | |
2680 | blobs2read_t& blobs2read, | |
2681 | vector<bufferlist>* compressed_blob_bls, | |
2682 | IOContext* ioc); | |
2683 | ||
2684 | int _generate_read_result_bl( | |
2685 | OnodeRef o, | |
2686 | uint64_t offset, | |
2687 | size_t length, | |
2688 | ready_regions_t& ready_regions, | |
2689 | vector<bufferlist>& compressed_blob_bls, | |
2690 | blobs2read_t& blobs2read, | |
2691 | bool buffered, | |
2692 | bool* csum_error, | |
2693 | bufferlist& bl); | |
2694 | ||
7c673cae FG |
2695 | int _do_read( |
2696 | Collection *c, | |
2697 | OnodeRef o, | |
2698 | uint64_t offset, | |
2699 | size_t len, | |
2700 | bufferlist& bl, | |
f64942e4 AA |
2701 | uint32_t op_flags = 0, |
2702 | uint64_t retry_count = 0); | |
7c673cae | 2703 | |
9f95a23c TL |
2704 | int _do_readv( |
2705 | Collection *c, | |
2706 | OnodeRef o, | |
2707 | const interval_set<uint64_t>& m, | |
2708 | bufferlist& bl, | |
2709 | uint32_t op_flags = 0, | |
2710 | uint64_t retry_count = 0); | |
2711 | ||
7c673cae FG |
2712 | int _fiemap(CollectionHandle &c_, const ghobject_t& oid, |
2713 | uint64_t offset, size_t len, interval_set<uint64_t>& destset); | |
2714 | public: | |
7c673cae FG |
2715 | int fiemap(CollectionHandle &c, const ghobject_t& oid, |
2716 | uint64_t offset, size_t len, bufferlist& bl) override; | |
7c673cae FG |
2717 | int fiemap(CollectionHandle &c, const ghobject_t& oid, |
2718 | uint64_t offset, size_t len, map<uint64_t, uint64_t>& destmap) override; | |
2719 | ||
9f95a23c TL |
2720 | int readv( |
2721 | CollectionHandle &c_, | |
2722 | const ghobject_t& oid, | |
2723 | interval_set<uint64_t>& m, | |
2724 | bufferlist& bl, | |
2725 | uint32_t op_flags) override; | |
2726 | ||
2727 | int dump_onode(CollectionHandle &c, const ghobject_t& oid, | |
2728 | const string& section_name, Formatter *f) override; | |
7c673cae | 2729 | |
7c673cae FG |
2730 | int getattr(CollectionHandle &c, const ghobject_t& oid, const char *name, |
2731 | bufferptr& value) override; | |
2732 | ||
7c673cae FG |
2733 | int getattrs(CollectionHandle &c, const ghobject_t& oid, |
2734 | map<string,bufferptr>& aset) override; | |
2735 | ||
2736 | int list_collections(vector<coll_t>& ls) override; | |
2737 | ||
2738 | CollectionHandle open_collection(const coll_t &c) override; | |
11fdf7f2 TL |
2739 | CollectionHandle create_new_collection(const coll_t& cid) override; |
2740 | void set_collection_commit_queue(const coll_t& cid, | |
2741 | ContextQueue *commit_queue) override; | |
7c673cae FG |
2742 | |
2743 | bool collection_exists(const coll_t& c) override; | |
11fdf7f2 TL |
2744 | int collection_empty(CollectionHandle& c, bool *empty) override; |
2745 | int collection_bits(CollectionHandle& c) override; | |
7c673cae | 2746 | |
7c673cae FG |
2747 | int collection_list(CollectionHandle &c, |
2748 | const ghobject_t& start, | |
2749 | const ghobject_t& end, | |
2750 | int max, | |
2751 | vector<ghobject_t> *ls, ghobject_t *next) override; | |
2752 | ||
f91f0fd5 TL |
2753 | int collection_list_legacy(CollectionHandle &c, |
2754 | const ghobject_t& start, | |
2755 | const ghobject_t& end, | |
2756 | int max, | |
2757 | vector<ghobject_t> *ls, | |
2758 | ghobject_t *next) override; | |
2759 | ||
7c673cae FG |
2760 | int omap_get( |
2761 | CollectionHandle &c, ///< [in] Collection containing oid | |
2762 | const ghobject_t &oid, ///< [in] Object containing omap | |
2763 | bufferlist *header, ///< [out] omap header | |
2764 | map<string, bufferlist> *out /// < [out] Key to value map | |
2765 | ) override; | |
9f95a23c TL |
2766 | int _omap_get( |
2767 | Collection *c, ///< [in] Collection containing oid | |
2768 | const ghobject_t &oid, ///< [in] Object containing omap | |
2769 | bufferlist *header, ///< [out] omap header | |
2770 | map<string, bufferlist> *out /// < [out] Key to value map | |
2771 | ); | |
2772 | int _onode_omap_get( | |
2773 | const OnodeRef &o, ///< [in] Object containing omap | |
2774 | bufferlist *header, ///< [out] omap header | |
2775 | map<string, bufferlist> *out /// < [out] Key to value map | |
2776 | ); | |
2777 | ||
7c673cae FG |
2778 | |
2779 | /// Get omap header | |
7c673cae FG |
2780 | int omap_get_header( |
2781 | CollectionHandle &c, ///< [in] Collection containing oid | |
2782 | const ghobject_t &oid, ///< [in] Object containing omap | |
2783 | bufferlist *header, ///< [out] omap header | |
2784 | bool allow_eio = false ///< [in] don't assert on eio | |
2785 | ) override; | |
2786 | ||
2787 | /// Get keys defined on oid | |
7c673cae FG |
2788 | int omap_get_keys( |
2789 | CollectionHandle &c, ///< [in] Collection containing oid | |
2790 | const ghobject_t &oid, ///< [in] Object containing omap | |
2791 | set<string> *keys ///< [out] Keys defined on oid | |
2792 | ) override; | |
2793 | ||
2794 | /// Get key values | |
7c673cae FG |
2795 | int omap_get_values( |
2796 | CollectionHandle &c, ///< [in] Collection containing oid | |
2797 | const ghobject_t &oid, ///< [in] Object containing omap | |
2798 | const set<string> &keys, ///< [in] Keys to get | |
2799 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
2800 | ) override; | |
2801 | ||
9f95a23c TL |
2802 | #ifdef WITH_SEASTAR |
2803 | int omap_get_values( | |
2804 | CollectionHandle &c, ///< [in] Collection containing oid | |
