MySQL 8.4.3
Source Code Documentation
buf0buf.h
Go to the documentation of this file.
1/*****************************************************************************
2
3Copyright (c) 1995, 2024, Oracle and/or its affiliates.
4
5This program is free software; you can redistribute it and/or modify it under
6the terms of the GNU General Public License, version 2.0, as published by the
7Free Software Foundation.
8
9This program is designed to work with certain software (including
10but not limited to OpenSSL) that is licensed under separate terms,
11as designated in a particular file or component or in included license
12documentation. The authors of MySQL hereby grant you an additional
13permission to link the program and your derivative works with the
14separately licensed software that they have either included with
15the program or referenced in the documentation.
16
17This program is distributed in the hope that it will be useful, but WITHOUT
18ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
19FOR A PARTICULAR PURPOSE. See the GNU General Public License, version 2.0,
20for more details.
21
22You should have received a copy of the GNU General Public License along with
23this program; if not, write to the Free Software Foundation, Inc.,
2451 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
25
26*****************************************************************************/
27
28/** @file include/buf0buf.h
29 The database buffer pool high-level routines
30
31 Created 11/5/1995 Heikki Tuuri
32 *******************************************************/
33
34#ifndef buf0buf_h
35#define buf0buf_h
36
37#include "buf0types.h"
38#include "fil0fil.h"
39#include "hash0hash.h"
40#include "mtr0types.h"
41#include "os0proc.h"
42#include "page0types.h"
43#include "srv0shutdown.h"
44#include "srv0srv.h"
45#include "univ.i"
46#include "ut0byte.h"
47#include "ut0rbt.h"
48
49#include "buf/buf.h"
50
51#include <ostream>
52
53// Forward declaration
54struct fil_addr_t;
55
56/** @name Modes for buf_page_get_gen */
57/** @{ */
58enum class Page_fetch {
59 /** Get always */
60 NORMAL,
61
62 /** Same as NORMAL, but hint that the fetch is part of a large scan.
63 Try not to flood the buffer pool with pages that may not be accessed again
64 any time soon. */
65 SCAN,
66
67 /** get if in pool */
69
70 /** get if in pool, do not make the block young in the LRU list */
72
73 /** get and bufferfix, but set no latch; we have separated this case, because
74 it is error-prone programming not to set a latch, and it should be used with
75 care */
77
78 /** Get the page only if it's in the buffer pool, if not then set a watch on
79 the page. */
81
82 /** Like Page_fetch::NORMAL, but do not mind if the file page has been
83 freed. */
85
86 /** Like Page_fetch::POSSIBLY_FREED, but do not initiate read ahead. */
88};
89/** @} */
90
91/** @name Modes for buf_page_get_known_nowait */
92
93/** @{ */
94enum class Cache_hint {
95 /** Move the block to the start of the LRU list if there is a danger that the
96 block would drift out of the buffer pool*/
97 MAKE_YOUNG = 51,
98
99 /** Preserve the current LRU position of the block. */
100 KEEP_OLD = 52
101};
102
103/** @} */
104
105/** Number of bits to representing a buffer pool ID */
107
108/** The maximum number of buffer pools that can be defined */
110
111/** Maximum number of concurrent buffer pool watches */
112#define BUF_POOL_WATCH_SIZE (srv_n_purge_threads + 1)
113
114/** The maximum number of page_hash locks */
115constexpr ulint MAX_PAGE_HASH_LOCKS = 1024;
116
117/** The buffer pools of the database */
119
120#ifdef UNIV_HOTBACKUP
121/** first block, for --apply-log */
122extern buf_block_t *back_block1;
123/** second block, for page reorganize */
124extern buf_block_t *back_block2;
125#endif /* UNIV_HOTBACKUP */
126
127/** @brief States of a control block
128@see buf_page_t
129
130The enumeration values must be 0..7. */
131enum buf_page_state : uint8_t {
132 /** A sentinel for the buffer pool watch, element of buf_pool->watch[] */
134 /** Contains a clean compressed page */
136 /** Contains a compressed page that is in the buf_pool->flush_list */
138
139 /** Is in the free list; must be after the BUF_BLOCK_ZIP_ constants for
140 compressed-only pages @see buf_block_state_valid() */
142
143 /** When buf_LRU_get_free_block returns a block, it is in this state */
145
146 /** Contains a buffered file page */
148
149 /** Contains some main memory object */
151
152 /** Hash index should be removed before putting to the free list */
155
156/** This structure defines information we will fetch from each buffer pool. It
157will be used to print table IO stats */
159 /* General buffer pool info */
160
161 /** Buffer Pool ID */
163 /** Buffer Pool size in pages */
165 /** Length of buf_pool->LRU */
167 /** buf_pool->LRU_old_len */
169 /** Length of buf_pool->free list */
171 /** Length of buf_pool->flush_list */
173 /** buf_pool->n_pend_unzip, pages pending decompress */
175 /** buf_pool->n_pend_reads, pages pending read */
177 /** Number of pages pending flush of given type */
178 std::array<size_t, BUF_FLUSH_N_TYPES> n_pending_flush;
179 /** number of pages made young */
181 /** number of pages not made young */
183 /** buf_pool->n_pages_read */
185 /** buf_pool->n_pages_created */
187 /** buf_pool->n_pages_written */
189 /** buf_pool->n_page_gets */
191 /** buf_pool->n_ra_pages_read_rnd, number of pages readahead */
193 /** buf_pool->n_ra_pages_read, number of pages readahead */
195 /** buf_pool->n_ra_pages_evicted, number of readahead pages evicted without
196 access */
198 /** num of buffer pool page gets since last printout */
200
201 /* Buffer pool access stats */
202 double page_made_young_rate; /*!< page made young rate in pages
203 per second */
204 double page_not_made_young_rate; /*!< page not made young rate
205 in pages per second */
206 double pages_read_rate; /*!< num of pages read per second */
207 double pages_created_rate; /*!< num of pages create per second */
208 double pages_written_rate; /*!< num of pages written per second */
209 ulint page_read_delta; /*!< num of pages read since last
210 printout */
211 ulint young_making_delta; /*!< num of pages made young since
212 last printout */
213 ulint not_young_making_delta; /*!< num of pages not make young since
214 last printout */
215
216 /* Statistics about read ahead algorithm. */
217 double pages_readahead_rnd_rate; /*!< random readahead rate in pages per
218 second */
219 double pages_readahead_rate; /*!< readahead rate in pages per
220 second */
221 double pages_evicted_rate; /*!< rate of readahead page evicted
222 without access, in pages per second */
223
224 /* Stats about LRU eviction */
225 ulint unzip_lru_len; /*!< length of buf_pool->unzip_LRU
226 list */
227 /* Counters for LRU policy */
228 ulint io_sum; /*!< buf_LRU_stat_sum.io */
229 ulint io_cur; /*!< buf_LRU_stat_cur.io, num of IO
230 for current interval */
231 ulint unzip_sum; /*!< buf_LRU_stat_sum.unzip */
232 ulint unzip_cur; /*!< buf_LRU_stat_cur.unzip, num
233 pages decompressed in current
234 interval */
235};
236
237/** The occupied bytes of lists in all buffer pools */
239 ulint LRU_bytes; /*!< LRU size in bytes */
240 ulint unzip_LRU_bytes; /*!< unzip_LRU size in bytes */
241 ulint flush_list_bytes; /*!< flush_list size in bytes */
242};
243
244#ifndef UNIV_HOTBACKUP
245/** Creates the buffer pool.
246@param[in] total_size Size of the total pool in bytes.
247@param[in] n_instances Number of buffer pool instances to create.
248@return DB_SUCCESS if success, DB_ERROR if not enough memory or error */
249dberr_t buf_pool_init(ulint total_size, ulint n_instances);
250
251/** Frees the buffer pool at shutdown. This must not be invoked before
252 freeing all mutexes. */
253void buf_pool_free_all();
254
255/** Determines if a block is intended to be withdrawn. The caller must ensure
256that there was a sufficient memory barrier to read curr_size and old_size.
257@param[in] buf_pool buffer pool instance
258@param[in] block pointer to control block
259@retval true if will be withdrawn */
260bool buf_block_will_withdrawn(buf_pool_t *buf_pool, const buf_block_t *block);
261
262/** Determines if a frame is intended to be withdrawn. The caller must ensure
263that there was a sufficient memory barrier to read curr_size and old_size.
264@param[in] buf_pool buffer pool instance
265@param[in] ptr pointer to a frame
266@retval true if will be withdrawn */
267bool buf_frame_will_withdrawn(buf_pool_t *buf_pool, const byte *ptr);
268
269/** This is the thread for resizing buffer pool. It waits for an event and
270when waked up either performs a resizing and sleeps again. */
271void buf_resize_thread();
272
273/** Checks if innobase_should_madvise_buf_pool() value has changed since we've
274last check and if so, then updates buf_pool_should_madvise and calls madvise
275for all chunks in all srv_buf_pool_instances.
276@see buf_pool_should_madvise comment for a longer explanation. */
278
279/** Clears the adaptive hash index on all pages in the buffer pool. */
281
282/** Gets the current size of buffer buf_pool in bytes.
283 @return size in bytes */
284static inline ulint buf_pool_get_curr_size(void);
285/** Gets the current size of buffer buf_pool in frames.
286 @return size in pages */
287static inline ulint buf_pool_get_n_pages(void);
288
289/** @return true if buffer pool resize is in progress. */
291
292#endif /* !UNIV_HOTBACKUP */
293
294/** Gets the smallest oldest_modification lsn among all of the earliest
295added pages in flush lists. In other words - takes the last dirty page
296from each flush list, and calculates minimum oldest_modification among
297all of them. Does not acquire global lock for the whole process, so the
298result might come from inconsistent view on flush lists.
299
300@note Note that because of the relaxed order in each flush list, this
301functions no longer returns the smallest oldest_modification among all
302of the dirty pages. If you wanted to have a safe lsn, which is smaller
303than every oldest_modification, you would need to use another function:
304 buf_pool_get_oldest_modification_lwm().
305
306Returns zero if there were no dirty pages (flush lists were empty).
307
308@return minimum oldest_modification of last pages from flush lists,
309 zero if flush lists were empty */
311
312/** Gets a safe low watermark for oldest_modification. It's guaranteed
313that there were no dirty pages with smaller oldest_modification in the
314whole flush lists.
315
316Returns zero if flush lists were empty, be careful in such case, because
317taking the newest lsn is probably not a good idea. If you wanted to rely
318on some lsn in such case, you would need to follow pattern:
319
320 dpa_lsn = log_buffer_dirty_pages_added_up_to_lsn(*log_sys);
321
322 lwm_lsn = buf_pool_get_oldest_modification_lwm();
323
324 if (lwm_lsn == 0) lwm_lsn = dpa_lsn;
325
326The order is important to avoid race conditions.
327
328@remarks
329It's guaranteed that the returned value will not be smaller than the
330last checkpoint lsn. It's not guaranteed that the returned value is
331the maximum possible. It's just the best effort for the low cost.
332It basically takes result of buf_pool_get_oldest_modification_approx()
333and subtracts maximum possible lag introduced by relaxed order in
334flush lists (srv_log_recent_closed_size).
335
336@return safe low watermark for oldest_modification of dirty pages,
337 or zero if flush lists were empty; if non-zero, it is then
338 guaranteed not to be at block boundary (and it points to lsn
339 inside data fragment of block) */
341
342#ifndef UNIV_HOTBACKUP
343
344/** Allocates a buf_page_t descriptor. This function must succeed. In case
345 of failure we assert in this function. */
347 MY_ATTRIBUTE((malloc));
348
349/** Free a buf_page_t descriptor.
350@param[in] bpage bpage descriptor to free */
352
353/** Allocates a buffer block.
354 @return own: the allocated block, in state BUF_BLOCK_MEMORY */
356 buf_pool_t *buf_pool); /*!< in: buffer pool instance,
357 or NULL for round-robin selection
358 of the buffer pool */
359/** Frees a buffer block which does not contain a file page. */
360static inline void buf_block_free(
361 buf_block_t *block); /*!< in, own: block to be freed */
362#endif /* !UNIV_HOTBACKUP */
363
364/** Copies contents of a buffer frame to a given buffer.
365@param[in] buf buffer to copy to
366@param[in] frame buffer frame
367@return buf */
368static inline byte *buf_frame_copy(byte *buf, const buf_frame_t *frame);
369
370#ifndef UNIV_HOTBACKUP
371/** This is the general function used to get optimistic access to a database
372page.
373@param[in] rw_latch RW_S_LATCH, RW_X_LATCH
374@param[in,out] block Guessed block
375@param[in] modify_clock Modify clock value
376@param[in] fetch_mode Fetch mode
377@param[in] file File name
378@param[in] line Line where called
379@param[in,out] mtr Mini-transaction
380@return true if success */
381bool buf_page_optimistic_get(ulint rw_latch, buf_block_t *block,
382 uint64_t modify_clock, Page_fetch fetch_mode,
383 const char *file, ulint line, mtr_t *mtr);
384
385/** This is used to get access to a known database page, when no waiting can be
386done.
387@param[in] rw_latch RW_S_LATCH or RW_X_LATCH.
388@param[in] block The known page.
389@param[in] hint Cache_hint::MAKE_YOUNG or Cache_hint::KEEP_OLD
390@param[in] file File name from where it was called.
391@param[in] line Line from where it was called.
392@param[in,out] mtr Mini-transaction covering the fetch
393@return true if success */
394bool buf_page_get_known_nowait(ulint rw_latch, buf_block_t *block,
395 Cache_hint hint, const char *file, ulint line,
396 mtr_t *mtr);
397
398/** Given a tablespace id and page number tries to get that page. If the
399page is not in the buffer pool it is not loaded and NULL is returned.
400Suitable for using when holding the lock_sys latches (as it avoids deadlock).
401@param[in] page_id page Id
402@param[in] location Location where called
403@param[in] mtr Mini-transaction
404@return pointer to a page or NULL */
405const buf_block_t *buf_page_try_get(const page_id_t &page_id,
406 ut::Location location, mtr_t *mtr);
407
408/** Get read access to a compressed page (usually of type
409FIL_PAGE_TYPE_ZBLOB or FIL_PAGE_TYPE_ZBLOB2).
410The page must be released with buf_page_release_zip().
411NOTE: the page is not protected by any latch. Mutual exclusion has to
412be implemented at a higher level. In other words, all possible
413accesses to a given page through this function must be protected by
414the same set of mutexes or latches.
415@param[in] page_id page id
416@param[in] page_size page size
417@return pointer to the block */
419 const page_size_t &page_size);
420
421/** This is the general function used to get access to a database page.
422@param[in] page_id Page id
423@param[in] page_size Page size
424@param[in] rw_latch RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH
425@param[in] guess Guessed block or NULL
426@param[in] mode Fetch mode.
427@param[in] location Location from where this method was called.