2805 | const ghobject_t &oid, ///< [in] Object containing omap | |
2806 | const std::optional<string> &start_after, ///< [in] Keys to get | |
2807 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
2808 | ) override; | |
2809 | #endif | |
2810 | ||
7c673cae | 2811 | /// Filters keys into out which are defined on oid |
7c673cae FG |
2812 | int omap_check_keys( |
2813 | CollectionHandle &c, ///< [in] Collection containing oid | |
2814 | const ghobject_t &oid, ///< [in] Object containing omap | |
2815 | const set<string> &keys, ///< [in] Keys to check | |
2816 | set<string> *out ///< [out] Subset of keys defined on oid | |
2817 | ) override; | |
2818 | ||
7c673cae FG |
2819 | ObjectMap::ObjectMapIterator get_omap_iterator( |
2820 | CollectionHandle &c, ///< [in] collection | |
2821 | const ghobject_t &oid ///< [in] object | |
2822 | ) override; | |
2823 | ||
2824 | void set_fsid(uuid_d u) override { | |
2825 | fsid = u; | |
2826 | } | |
2827 | uuid_d get_fsid() override { | |
2828 | return fsid; | |
2829 | } | |
2830 | ||
2831 | uint64_t estimate_objects_overhead(uint64_t num_objects) override { | |
2832 | return num_objects * 300; //assuming per-object overhead is 300 bytes | |
2833 | } | |
2834 | ||
2835 | struct BSPerfTracker { | |
11fdf7f2 TL |
2836 | PerfCounters::avg_tracker<uint64_t> os_commit_latency_ns; |
2837 | PerfCounters::avg_tracker<uint64_t> os_apply_latency_ns; | |
7c673cae FG |
2838 | |
2839 | objectstore_perf_stat_t get_cur_stats() const { | |
2840 | objectstore_perf_stat_t ret; | |
11fdf7f2 TL |
2841 | ret.os_commit_latency_ns = os_commit_latency_ns.current_avg(); |
2842 | ret.os_apply_latency_ns = os_apply_latency_ns.current_avg(); | |
7c673cae FG |
2843 | return ret; |
2844 | } | |
2845 | ||
2846 | void update_from_perfcounters(PerfCounters &logger); | |
2847 | } perf_tracker; | |
2848 | ||
2849 | objectstore_perf_stat_t get_cur_stats() override { | |
2850 | perf_tracker.update_from_perfcounters(*logger); | |
2851 | return perf_tracker.get_cur_stats(); | |
2852 | } | |
2853 | const PerfCounters* get_perf_counters() const override { | |
2854 | return logger; | |
2855 | } | |
9f95a23c TL |
2856 | const PerfCounters* get_bluefs_perf_counters() const { |
2857 | return bluefs->get_perf_counters(); | |
2858 | } | |
7c673cae FG |
2859 | |
2860 | int queue_transactions( | |
11fdf7f2 | 2861 | CollectionHandle& ch, |
7c673cae FG |
2862 | vector<Transaction>& tls, |
2863 | TrackedOpRef op = TrackedOpRef(), | |
2864 | ThreadPool::TPHandle *handle = NULL) override; | |
2865 | ||
2866 | // error injection | |
2867 | void inject_data_error(const ghobject_t& o) override { | |
9f95a23c | 2868 | std::unique_lock l(debug_read_error_lock); |
7c673cae FG |
2869 | debug_data_error_objects.insert(o); |
2870 | } | |
2871 | void inject_mdata_error(const ghobject_t& o) override { | |
9f95a23c | 2872 | std::unique_lock l(debug_read_error_lock); |
7c673cae FG |
2873 | debug_mdata_error_objects.insert(o); |
2874 | } | |
11fdf7f2 TL |
2875 | |
2876 | /// methods to inject various errors fsck can repair | |
2877 | void inject_broken_shared_blob_key(const string& key, | |
2878 | const bufferlist& bl); | |
2879 | void inject_leaked(uint64_t len); | |
2880 | void inject_false_free(coll_t cid, ghobject_t oid); | |
2881 | void inject_statfs(const string& key, const store_statfs_t& new_statfs); | |
eafe8130 | 2882 | void inject_global_statfs(const store_statfs_t& new_statfs); |
11fdf7f2 TL |
2883 | void inject_misreference(coll_t cid1, ghobject_t oid1, |
2884 | coll_t cid2, ghobject_t oid2, | |
2885 | uint64_t offset); | |
9f95a23c TL |
2886 | // resets global per_pool_omap in DB |
2887 | void inject_legacy_omap(); | |
2888 | // resets per_pool_omap | pgmeta_omap for onode | |
2889 | void inject_legacy_omap(coll_t cid, ghobject_t oid); | |
11fdf7f2 | 2890 | |
224ce89b | 2891 | void compact() override { |
11fdf7f2 | 2892 | ceph_assert(db); |
224ce89b WB |
2893 | db->compact(); |
2894 | } | |
28e407b8 AA |
2895 | bool has_builtin_csum() const override { |
2896 | return true; | |
2897 | } | |
2898 | ||
11fdf7f2 TL |
2899 | /* |
2900 | Allocate space for BlueFS from slow device. | |
2901 | Either automatically applies allocated extents to underlying | |
2902 | BlueFS (extents == nullptr) or just return them (non-null extents) provided | |
2903 | */ | |
2904 | int allocate_bluefs_freespace( | |
2905 | uint64_t min_size, | |
2906 | uint64_t size, | |
2907 | PExtentVector* extents); | |
2908 | ||
494da23a TL |
2909 | inline void log_latency(const char* name, |
2910 | int idx, | |
2911 | const ceph::timespan& lat, | |
2912 | double lat_threshold, | |
2913 | const char* info = "") const; | |
2914 | ||
2915 | inline void log_latency_fn(const char* name, | |
2916 | int idx, | |
2917 | const ceph::timespan& lat, | |
2918 | double lat_threshold, | |
2919 | std::function<string (const ceph::timespan& lat)> fn) const; | |
11fdf7f2 | 2920 | |
7c673cae FG |
2921 | private: |
2922 | bool _debug_data_eio(const ghobject_t& o) { | |
2923 | if (!cct->_conf->bluestore_debug_inject_read_err) { | |
2924 | return false; | |
2925 | } | |
9f95a23c | 2926 | std::shared_lock l(debug_read_error_lock); |
7c673cae FG |
2927 | return debug_data_error_objects.count(o); |
2928 | } | |
2929 | bool _debug_mdata_eio(const ghobject_t& o) { | |
2930 | if (!cct->_conf->bluestore_debug_inject_read_err) { | |
2931 | return false; | |
2932 | } | |
9f95a23c | 2933 | std::shared_lock l(debug_read_error_lock); |
7c673cae FG |
2934 | return debug_mdata_error_objects.count(o); |
2935 | } | |
2936 | void _debug_obj_on_delete(const ghobject_t& o) { | |
2937 | if (cct->_conf->bluestore_debug_inject_read_err) { | |
9f95a23c | 2938 | std::unique_lock l(debug_read_error_lock); |
7c673cae FG |
2939 | debug_data_error_objects.erase(o); |
2940 | debug_mdata_error_objects.erase(o); | |
2941 | } | |
2942 | } | |
11fdf7f2 TL |
2943 | private: |
2944 | ceph::mutex qlock = ceph::make_mutex("BlueStore::Alerts::qlock"); | |
2945 | string failed_cmode; | |
2946 | set<string> failed_compressors; | |
2947 | string spillover_alert; | |
81eedcae | 2948 | string legacy_statfs_alert; |
9f95a23c | 2949 | string no_per_pool_omap_alert; |
81eedcae | 2950 | string disk_size_mismatch_alert; |
11fdf7f2 TL |
2951 | |
2952 | void _log_alerts(osd_alert_list_t& alerts); | |
2953 | bool _set_compression_alert(bool cmode, const char* s) { | |
2954 | std::lock_guard l(qlock); | |
2955 | if (cmode) { | |
2956 | bool ret = failed_cmode.empty(); | |
2957 | failed_cmode = s; | |
2958 | return ret; | |
2959 | } | |
2960 | return failed_compressors.emplace(s).second; | |
2961 | } | |
2962 | void _clear_compression_alert() { | |
2963 | std::lock_guard l(qlock); | |
2964 | failed_compressors.clear(); | |
2965 | failed_cmode.clear(); | |
2966 | } | |
2967 | ||
2968 | void _set_spillover_alert(const string& s) { | |
2969 | std::lock_guard l(qlock); | |
2970 | spillover_alert = s; | |
2971 | } | |
2972 | void _clear_spillover_alert() { | |
2973 | std::lock_guard l(qlock); | |
2974 | spillover_alert.clear(); | |
2975 | } | |
7c673cae | 2976 | |
81eedcae | 2977 | void _check_legacy_statfs_alert(); |
9f95a23c | 2978 | void _check_no_per_pool_omap_alert(); |
81eedcae TL |
2979 | void _set_disk_size_mismatch_alert(const string& s) { |
2980 | std::lock_guard l(qlock); | |
2981 | disk_size_mismatch_alert = s; | |
2982 | } | |
2983 | ||
7c673cae FG |
2984 | private: |
2985 | ||
2986 | // -------------------------------------------------------- | |
2987 | // read processing internal methods | |
2988 | int _verify_csum( | |
2989 | OnodeRef& o, | |
2990 | const bluestore_blob_t* blob, | |
2991 | uint64_t blob_xoffset, | |
2992 | const bufferlist& bl, | |
2993 | uint64_t logical_offset) const; | |
2994 | int _decompress(bufferlist& source, bufferlist* result); | |
2995 | ||
2996 | ||
2997 | // -------------------------------------------------------- | |
2998 | // write ops | |
2999 | ||
3000 | struct WriteContext { | |
3001 | bool buffered = false; ///< buffered write | |
3002 | bool compress = false; ///< compressed write | |
3003 | uint64_t target_blob_size = 0; ///< target (max) blob size | |
3004 | unsigned csum_order = 0; ///< target checksum chunk order | |
3005 | ||
3006 | old_extent_map_t old_extents; ///< must deref these blobs | |
eafe8130 | 3007 | interval_set<uint64_t> extents_to_gc; ///< extents for garbage collection |
7c673cae FG |
3008 | |
3009 | struct write_item { | |
3010 | uint64_t logical_offset; ///< write logical offset | |
3011 | BlobRef b; | |
3012 | uint64_t blob_length; | |
3013 | uint64_t b_off; | |
3014 | bufferlist bl; | |
3015 | uint64_t b_off0; ///< original offset in a blob prior to padding | |
3016 | uint64_t length0; ///< original data length prior to padding | |
3017 | ||
3018 | bool mark_unused; | |
3019 | bool new_blob; ///< whether new blob was created | |
3020 | ||
3efd9988 FG |
3021 | bool compressed = false; |
3022 | bufferlist compressed_bl; | |
3023 | size_t compressed_len = 0; | |
3024 | ||
7c673cae FG |
3025 | write_item( |
3026 | uint64_t logical_offs, | |
3027 | BlobRef b, | |
3028 | uint64_t blob_len, | |
3029 | uint64_t o, | |
3030 | bufferlist& bl, | |
3031 | uint64_t o0, | |
3032 | uint64_t l0, | |
3033 | bool _mark_unused, | |
3034 | bool _new_blob) | |
3035 | : | |
3036 | logical_offset(logical_offs), | |
3037 | b(b), | |
3038 | blob_length(blob_len), | |
3039 | b_off(o), | |
3040 | bl(bl), | |
3041 | b_off0(o0), | |
3042 | length0(l0), | |
3043 | mark_unused(_mark_unused), | |
3044 | new_blob(_new_blob) {} | |
3045 | }; | |
3046 | vector<write_item> writes; ///< blobs we're writing | |
3047 | ||
3048 | /// partial clone of the context | |
3049 | void fork(const WriteContext& other) { | |
3050 | buffered = other.buffered; | |
3051 | compress = other.compress; | |
3052 | target_blob_size = other.target_blob_size; | |
3053 | csum_order = other.csum_order; | |
3054 | } | |
3055 | void write( | |
3056 | uint64_t loffs, | |
3057 | BlobRef b, | |
3058 | uint64_t blob_len, | |
3059 | uint64_t o, | |
3060 | bufferlist& bl, | |
3061 | uint64_t o0, | |
3062 | uint64_t len0, | |
3063 | bool _mark_unused, | |
3064 | bool _new_blob) { | |
3065 | writes.emplace_back(loffs, | |
3066 | b, | |
3067 | blob_len, | |
3068 | o, | |
3069 | bl, | |
3070 | o0, | |
3071 | len0, | |
3072 | _mark_unused, | |
3073 | _new_blob); | |
3074 | } | |
3075 | /// Checks for writes to the same pextent within a blob | |
3076 | bool has_conflict( | |
3077 | BlobRef b, | |
3078 | uint64_t loffs, | |
3079 | uint64_t loffs_end, | |
3080 | uint64_t min_alloc_size); | |
3081 | }; | |