428@param[in] mtr Mini-transaction
429@param[in] dirty_with_no_latch Mark page as dirty even if page is being
430 pinned without any latch
431@return pointer to the block or NULL */
433 const page_size_t &page_size, ulint rw_latch,
435 ut::Location location, mtr_t *mtr,
436 bool dirty_with_no_latch = false);
437
438/** NOTE! The following macros should be used instead of buf_page_get_gen,
439 to improve debugging. Only values RW_S_LATCH and RW_X_LATCH are allowed
440 in LA! */
442 ulint latch, ut::Location location,
443 mtr_t *mtr) {
444 return buf_page_get_gen(id, size, latch, nullptr, Page_fetch::NORMAL,
445 location, mtr);
446}
447/** Use these macros to bufferfix a page with no latching. Remember not to
448 read the contents of the page unless you know it is safe. Do not modify
449 the contents of the page! We have separated this case, because it is
450 error-prone programming not to set a latch, and it should be used
451 with care. */
453 const page_size_t &size,
454 ut::Location location,
455 mtr_t *mtr) {
457 location, mtr);
458}
459
460/** Initializes a page to the buffer buf_pool. The page is usually not read
461from a file even if it cannot be found in the buffer buf_pool. This is one
462of the functions which perform to a block a state transition NOT_USED =>
463FILE_PAGE (the other is buf_page_get_gen). The page is latched by passed mtr.
464@param[in] page_id Page id
465@param[in] page_size Page size
466@param[in] rw_latch RW_SX_LATCH, RW_X_LATCH
467@param[in] mtr Mini-transaction
468@return pointer to the block, page bufferfixed */
470 const page_size_t &page_size,
471 rw_lock_type_t rw_latch, mtr_t *mtr);
472
473#else /* !UNIV_HOTBACKUP */
474
475/** Inits a page to the buffer buf_pool, for use in mysqlbackup --restore.
476@param[in] page_id page id
477@param[in] page_size page size
478@param[in,out] block block to init */
479void meb_page_init(const page_id_t &page_id, const page_size_t &page_size,
480 buf_block_t *block);
481#endif /* !UNIV_HOTBACKUP */
482
483#ifndef UNIV_HOTBACKUP
484/** Releases a compressed-only page acquired with buf_page_get_zip(). */
485static inline void buf_page_release_zip(
486 buf_page_t *bpage); /*!< in: buffer block */
487
488/** Releases a latch, if specified.
489@param[in] block buffer block
490@param[in] rw_latch RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH */
491static inline void buf_page_release_latch(buf_block_t *block, ulint rw_latch);
492
493/** Moves a page to the start of the buffer pool LRU list. This high-level
494function can be used to prevent an important page from slipping out of
495the buffer pool.
496@param[in,out] bpage buffer block of a file page */
498
499/** Moved a page to the end of the buffer pool LRU list so that it can be
500flushed out at the earliest.
501@param[in] bpage buffer block of a file page */
502void buf_page_make_old(buf_page_t *bpage);
503
504/** Returns true if the page can be found in the buffer pool hash table.
505NOTE that it is possible that the page is not yet read from disk,
506though.
507@param[in] page_id page id
508@return true if found in the page hash table */
509static inline bool buf_page_peek(const page_id_t &page_id);
510
511#ifdef UNIV_DEBUG
512
513/** Sets file_page_was_freed true if the page is found in the buffer pool.
514This function should be called when we free a file page and want the
515debug version to check that it is not accessed any more unless
516reallocated.
517@param[in] page_id page id
518@return control block if found in page hash table, otherwise NULL */
520
521/** Sets file_page_was_freed false if the page is found in the buffer pool.
522This function should be called when we free a file page and want the
523debug version to check that it is not accessed any more unless
524reallocated.
525@param[in] page_id page id
526@return control block if found in page hash table, otherwise NULL */
528
529#endif /* UNIV_DEBUG */
530/** Reads the freed_page_clock of a buffer block.
531 @return freed_page_clock */
532[[nodiscard]] static inline ulint buf_page_get_freed_page_clock(
533 const buf_page_t *bpage); /*!< in: block */
534/** Reads the freed_page_clock of a buffer block.
535 @return freed_page_clock */
536[[nodiscard]] static inline ulint buf_block_get_freed_page_clock(
537 const buf_block_t *block); /*!< in: block */
538
539/** Tells, for heuristics, if a block is still close enough to the MRU end of
540the LRU list meaning that it is not in danger of getting evicted and also
541implying that it has been accessed recently.
542The page must be either buffer-fixed, either its page hash must be locked.
543@param[in] bpage block
544@return true if block is close to MRU end of LRU */
545static inline bool buf_page_peek_if_young(const buf_page_t *bpage);
546
547/** Recommends a move of a block to the start of the LRU list if there is
548danger of dropping from the buffer pool.
549NOTE: does not reserve the LRU list mutex.
550@param[in] bpage block to make younger
551@return true if should be made younger */
552static inline bool buf_page_peek_if_too_old(const buf_page_t *bpage);
553
554/** Gets the youngest modification log sequence number for a frame.
555 Returns zero if not file page or no modification occurred yet.
556 @return newest modification to page */
558 const buf_page_t *bpage); /*!< in: block containing the
559 page frame */
560
561/** Increment the modify clock.
562The caller must
563(1) own the buf_pool->mutex and block bufferfix count has to be zero,
564(2) own X or SX latch on the block->lock, or
565(3) operate on a thread-private temporary table
566@param[in,out] block buffer block */
567static inline void buf_block_modify_clock_inc(buf_block_t *block);
568
569/** Increments the bufferfix count.
570@param[in] location location
571@param[in,out] block block to bufferfix */
573 buf_block_t *block);
574
575/** Increments the bufferfix count.
576@param[in,out] bpage block to bufferfix
577@return the count */
578static inline ulint buf_block_fix(buf_page_t *bpage);
579
580/** Increments the bufferfix count.
581@param[in,out] block block to bufferfix
582@return the count */
583static inline ulint buf_block_fix(buf_block_t *block);
584
585/** Decrements the bufferfix count.
586@param[in,out] bpage block to bufferunfix
587@return the remaining buffer-fix count */
588static inline ulint buf_block_unfix(buf_page_t *bpage);
589
590/** Decrements the bufferfix count.
591@param[in,out] block block to bufferunfix
592@return the remaining buffer-fix count */
593static inline ulint buf_block_unfix(buf_block_t *block);
594
595/** Unfixes the page, unlatches the page,
596removes it from page_hash and removes it from LRU.
597@param[in,out] bpage pointer to the block */
599
600/** Increments the bufferfix count.
601@param[in,out] b block to bufferfix
602@param[in] l location where requested */
604 ut::Location l [[maybe_unused]]) {
606}
607#else /* !UNIV_HOTBACKUP */
608static inline void buf_block_modify_clock_inc(buf_block_t *block) {}
609#endif /* !UNIV_HOTBACKUP */
610
611#ifndef UNIV_HOTBACKUP
612
613/** Gets the space id, page offset, and byte offset within page of a pointer
614pointing to a buffer frame containing a file page.
615@param[in] ptr pointer to a buffer frame
616@param[out] space space id
617@param[out] addr page offset and byte offset */
618static inline void buf_ptr_get_fsp_addr(const void *ptr, space_id_t *space,
619 fil_addr_t *addr);
620
621#ifdef UNIV_DEBUG
622/** Finds a block in the buffer pool that points to a
623given compressed page. Used only to confirm that buffer pool does not contain a
624given pointer, thus protected by zip_free_mutex.
625@param[in] buf_pool buffer pool instance
626@param[in] data pointer to compressed page
627@return buffer block pointing to the compressed page, or NULL */
628buf_block_t *buf_pool_contains_zip(buf_pool_t *buf_pool, const void *data);
629#endif /* UNIV_DEBUG */
630
631/***********************************************************************
632FIXME_FTS: Gets the frame the pointer is pointing to. */
634 /* out: pointer to frame */
635 byte *ptr); /* in: pointer to a frame */
636
637#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
638/** Validates the buffer pool data structure.
639 @return true */
640bool buf_validate(void);
641#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
642#if defined UNIV_DEBUG_PRINT || defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
643/** Prints info of the buffer pool data structure. */
644void buf_print(void);
645#endif /* UNIV_DEBUG_PRINT || UNIV_DEBUG || UNIV_BUF_DEBUG */
646#endif /* !UNIV_HOTBACKUP */
648 /** Do not crash at the end of buf_page_print(). */
650 /** Do not print the full page dump. */
653
654/** Prints a page to stderr.
655@param[in] read_buf a database page
656@param[in] page_size page size
657@param[in] flags 0 or BUF_PAGE_PRINT_NO_CRASH or
658BUF_PAGE_PRINT_NO_FULL */
659void buf_page_print(const byte *read_buf, const page_size_t &page_size,
660 ulint flags);
661
662/** Decompress a block.
663 @return true if successful */
664bool buf_zip_decompress(buf_block_t *block, /*!< in/out: block */
665 bool check); /*!< in: true=verify the page checksum */
666#ifndef UNIV_HOTBACKUP
667#ifdef UNIV_DEBUG
668/** Returns the number of latched pages in the buffer pool.
669 @return number of latched pages */
671#endif /* UNIV_DEBUG */
672/** Returns the number of pending buf pool read ios.
673 @return number of pending read I/O operations */
675/** Prints info of the buffer i/o. */
676void buf_print_io(FILE *file); /*!< in: file where to print */
677/** Collect buffer pool stats information for a buffer pool. Also
678 record aggregated stats if there are more than one buffer pool
679 in the server */
681 buf_pool_t *buf_pool, /*!< in: buffer pool */
682 ulint pool_id, /*!< in: buffer pool ID */
683 buf_pool_info_t *all_pool_info); /*!< in/out: buffer pool info
684 to fill */
685/** Return the ratio in percents of modified pages in the buffer pool /
686database pages in the buffer pool.
687@return modified page percentage ratio */
688double buf_get_modified_ratio_pct(void);
689/** Refresh the statistics used to print per-second averages. */
691
692/** Assert that all file pages in the buffer are in a replaceable state. */
693void buf_must_be_all_freed(void);
694
695/** Computes number of pending I/O read operations for the buffer pool.
696@return number of pending i/o reads */
698
699/** Computes number of pending I/O write operations for the buffer pool.
700@return number of pending i/o writes */
702
703/** Waits until there are no pending I/O operations for the buffer pool.
704Keep waiting in loop with sleeps, emitting information every minute.
705This is used to avoid risk of some pending async IO (e.g. enqueued by
706the linear read-ahead), which would involve ibuf merge and create new
707redo records. */
709
710/** Invalidates the file pages in the buffer pool when an archive recovery is
711 completed. All the file pages buffered must be in a replaceable state when
712 this function is called: not latched and not modified. */
713void buf_pool_invalidate(void);
714
715/*========================================================================
716--------------------------- LOWER LEVEL ROUTINES -------------------------
717=========================================================================*/
718
719#ifdef UNIV_DEBUG
720/** Adds latch level info for the rw-lock protecting the buffer frame. This
721should be called in the debug version after a successful latching of a page if
722we know the latching order level of the acquired latch.
723@param[in] block buffer page where we have acquired latch
724@param[in] level latching order level */
725static inline void buf_block_dbg_add_level(buf_block_t *block,
726 latch_level_t level);
727#else /* UNIV_DEBUG */
728#define buf_block_dbg_add_level(block, level) /* nothing */
729#endif /* UNIV_DEBUG */
730
731#endif /* !UNIV_HOTBACKUP */
732
733/** Gets the state of a block.
734 @return state */
736 const buf_page_t *bpage); /*!< in: pointer to the control block */
737/** Gets the state of a block.
738 @return state */
739[[nodiscard]] static inline enum buf_page_state buf_block_get_state(
740 const buf_block_t *block); /*!< in: pointer to the control block */
741
742/** Sets the state of a block.
743@param[in,out] bpage pointer to control block
744@param[in] state state */
745static inline void buf_page_set_state(buf_page_t *bpage,
746 enum buf_page_state state);
747
748/** Sets the state of a block.
749@param[in,out] block pointer to control block
750@param[in] state state */
751static inline void buf_block_set_state(buf_block_t *block,
752 enum buf_page_state state);
753
754/** Determines if a block is mapped to a tablespace.
755 @return true if mapped */
756[[nodiscard]] static inline bool buf_page_in_file(
757 const buf_page_t *bpage); /*!< in: pointer to control block */
758#ifndef UNIV_HOTBACKUP
759/** Determines if a block should be on unzip_LRU list.
760 @return true if block belongs to unzip_LRU */
761[[nodiscard]] static inline bool buf_page_belongs_to_unzip_LRU(
762 const buf_page_t *bpage); /*!< in: pointer to control block */
763
764/** Gets the mutex of a block.
765 @return pointer to mutex protecting bpage */
766[[nodiscard]] static inline BPageMutex *buf_page_get_mutex(
767 const buf_page_t *bpage); /*!< in: pointer to control block */
768
769/** Get the flush type of a page.
770 @return flush type */
771[[nodiscard]] static inline buf_flush_t buf_page_get_flush_type(
772 const buf_page_t *bpage); /*!< in: buffer page */
773
774/** Set the flush type of a page.
775@param[in] bpage buffer page
776@param[in] flush_type flush type */
777static inline void buf_page_set_flush_type(buf_page_t *bpage,
779
780/** Map a block to a file page.
781@param[in,out] block pointer to control block
782@param[in] page_id page id */
783static inline void buf_block_set_file_page(buf_block_t *block,
784 const page_id_t &page_id);
785
786/** Gets the io_fix state of a block.
787 @return io_fix state */
788[[nodiscard]] static inline enum buf_io_fix buf_page_get_io_fix(
789 const buf_page_t *bpage); /*!< in: pointer to the control block */
790/** Gets the io_fix state of a block.
791 @return io_fix state */
792[[nodiscard]] static inline enum buf_io_fix buf_block_get_io_fix(
793 const buf_block_t *block); /*!< in: pointer to the control block */
794
795/** Sets the io_fix state of a block.
796@param[in,out] bpage control block
797@param[in] io_fix io_fix state */
798static inline void buf_page_set_io_fix(buf_page_t *bpage,
799 enum buf_io_fix io_fix);
800
801/** Sets the io_fix state of a block.
802@param[in,out] block control block
803@param[in] io_fix io_fix state */
804static inline void buf_block_set_io_fix(buf_block_t *block,
805 enum buf_io_fix io_fix);
806
807/** Makes a block sticky. A sticky block implies that even after we release
808the buf_pool->LRU_list_mutex and the block->mutex:
809* it cannot be removed from the flush_list
810* the block descriptor cannot be relocated
811* it cannot be removed from the LRU list
812Note that:
813* the block can still change its position in the LRU list
814* the next and previous pointers can change.
815@param[in,out] bpage control block */
816static inline void buf_page_set_sticky(buf_page_t *bpage);
817
818/** Removes stickiness of a block. */
819static inline void buf_page_unset_sticky(
820 buf_page_t *bpage); /*!< in/out: control block */
821/** Determine if a buffer block can be relocated in memory. The block
822 can be dirty, but it must not be I/O-fixed or bufferfixed. */
823[[nodiscard]] static inline bool buf_page_can_relocate(
824 const buf_page_t *bpage); /*!< control block being relocated */
825
826/** Determine if a block has been flagged old.
827@param[in] bpage control block
828@return true if old */
829[[nodiscard]] static inline bool buf_page_is_old(const buf_page_t *bpage);
830
831/** Flag a block old.
832@param[in,out] bpage control block
833@param[in] old old */
834static inline void buf_page_set_old(buf_page_t *bpage, bool old);
835
836/** Determine the time of first access of a block in the buffer pool.
837 @return Time of first access, zero if not accessed
838 */
839[[nodiscard]] static inline std::chrono::steady_clock::time_point
840buf_page_is_accessed(const buf_page_t *bpage); /*!< in: control block */
841/** Flag a block accessed. */
842static inline void buf_page_set_accessed(
843 buf_page_t *bpage); /*!< in/out: control block */
844
845/** Gets the buf_block_t handle of a buffered file block if an uncompressed
846page frame exists, or NULL. page frame exists, or NULL. The caller must hold
847either the appropriate hash lock in any mode, either the LRU list mutex. Note:
848even though bpage is not declared a const we don't update its value. It is safe
849to make this pure.