3082 | ||
3083 | void _do_write_small( | |
3084 | TransContext *txc, | |
3085 | CollectionRef &c, | |
3086 | OnodeRef o, | |
3087 | uint64_t offset, uint64_t length, | |
3088 | bufferlist::iterator& blp, | |
3089 | WriteContext *wctx); | |
3090 | void _do_write_big( | |
3091 | TransContext *txc, | |
3092 | CollectionRef &c, | |
3093 | OnodeRef o, | |
3094 | uint64_t offset, uint64_t length, | |
3095 | bufferlist::iterator& blp, | |
3096 | WriteContext *wctx); | |
3097 | int _do_alloc_write( | |
3098 | TransContext *txc, | |
3099 | CollectionRef c, | |
3100 | OnodeRef o, | |
3101 | WriteContext *wctx); | |
3102 | void _wctx_finish( | |
3103 | TransContext *txc, | |
3104 | CollectionRef& c, | |
3105 | OnodeRef o, | |
31f18b77 FG |
3106 | WriteContext *wctx, |
3107 | set<SharedBlob*> *maybe_unshared_blobs=0); | |
7c673cae | 3108 | |
7c673cae FG |
3109 | int _write(TransContext *txc, |
3110 | CollectionRef& c, | |
3111 | OnodeRef& o, | |
3112 | uint64_t offset, size_t len, | |
3113 | bufferlist& bl, | |
3114 | uint32_t fadvise_flags); | |
3115 | void _pad_zeros(bufferlist *bl, uint64_t *offset, | |
3116 | uint64_t chunk_size); | |
3117 | ||
31f18b77 FG |
3118 | void _choose_write_options(CollectionRef& c, |
3119 | OnodeRef o, | |
3120 | uint32_t fadvise_flags, | |
3121 | WriteContext *wctx); | |
3122 | ||
3123 | int _do_gc(TransContext *txc, | |
3124 | CollectionRef& c, | |
3125 | OnodeRef o, | |
31f18b77 FG |
3126 | const WriteContext& wctx, |
3127 | uint64_t *dirty_start, | |
3128 | uint64_t *dirty_end); | |
3129 | ||
7c673cae FG |
3130 | int _do_write(TransContext *txc, |
3131 | CollectionRef &c, | |
3132 | OnodeRef o, | |
3133 | uint64_t offset, uint64_t length, | |
3134 | bufferlist& bl, | |
3135 | uint32_t fadvise_flags); | |
3136 | void _do_write_data(TransContext *txc, | |
3137 | CollectionRef& c, | |
3138 | OnodeRef o, | |
3139 | uint64_t offset, | |
3140 | uint64_t length, | |
3141 | bufferlist& bl, | |
3142 | WriteContext *wctx); | |
3143 | ||
3144 | int _touch(TransContext *txc, | |
3145 | CollectionRef& c, | |
3146 | OnodeRef& o); | |
3147 | int _do_zero(TransContext *txc, | |
3148 | CollectionRef& c, | |
3149 | OnodeRef& o, | |
3150 | uint64_t offset, size_t len); | |
3151 | int _zero(TransContext *txc, | |
3152 | CollectionRef& c, | |
3153 | OnodeRef& o, | |
3154 | uint64_t offset, size_t len); | |
3155 | void _do_truncate(TransContext *txc, | |
3156 | CollectionRef& c, | |
3157 | OnodeRef o, | |
31f18b77 FG |
3158 | uint64_t offset, |
3159 | set<SharedBlob*> *maybe_unshared_blobs=0); | |
35e4c445 | 3160 | int _truncate(TransContext *txc, |
7c673cae FG |
3161 | CollectionRef& c, |
3162 | OnodeRef& o, | |
3163 | uint64_t offset); | |
3164 | int _remove(TransContext *txc, | |
3165 | CollectionRef& c, | |
3166 | OnodeRef& o); | |
3167 | int _do_remove(TransContext *txc, | |
3168 | CollectionRef& c, | |
3169 | OnodeRef o); | |
3170 | int _setattr(TransContext *txc, | |
3171 | CollectionRef& c, | |
3172 | OnodeRef& o, | |
3173 | const string& name, | |
3174 | bufferptr& val); | |
3175 | int _setattrs(TransContext *txc, | |
3176 | CollectionRef& c, | |
3177 | OnodeRef& o, | |
3178 | const map<string,bufferptr>& aset); | |
3179 | int _rmattr(TransContext *txc, | |
3180 | CollectionRef& c, | |
3181 | OnodeRef& o, | |
3182 | const string& name); | |
3183 | int _rmattrs(TransContext *txc, | |
3184 | CollectionRef& c, | |
3185 | OnodeRef& o); | |
9f95a23c | 3186 | void _do_omap_clear(TransContext *txc, OnodeRef &o); |
7c673cae FG |
3187 | int _omap_clear(TransContext *txc, |
3188 | CollectionRef& c, | |
3189 | OnodeRef& o); | |
3190 | int _omap_setkeys(TransContext *txc, | |
3191 | CollectionRef& c, | |
3192 | OnodeRef& o, | |
3193 | bufferlist& bl); | |
3194 | int _omap_setheader(TransContext *txc, | |
3195 | CollectionRef& c, | |
3196 | OnodeRef& o, | |
3197 | bufferlist& header); | |
3198 | int _omap_rmkeys(TransContext *txc, | |
3199 | CollectionRef& c, | |
3200 | OnodeRef& o, | |
3201 | bufferlist& bl); | |
3202 | int _omap_rmkey_range(TransContext *txc, | |
3203 | CollectionRef& c, | |
3204 | OnodeRef& o, | |
3205 | const string& first, const string& last); | |
3206 | int _set_alloc_hint( | |
3207 | TransContext *txc, | |
3208 | CollectionRef& c, | |
3209 | OnodeRef& o, | |
3210 | uint64_t expected_object_size, | |
3211 | uint64_t expected_write_size, | |
3212 | uint32_t flags); | |
3213 | int _do_clone_range(TransContext *txc, | |
3214 | CollectionRef& c, | |
3215 | OnodeRef& oldo, | |
3216 | OnodeRef& newo, | |
3217 | uint64_t srcoff, uint64_t length, uint64_t dstoff); | |
3218 | int _clone(TransContext *txc, | |
3219 | CollectionRef& c, | |
3220 | OnodeRef& oldo, | |
3221 | OnodeRef& newo); | |
3222 | int _clone_range(TransContext *txc, | |
3223 | CollectionRef& c, | |
3224 | OnodeRef& oldo, | |
3225 | OnodeRef& newo, | |
3226 | uint64_t srcoff, uint64_t length, uint64_t dstoff); | |
3227 | int _rename(TransContext *txc, | |
3228 | CollectionRef& c, | |
3229 | OnodeRef& oldo, | |
3230 | OnodeRef& newo, | |
3231 | const ghobject_t& new_oid); | |
3232 | int _create_collection(TransContext *txc, const coll_t &cid, | |
3233 | unsigned bits, CollectionRef *c); | |
3234 | int _remove_collection(TransContext *txc, const coll_t &cid, | |
3235 | CollectionRef *c); | |
11fdf7f2 | 3236 | void _do_remove_collection(TransContext *txc, CollectionRef *c); |
7c673cae FG |
3237 | int _split_collection(TransContext *txc, |
3238 | CollectionRef& c, | |
3239 | CollectionRef& d, | |
3240 | unsigned bits, int rem); | |
11fdf7f2 TL |
3241 | int _merge_collection(TransContext *txc, |
3242 | CollectionRef *c, | |
3243 | CollectionRef& d, | |
3244 | unsigned bits); | |
3245 | ||
9f95a23c TL |
3246 | void _collect_allocation_stats(uint64_t need, uint32_t alloc_size, |
3247 | size_t extents); | |
3248 | void _record_allocation_stats(); | |
11fdf7f2 | 3249 | private: |
9f95a23c TL |
3250 | uint64_t probe_count = 0; |
3251 | std::atomic<uint64_t> alloc_stats_count = {0}; | |
3252 | std::atomic<uint64_t> alloc_stats_fragments = { 0 }; | |
3253 | std::atomic<uint64_t> alloc_stats_size = { 0 }; | |
3254 | // | |
3255 | std::array<std::tuple<uint64_t, uint64_t, uint64_t>, 5> alloc_stats_history = | |
3256 | { std::make_tuple(0ul, 0ul, 0ul) }; | |
3257 | ||
11fdf7f2 TL |
3258 | std::atomic<uint64_t> out_of_sync_fm = {0}; |
3259 | // -------------------------------------------------------- | |
3260 | // BlueFSDeviceExpander implementation | |
3261 | uint64_t get_recommended_expansion_delta(uint64_t bluefs_free, | |
3262 | uint64_t bluefs_total) override { | |
3263 | auto delta = _get_bluefs_size_delta(bluefs_free, bluefs_total); | |
3264 | return delta > 0 ? delta : 0; | |
3265 | } | |
3266 | int allocate_freespace( | |
3267 | uint64_t min_size, | |
3268 | uint64_t size, | |
3269 | PExtentVector& extents) override { | |
3270 | return allocate_bluefs_freespace(min_size, size, &extents); | |
3271 | }; | |
9f95a23c TL |
3272 | uint64_t available_freespace(uint64_t alloc_size) override; |
3273 | inline bool _use_rotational_settings(); | |
eafe8130 TL |
3274 | |
3275 | public: | |
3276 | struct sb_info_t { | |
3277 | coll_t cid; | |
3278 | int64_t pool_id = INT64_MIN; | |
3279 | list<ghobject_t> oids; | |
3280 | BlueStore::SharedBlobRef sb; | |
3281 | bluestore_extent_ref_map_t ref_map; | |
3282 | bool compressed = false; | |
3283 | bool passed = false; | |
3284 | bool updated = false; | |
3285 | }; | |
3286 | typedef btree::btree_set< | |
3287 | uint64_t, std::less<uint64_t>, | |
3288 | mempool::bluestore_fsck::pool_allocator<uint64_t>> uint64_t_btree_t; | |
3289 | ||
3290 | typedef mempool::bluestore_fsck::map<uint64_t, sb_info_t> sb_info_map_t; | |
3291 | struct FSCK_ObjectCtx { | |
3292 | int64_t& errors; | |
3293 | int64_t& warnings; | |
3294 | uint64_t& num_objects; | |
3295 | uint64_t& num_extents; | |
3296 | uint64_t& num_blobs; | |
3297 | uint64_t& num_sharded_objects; | |
3298 | uint64_t& num_spanning_blobs; | |
3299 | ||
3300 | mempool_dynamic_bitset* used_blocks; | |
3301 | uint64_t_btree_t* used_omap_head; | |
eafe8130 TL |
3302 | |
3303 | ceph::mutex* sb_info_lock; | |
3304 | sb_info_map_t& sb_info; | |
3305 | ||
3306 | store_statfs_t& expected_store_statfs; | |
3307 | per_pool_statfs& expected_pool_statfs; | |
3308 | BlueStoreRepairer* repairer; | |
3309 | ||
3310 | FSCK_ObjectCtx(int64_t& e, | |
3311 | int64_t& w, | |
3312 | uint64_t& _num_objects, | |
3313 | uint64_t& _num_extents, | |
3314 | uint64_t& _num_blobs, | |
3315 | uint64_t& _num_sharded_objects, | |
3316 | uint64_t& _num_spanning_blobs, | |
3317 | mempool_dynamic_bitset* _ub, | |
3318 | uint64_t_btree_t* _used_omap_head, | |
eafe8130 TL |
3319 | ceph::mutex* _sb_info_lock, |
3320 | sb_info_map_t& _sb_info, | |
3321 | store_statfs_t& _store_statfs, | |
3322 | per_pool_statfs& _pool_statfs, | |
3323 | BlueStoreRepairer* _repairer) : | |
3324 | errors(e), | |
3325 | warnings(w), | |
3326 | num_objects(_num_objects), | |
3327 | num_extents(_num_extents), | |
3328 | num_blobs(_num_blobs), | |
3329 | num_sharded_objects(_num_sharded_objects), | |
3330 | num_spanning_blobs(_num_spanning_blobs), | |
3331 | used_blocks(_ub), | |
3332 | used_omap_head(_used_omap_head), | |
eafe8130 TL |
3333 | sb_info_lock(_sb_info_lock), |
3334 | sb_info(_sb_info), | |
3335 | expected_store_statfs(_store_statfs), | |
3336 | expected_pool_statfs(_pool_statfs), | |
3337 | repairer(_repairer) { | |
3338 | } | |
3339 | }; | |
3340 | ||
3341 | OnodeRef fsck_check_objects_shallow( | |
3342 | FSCKDepth depth, | |
3343 | int64_t pool_id, | |
3344 | CollectionRef c, | |
3345 | const ghobject_t& oid, | |
3346 | const string& key, | |
3347 | const bufferlist& value, | |
9f95a23c | 3348 | mempool::bluestore_fsck::list<string>* expecting_shards, |
eafe8130 TL |
3349 | map<BlobRef, bluestore_blob_t::unused_t>* referenced, |
3350 | const BlueStore::FSCK_ObjectCtx& ctx); | |
3351 | ||
3352 | private: | |
9f95a23c TL |
3353 | void _fsck_check_object_omap(FSCKDepth depth, |
3354 | OnodeRef& o, | |
3355 | const BlueStore::FSCK_ObjectCtx& ctx); | |
3356 | ||
eafe8130 TL |
3357 | void _fsck_check_objects(FSCKDepth depth, |
3358 | FSCK_ObjectCtx& ctx); | |
7c673cae FG |
3359 | }; |
3360 | ||
11fdf7f2 TL |
3361 | inline ostream& operator<<(ostream& out, const BlueStore::volatile_statfs& s) { |
3362 | return out | |
3363 | << " allocated:" | |
3364 | << s.values[BlueStore::volatile_statfs::STATFS_ALLOCATED] | |
3365 | << " stored:" | |
3366 | << s.values[BlueStore::volatile_statfs::STATFS_STORED] | |
3367 | << " compressed:" | |