850@param[in] bpage control block, or NULL
851@return control block, or NULL */
852[[nodiscard]] static inline buf_block_t *buf_page_get_block(buf_page_t *bpage);
853#ifdef UNIV_DEBUG
854/** Gets a pointer to the memory frame of a block.
855 @return pointer to the frame */
856[[nodiscard]] static inline buf_frame_t *buf_block_get_frame(
857 const buf_block_t *block); /*!< in: pointer to the control block */
858#else /* UNIV_DEBUG */
859#define buf_block_get_frame(block) (block)->frame
860#endif /* UNIV_DEBUG */
861#else /* !UNIV_HOTBACKUP */
862#define buf_block_get_frame(block) (block)->frame
863#endif /* !UNIV_HOTBACKUP */
864
865/** Get a buffer block from an adaptive hash index pointer.
866This function does not return if the block is not identified.
867@param[in] ptr pointer to within a page frame
868@return pointer to block, never NULL */
869buf_block_t *buf_block_from_ahi(const byte *ptr);
870
871/** Find out if a block pointer points into one of currently used chunks of
872the buffer pool. This is useful if you stored the pointer some time ago, and
873want to dereference it now, and are afraid that buffer pool resize could free
874the memory pointed by it. Thus calling this function requires holding at least
875one of the latches which prevent freeing memory from buffer pool for the
876duration of the call and until you pin the block in some other way, as otherwise
877the result of this function might be obsolete by the time you dereference the
878block (an s-latch on buf_page_hash_lock_get for any hash cell is enough).
879@param buf_pool The buffer pool instance to search in.
880@param ptr A pointer which you want to check. This function will not
881 dereference it.
882@return true iff `block` points inside one of the chunks of the `buf_pool`
883*/
884bool buf_is_block_in_instance(const buf_pool_t *buf_pool,
885 const buf_block_t *ptr);
886
887#ifndef UNIV_HOTBACKUP
888
889/** Inits a page for read to the buffer buf_pool. If the page is
890(1) already in buf_pool, or
891(2) if we specify to read only ibuf pages and the page is not an ibuf page, or
892(3) if the space is deleted or being deleted,
893then this function does nothing.
894Sets the io_fix flag to BUF_IO_READ and sets a non-recursive exclusive lock
895on the buffer frame. The io-handler must take care that the flag is cleared
896and the lock released later.
897@param[in] mode BUF_READ_IBUF_PAGES_ONLY, ...
898@param[in] page_id page id
899@param[in] page_size page size
900@param[in] unzip true=request uncompressed page
901@return pointer to the block or NULL */
903 const page_size_t &page_size, bool unzip);
904
905/** Completes an asynchronous read or write request of a file page to or from
906the buffer pool.
907@param[in] bpage pointer to the block in question
908@param[in] evict whether or not to evict the page from LRU list
909@param[in] type i/o request type for which this completion routine is
910 called.
911@param[in] node file node in which the disk copy of the page exists.
912@return true if successful */
913bool buf_page_io_complete(buf_page_t *bpage, bool evict,
914 IORequest *type = nullptr,
915 fil_node_t *node = nullptr);
916
917/** Free a stale page. Caller must hold the LRU mutex. Upon successful page
918free the LRU mutex will be released.
919@param[in,out] buf_pool Buffer pool the page belongs to.
920@param[in,out] bpage Page to free.
921@return true if page was freed. */
922bool buf_page_free_stale(buf_pool_t *buf_pool, buf_page_t *bpage) noexcept;
923
924/** Evict a page from the buffer pool.
925@param[in] page_id page to be evicted.
926@param[in] page_size page size of the tablespace.
927@param[in] dirty_is_ok if true, it is OK for the page to be dirty. */
928void buf_page_force_evict(const page_id_t &page_id,
929 const page_size_t &page_size,
930 const bool dirty_is_ok = true) noexcept;
931
932/** Free a stale page. Caller must be holding the hash_lock in S mode if
933hash_lock parameter is not nullptr. The hash lock will be released upon return
934always. Caller must hold the LRU mutex if and only if the hash_lock parameter
935is nullptr. Upon unsuccessful page free the LRU mutex will not be released if
936hash_lock is nullptr.
937@param[in,out] buf_pool Buffer pool the page belongs to.
938@param[in,out] bpage Page to free.
939@param[in,out] hash_lock Hash lock covering the fetch from the hash table if
940latched in S mode. nullptr otherwise.
941@return true if page was freed. */
942bool buf_page_free_stale(buf_pool_t *buf_pool, buf_page_t *bpage,
943 rw_lock_t *hash_lock) noexcept;
944
945/** Free a stale page that is being written. The caller must be within the
946page's write code path.
947@param[in,out] bpage Page to free.
948@param[in] owns_sx_lock SX lock on block->lock is set. */
950 bool owns_sx_lock = false) noexcept;
951
952/** Calculates the index of a buffer pool to the buf_pool[] array.
953 @return the position of the buffer pool in buf_pool[] */
954[[nodiscard]] static inline ulint buf_pool_index(
955 const buf_pool_t *buf_pool); /*!< in: buffer pool */
956/** Returns the buffer pool instance given a page instance
957 @return buf_pool */
959 const buf_page_t *bpage); /*!< in: buffer pool page */
960/** Returns the buffer pool instance given a block instance
961 @return buf_pool */
963 const buf_block_t *block); /*!< in: block */
964
965/** Returns the buffer pool instance given a page id.
966@param[in] page_id page id
967@return buffer pool */
968static inline buf_pool_t *buf_pool_get(const page_id_t &page_id);
969
970/** Returns the buffer pool instance given its array index
971 @return buffer pool */
973 ulint index); /*!< in: array index to get
974 buffer pool instance from */
975
976/** Returns the control block of a file page, NULL if not found.
977@param[in] buf_pool buffer pool instance
978@param[in] page_id page id
979@return block, NULL if not found */
981 const page_id_t &page_id);
982
983/** Returns the control block of a file page, NULL if not found.
984If the block is found and lock is not NULL then the appropriate
985page_hash lock is acquired in the specified lock mode. Otherwise,
986mode value is ignored. It is up to the caller to release the
987lock. If the block is found and the lock is NULL then the page_hash
988lock is released by this function.
989@param[in] buf_pool buffer pool instance
990@param[in] page_id page id
991@param[in,out] lock lock of the page hash acquired if bpage is
992found, NULL otherwise. If NULL is passed then the hash_lock is released by
993this function.
994@param[in] lock_mode RW_LOCK_X or RW_LOCK_S. Ignored if
995lock == NULL
996@param[in] watch if true, return watch sentinel also.
997@return pointer to the bpage or NULL; if NULL, lock is also NULL or
998a watch sentinel. */
1000 const page_id_t &page_id,
1001 rw_lock_t **lock,
1003 bool watch = false);
1004
1005/** Returns the control block of a file page, NULL if not found.
1006If the block is found and lock is not NULL then the appropriate
1007page_hash lock is acquired in the specified lock mode. Otherwise,
1008mode value is ignored. It is up to the caller to release the
1009lock. If the block is found and the lock is NULL then the page_hash
1010lock is released by this function.
1011@param[in] buf_pool buffer pool instance
1012@param[in] page_id page id
1013@param[in,out] lock lock of the page hash acquired if bpage is
1014found, NULL otherwise. If NULL is passed then the hash_lock is released by
1015this function.
1016@param[in] lock_mode RW_LOCK_X or RW_LOCK_S. Ignored if
1017lock == NULL
1018@return pointer to the block or NULL; if NULL, lock is also NULL. */
1020 const page_id_t &page_id,
1021 rw_lock_t **lock,
1023
1024/* There are four different ways we can try to get a bpage or block
1025from the page hash:
10261) Caller already holds the appropriate page hash lock: in the case call
1027buf_page_hash_get_low() function.
10282) Caller wants to hold page hash lock in x-mode
10293) Caller wants to hold page hash lock in s-mode
10304) Caller doesn't want to hold page hash lock */
1032 const page_id_t &page_id,
1033 rw_lock_t **l) {
1034 return buf_page_hash_get_locked(b, page_id, l, RW_LOCK_S);
1035}
1037 const page_id_t &page_id,
1038 rw_lock_t **l) {
1039 return buf_page_hash_get_locked(b, page_id, l, RW_LOCK_X);
1040}
1042 return buf_page_hash_get_locked(b, page_id, nullptr, 0);
1043}
1045 const page_id_t &page_id) {
1046 return buf_page_hash_get_locked(b, page_id, nullptr, 0, true);
1047}
1048
1050 const page_id_t &page_id,
1051 rw_lock_t **l) {
1052 return buf_block_hash_get_locked(b, page_id, l, RW_LOCK_S);
1053}
1055 const page_id_t &page_id,
1056 rw_lock_t **l) {
1057 return buf_block_hash_get_locked(b, page_id, l, RW_LOCK_X);
1058}
1060 const page_id_t &page_id) {
1061 return buf_block_hash_get_locked(b, page_id, nullptr, 0);
1062}
1063
1064/** Gets the current length of the free list of buffer blocks.
1065 @return length of the free list */
1067
1068/** Determine if a block is a sentinel for a buffer pool watch.
1069@param[in] buf_pool buffer pool instance
1070@param[in] bpage block
1071@return true if a sentinel for a buffer pool watch, false if not */
1072[[nodiscard]] bool buf_pool_watch_is_sentinel(const buf_pool_t *buf_pool,
1073 const buf_page_t *bpage);
1074
1075/** Stop watching if the page has been read in.
1076buf_pool_watch_set(same_page_id) must have returned NULL before.
1077@param[in] page_id page id */
1078void buf_pool_watch_unset(const page_id_t &page_id);
1079
1080/** Check if the page has been read in.
1081This may only be called after buf_pool_watch_set(same_page_id)
1082has returned NULL and before invoking buf_pool_watch_unset(same_page_id).
1083@param[in] page_id page id
1084@return false if the given page was not read in, true if it was */
1085[[nodiscard]] bool buf_pool_watch_occurred(const page_id_t &page_id);
1086
1087/** Get total buffer pool statistics.
1088@param[out] LRU_len Length of all lru lists
1089@param[out] free_len Length of all free lists
1090@param[out] flush_list_len Length of all flush lists */
1091void buf_get_total_list_len(ulint *LRU_len, ulint *free_len,
1092 ulint *flush_list_len);
1093
1094/** Get total list size in bytes from all buffer pools. */
1096 buf_pools_list_size_t *buf_pools_list_size); /*!< out: list sizes
1097 in all buffer pools */
1098/** Get total buffer pool statistics. */
1100 buf_pool_stat_t *tot_stat); /*!< out: buffer pool stats */
1101
1102/** Get the nth chunk's buffer block in the specified buffer pool.
1103@param[in] buf_pool buffer pool instance
1104@param[in] n nth chunk in the buffer pool
1105@param[in] chunk_size chunk_size
1106@return the nth chunk's buffer block. */
1107static inline buf_block_t *buf_get_nth_chunk_block(const buf_pool_t *buf_pool,
1108 ulint n, ulint *chunk_size);
1109
1110/** Calculate aligned buffer pool size based on srv_buf_pool_chunk_unit,
1111if needed.
1112@param[in] size size in bytes
1113@return aligned size */
1115
1116/** Adjust the proposed chunk unit size so that it satisfies all invariants
1117@param[in] size proposed size of buffer pool chunk unit in bytes
1118@return adjusted size which meets invariants */
1120
1121/** Calculate the checksum of a page from compressed table and update the
1122page.
1123@param[in,out] page page to update
1124@param[in] size compressed page size
1125@param[in] lsn LSN to stamp on the page
1126@param[in] skip_lsn_check true to skip check for lsn (in DEBUG) */
1128 bool skip_lsn_check);
1129
1130/** Return how many more pages must be added to the withdraw list to reach the
1131withdraw target of the currently ongoing buffer pool resize.
1132@param[in] buf_pool buffer pool instance
1133@return page count to be withdrawn or zero if the target is already achieved or
1134if the buffer pool is not currently being resized. */
1135static inline ulint buf_get_withdraw_depth(buf_pool_t *buf_pool);
1136
1137#endif /* !UNIV_HOTBACKUP */
1138
1139/** The common buffer control block structure
1140for compressed and uncompressed frames */
1141
1142/** Number of bits used for buffer page states. */
1143constexpr uint32_t BUF_PAGE_STATE_BITS = 3;
1144
1145template <typename T>
1146class copyable_atomic_t : public std::atomic<T> {
1147 public:
1149 : std::atomic<T>(other.load(std::memory_order_relaxed)) {}
1150};
1151
1154 public:
1155 /** Copy constructor.
1156 @param[in] other Instance to copy from. */
1158 : id(other.id),
1159 size(other.size),
1161 io_fix(other.io_fix),
1162 state(other.state),
1163 flush_type(other.flush_type),
1165#ifndef UNIV_HOTBACKUP
1166 hash(other.hash),
1167#endif /* !UNIV_HOTBACKUP */
1168 list(other.list),
1171 LRU(other.LRU),
1172 zip(other.zip)
1173#ifndef UNIV_HOTBACKUP
1174 ,
1176 m_space(other.m_space),
1178 m_version(other.m_version),
1179 access_time(other.access_time),
1180 m_dblwr_id(other.m_dblwr_id),
1181 old(other.old)
1182#ifdef UNIV_DEBUG
1183 ,
1187 in_LRU_list(other.in_LRU_list),
1190#endif /* UNIV_DEBUG */
1191#endif /* !UNIV_HOTBACKUP */
1192 {
1193#ifndef UNIV_HOTBACKUP
1194 m_space->inc_ref();
1195#endif /* !UNIV_HOTBACKUP */
1196 }
1197
1198 public:
1199 /** Check if the given ptr lies in a memory block of type BUF_BLOCK_MEMORY.
1200 This is checked by looking at the FIL_PAGE_LSN. If the FIL_PAGE_LSN is zero,
1201 then the block state is assumed to be BUF_BLOCK_MEMORY.
1202 @return true if the FIL_PAGE_LSN is zero, false otherwise. */
1203 [[nodiscard]] static bool is_memory(const page_t *const ptr) noexcept;
1204
1205 /** Check if the state of this page is BUF_BLOCK_MEMORY.
1206 @return true if the state is BUF_BLOCK_MEMORY, or false. */
1207 [[nodiscard]] bool is_memory() const noexcept {
1208 return state == BUF_BLOCK_MEMORY;
1209 }
1210
1211#ifndef UNIV_HOTBACKUP
1212 /** Set the doublewrite buffer ID.
1213 @param[in] batch_id Double write batch ID that flushed the page. */
1214 void set_dblwr_batch_id(uint16_t batch_id) { m_dblwr_id = batch_id; }
1215
1216 /** @return the double write batch id, or uint16_t max if undefined. */
1217 [[nodiscard]] uint16_t get_dblwr_batch_id() const { return (m_dblwr_id); }
1218
1219 /** Retrieve the tablespace id.
1220 @return tablespace id */
1221 [[nodiscard]] space_id_t space() const noexcept { return id.space(); }
1222
1223 /** Retrieve the page number.
1224 @return page number */
1225 [[nodiscard]] page_no_t page_no() const noexcept { return id.page_no(); }
1226
1227 /** Checks if this space reference saved during last page ID initialization
1228 was deleted or truncated since.