3368 | << s.values[BlueStore::volatile_statfs::STATFS_COMPRESSED] | |
3369 | << " compressed_orig:" | |
3370 | << s.values[BlueStore::volatile_statfs::STATFS_COMPRESSED_ORIGINAL] | |
3371 | << " compressed_alloc:" | |
3372 | << s.values[BlueStore::volatile_statfs::STATFS_COMPRESSED_ALLOCATED]; | |
7c673cae FG |
3373 | } |
3374 | ||
3375 | static inline void intrusive_ptr_add_ref(BlueStore::Onode *o) { | |
3376 | o->get(); | |
3377 | } | |
3378 | static inline void intrusive_ptr_release(BlueStore::Onode *o) { | |
3379 | o->put(); | |
3380 | } | |
3381 | ||
3382 | static inline void intrusive_ptr_add_ref(BlueStore::OpSequencer *o) { | |
3383 | o->get(); | |
3384 | } | |
3385 | static inline void intrusive_ptr_release(BlueStore::OpSequencer *o) { | |
3386 | o->put(); | |
3387 | } | |
3388 | ||
11fdf7f2 TL |
3389 | class BlueStoreRepairer |
3390 | { | |
3391 | public: | |
3392 | // to simplify future potential migration to mempools | |
3393 | using fsck_interval = interval_set<uint64_t>; | |
3394 | ||
3395 | // Structure to track what pextents are used for specific cid/oid. | |
3396 | // Similar to Bloom filter positive and false-positive matches are | |
3397 | // possible only. | |
3398 | // Maintains two lists of bloom filters for both cids and oids | |
3399 | // where each list entry is a BF for specific disk pextent | |
3400 | // The length of the extent per filter is measured on init. | |
3401 | // Allows to filter out 'uninteresting' pextents to speadup subsequent | |
3402 | // 'is_used' access. | |
3403 | struct StoreSpaceTracker { | |
3404 | const uint64_t BLOOM_FILTER_SALT_COUNT = 2; | |
3405 | const uint64_t BLOOM_FILTER_TABLE_SIZE = 32; // bytes per single filter | |
3406 | const uint64_t BLOOM_FILTER_EXPECTED_COUNT = 16; // arbitrary selected | |
3407 | static const uint64_t DEF_MEM_CAP = 128 * 1024 * 1024; | |
3408 | ||
3409 | typedef mempool::bluestore_fsck::vector<bloom_filter> bloom_vector; | |
3410 | bloom_vector collections_bfs; | |
3411 | bloom_vector objects_bfs; | |
3412 | ||
3413 | bool was_filtered_out = false; | |
3414 | uint64_t granularity = 0; // extent length for a single filter | |
3415 | ||
3416 | StoreSpaceTracker() { | |
3417 | } | |
3418 | StoreSpaceTracker(const StoreSpaceTracker& from) : | |
3419 | collections_bfs(from.collections_bfs), | |
3420 | objects_bfs(from.objects_bfs), | |
3421 | granularity(from.granularity) { | |
3422 | } | |
3423 | ||
3424 | void init(uint64_t total, | |
3425 | uint64_t min_alloc_size, | |
3426 | uint64_t mem_cap = DEF_MEM_CAP) { | |
3427 | ceph_assert(!granularity); // not initialized yet | |
3428 | ceph_assert(min_alloc_size && isp2(min_alloc_size)); | |
3429 | ceph_assert(mem_cap); | |
3430 | ||
3431 | total = round_up_to(total, min_alloc_size); | |
3432 | granularity = total * BLOOM_FILTER_TABLE_SIZE * 2 / mem_cap; | |
3433 | ||
3434 | if (!granularity) { | |
3435 | granularity = min_alloc_size; | |
3436 | } else { | |
3437 | granularity = round_up_to(granularity, min_alloc_size); | |
3438 | } | |
3439 | ||
3440 | uint64_t entries = round_up_to(total, granularity) / granularity; | |
3441 | collections_bfs.resize(entries, | |
3442 | bloom_filter(BLOOM_FILTER_SALT_COUNT, | |
3443 | BLOOM_FILTER_TABLE_SIZE, | |
3444 | 0, | |
3445 | BLOOM_FILTER_EXPECTED_COUNT)); | |
3446 | objects_bfs.resize(entries, | |
3447 | bloom_filter(BLOOM_FILTER_SALT_COUNT, | |
3448 | BLOOM_FILTER_TABLE_SIZE, | |
3449 | 0, | |
3450 | BLOOM_FILTER_EXPECTED_COUNT)); | |
3451 | } | |
3452 | inline uint32_t get_hash(const coll_t& cid) const { | |
3453 | return cid.hash_to_shard(1); | |
3454 | } | |
3455 | inline void set_used(uint64_t offset, uint64_t len, | |
3456 | const coll_t& cid, const ghobject_t& oid) { | |
3457 | ceph_assert(granularity); // initialized | |
3458 | ||
3459 | // can't call this func after filter_out has been applied | |
3460 | ceph_assert(!was_filtered_out); | |
3461 | if (!len) { | |
3462 | return; | |
3463 | } | |
3464 | auto pos = offset / granularity; | |
3465 | auto end_pos = (offset + len - 1) / granularity; | |
3466 | while (pos <= end_pos) { | |
3467 | collections_bfs[pos].insert(get_hash(cid)); | |
3468 | objects_bfs[pos].insert(oid.hobj.get_hash()); | |
3469 | ++pos; | |
3470 | } | |
3471 | } | |
3472 | // filter-out entries unrelated to the specified(broken) extents. | |
3473 | // 'is_used' calls are permitted after that only | |
3474 | size_t filter_out(const fsck_interval& extents); | |
3475 | ||
3476 | // determines if collection's present after filtering-out | |
3477 | inline bool is_used(const coll_t& cid) const { | |
3478 | ceph_assert(was_filtered_out); | |
3479 | for(auto& bf : collections_bfs) { | |
3480 | if (bf.contains(get_hash(cid))) { | |
3481 | return true; | |
3482 | } | |
3483 | } | |
3484 | return false; | |
3485 | } | |
3486 | // determines if object's present after filtering-out | |
3487 | inline bool is_used(const ghobject_t& oid) const { | |
3488 | ceph_assert(was_filtered_out); | |
3489 | for(auto& bf : objects_bfs) { | |
3490 | if (bf.contains(oid.hobj.get_hash())) { | |
3491 | return true; | |
3492 | } | |
3493 | } | |
3494 | return false; | |
3495 | } | |
3496 | // determines if collection's present before filtering-out | |
3497 | inline bool is_used(const coll_t& cid, uint64_t offs) const { | |