1229 @return true when space reference stored leads was deleted or truncated and
1230 this page should be discarded. Result is up to date until the fil_shard mutex
1231 is released. */
1232 inline bool is_stale() const {
1233 ut_a(m_space != nullptr);
1234 ut_a(id.space() == m_space->id);
1235 ut_a(m_version <= m_space->get_current_version());
1237 ut_a(!m_space->is_deleted());
1238 return false;
1239 } else {
1240 return true;
1241 }
1242 }
1243
1244 /** Checks if this space reference saved during last page ID initialization
1245 was deleted or truncated since.
1246 @return true when space reference stored leads was deleted or truncated and
1247 this page should be discarded. When false is returned, the status of stale is
1248 checked to be guaranteed. */
1249 inline bool was_stale() const {
1250 ut_a(m_space != nullptr);
1251 ut_a(id.space() == m_space->id);
1252 /* If the the version is OK, then the space must not be deleted.
1253 However, version is modified before the deletion flag is set, so reading
1254 these values need to be executed in reversed order. The atomic reads
1255 cannot be relaxed for it to work. */
1256 bool was_not_deleted = m_space->was_not_deleted();
1258 ut_a(was_not_deleted);
1259 return false;
1260 } else {
1261 return true;
1262 }
1263 }
1264
1265 /** Retrieve the tablespace object if one was available during page ID
1266 initialization. The returned object is safe to use as long as this buf_page_t
1267 object is not changed. Caller should have a IO fix, buffer fix, mutex or any
1268 other mean to assure the page will not be freed. After that is released the
1269 space object may be freed.
1270 @return tablespace object */
1271 inline fil_space_t *get_space() const { return m_space; }
1272
1273 /** Set the stored page id to a new value. This is used only on a buffer
1274 block with BUF_BLOCK_MEMORY state.
1275 @param[in] page_id the new value of the page id. */
1276 void set_page_id(const page_id_t page_id) {
1278 id = page_id;
1279 }
1280
1281 /** Set the page size to a new value. This can be used during initialization
1282 of a newly allocated buffer page.
1283 @param[in] page_size the new value of the page size. */
1284 void set_page_size(const page_size_t &page_size) {
1286 size = page_size;
1287 }
1288
1289 /** Sets stored page ID to the new value. Handles space object reference
1290 count.
1291 @param[in] new_page_id new page ID to be set. */
1292 inline void reset_page_id(page_id_t new_page_id) {
1293 if (m_space != nullptr) {
1294 /* If we reach this line through a call chain:
1295 srv_shutdown -> buf_pool_free_all -> buf_pool_free_instance ->
1296 buf_page_free_descriptor, then we are already past the fil system
1297 shutdown, and all fil_space_t objects were already freed. */
1299 m_space->dec_ref();
1300 }
1301 }
1302 id = new_page_id;
1304 }
1305
1306 /** Sets stored value to invalid/empty value. Handles space object reference
1307 count. */
1308 inline void reset_page_id() {
1310 }
1311
1312 private:
1313 /** Updates new space reference and acquires "reference count latch" and the
1314 current version of the space object. */
1315 inline void space_id_changed() {
1316 m_space = nullptr;
1317 m_version = 0;
1318 if (id.space() != UINT32_UNDEFINED) {
1319 m_space = fil_space_get(id.space());
1320 /* There could be non-existent tablespace while importing it */
1321 if (m_space) {
1322 m_space->inc_ref();
1323 /* We don't have a way to check the MDL locks, which are guarding the
1324 version number, so we don't use get_current_version(). */
1326 }
1327 }
1328 }
1329
1330 public:
1331 /** @return the flush observer instance. */
1333
1334 /** Set the flush observer for the page.
1335 @param[in] flush_observer The flush observer to set. */
1336 void set_flush_observer(Flush_observer *flush_observer) noexcept {
1337 /* Don't allow to set flush observer from non-null to null, or from one
1338 observer to another. */
1339 ut_a(m_flush_observer == nullptr || m_flush_observer == flush_observer);
1340 m_flush_observer = flush_observer;
1341 }
1342
1343 /** Remove the flush observer. */
1344 void reset_flush_observer() noexcept { m_flush_observer = nullptr; }
1345#endif /* !UNIV_HOTBACKUP */
1346
1347 /** @return the LSN of the latest modification. */
1348 lsn_t get_newest_lsn() const noexcept { return newest_modification; }
1349
1350 /** @return the LSN of the first modification since the last time
1351 it was clean. */
1352 lsn_t get_oldest_lsn() const noexcept { return oldest_modification; }
1353
1354 /** @return true if the page is dirty. */
1355 bool is_dirty() const noexcept { return get_oldest_lsn() > 0; }
1356
1357 /** Set the latest modification LSN.
1358 @param[in] lsn Latest modification lSN. */
1360
1361 /** Set the LSN when the page is modified for the first time.
1362 @param[in] lsn First modification LSN. */
1363 void set_oldest_lsn(lsn_t lsn) noexcept;
1364
1365 /** Set page to clean state. */
1366 void set_clean() noexcept { set_oldest_lsn(0); }
1367
1368 /** @name General fields
1369 None of these bit-fields must be modified without holding
1370 buf_page_get_mutex() [buf_block_t::mutex or
1371 buf_pool->zip_mutex], since they can be stored in the same
1372 machine word. */
1373 /** @{ */
1374
1375 /** Page id. */
1377
1378 /** Page size. */
1380
1381 /** Count of how many fold this block is currently bufferfixed. */
1383
1384 private:
1385 /** Type of pending I/O operation.
1386 Modified under protection of buf_page_get_mutex(this).
1387 Read under protection of rules described in @see Buf_io_fix_latching_rules */
1389
1390#ifdef UNIV_DEBUG
1391 public:
1392 /** Checks if io_fix has any of the known enum values.
1393 @param[in] io_fix the value to test
1394 @return true iff io_fix has any of the known enum values
1395 */
1397 switch (io_fix) {
1398 case BUF_IO_NONE:
1399 case BUF_IO_READ:
1400 case BUF_IO_WRITE:
1401 case BUF_IO_PIN:
1402 return true;
1403 }
1404 return false;
1405 }
1406
1407 private:
1408 /** Checks if io_fix has any of the known enum values.
1409 @return true iff io_fix has any of the known enum values
1410 */
1413 }
1414 /* Helper debug-only functions related latching rules are moved to a separate
1415 class so that this header doesn't have to pull in Stateful_latching_rules.*/
1418
1419 /* Helper debug-only class used to track which thread is currently responsible
1420 for performing I/O operation on this page. There's at most one such thread and
1421 the responsibility might be passed from one to another during async I/O. This
1422 is used to prove correctness of io_fix state transitions and checking it
1423 without a latch in the io_completion threads. */
1425 /** The thread responsible for I/O on this page, or an impossible value if
1426 no thread is currently responsible for I/O*/
1427 std::thread::id responsible_thread{std::thread().get_id()};
1428
1429 public:
1430 /** Checks if there is any thread responsible for I/O on this page now.
1431 @return true iff there is a thread responsible for I/O on this page.*/
1433 return responsible_thread != std::thread().get_id();
1434 }
1435
1436 /** Checks if the current thread is responsible for I/O on this page now.
1437 @return true iff the current thread is responsible for I/O on this page.*/
1440 }
1441
1442 /** Called by the thread responsible for I/O on this page to release its
1443 responsibility. */
1444 void release() {
1446 responsible_thread = std::thread().get_id();
1447 }
1448
1449 /** Called by the thread which becomes responsible for I/O on this page to
1450 indicate that it takes the responsibility. */
1451 void take() {
1454 }
1455 };
1456 /** Tracks which thread is responsible for I/O on this page. */
1458
1459 public:
1460 /** Checks if there is any thread responsible for I/O on this page now.
1461 @return true iff there is a thread responsible for I/O on this page.*/
1464 }
1465
1466 /** Checks if the current thread is responsible for I/O on this page now.
1467 @return true iff the current thread is responsible for I/O on this page.*/
1470 }
1471
1472 /** Called by the thread responsible for I/O on this page to release its
1473 responsibility. */
1475
1476 /** Called by the thread which becomes responsible for I/O on this page to
1477 indicate that it takes the responsibility. */
1480 io_fix.load(std::memory_order_relaxed) == BUF_IO_WRITE ||
1481 io_fix.load(std::memory_order_relaxed) == BUF_IO_READ);
1483 }
1484#endif /* UNIV_DEBUG */
1485 private:
1486 /** Retrieves a value of io_fix without requiring or acquiring any latches.
1487 Note that this implies that the value might be stale unless caller establishes
1488 happens-before relation in some other way.
1489 This is a low-level function which shouldn't be used directly, but
1490 rather via wrapper methods which check if proper latches are taken or via one
1491 of the many `was_io_fix_something()` methods with name explicitly warning the
1492 developer about the uncertainty involved.
1493 @return the value of io_fix at some moment "during" the call */
1496 return io_fix.load(std::memory_order_relaxed);
1497 }
1498
1499 public:
1500 /** This is called only when having full ownership of the page object and no
1501 other thread can reach it. This currently happens during buf_pool_create(),
1502 buf_pool_resize() (which latch quite a lot) or from fil_tablespace_iterate()
1503 which creates a fake, private block which is not really a part of the buffer
1504 pool.
1505 Therefore we allow this function to set io_fix without checking for any
1506 latches.
1507 Please use set_io_fix(BUF_IO_NONE) to change state in a regular situation. */
1509 io_fix.store(BUF_IO_NONE, std::memory_order_relaxed);
1510 /* This is only needed because places which call init_io_fix() do not call
1511 buf_page_t's constructor */
1513 }
1514
1515 /** This is called only when having full ownership of the page object and no
1516 other thread can reach it. This currently happens during buf_page_init_low()
1517 under buf_page_get_mutex(this), on a previously initialized page for reuse,
1518 yet should be treated as initialization of the field, not a state transition.
1519 Please use set_io_fix(BUF_IO_NONE) to change state in a regular situation. */
1521 ut_ad(io_fix.load(std::memory_order_relaxed) == BUF_IO_NONE);
1523 io_fix.store(BUF_IO_NONE, std::memory_order_relaxed);
1524 }
1525
1526 /** Sets io_fix to specified value.
1527 Assumes the caller holds buf_page_get_mutex(this).
1528 Might require additional latches depending on particular state transition.
1529 Calls take_io_responsibility() or release_io_responsibility() as needed.
1530 @see Buf_io_fix_latching_rules for specific rules. */
1532
1533 /** Retrieves the current value of io_fix.
1534 Assumes the caller holds buf_page_get_mutex(this).
1535 @return the current value of io_fix */
1538 return get_io_fix_snapshot();
1539 }
1540
1541 /** Checks if the current value of io_fix is BUF_IO_WRITE.
1542 Assumes the caller holds buf_page_get_mutex(this) or some other latches which
1543 prevent state transition from/to BUF_IO_WRITE.
1544 @see Buf_io_fix_latching_rules for specific rules.
1545 @return true iff the current value of io_fix == BUF_IO_WRITE */
1546 bool is_io_fix_write() const;
1547
1548 /** Checks if the current value of io_fix is BUF_IO_READ.
1549 Assumes the caller holds buf_page_get_mutex(this) or some other latches which
1550 prevent state transition from/to BUF_IO_READ.
1551 @see Buf_io_fix_latching_rules for specific rules.
1552 @return true iff the current value of io_fix == BUF_IO_READ */
1553 bool is_io_fix_read() const;
1554
1555 /** Assuming that io_fix is either BUF_IO_READ or BUF_IO_WRITE determines
1556 which of the two it is. Additionally it assumes the caller holds
1557 buf_page_get_mutex(this) or some other latches which prevent state transition
1558 from BUF_IO_READ or from BUF_IO_WRITE to another state.
1559 @see Buf_io_fix_latching_rules for specific rules.
1560 @return true iff the current value of io_fix == BUF_IO_READ */
1562
1563 /** Checks if io_fix is BUF_IO_READ without requiring or acquiring any
1564 latches.
1565 Note that this implies calling this function twice in a row could produce
1566 different results.
1567 @return true iff io_fix equal to BUF_IO_READ was noticed*/
1569
1570 /** Checks if io_fix is BUF_IO_FIX or BUF_IO_READ or BUF_IO_WRITE without
1571 requiring or acquiring any latches.
1572 Note that this implies calling this function twice in a row could produce
1573 different results.
1574 @return true iff io_fix not equal to BUF_IO_NONE was noticed */
1575 bool was_io_fixed() const { return get_io_fix_snapshot() != BUF_IO_NONE; }
1576
1577 /** Checks if io_fix is BUF_IO_NONE without requiring or acquiring any
1578 latches.
1579 Note that this implies calling this function twice in a row could produce
1580 different results.
1581 Please, prefer this function over !was_io_fixed() to avoid the misleading
1582 interpretation as "not(Exists time such that io_fix(time))", while in fact we
1583 want and get "Exists time such that !io_fix(time)".
1584 @return true iff io_fix equal to BUF_IO_NONE was noticed */
1586
1587 /** Block state. @see buf_page_in_file */
1589
1590 /** If this block is currently being flushed to disk, this tells
1591 the flush_type. @see buf_flush_t */
1593
1594 /** Index number of the buffer pool that this block belongs to */
1596
1597 static_assert(MAX_BUFFER_POOLS <= 64,
1598 "MAX_BUFFER_POOLS > 64; redefine buf_pool_index");
1599
1600 /** @} */
1601#ifndef UNIV_HOTBACKUP
1602 /** Node used in chaining to buf_pool->page_hash or buf_pool->zip_hash */
1604#endif /* !UNIV_HOTBACKUP */
1605
1606 /** @name Page flushing fields
1607 All these are protected by buf_pool->mutex. */
1608 /** @{ */
1609
1610 /** Based on state, this is a list node, protected by the corresponding list
1611 mutex, in one of the following lists in buf_pool:
1612
1613 - BUF_BLOCK_NOT_USED: free, withdraw
1614 - BUF_BLOCK_FILE_PAGE: flush_list
1615 - BUF_BLOCK_ZIP_DIRTY: flush_list
1616 - BUF_BLOCK_ZIP_PAGE: zip_clean
1617
1618 The node pointers are protected by the corresponding list mutex.
1619
1620 The contents of the list node is undefined if !in_flush_list &&
1621 state == BUF_BLOCK_FILE_PAGE, or if state is one of
1622 BUF_BLOCK_MEMORY,
1623 BUF_BLOCK_REMOVE_HASH or
1624 BUF_BLOCK_READY_IN_USE. */
1625
1627
1628 private:
1629 /** The flush LSN, LSN when this page was written to the redo log. For
1630 non redo logged pages this is set using: buf_flush_borrow_lsn() */
1632
1633 /** log sequence number of the youngest modification to this block, zero
1634 if not modified. Protected by block mutex */
1636
1637 public:
1638 /** log sequence number of the START of the log entry written of the oldest
1639 modification to this block which has not yet been flushed on disk; zero if
1640 all modifications are on disk. Writes to this field must be covered by both
1641 block->mutex and buf_pool->flush_list_mutex. Hence reads can happen while
1642 holding any one of the two mutexes */
1643 /** @} */
1644
1645 /** @name LRU replacement algorithm fields
1646 These fields are protected by both buf_pool->LRU_list_mutex and the
1647 block mutex. */
1648 /** @{ */
1649
1650 /** node of the LRU list */
1652
1653 /** compressed page; zip.data (but not the data it points to) is
1654 protected by buf_pool->zip_mutex; state == BUF_BLOCK_ZIP_PAGE and
1655 zip.data == NULL means an active buf_pool->watch */
1657
1658#ifndef UNIV_HOTBACKUP
1659 /** Flush observer instance. */
1661
1662 /** Tablespace instance that this page belongs to. */
1664
1665 /** The value of buf_pool->freed_page_clock when this block was the last
1666 time put to the head of the LRU list; a thread is allowed to read this
1667 for heuristic purposes without holding any mutex or latch */
1669
1670 /** @} */
1671 /** Version of fil_space_t when the page was updated. It can also be viewed as
1672 the truncation number. */
1673 uint32_t m_version{};
1674
1675 /** Time of first access, or 0 if the block was never accessed in the
1676 buffer pool. Protected by block mutex */
1677 std::chrono::steady_clock::time_point access_time;
1678
1679 private:
1680 /** Double write instance ordinal value during writes. This is used
1681 by IO completion (writes) to select the double write instance.*/
1682 uint16_t m_dblwr_id{};
1683
1684 public:
1685 /** true if the block is in the old blocks in buf_pool->LRU_old */
1686 bool old;
1687
1688#ifdef UNIV_DEBUG
1689 /** This is set to true when fsp frees a page in buffer pool;
1690 protected by buf_pool->zip_mutex or buf_block_t::mutex. */
1692
1693 /** true if in buf_pool->flush_list; when buf_pool->flush_list_mutex
1694 is free, the following should hold:
1695 in_flush_list == (state == BUF_BLOCK_FILE_PAGE ||
1696 state == BUF_BLOCK_ZIP_DIRTY)
1697 Writes to this field must be covered by both buf_pool->flush_list_mutex
1698 and block->mutex. Hence reads can happen while holding any one of the
1699 two mutexes */
1701
1702 /** true if in buf_pool->free; when buf_pool->free_list_mutex is free, the
1703 following should hold: in_free_list == (state == BUF_BLOCK_NOT_USED) */
1705
1706 /** true if the page is in the LRU list; used in debugging */
1708
1709 /** true if in buf_pool->page_hash */
1711
1712 /** true if in buf_pool->zip_hash */
1714#endif /* UNIV_DEBUG */
1715
1716#endif /* !UNIV_HOTBACKUP */
1717};
1718
1719/** Structure used by AHI to contain information on record prefixes to be
1720considered in hash index subsystem. It is meant for using as a single 64bit
1721atomic value, thus it needs to be aligned properly. */
1722struct alignas(alignof(uint64_t)) btr_search_prefix_info_t {
1723 /** recommended prefix: number of bytes in an incomplete field
1724 @see BTR_PAGE_MAX_REC_SIZE */
1725 uint32_t n_bytes;
1726 /** recommended prefix length for hash search: number of full fields */
1727 uint16_t n_fields;
1728 /** true or false, depending on whether the leftmost record of several records
1729 with the same prefix should be indexed in the hash index */
1731
1733 return n_bytes == other.n_bytes && n_fields == other.n_fields;
1734 }
1735
1736 bool operator==(const btr_search_prefix_info_t &other) const {
1737 return n_bytes == other.n_bytes && n_fields == other.n_fields &&
1738 left_side == other.left_side;
1739 }
1740
1741 bool operator!=(const btr_search_prefix_info_t &other) const {
1742 return !(*this == other);
1743 }
1744};
1745
1746/** The buffer control block structure */
1748 /** @name General fields */
1749 /** @{ */
1750
1751 /** page information; this must be the first field, so
1752 that buf_pool->page_hash can point to buf_page_t or buf_block_t */
1754
1755#ifndef UNIV_HOTBACKUP
1756 /** read-write lock of the buffer frame */
1758
1759#ifdef UNIV_DEBUG
1760 /** Check if the buffer block was freed.
1761 @return true if the block was freed, false otherwise. */
1762 bool was_freed() const { return page.file_page_was_freed; }
1763#endif /* UNIV_DEBUG */
1764
1765#endif /* UNIV_HOTBACKUP */
1766
1767 /** pointer to buffer frame which is of size UNIV_PAGE_SIZE, and aligned
1768 to an address divisible by UNIV_PAGE_SIZE */
1769 byte *frame;
1770
1771 /** Determine whether the page is in new-style compact format.
1772 @return true if the page is in compact format
1773 @return false if it is in old-style format */
1774 bool is_compact() const;
1775
1776 /** node of the decompressed LRU list; a block is in the unzip_LRU list if
1777 page.state == BUF_BLOCK_FILE_PAGE and page.zip.data != NULL. Protected by
1778 both LRU_list_mutex and the block mutex. */
1780#ifdef UNIV_DEBUG
1781 /** true if the page is in the decompressed LRU list; used in debugging */
1783
1785#endif /* UNIV_DEBUG */
1786
1787 /** @} */
1788
1789 /** Structure that holds most AHI-related fields. */
1790 struct ahi_t {
1791 public:
1792 /** Recommended prefix info for hash search. It is atomically copied
1793 from the index's current recommendation for the prefix info and should
1794 eventually get to the block's actual prefix info used. It is used to decide
1795 when the n_hash_helps should be reset. It is modified only while having S-
1796 or X- latch on block's lock. */
1797 std::atomic<btr_search_prefix_info_t> recommended_prefix_info;
1798 /** Prefix info that was used for building hash index. It cannot be modified
1799 while there are any record entries added in the AHI. It's invariant that all
1800 records added to AHI from this block were folded using this prefix info. It
1801 may only be modified when we are holding the appropriate X-latch in
1802 btr_search_sys->parts[]->latch. Also, it happens that it is modified
1803 to not-empty value only when the block is held in private or the block's
1804 lock is S- or X-latched. This implies that the field's non-empty value may
1805 be read and use reliably when the appropriate
1806 btr_search_sys->parts[]->latch S-latch or X-latch is being held, or
1807 the block's lock is X-latched. */
1808 std::atomic<btr_search_prefix_info_t> prefix_info;
1809 static_assert(decltype(prefix_info)::is_always_lock_free);
1810
1811 /** Index for which the adaptive hash index has been created, or nullptr if
1812 the page does not exist in the index. Note that it does not guarantee that
1813 the AHI index is complete, though: there may have been hash collisions etc.
1814 It may be modified:
1815 - to nullptr if btr_search_enabled is false and block's mutex is held and
1816 block's state is BUF_BLOCK_FILE_PAGE and btr_search_enabled_mutex is
1817 owned, protecting the btr_search_enabled from being changed,
1818 - to nullptr if btr_search_enabled is false and block is held in private in
1819 BUF_BLOCK_REMOVE_HASH state in buf_LRU_free_page().
1820 - to any value under appropriate X-latch in btr_search_sys->parts[]->latch
1821 if btr_search_enabled is true (and setting btr_search_enabled to false in
1822 turn is protected by having all btr_search_sys->parts[]->latch X-latched).
1823 */
1824 std::atomic<dict_index_t *> index;
1825
1826#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
1827 /** Used in debugging. The number of pointers in the adaptive hash index
1828 pointing to this frame; Modified under appropriate X-latch in
1829 btr_search_sys->parts[]->latch. */
1830 std::atomic<uint16_t> n_pointers;
1831
1832 inline void validate() const {
1833 /* These fields are read without holding any AHI latches. Adding or
1834 removing a block from AHI requires having only an appropriate AHI part
1835 X-latched. If we have at least S-latched the correct AHI part (for which
1836 we would need at least S-latch on block for the block's index to not be
1837 changed in meantime) this check is certain. If we don't have necessary AHI
1838 latches, then:
1839 - it can't happen that the check fails while the block is removed from
1840 AHI. Both btr_search_drop_page_hash_index() and
1841 buf_pool_clear_hash_index() will first make the n_pointers be 0 and then
1842 set index to nullptr. As the index is an atomic variable, so if we
1843 synchronize with a reset to nullptr which is sequenced after the reset of
1844 n_pointers, we should see the n_pointers set to 0 here.
1845 - it can happen that the check fails while the block is added to the AHI
1846 right after we read the index is nullptr. In such case, if the n_pointers
1847 is not 0, we double check the index member. It can still be nullptr, if
1848 the block is removed after reading the n_pointers, but that should be near
1849 impossible. */
1850 ut_a(this->index.load() != nullptr || this->n_pointers.load() == 0 ||
1851 this->index.load() != nullptr);
1852 }
1853
1854 inline void assert_empty() const { ut_a(this->n_pointers.load() == 0); }
1855
1856 inline void assert_empty_on_init() const {
1857 UNIV_MEM_VALID(&this->n_pointers, sizeof(this->n_pointers));
1858 assert_empty();
1859 }
1860#else
1861 inline void validate() const {}
1862
1863 inline void assert_empty() const {}
1864
1865 inline void assert_empty_on_init() const {}
1866#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
1868
1869 /** Counter which controls how many times the current prefix recommendation
1870 would help in searches. If it is helpful enough, it will be used as the
1871 actual prefix to build hash for this block. It is modified similarly as
1872 recommended_prefix_info, that is only while having S- or X- latch on block's
1873 lock. Because it is modified concurrently, it may not have fully reliable
1874 count, but it is enough for this use case.
1875 Mind the n_hash_helps is AHI-related, and should be in the ahi_t struct above,
1876 but having it outside causes the made_dirty_with_no_latch to occupy the common
1877 8byte aligned 8byte long space, so basically it saves us 8bytes of the object
1878 that is used in high volumes. */
1879 std::atomic<uint32_t> n_hash_helps;
1880 /** true if block has been made dirty without acquiring X/SX latch as the
1881 block belongs to temporary tablespace and block is always accessed by a
1882 single thread. */
1884
1885#ifndef UNIV_HOTBACKUP
1886#ifdef UNIV_DEBUG
1887 /** @name Debug fields */
1888 /** @{ */
1889 /** In the debug version, each thread which bufferfixes the block acquires
1890 an s-latch here; so we can use the debug utilities in sync0rw */
1892 /** @} */
1893#endif /* UNIV_DEBUG */
1894#endif /* !UNIV_HOTBACKUP */
1895
1896 /** @name Optimistic search field */
1897 /** @{ */
1898
1899 /** This clock is incremented every time a pointer to a record on the page
1900 may become obsolete; this is used in the optimistic cursor positioning: if
1901 the modify clock has not changed, we know that the pointer is still valid;
1902 this field may be changed if the thread (1) owns the LRU list mutex and the
1903 page is not bufferfixed, or (2) the thread has an x-latch on the block,
1904 or (3) the block must belong to an intrinsic table */
1906
1907 /** @} */
1908
1909 /** mutex protecting this block: state (also protected by the buffer
1910 pool mutex), io_fix, buf_fix_count, and accessed; we introduce this
1911 new mutex in InnoDB-5.1 to relieve contention on the buffer pool mutex */
1913
1914 /** Get the modified clock (version) value.
1915 @param[in] single_threaded Thread can only be written to or read by a
1916 single thread
1917 @return the modified clock vlue. */
1918 uint64_t get_modify_clock(IF_DEBUG(bool single_threaded)) const noexcept {
1919#if defined(UNIV_DEBUG) && !defined(UNIV_LIBRARY) && !defined(UNIV_HOTBACKUP)
1920 /* No block latch is acquired when blocks access is guaranteed to be
1921 in single threaded mode. */
1923 ut_ad(single_threaded || rw_lock_own_flagged(&lock, mode));
1924#endif /* UNIV_DEBUG && !UNIV_LIBRARY */
1925
1926 return modify_clock;
1927 }
1928
1929 /** Get the page number and space id of the current buffer block.
1930 @return page number of the current buffer block. */
1931 const page_id_t &get_page_id() const { return page.id; }
1932
1933 /** Get the page number of the current buffer block.
1934 @return page number of the current buffer block. */
1935 page_no_t get_page_no() const { return (page.id.page_no()); }
1936
1937 /** Get the next page number of the current buffer block.
1938 @return next page number of the current buffer block. */
1941 }
1942
1943 /** Get the prev page number of the current buffer block.
1944 @return prev page number of the current buffer block. */
1947 }
1948
1949 /** Get the page type of the current buffer block.
1950 @return page type of the current buffer block. */
1953 }
1954
1955#ifndef UNIV_HOTBACKUP
1956 /** Mark the frame with jumbled page_id, while initiating i/o read
1957 (BUF_IO_READ).*/
1959 memset(frame, 0x00, page.size.logical());
1962 }
1963#endif /* UNIV_HOTBACKUP */
1964
1965 uint16_t get_page_level() const;
1966 bool is_leaf() const;
1967 bool is_root() const;
1968 bool is_index_page() const;
1969
1970 /** Check if this index page is empty. An index page is considered empty
1971 if the next record of an infimum record is supremum record. Presence of
1972 del-marked records will make the page non-empty.
1973 @return true if this index page is empty. */
1974 bool is_empty() const;
1975
1976 /** Get the page type of the current buffer block as string.
1977 @return page type of the current buffer block as string. */
1978 [[nodiscard]] const char *get_page_type_str() const noexcept;
1979
1980 /** Gets the compressed page descriptor corresponding to an uncompressed page
1981 if applicable.
1982 @return page descriptor or nullptr. */
1984 return page.zip.data != nullptr ? &page.zip : nullptr;
1985 }
1986
1987 /** Const version.
1988 @return page descriptor or nullptr. */
1989 page_zip_des_t const *get_page_zip() const noexcept {
1990 return page.zip.data != nullptr ? &page.zip : nullptr;
1991 }
1992
1993 [[nodiscard]] bool is_memory() const noexcept { return page.is_memory(); }
1994};
1995
1996inline bool buf_block_t::is_root() const {
1997 return ((get_next_page_no() == FIL_NULL) && (get_prev_page_no() == FIL_NULL));
1998}
1999
2000inline bool buf_block_t::is_leaf() const { return get_page_level() == 0; }
2001
2002inline bool buf_block_t::is_index_page() const {
2003 return get_page_type() == FIL_PAGE_INDEX;
2004}
2005
2006/** Check if a buf_block_t object is in a valid state
2007@param block buffer block
2008@return true if valid */
2010 return buf_block_get_state(block) >= BUF_BLOCK_NOT_USED &&
2012}
2013
2014/** Compute the hash value for blocks in buf_pool->zip_hash. */
2015/** @{ */
2016static inline uint64_t buf_pool_hash_zip_frame(void *ptr) {
2017 return ut::hash_uint64(reinterpret_cast<uintptr_t>(ptr) >>
2019}
2020static inline uint64_t buf_pool_hash_zip(buf_block_t *b) {
2021 return buf_pool_hash_zip_frame(b->frame);
2022}
2023/** @} */
2024
2025/** A "Hazard Pointer" class used to iterate over page lists
2026inside the buffer pool. A hazard pointer is a buf_page_t pointer
2027which we intend to iterate over next and we want it remain valid
2028even after we release the buffer pool mutex. */
2030 public:
2031 /** Constructor
2032 @param buf_pool buffer pool instance
2033 @param mutex mutex that is protecting the hp. */
2034 HazardPointer(const buf_pool_t *buf_pool, const ib_mutex_t *mutex)
2035 : m_buf_pool(buf_pool) IF_DEBUG(, m_mutex(mutex)), m_hp() {}
2036
2037 /** Destructor */
2038 virtual ~HazardPointer() = default;
2039
2040 /** Get current value */
2041 buf_page_t *get() const {
2043 return (m_hp);
2044 }
2045
2046 /** Set current value
2047 @param bpage buffer block to be set as hp */
2048 void set(buf_page_t *bpage);
2049
2050 /** Checks if a bpage is the hp
2051 @param bpage buffer block to be compared
2052 @return true if it is hp */
2053 bool is_hp(const buf_page_t *bpage);
2054
2055 /** Adjust the value of hp. This happens when some
2056 other thread working on the same list attempts to
2057 remove the hp from the list. Must be implemented
2058 by the derived classes.