3498 | ceph_assert(granularity); // initialized | |
3499 | ceph_assert(!was_filtered_out); | |
3500 | auto &bf = collections_bfs[offs / granularity]; | |
3501 | if (bf.contains(get_hash(cid))) { | |
3502 | return true; | |
3503 | } | |
3504 | return false; | |
3505 | } | |
3506 | // determines if object's present before filtering-out | |
3507 | inline bool is_used(const ghobject_t& oid, uint64_t offs) const { | |
3508 | ceph_assert(granularity); // initialized | |
3509 | ceph_assert(!was_filtered_out); | |
3510 | auto &bf = objects_bfs[offs / granularity]; | |
3511 | if (bf.contains(oid.hobj.get_hash())) { | |
3512 | return true; | |
3513 | } | |
3514 | return false; | |
3515 | } | |
3516 | }; | |
3517 | public: | |
9f95a23c | 3518 | void fix_per_pool_omap(KeyValueDB *db); |
11fdf7f2 TL |
3519 | bool remove_key(KeyValueDB *db, const string& prefix, const string& key); |
3520 | bool fix_shared_blob(KeyValueDB *db, | |
3521 | uint64_t sbid, | |
3522 | const bufferlist* bl); | |
3523 | bool fix_statfs(KeyValueDB *db, const string& key, | |
3524 | const store_statfs_t& new_statfs); | |
3525 | ||
3526 | bool fix_leaked(KeyValueDB *db, | |
3527 | FreelistManager* fm, | |
3528 | uint64_t offset, uint64_t len); | |
3529 | bool fix_false_free(KeyValueDB *db, | |
3530 | FreelistManager* fm, | |
3531 | uint64_t offset, uint64_t len); | |
3532 | bool fix_bluefs_extents(std::atomic<uint64_t>& out_of_sync_flag); | |
3533 | ||
3534 | void init(uint64_t total_space, uint64_t lres_tracking_unit_size); | |
3535 | ||
3536 | bool preprocess_misreference(KeyValueDB *db); | |
3537 | ||
3538 | unsigned apply(KeyValueDB* db); | |
3539 | ||
3540 | void note_misreference(uint64_t offs, uint64_t len, bool inc_error) { | |
3541 | misreferenced_extents.union_insert(offs, len); | |
3542 | if (inc_error) { | |
3543 | ++to_repair_cnt; | |
3544 | } | |
3545 | } | |
9f95a23c | 3546 | // In fact this is the only repairer's method which is thread-safe!! |
eafe8130 TL |
3547 | void inc_repaired() { |
3548 | ++to_repair_cnt; | |
9f95a23c | 3549 | } |
11fdf7f2 TL |
3550 | |
3551 | StoreSpaceTracker& get_space_usage_tracker() { | |
3552 | return space_usage_tracker; | |
3553 | } | |
3554 | const fsck_interval& get_misreferences() const { | |
3555 | return misreferenced_extents; | |
3556 | } | |
3557 | KeyValueDB::Transaction get_fix_misreferences_txn() { | |
3558 | return fix_misreferences_txn; | |
3559 | } | |
3560 | ||
3561 | private: | |
9f95a23c TL |
3562 | std::atomic<unsigned> to_repair_cnt = { 0 }; |
3563 | KeyValueDB::Transaction fix_per_pool_omap_txn; | |
11fdf7f2 TL |
3564 | KeyValueDB::Transaction fix_fm_leaked_txn; |
3565 | KeyValueDB::Transaction fix_fm_false_free_txn; | |
3566 | KeyValueDB::Transaction remove_key_txn; | |
3567 | KeyValueDB::Transaction fix_statfs_txn; | |
3568 | KeyValueDB::Transaction fix_shared_blob_txn; | |
3569 | ||
3570 | KeyValueDB::Transaction fix_misreferences_txn; | |
3571 | ||
3572 | StoreSpaceTracker space_usage_tracker; | |
3573 | ||
3574 | // non-shared extents with multiple references | |
3575 | fsck_interval misreferenced_extents; | |
3576 | ||
3577 | }; | |
9f95a23c TL |
3578 | |
3579 | class RocksDBBlueFSVolumeSelector : public BlueFSVolumeSelector | |
3580 | { | |
3581 | template <class T, size_t MaxX, size_t MaxY> | |
3582 | class matrix_2d { | |
3583 | T values[MaxX][MaxY]; | |
3584 | public: | |
3585 | matrix_2d() { | |
3586 | clear(); | |
3587 | } | |
3588 | T& at(size_t x, size_t y) { | |
3589 | ceph_assert(x < MaxX); | |
3590 | ceph_assert(y < MaxY); | |
3591 | ||
3592 | return values[x][y]; | |
3593 | } | |
3594 | size_t get_max_x() const { | |
3595 | return MaxX; | |
3596 | } | |
3597 | size_t get_max_y() const { | |
3598 | return MaxY; | |
3599 | } | |
3600 | void clear() { | |
3601 | memset(values, 0, sizeof(values)); | |
3602 | } | |
3603 | }; | |
3604 | ||
3605 | enum { | |
3606 | // use 0/nullptr as unset indication | |
3607 | LEVEL_FIRST = 1, | |
f6b5b4d7 TL |
3608 | LEVEL_LOG = LEVEL_FIRST, // BlueFS log |
3609 | LEVEL_WAL, | |
9f95a23c TL |
3610 | LEVEL_DB, |
3611 | LEVEL_SLOW, | |
3612 | LEVEL_MAX | |
3613 | }; | |
3614 | // add +1 row for corresponding per-device totals | |
3615 | // add +1 column for per-level actual (taken from file size) total | |
3616 | typedef matrix_2d<uint64_t, BlueFS::MAX_BDEV + 1, LEVEL_MAX - LEVEL_FIRST + 1> per_level_per_dev_usage_t; | |
3617 | ||
3618 | per_level_per_dev_usage_t per_level_per_dev_usage; | |
f6b5b4d7 TL |
3619 | // file count per level, add +1 to keep total file count |
3620 | uint64_t per_level_files[LEVEL_MAX - LEVEL_FIRST + 1] = { 0 }; | |
9f95a23c TL |
3621 | |
3622 | // Note: maximum per-device totals below might be smaller than corresponding | |
3623 | // perf counters by up to a single alloc unit (1M) due to superblock extent. | |
3624 | // The later is not accounted here. | |
3625 | per_level_per_dev_usage_t per_level_per_dev_max; | |
3626 | ||
3627 | uint64_t l_totals[LEVEL_MAX - LEVEL_FIRST]; | |
3628 | uint64_t db_avail4slow = 0; | |
3629 | enum { | |
3630 | OLD_POLICY, | |
3631 | USE_SOME_EXTRA | |
3632 | }; | |
3633 | ||
3634 | public: | |
3635 | RocksDBBlueFSVolumeSelector( | |
3636 | uint64_t _wal_total, | |
3637 | uint64_t _db_total, | |
3638 | uint64_t _slow_total, | |
3639 | uint64_t _level0_size, | |