2059 @param bpage buffer block to be compared */
2060 virtual void adjust(const buf_page_t *bpage) = 0;
2061
2062 /** Adjust the value of hp for moving. This happens
2063 when some other thread working on the same list
2064 attempts to relocate the hp of the page.
2065 @param bpage buffer block to be compared
2066 @param dpage buffer block to be moved to */
2067 void move(const buf_page_t *bpage, buf_page_t *dpage);
2068
2069 protected:
2070 /** Disable copying */
2073
2074 /** Buffer pool instance */
2076
2077#ifdef UNIV_DEBUG
2078 /** mutex that protects access to the m_hp. */
2079 const ib_mutex_t *m_mutex;
2080#endif /* UNIV_DEBUG */
2081
2082 /** hazard pointer. */
2084};
2085
2086/** Class implementing buf_pool->flush_list hazard pointer */
2087class FlushHp : public HazardPointer {
2088 public:
2089 /** Constructor
2090 @param buf_pool buffer pool instance
2091 @param mutex mutex that is protecting the hp. */
2092 FlushHp(const buf_pool_t *buf_pool, const ib_mutex_t *mutex)
2093 : HazardPointer(buf_pool, mutex) {}
2094
2095 /** Destructor */
2096 ~FlushHp() override = default;
2097
2098 /** Adjust the value of hp. This happens when some
2099 other thread working on the same list attempts to
2100 remove the hp from the list.
2101 @param bpage buffer block to be compared */
2102 void adjust(const buf_page_t *bpage) override;
2103};
2104
2105/** Class implementing buf_pool->LRU hazard pointer */
2106class LRUHp : public HazardPointer {
2107 public:
2108 /** Constructor
2109 @param buf_pool buffer pool instance
2110 @param mutex mutex that is protecting the hp. */
2111 LRUHp(const buf_pool_t *buf_pool, const ib_mutex_t *mutex)
2112 : HazardPointer(buf_pool, mutex) {}
2113
2114 /** Destructor */
2115 ~LRUHp() override = default;
2116
2117 /** Adjust the value of hp. This happens when some
2118 other thread working on the same list attempts to
2119 remove the hp from the list.
2120 @param bpage buffer block to be compared */
2121 void adjust(const buf_page_t *bpage) override;
2122};
2123
2124/** Special purpose iterators to be used when scanning the LRU list.
2125The idea is that when one thread finishes the scan it leaves the
2126itr in that position and the other thread can start scan from
2127there */
2128class LRUItr : public LRUHp {
2129 public:
2130 /** Constructor
2131 @param buf_pool buffer pool instance
2132 @param mutex mutex that is protecting the hp. */
2133 LRUItr(const buf_pool_t *buf_pool, const ib_mutex_t *mutex)
2134 : LRUHp(buf_pool, mutex) {}
2135
2136 /** Destructor */
2137 ~LRUItr() override = default;
2138
2139 /** Selects from where to start a scan. If we have scanned
2140 too deep into the LRU list it resets the value to the tail
2141 of the LRU list.
2142 @return buf_page_t from where to start scan. */
2143 buf_page_t *start();
2144};
2145
2146/** Struct that is embedded in the free zip blocks */
2148 union {
2149 ulint size; /*!< size of the block */
2151 /*!< stamp[FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID]
2152 == BUF_BUDDY_FREE_STAMP denotes a free
2153 block. If the space_id field of buddy
2154 block != BUF_BUDDY_FREE_STAMP, the block
2155 is not in any zip_free list. If the
2156 space_id is BUF_BUDDY_FREE_STAMP then
2157 stamp[0] will contain the
2158 buddy block size. */
2160
2161 buf_page_t bpage; /*!< Embedded bpage descriptor */
2163 /*!< Node of zip_free list */
2164};
2165
2166/** @brief The buffer pool statistics structure. */
2169
2170 /** Number of page gets performed; also successful searches through the
2171 adaptive hash index are counted as page gets; this field is NOT protected
2172 by the buffer pool mutex */
2174
2175 /** Number of read operations. */
2176 std::atomic<uint64_t> n_pages_read;
2177
2178 /** Number of write operations. */
2179 std::atomic<uint64_t> n_pages_written;
2180
2181 /** number of pages created in the pool with no read. */
2182 std::atomic<uint64_t> n_pages_created;
2183
2184 /** Number of pages read in as part of random read ahead. */
2185 std::atomic<uint64_t> n_ra_pages_read_rnd;
2186
2187 /** Number of pages read in as part of read ahead. */
2188 std::atomic<uint64_t> n_ra_pages_read;
2189
2190 /** Number of read ahead pages that are evicted without being accessed.
2191 Protected by LRU_list_mutex. */
2193
2194 /** Number of pages made young, in calls to buf_LRU_make_block_young().
2195 Protected by LRU_list_mutex. */
2197
2198 /** Number of pages not made young because the first access was not long
2199 enough ago, in buf_page_peek_if_too_old(). Not protected. */
2201
2202 /** LRU size in bytes. Protected by LRU_list_mutex. */
2203 uint64_t LRU_bytes;
2204
2205 /** Flush_list size in bytes. Protected by flush_list_mutex */
2207
2208 static void copy(buf_pool_stat_t &dst, const buf_pool_stat_t &src) noexcept {
2209 Counter::copy(dst.m_n_page_gets, src.m_n_page_gets);
2210
2211 dst.n_pages_read.store(src.n_pages_read.load());
2212
2213 dst.n_pages_written.store(src.n_pages_written.load());
2214
2215 dst.n_pages_created.store(src.n_pages_created.load());
2216
2217 dst.n_ra_pages_read_rnd.store(src.n_ra_pages_read_rnd.load());
2218
2219 dst.n_ra_pages_read.store(src.n_ra_pages_read.load());
2220
2221 dst.n_ra_pages_evicted = src.n_ra_pages_evicted;
2222
2223 dst.n_pages_made_young = src.n_pages_made_young;
2224
2225 dst.n_pages_not_made_young = src.n_pages_not_made_young;
2226
2227 dst.LRU_bytes = src.LRU_bytes;
2228
2229 dst.flush_list_bytes = src.flush_list_bytes;
2230 }
2231
2232 void reset() {
2234
2235 n_pages_read = 0;
2236 n_pages_written = 0;
2237 n_pages_created = 0;
2239 n_ra_pages_read = 0;
2243 LRU_bytes = 0;
2244 flush_list_bytes = 0;
2245 }
2246};
2247
2248/** Statistics of buddy blocks of a given size. */
2250 /** Number of blocks allocated from the buddy system. */
2251 std::atomic<ulint> used;
2252 /** Number of blocks relocated by the buddy system.
2253 Protected by buf_pool zip_free_mutex. */
2254 uint64_t relocated;
2255 /** Total duration of block relocations.
2256 Protected by buf_pool zip_free_mutex. */
2257 std::chrono::steady_clock::duration relocated_duration;
2258
2259 struct snapshot_t {
2261 uint64_t relocated;
2262 std::chrono::steady_clock::duration relocated_duration;
2263 };
2264
2266 return {used.load(), relocated, relocated_duration};
2267 }
2268};
2269
2270/** @brief The buffer pool structure.
2271
2272NOTE! The definition appears here only for other modules of this
2273directory (buf) to see it. Do not use from outside! */
2274
2276 /** @name General fields */
2277 /** @{ */
2278 /** protects (de)allocation of chunks:
2279 - changes to chunks, n_chunks are performed while holding this latch,
2280 - reading buf_pool_should_madvise requires holding this latch for any
2281 buf_pool_t
2282 - writing to buf_pool_should_madvise requires holding these latches
2283 for all buf_pool_t-s */
2285
2286 /** LRU list mutex */
2288
2289 /** free and withdraw list mutex */
2291
2292 /** buddy allocator mutex */
2294
2295 /** zip_hash mutex */
2297
2298 /** Flush state protection mutex */
2300
2301 /** Zip mutex of this buffer pool instance, protects compressed only pages (of
2302 type buf_page_t, not buf_block_t */
2304
2305 /** Array index of this buffer pool instance */
2307
2308 /** Current pool size in bytes */
2310
2311 /** Reserve this much of the buffer pool for "old" blocks */
2313#ifdef UNIV_DEBUG
2314 /** Number of frames allocated from the buffer pool to the buddy system.
2315 Protected by zip_hash_mutex. */
2317#endif
2318
2319 /** Number of buffer pool chunks */
2320 volatile ulint n_chunks;
2321
2322 /** New number of buffer pool chunks */
2324
2325 /** buffer pool chunks */
2327
2328 /** old buffer pool chunks to be freed after resizing buffer pool */
2330
2331 /** Current pool size in pages */
2333
2334 /** Previous pool size in pages */
2336
2337 /** Size in pages of the area which the read-ahead algorithms read
2338 if invoked */
2340
2341 /** Hash table of buf_page_t or buf_block_t file pages, buf_page_in_file() ==
2342 true, indexed by (space_id, offset). page_hash is protected by an array of
2343 mutexes. */
2345
2346 /** Hash table of buf_block_t blocks whose frames are allocated to the zip
2347 buddy system, indexed by block->frame */
2349
2350 /** Number of pending read operations. Accessed atomically */
2351 std::atomic<ulint> n_pend_reads;
2352
2353 /** number of pending decompressions. Accessed atomically. */
2354 std::atomic<ulint> n_pend_unzip;
2355
2356 /** when buf_print_io was last time called. Accesses not protected. */
2357 std::chrono::steady_clock::time_point last_printout_time;
2358
2359 /** Statistics of buddy system, indexed by block size. Protected by zip_free
2360 mutex, except for the used field, which is also accessed atomically */
2362
2363 /** Current statistics */
2365
2366 /** Old statistics */
2368
2369 /** @} */
2370
2371 /** @name Page flushing algorithm fields */
2372
2373 /** @{ */
2374
2375 /** Mutex protecting the flush list access. This mutex protects flush_list,
2376 flush_rbt and bpage::list pointers when the bpage is on flush_list. It also
2377 protects writes to bpage::oldest_modification and flush_list_hp */
2379
2380 /** "Hazard pointer" used during scan of flush_list while doing flush list
2381 batch. Protected by flush_list_mutex */
2383
2384 /** Entry pointer to scan the oldest page except for system temporary */
2386
2387 /** Base node of the modified block list */
2389
2390 /** This is true when a flush of the given type is being initialized.
2391 Protected by flush_state_mutex. */
2393
2394 /** This is the number of pending writes in the given flush type. Protected
2395 by flush_state_mutex. */
2396 std::array<size_t, BUF_FLUSH_N_TYPES> n_flush;
2397
2398 /** This is in the set state when there is no flush batch of the given type
2399 running. Protected by flush_state_mutex. */
2401
2402 /** A red-black tree is used exclusively during recovery to speed up
2403 insertions in the flush_list. This tree contains blocks in order of
2404 oldest_modification LSN and is kept in sync with the flush_list. Each
2405 member of the tree MUST also be on the flush_list. This tree is relevant
2406 only in recovery and is set to NULL once the recovery is over. Protected
2407 by flush_list_mutex */
2409
2410 /** A sequence number used to count the number of buffer blocks removed from
2411 the end of the LRU list; NOTE that this counter may wrap around at 4
2412 billion! A thread is allowed to read this for heuristic purposes without
2413 holding any mutex or latch. For non-heuristic purposes protected by
2414 LRU_list_mutex */
2416
2417 /** Set to false when an LRU scan for free block fails. This flag is used to
2418 avoid repeated scans of LRU list when we know that there is no free block
2419 available in the scan depth for eviction. Set to true whenever we flush a
2420 batch from the buffer pool. Accessed protected by memory barriers. */
2422
2423 /** Page Tracking start LSN. */
2425
2426 /** Check if the page modifications are tracked.
2427 @return true if page modifications are tracked, false otherwise. */
2428 bool is_tracking() { return track_page_lsn != LSN_MAX; }
2429
2430 /** Maximum LSN for which write io has already started. */
2432
2433 /** @} */
2434
2435 /** @name LRU replacement algorithm fields */
2436 /** @{ */
2437
2438 /** Base node of the free block list */
2440
2441 /** base node of the withdraw block list. It is only used during shrinking
2442 buffer pool size, not to reuse the blocks will be removed. Protected by
2443 free_list_mutex */
2445
2446 /** Target length of withdraw block list, when withdrawing */
2448
2449 /** "hazard pointer" used during scan of LRU while doing
2450 LRU list batch. Protected by buf_pool::LRU_list_mutex */
2452
2453 /** Iterator used to scan the LRU list when searching for
2454 replaceable victim. Protected by buf_pool::LRU_list_mutex. */
2456
2457 /** Iterator used to scan the LRU list when searching for
2458 single page flushing victim. Protected by buf_pool::LRU_list_mutex. */
2460
2461 /** Base node of the LRU list */
2463
2464 /** Pointer to the about LRU_old_ratio/BUF_LRU_OLD_RATIO_DIV oldest blocks in
2465 the LRU list; NULL if LRU length less than BUF_LRU_OLD_MIN_LEN; NOTE: when
2466 LRU_old != NULL, its length should always equal LRU_old_len */
2468
2469 /** Length of the LRU list from the block to which LRU_old points onward,
2470 including that block; see buf0lru.cc for the restrictions on this value; 0
2471 if LRU_old == NULL; NOTE: LRU_old_len must be adjusted whenever LRU_old
2472 shrinks or grows! */
2474
2475 /** Base node of the unzip_LRU list. The list is protected by the
2476 LRU_list_mutex. */
2477 UT_LIST_BASE_NODE_T(buf_block_t, unzip_LRU) unzip_LRU;
2478
2479 /** @} */
2480 /** @name Buddy allocator fields
2481 The buddy allocator is used for allocating compressed page
2482 frames and buf_page_t descriptors of blocks that exist
2483 in the buffer pool only in compressed form. */
2484 /** @{ */
2485#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
2486 /** Unmodified compressed pages */
2488#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
2489
2490 /** Buddy free lists */
2492
2493 /** Sentinel records for buffer pool watches. Scanning the array is protected
2494 by taking all page_hash latches in X. Updating or reading an individual
2495 watch page is protected by a corresponding individual page_hash latch. */
2497
2498 /** A wrapper for buf_pool_t::allocator.alocate_large which also advices the
2499 OS that this chunk should not be dumped to a core file if that was requested.
2500 Emits a warning to the log and disables @@global.core_file if advising was
2501 requested but could not be performed, but still return true as the allocation
2502 itself succeeded.
2503 @param[in] mem_size number of bytes to allocate
2504 @param[in,out] chunk mem and mem_pfx fields of this chunk will be updated
2505 to contain information about allocated memory region
2506 @return true iff allocated successfully */
2507 bool allocate_chunk(ulonglong mem_size, buf_chunk_t *chunk);
2508
2509 /** A wrapper for buf_pool_t::allocator.deallocate_large which also advices
2510 the OS that this chunk can be dumped to a core file.
2511 Emits a warning to the log and disables @@global.core_file if advising was
2512 requested but could not be performed.
2513 @param[in] chunk mem and mem_pfx fields of this chunk will be used to
2514 locate the memory region to free */
2515 void deallocate_chunk(buf_chunk_t *chunk);
2516
2517 /** Advices the OS that all chunks in this buffer pool instance can be dumped
2518 to a core file.
2519 Emits a warning to the log if could not succeed.
2520 @return true iff succeeded, false if no OS support or failed */
2521 bool madvise_dump();
2522
2523 /** Advices the OS that all chunks in this buffer pool instance should not
2524 be dumped to a core file.
2525 Emits a warning to the log if could not succeed.
2526 @return true iff succeeded, false if no OS support or failed */
2527 bool madvise_dont_dump();
2528
2529 /** Checks if the batch is running, which is basically equivalent to
2530 !os_event_is_set(no_flush[type]) if you hold flush_state_mutex.
2531 It is used as source of truth to know when to set or reset this event.
2532 Caller should hold flush_state_mutex.
2533 @param[in] flush_type The type of the flush we are interested in
2534 @return Should no_flush[type] be in the "unset" state? */
2537 return init_flush[flush_type] || 0 < n_flush[flush_type];
2538 }
2539
2540#ifndef UNIV_HOTBACKUP
2541 /** Executes change() which modifies fields protected by flush_state_mutex.
2542 If it caused a change to is_flushing(flush_type) then it sets or resets the
2543 no_flush[flush_type] to keep it in sync.
2544 @param[in] flush_type The type of the flush this change of state concerns
2545 @param[in] change A callback to execute within flush_state_mutex
2546 */
2547 template <typename F>
2550 const bool was_set = !is_flushing(flush_type);
2552 std::forward<F>(change)();
2553 const bool should_be_set = !is_flushing(flush_type);
2554 if (was_set && !should_be_set) {
2556 } else if (!was_set && should_be_set) {
2558 }
2559 ut_ad(should_be_set == os_event_is_set(no_flush[flush_type]));
2561 }
2562#endif /*! UNIV_HOTBACKUP */
2563
2564 static_assert(BUF_BUDDY_LOW <= UNIV_ZIP_SIZE_MIN,
2565 "BUF_BUDDY_LOW > UNIV_ZIP_SIZE_MIN");
2566 /** @} */
2567};
2568
2569/** Print the given buf_pool_t object.
2570@param[in,out] out the output stream
2571@param[in] buf_pool the buf_pool_t object to be printed
2572@return the output stream */
2573std::ostream &operator<<(std::ostream &out, const buf_pool_t &buf_pool);
2574
2575/** @name Accessors for buffer pool mutexes
2576Use these instead of accessing buffer pool mutexes directly. */
2577/** @{ */
2578
2579#ifndef UNIV_HOTBACKUP
2580/** Test if flush list mutex is owned. */
2581#define buf_flush_list_mutex_own(b) mutex_own(&(b)->flush_list_mutex)
2582
2583/** Acquire the flush list mutex. */
2584#define buf_flush_list_mutex_enter(b) \
2585 do { \
2586 mutex_enter(&(b)->flush_list_mutex); \
2587 } while (0)
2588/** Release the flush list mutex. */
2589#define buf_flush_list_mutex_exit(b) \
2590 do { \
2591 mutex_exit(&(b)->flush_list_mutex); \
2592 } while (0)
2593/** Acquire the block->mutex. */
2594#define buf_page_mutex_enter(b) \
2595 do { \
2596 mutex_enter(&(b)->mutex); \
2597 } while (0)
2598
2599/** Release the block->mutex. */
2600#define buf_page_mutex_exit(b) \
2601 do { \
2602 (b)->mutex.exit(); \
2603 } while (0)
2604
2605/** Get appropriate page_hash_lock. */
2607 const page_id_t page_id) {
2608 return hash_get_lock(buf_pool->page_hash, page_id.hash());
2609}
2610
2611/** If not appropriate page_hash_lock, relock until appropriate. */
2613 const buf_pool_t *buf_pool,
2614 const page_id_t page_id) {
2615 return hash_lock_s_confirm(hash_lock, buf_pool->page_hash, page_id.hash());
2616}
2617
2619 buf_pool_t *buf_pool,
2620 const page_id_t &page_id) {
2621 return hash_lock_x_confirm(hash_lock, buf_pool->page_hash, page_id.hash());
2622}
2623#endif /* !UNIV_HOTBACKUP */
2624
2625#if defined(UNIV_DEBUG) && !defined(UNIV_HOTBACKUP)
2626/** Test if page_hash lock is held in s-mode. */
2628 const buf_page_t *bpage) {
2629 return rw_lock_own(buf_page_hash_lock_get(buf_pool, bpage->id), RW_LOCK_S);
2630}
2631
2632/** Test if page_hash lock is held in x-mode. */
2634 const buf_page_t *bpage) {
2635 return rw_lock_own(buf_page_hash_lock_get((buf_pool), (bpage)->id),
2636 RW_LOCK_X);
2637}
2638
2639/** Test if page_hash lock is held in x or s-mode. */
2640inline bool buf_page_hash_lock_held_s_or_x(const buf_pool_t *buf_pool,
2641 const buf_page_t *bpage) {
2642 return buf_page_hash_lock_held_s(buf_pool, bpage) ||
2643 buf_page_hash_lock_held_x(buf_pool, bpage);
2644}
2645
2646inline bool buf_block_hash_lock_held_s(const buf_pool_t *buf_pool,
2647 const buf_block_t *block) {
2648 return buf_page_hash_lock_held_s(buf_pool, &block->page);
2649}
2650
2651inline bool buf_block_hash_lock_held_x(const buf_pool_t *buf_pool,
2652 const buf_block_t *block) {
2653 return buf_page_hash_lock_held_x(buf_pool, &block->page);
2654}
2655
2657 const buf_block_t *block) {
2658 return buf_page_hash_lock_held_s_or_x(buf_pool, &block->page);
2659}
2660#else /* UNIV_DEBUG && !UNIV_HOTBACKUP */
2661#define buf_page_hash_lock_held_s(b, p) (true)
2662#define buf_page_hash_lock_held_x(b, p) (true)
2663#define buf_page_hash_lock_held_s_or_x(b, p) (true)
2664#define buf_block_hash_lock_held_s(b, p) (true)
2665#define buf_block_hash_lock_held_x(b, p) (true)
2666#define buf_block_hash_lock_held_s_or_x(b, p) (true)
2667#endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
2668
2669/** @} */
2670
2671/**********************************************************************
2672Let us list the consistency conditions for different control block states.
2673
2674NOT_USED: is in free list, not in LRU list, not in flush list, nor
2675 page hash table
2676READY_FOR_USE: is not in free list, LRU list, or flush list, nor page
2677 hash table
2678MEMORY: is not in free list, LRU list, or flush list, nor page
2679 hash table
2680FILE_PAGE: space and offset are defined, is in page hash table
2681 if io_fix == BUF_IO_WRITE,
2682 pool: no_flush[flush_type] is in reset state,
2683 pool: n_flush[flush_type] > 0
2684
2685 (1) if buf_fix_count == 0, then
2686 is in LRU list, not in free list
2687 is in flush list,
2688 if and only if oldest_modification > 0
2689 is x-locked,
2690 if and only if io_fix == BUF_IO_READ
2691 is s-locked,
2692 if and only if io_fix == BUF_IO_WRITE
2693
2694 (2) if buf_fix_count > 0, then
2695 is not in LRU list, not in free list
2696 is in flush list,
2697 if and only if oldest_modification > 0
2698 if io_fix == BUF_IO_READ,
2699 is x-locked
2700 if io_fix == BUF_IO_WRITE,
2701 is s-locked
2702
2703State transitions:
2704
2705NOT_USED => READY_FOR_USE
2706READY_FOR_USE => MEMORY
2707READY_FOR_USE => FILE_PAGE
2708MEMORY => NOT_USED
2709FILE_PAGE => NOT_USED NOTE: This transition is allowed if and only if
2710 (1) buf_fix_count == 0,
2711 (2) oldest_modification == 0, and
2712 (3) io_fix == 0.
2713*/
2714
2715#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
2716#ifndef UNIV_HOTBACKUP
2717/** Functor to validate the LRU list. */
2719 void operator()(const buf_page_t *elem) const { ut_a(elem->in_LRU_list); }
2720
2721 static void validate(const buf_pool_t *buf_pool) {
2722 CheckInLRUList check;
2723 ut_list_validate(buf_pool->LRU, check);
2724 }
2725};
2726
2727/** Functor to validate the LRU list. */
2729 void operator()(const buf_page_t *elem) const { ut_a(elem->in_free_list); }
2730
2731 static void validate(const buf_pool_t *buf_pool) {
2732 CheckInFreeList check;
2733 ut_list_validate(buf_pool->free, check);
2734 }
2735};
2736
2738 void operator()(const buf_block_t *elem) const {
2739 ut_a(elem->page.in_LRU_list);
2740 ut_a(elem->in_unzip_LRU_list);
2741 }
2742
2743 static void validate(const buf_pool_t *buf_pool) {
2745 ut_list_validate(buf_pool->unzip_LRU, check);
2746 }
2747};
2748#endif /* !UNIV_HOTBACKUP */
2749#endif /* UNIV_DEBUG || defined UNIV_BUF_DEBUG */
2750
2751#ifndef UNIV_HOTBACKUP
2752/** Prepare a page before adding to the free list.
2753@param[in,out] bpage Buffer page to prepare for freeing. */
2754inline void buf_page_prepare_for_free(buf_page_t *bpage) noexcept {
2755 bpage->reset_page_id();
2756}
2757#endif /* !UNIV_HOTBACKUP */
2758
2759/** Gets the compressed page descriptor corresponding to an uncompressed
2760page if applicable.
2761@param[in] block Get the zip descriptor for this block. */
2763 return block->get_page_zip();
2764}
2765
2766/** Gets the compressed page descriptor corresponding to an uncompressed
2767page if applicable. Const version.
2768@param[in] block Get the zip descriptor for this block.
2769@return page descriptor or nullptr. */
2771 const buf_block_t *block) noexcept {
2772 return block->get_page_zip();
2773}
2774
2775inline bool buf_page_in_memory(const buf_page_t *bpage) {
2776 switch (buf_page_get_state(bpage)) {
2777 case BUF_BLOCK_MEMORY:
2778 return true;
2779 default:
2780 break;
2781 }
2782 return false;
2783}
2784
2785/** Verify the page contained by the block. If there is page type
2786mismatch then reset it to expected page type. Data files created
2787before MySQL 5.7 GA may contain garbage in the FIL_PAGE_TYPE field.
2788@param[in,out] block block that may possibly have invalid
2789 FIL_PAGE_TYPE
2790@param[in] type Expected page type
2791@param[in,out] mtr Mini-transaction */
2794 mtr_t &mtr) {
2795 byte *page = block.frame;
2796 page_type_t page_type = fil_page_get_type(page);
2797 if (page_type != type) {
2798 const page_id_t &page_id = block.page.id;
2799 fil_page_reset_type(page_id, page, type, &mtr);
2800 }
2801}
2802
2803#include "buf0buf.ic"
2804
2805#endif /* !buf0buf_h */
uint32_t space_id_t
Tablespace identifier.
Definition: api0api.h:48
uint32_t page_no_t
Page number.
Definition: api0api.h:46
Cache_hint
Definition: buf0buf.h:94
@ MAKE_YOUNG
Move the block to the start of the LRU list if there is a danger that the block would drift out of th...
@ KEEP_OLD
Preserve the current LRU position of the block.
static void buf_block_dbg_add_level(buf_block_t *block, latch_level_t level)
Adds latch level info for the rw-lock protecting the buffer frame.
bool buf_block_hash_lock_held_x(const buf_pool_t *buf_pool, const buf_block_t *block)
Definition: buf0buf.h:2651
static buf_block_t * buf_page_get_block(buf_page_t *bpage)
Gets the buf_block_t handle of a buffered file block if an uncompressed page frame exists,...
static bool buf_page_belongs_to_unzip_LRU(const buf_page_t *bpage)
Determines if a block should be on unzip_LRU list.
static bool buf_page_can_relocate(const buf_page_t *bpage)
Determine if a buffer block can be relocated in memory.
buf_page_t * buf_page_get_zip(const page_id_t &page_id, const page_size_t &page_size)
Get read access to a compressed page (usually of type FIL_PAGE_TYPE_ZBLOB or FIL_PAGE_TYPE_ZBLOB2).
Definition: buf0buf.cc:3248
dberr_t buf_pool_init(ulint total_size, ulint n_instances)
Creates the buffer pool.
Definition: buf0buf.cc:1445
void buf_page_print(const byte *read_buf, const page_size_t &page_size, ulint flags)
Prints a page to stderr.
Definition: buf0buf.cc:550
ulint buf_get_n_pending_read_ios(void)
Returns the number of pending buf pool read ios.
Definition: buf0buf.cc:6513
bool buf_page_hash_lock_held_s_or_x(const buf_pool_t *buf_pool, const buf_page_t *bpage)
Test if page_hash lock is held in x or s-mode.
Definition: buf0buf.h:2640
static ulint buf_pool_get_curr_size(void)
Gets the current size of buffer buf_pool in bytes.
ulint buf_page_hash_lock_held_x(const buf_pool_t *buf_pool, const buf_page_t *bpage)
Test if page_hash lock is held in x-mode.
Definition: buf0buf.h:2633
void buf_page_force_evict(const page_id_t &page_id, const page_size_t &page_size, const bool dirty_is_ok=true) noexcept
Evict a page from the buffer pool.
Definition: buf0buf.cc:5365
bool buf_validate(void)
Validates the buffer pool data structure.
Definition: buf0buf.cc:6299
void buf_flush_update_zip_checksum(buf_frame_t *page, ulint size, lsn_t lsn, bool skip_lsn_check)
Calculate the checksum of a page from compressed table and update the page.
Definition: buf0flu.cc:941
static enum buf_page_state buf_block_get_state(const buf_block_t *block)
Gets the state of a block.
static void buf_page_set_state(buf_page_t *bpage, enum buf_page_state state)
Sets the state of a block.
buf_block_t * buf_block_from_ahi(const byte *ptr)
Get a buffer block from an adaptive hash index pointer.
Definition: buf0buf.cc:3458
bool buf_page_io_complete(buf_page_t *bpage, bool evict, IORequest *type=nullptr, fil_node_t *node=nullptr)
Completes an asynchronous read or write request of a file page to or from the buffer pool.
Definition: buf0buf.cc:5741
lsn_t buf_pool_get_oldest_modification_lwm(void)
Gets a safe low watermark for oldest_modification.
Definition: buf0buf.cc:423
bool buf_frame_will_withdrawn(buf_pool_t *buf_pool, const byte *ptr)
Determines if a frame is intended to be withdrawn.
Definition: buf0buf.cc:1821
void buf_pool_update_madvise()
Checks if innobase_should_madvise_buf_pool() value has changed since we've last check and if so,...
Definition: buf0buf.cc:955
ulonglong buf_pool_adjust_chunk_unit(ulonglong size)
Adjust the proposed chunk unit size so that it satisfies all invariants.
Definition: buf0buf.cc:2090
static buf_page_t * buf_page_hash_get_locked(buf_pool_t *buf_pool, const page_id_t &page_id, rw_lock_t **lock, ulint lock_mode, bool watch=false)
Returns the control block of a file page, NULL if not found.
static enum buf_io_fix buf_page_get_io_fix(const buf_page_t *bpage)
Gets the io_fix state of a block.
static void buf_page_set_io_fix(buf_page_t *bpage, enum buf_io_fix io_fix)
Sets the io_fix state of a block.
buf_page_t * buf_page_set_file_page_was_freed(const page_id_t &page_id)
Sets file_page_was_freed true if the page is found in the buffer pool.