3640 | uint64_t _level_base, | |
3641 | uint64_t _level_multiplier, | |
3642 | double reserved_factor, | |
3643 | uint64_t reserved, | |
3644 | bool new_pol) | |
3645 | { | |
f6b5b4d7 | 3646 | l_totals[LEVEL_LOG - LEVEL_FIRST] = 0; // not used at the moment |
9f95a23c TL |
3647 | l_totals[LEVEL_WAL - LEVEL_FIRST] = _wal_total; |
3648 | l_totals[LEVEL_DB - LEVEL_FIRST] = _db_total; | |
3649 | l_totals[LEVEL_SLOW - LEVEL_FIRST] = _slow_total; | |
3650 | ||
3651 | if (!new_pol) { | |
3652 | return; | |
3653 | } | |
3654 | ||
3655 | // Calculating how much extra space is available at DB volume. | |
3656 | // Depending on the presence of explicit reserved size specification it might be either | |
3657 | // * DB volume size - reserved | |
3658 | // or | |
3659 | // * DB volume size - sum_max_level_size(0, L-1) - max_level_size(L) * reserved_factor | |
3660 | if (!reserved) { | |
3661 | uint64_t prev_levels = _level0_size; | |
3662 | uint64_t cur_level = _level_base; | |
3663 | uint64_t cur_threshold = 0; | |
3664 | do { | |
3665 | uint64_t next_level = cur_level * _level_multiplier; | |
3666 | uint64_t next_threshold = prev_levels + cur_level + next_level * reserved_factor; | |
3667 | if (_db_total <= next_threshold) { | |
3668 | db_avail4slow = cur_threshold ? _db_total - cur_threshold : 0; | |
3669 | break; | |
3670 | } else { | |
3671 | prev_levels += cur_level; | |
3672 | cur_level = next_level; | |
3673 | cur_threshold = next_threshold; | |
3674 | } | |
3675 | } while (true); | |
3676 | } else { | |
3677 | db_avail4slow = _db_total - reserved; | |
3678 | } | |
3679 | } | |
3680 | ||
f6b5b4d7 TL |
3681 | void* get_hint_for_log() const override { |
3682 | return reinterpret_cast<void*>(LEVEL_LOG); | |
9f95a23c TL |
3683 | } |
3684 | void* get_hint_by_dir(const string& dirname) const override; | |
3685 | ||
3686 | void add_usage(void* hint, const bluefs_fnode_t& fnode) override { | |
3687 | if (hint == nullptr) | |
3688 | return; | |
3689 | size_t pos = (size_t)hint - LEVEL_FIRST; | |
3690 | for (auto& p : fnode.extents) { | |
3691 | auto& cur = per_level_per_dev_usage.at(p.bdev, pos); | |
3692 | auto& max = per_level_per_dev_max.at(p.bdev, pos); | |
3693 | cur += p.length; | |
3694 | if (cur > max) { | |
3695 | max = cur; | |
3696 | } | |
3697 | { | |
3698 | //update per-device totals | |
3699 | auto& cur = per_level_per_dev_usage.at(p.bdev, LEVEL_MAX - LEVEL_FIRST); | |
3700 | auto& max = per_level_per_dev_max.at(p.bdev, LEVEL_MAX - LEVEL_FIRST); | |
3701 | cur += p.length; | |
3702 | if (cur > max) { | |
3703 | max = cur; | |
3704 | } | |
3705 | } | |
3706 | } | |
3707 | { | |
3708 | //update per-level actual totals | |
3709 | auto& cur = per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos); | |
3710 | auto& max = per_level_per_dev_max.at(BlueFS::MAX_BDEV, pos); | |
3711 | cur += fnode.size; | |
3712 | if (cur > max) { | |
3713 | max = cur; | |
3714 | } | |
3715 | } | |
f6b5b4d7 TL |
3716 | ++per_level_files[pos]; |
3717 | ++per_level_files[LEVEL_MAX - LEVEL_FIRST]; | |
9f95a23c TL |
3718 | } |
3719 | void sub_usage(void* hint, const bluefs_fnode_t& fnode) override { | |
3720 | if (hint == nullptr) | |
3721 | return; | |
3722 | size_t pos = (size_t)hint - LEVEL_FIRST; | |
3723 | for (auto& p : fnode.extents) { | |
3724 | auto& cur = per_level_per_dev_usage.at(p.bdev, pos); | |
3725 | ceph_assert(cur >= p.length); | |
3726 | cur -= p.length; | |
3727 | ||
3728 | //update per-device totals | |
3729 | auto& cur2 = per_level_per_dev_usage.at(p.bdev, LEVEL_MAX - LEVEL_FIRST); | |
3730 | ceph_assert(cur2 >= p.length); | |
3731 | cur2 -= p.length; | |
3732 | } | |
3733 | //update per-level actual totals | |
3734 | auto& cur = per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos); | |
3735 | ceph_assert(cur >= fnode.size); | |
3736 | cur -= fnode.size; | |
f6b5b4d7 TL |
3737 | ceph_assert(per_level_files[pos] > 0); |
3738 | --per_level_files[pos]; | |
3739 | ceph_assert(per_level_files[LEVEL_MAX - LEVEL_FIRST] > 0); | |
3740 | --per_level_files[LEVEL_MAX - LEVEL_FIRST]; | |
9f95a23c TL |
3741 | } |
3742 | void add_usage(void* hint, uint64_t fsize) override { | |
3743 | if (hint == nullptr) | |
3744 | return; | |
3745 | size_t pos = (size_t)hint - LEVEL_FIRST; | |
3746 | //update per-level actual totals | |
3747 | auto& cur = per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos); | |
3748 | auto& max = per_level_per_dev_max.at(BlueFS::MAX_BDEV, pos); | |
3749 | cur += fsize; | |
3750 | if (cur > max) { | |
3751 | max = cur; | |
3752 | } | |
3753 | } | |
3754 | void sub_usage(void* hint, uint64_t fsize) override { | |
3755 | if (hint == nullptr) | |
3756 | return; | |
3757 | size_t pos = (size_t)hint - LEVEL_FIRST; | |
3758 | //update per-level actual totals | |
3759 | auto& cur = per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos); | |
3760 | ceph_assert(cur >= fsize); | |
3761 | per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos) -= fsize; | |
3762 | } | |
3763 | ||
3764 | uint8_t select_prefer_bdev(void* h) override; | |
3765 | void get_paths( | |
3766 | const std::string& base, | |
3767 | BlueFSVolumeSelector::paths& res) const override; | |
3768 | ||
3769 | void dump(ostream& sout) override; | |
3770 | }; | |
3771 | ||
7c673cae | 3772 | #endif |