Definition: buf0buf.cc:3161
static void buf_page_set_accessed(buf_page_t *bpage)
Flag a block accessed.
static void buf_block_set_io_fix(buf_block_t *block, enum buf_io_fix io_fix)
Sets the io_fix state of a block.
void buf_page_make_young(buf_page_t *bpage)
Moves a page to the start of the buffer pool LRU list.
Definition: buf0buf.cc:3115
buf_pool_t * buf_pool_ptr
The buffer pools of the database.
Definition: buf0buf.cc:304
void buf_pool_clear_hash_index(void)
Clears the adaptive hash index on all pages in the buffer pool.
Definition: buf0buf.cc:2653
void buf_get_total_stat(buf_pool_stat_t *tot_stat)
Get total buffer pool statistics.
Definition: buf0buf.cc:495
static buf_pool_t * buf_pool_from_bpage(const buf_page_t *bpage)
Returns the buffer pool instance given a page instance.
static buf_flush_t buf_page_get_flush_type(const buf_page_t *bpage)
Get the flush type of a page.
static ulint buf_page_get_freed_page_clock(const buf_page_t *bpage)
Reads the freed_page_clock of a buffer block.
ulint buf_page_hash_lock_held_s(const buf_pool_t *buf_pool, const buf_page_t *bpage)
Test if page_hash lock is held in s-mode.
Definition: buf0buf.h:2627
static std::chrono::steady_clock::time_point buf_page_is_accessed(const buf_page_t *bpage)
Determine the time of first access of a block in the buffer pool.
constexpr ulint MAX_PAGE_HASH_LOCKS
The maximum number of page_hash locks.
Definition: buf0buf.h:115
bool buf_page_optimistic_get(ulint rw_latch, buf_block_t *block, uint64_t modify_clock, Page_fetch fetch_mode, const char *file, ulint line, mtr_t *mtr)
This is the general function used to get optimistic access to a database page.
Definition: buf0buf.cc:4447
void buf_pool_watch_unset(const page_id_t &page_id)
Stop watching if the page has been read in.
Definition: buf0buf.cc:3064
static byte * buf_frame_copy(byte *buf, const buf_frame_t *frame)
Copies contents of a buffer frame to a given buffer.
bool buf_pool_watch_occurred(const page_id_t &page_id)
Check if the page has been read in.
Definition: buf0buf.cc:3091
buf_page_t * buf_page_init_for_read(ulint mode, const page_id_t &page_id, const page_size_t &page_size, bool unzip)
Inits a page for read to the buffer buf_pool.
Definition: buf0buf.cc:4811
static bool buf_page_peek_if_young(const buf_page_t *bpage)
Tells, for heuristics, if a block is still close enough to the MRU end of the LRU list meaning that i...
bool buf_page_free_stale(buf_pool_t *buf_pool, buf_page_t *bpage) noexcept
Free a stale page.
Definition: buf0buf.cc:5332
buf_page_t * buf_page_hash_get_x_locked(buf_pool_t *b, const page_id_t &page_id, rw_lock_t **l)
Definition: buf0buf.h:1036
static void buf_page_set_old(buf_page_t *bpage, bool old)
Flag a block old.
size_t buf_pool_pending_io_reads_count()
Computes number of pending I/O read operations for the buffer pool.
Definition: buf0buf.cc:6873
size_t buf_pool_pending_io_writes_count()
Computes number of pending I/O write operations for the buffer pool.
Definition: buf0buf.cc:6881
void buf_read_page_handle_error(buf_page_t *bpage)
Unfixes the page, unlatches the page, removes it from page_hash and removes it from LRU.
Definition: buf0buf.cc:5297
static ulint buf_block_unfix(buf_page_t *bpage)
Decrements the bufferfix count.
static void buf_page_set_flush_type(buf_page_t *bpage, buf_flush_t flush_type)
Set the flush type of a page.
void buf_page_free_stale_during_write(buf_page_t *bpage, bool owns_sx_lock=false) noexcept
Free a stale page that is being written.
Definition: buf0buf.cc:5473
static buf_pool_t * buf_pool_from_block(const buf_block_t *block)
Returns the buffer pool instance given a block instance.
bool buf_is_block_in_instance(const buf_pool_t *buf_pool, const buf_block_t *ptr)
Find out if a block pointer points into one of currently used chunks of the buffer pool.
Definition: buf0buf.cc:3494
rw_lock_t * buf_page_hash_lock_get(const buf_pool_t *buf_pool, const page_id_t page_id)
Get appropriate page_hash_lock.
Definition: buf0buf.h:2606
std::ostream & operator<<(std::ostream &out, const buf_pool_t &buf_pool)
Print the given buf_pool_t object.
Definition: buf0buf.cc:6954
static BPageMutex * buf_page_get_mutex(const buf_page_t *bpage)
Gets the mutex of a block.
static uint64_t buf_pool_hash_zip_frame(void *ptr)
Compute the hash value for blocks in buf_pool->zip_hash.
Definition: buf0buf.h:2016
static buf_frame_t * buf_block_get_frame(const buf_block_t *block)
Gets a pointer to the memory frame of a block.
void buf_block_reset_page_type_on_mismatch(buf_block_t &block, page_type_t type, mtr_t &mtr)
Verify the page contained by the block.
Definition: buf0buf.h:2792
void buf_pool_free_all()
Frees the buffer pool at shutdown.
Definition: buf0buf.cc:6988
static enum buf_page_state buf_page_get_state(const buf_page_t *bpage)
Gets the state of a block.
static ulint buf_pool_index(const buf_pool_t *buf_pool)
Calculates the index of a buffer pool to the buf_pool[] array.
buf_block_t * buf_page_get_gen(const page_id_t &page_id, const page_size_t &page_size, ulint rw_latch, buf_block_t *guess, Page_fetch mode, ut::Location location, mtr_t *mtr, bool dirty_with_no_latch=false)
This is the general function used to get access to a database page.
Definition: buf0buf.cc:4380
ulint buf_get_free_list_len(void)
Gets the current length of the free list of buffer blocks.
rw_lock_t * buf_page_hash_lock_x_confirm(rw_lock_t *hash_lock, buf_pool_t *buf_pool, const page_id_t &page_id)
Definition: buf0buf.h:2618
static lsn_t buf_page_get_newest_modification(const buf_page_t *bpage)
Gets the youngest modification log sequence number for a frame.
ulint buf_get_latched_pages_number(void)
Returns the number of latched pages in the buffer pool.
Definition: buf0buf.cc:6494
static buf_page_t * buf_page_hash_get_low(buf_pool_t *buf_pool, const page_id_t &page_id)
Returns the control block of a file page, NULL if not found.
static ulint buf_block_fix(buf_page_t *bpage)
Increments the bufferfix count.
constexpr ulint MAX_BUFFER_POOLS
The maximum number of buffer pools that can be defined.
Definition: buf0buf.h:109
static buf_block_t * buf_block_hash_get_locked(buf_pool_t *buf_pool, const page_id_t &page_id, rw_lock_t **lock, ulint lock_mode)
Returns the control block of a file page, NULL if not found.
void buf_print_io(FILE *file)
Prints info of the buffer i/o.
Definition: buf0buf.cc:6790
static buf_pool_t * buf_pool_from_array(ulint index)
Returns the buffer pool instance given its array index.
static void buf_page_release_latch(buf_block_t *block, ulint rw_latch)
Releases a latch, if specified.
void buf_stats_get_pool_info(buf_pool_t *buf_pool, ulint pool_id, buf_pool_info_t *all_pool_info)
Collect buffer pool stats information for a buffer pool.
Definition: buf0buf.cc:6597
void buf_resize_thread()
This is the thread for resizing buffer pool.
Definition: buf0buf.cc:2626
void buf_print(void)
Prints info of the buffer pool data structure.
Definition: buf0buf.cc:6400
const buf_block_t * buf_page_try_get(const page_id_t &page_id, ut::Location location, mtr_t *mtr)
Given a tablespace id and page number tries to get that page.
Definition: buf0buf.cc:4640
static ulint buf_block_get_freed_page_clock(const buf_block_t *block)
Reads the freed_page_clock of a buffer block.
static void buf_ptr_get_fsp_addr(const void *ptr, space_id_t *space, fil_addr_t *addr)
Gets the space id, page offset, and byte offset within page of a pointer pointing to a buffer frame c...
static bool buf_page_peek_if_too_old(const buf_page_t *bpage)
Recommends a move of a block to the start of the LRU list if there is danger of dropping from the buf...
bool buf_block_will_withdrawn(buf_pool_t *buf_pool, const buf_block_t *block)
Determines if a block is intended to be withdrawn.
Definition: buf0buf.cc:1800
void buf_page_make_old(buf_page_t *bpage)
Moved a page to the end of the buffer pool LRU list so that it can be flushed out at the earliest.
Definition: buf0buf.cc:3127
lsn_t buf_pool_get_oldest_modification_approx(void)
Gets the smallest oldest_modification lsn among all of the earliest added pages in flush lists.
Definition: buf0buf.cc:370
buf_block_t * buf_block_hash_get_s_locked(buf_pool_t *b, const page_id_t &page_id, rw_lock_t **l)
Definition: buf0buf.h:1049
void buf_must_be_all_freed(void)
Assert that all file pages in the buffer are in a replaceable state.
Definition: buf0buf.cc:6863
Page_fetch
Definition: buf0buf.h:58
@ NO_LATCH
get and bufferfix, but set no latch; we have separated this case, because it is error-prone programmi...
@ NORMAL
Get always.
@ IF_IN_POOL
get if in pool
@ POSSIBLY_FREED_NO_READ_AHEAD
Like Page_fetch::POSSIBLY_FREED, but do not initiate read ahead.
@ IF_IN_POOL_OR_WATCH
Get the page only if it's in the buffer pool, if not then set a watch on the page.
@ PEEK_IF_IN_POOL
get if in pool, do not make the block young in the LRU list
@ POSSIBLY_FREED
Like Page_fetch::NORMAL, but do not mind if the file page has been freed.
@ SCAN
Same as NORMAL, but hint that the fetch is part of a large scan.
rw_lock_t * buf_page_hash_lock_s_confirm(rw_lock_t *hash_lock, const buf_pool_t *buf_pool, const page_id_t page_id)
If not appropriate page_hash_lock, relock until appropriate.
Definition: buf0buf.h:2612
static void buf_page_unset_sticky(buf_page_t *bpage)
Removes stickiness of a block.
buf_page_print_flags
Definition: buf0buf.h:647
@ BUF_PAGE_PRINT_NO_FULL
Do not print the full page dump.
Definition: buf0buf.h:651
@ BUF_PAGE_PRINT_NO_CRASH
Do not crash at the end of buf_page_print().
Definition: buf0buf.h:649
buf_page_t * buf_page_get_also_watch(buf_pool_t *b, const page_id_t &page_id)
Definition: buf0buf.h:1044
bool buf_pool_watch_is_sentinel(const buf_pool_t *buf_pool, const buf_page_t *bpage)
Determine if a block is a sentinel for a buffer pool watch.
Definition: buf0buf.cc:2909
buf_block_t * buf_page_get_with_no_latch(const page_id_t &id, const page_size_t &size, ut::Location location, mtr_t *mtr)
Use these macros to bufferfix a page with no latching.
Definition: buf0buf.h:452
constexpr ulint MAX_BUFFER_POOLS_BITS
Number of bits to representing a buffer pool ID.
Definition: buf0buf.h:106
bool buf_block_hash_lock_held_s(const buf_pool_t *buf_pool, const buf_block_t *block)
Definition: buf0buf.h:2646
buf_block_t * buf_block_hash_get(buf_pool_t *b, const page_id_t &page_id)
Definition: buf0buf.h:1059
bool buf_page_in_memory(const buf_page_t *bpage)
Definition: buf0buf.h:2775
bool is_buffer_pool_resize_in_progress()
Definition: buf0buf.cc:6998
static bool buf_page_peek(const page_id_t &page_id)
Returns true if the page can be found in the buffer pool hash table.
bool buf_page_get_known_nowait(ulint rw_latch, buf_block_t *block, Cache_hint hint, const char *file, ulint line, mtr_t *mtr)
This is used to get access to a known database page, when no waiting can be done.
Definition: buf0buf.cc:4547
buf_page_t * buf_page_hash_get_s_locked(buf_pool_t *b, const page_id_t &page_id, rw_lock_t **l)
Definition: buf0buf.h:1031
static void buf_page_release_zip(buf_page_t *bpage)
Releases a compressed-only page acquired with buf_page_get_zip().
static void buf_block_free(buf_block_t *block)
Frees a buffer block which does not contain a file page.
static uint64_t buf_pool_hash_zip(buf_block_t *b)
Definition: buf0buf.h:2020
buf_page_t * buf_page_hash_get(buf_pool_t *b, const page_id_t &page_id)
Definition: buf0buf.h:1041
static void buf_block_set_state(buf_block_t *block, enum buf_page_state state)
Sets the state of a block.
constexpr uint32_t BUF_PAGE_STATE_BITS
The common buffer control block structure for compressed and uncompressed frames.
Definition: buf0buf.h:1143
static void buf_block_buf_fix_inc_func(ut::Location location, buf_block_t *block)
Increments the bufferfix count.
static buf_pool_t * buf_pool_get(const page_id_t &page_id)
Returns the buffer pool instance given a page id.
static ulint buf_get_withdraw_depth(buf_pool_t *buf_pool)
Return how many more pages must be added to the withdraw list to reach the withdraw target of the cur...
buf_page_state
States of a control block.
Definition: buf0buf.h:131
@ BUF_BLOCK_NOT_USED
Is in the free list; must be after the BUF_BLOCK_ZIP_ constants for compressed-only pages.
Definition: buf0buf.h:141
@ BUF_BLOCK_ZIP_PAGE
Contains a clean compressed page.
Definition: buf0buf.h:135
@ BUF_BLOCK_REMOVE_HASH
Hash index should be removed before putting to the free list.
Definition: buf0buf.h:153
@ BUF_BLOCK_MEMORY
Contains some main memory object.
Definition: buf0buf.h:150
@ BUF_BLOCK_ZIP_DIRTY
Contains a compressed page that is in the buf_pool->flush_list.
Definition: buf0buf.h:137
@ BUF_BLOCK_POOL_WATCH
A sentinel for the buffer pool watch, element of buf_pool->watch[].
Definition: buf0buf.h:133
@ BUF_BLOCK_READY_FOR_USE
When buf_LRU_get_free_block returns a block, it is in this state.
Definition: buf0buf.h:144
@ BUF_BLOCK_FILE_PAGE
Contains a buffered file page.
Definition: buf0buf.h:147
void buf_refresh_io_stats_all()
Refresh the statistics used to print per-second averages.
Definition: buf0buf.cc:6852
static ulint buf_pool_get_n_pages(void)
Gets the current size of buffer buf_pool in frames.
buf_block_t * buf_block_alloc(buf_pool_t *buf_pool)
Allocates a buffer block.
Definition: buf0buf.cc:521
static buf_block_t * buf_get_nth_chunk_block(const buf_pool_t *buf_pool, ulint n, ulint *chunk_size)
Get the nth chunk's buffer block in the specified buffer pool.
static ulint buf_pool_size_align(ulint size)
Calculate aligned buffer pool size based on srv_buf_pool_chunk_unit, if needed.
buf_block_t * buf_page_get(const page_id_t &id, const page_size_t &size, ulint latch, ut::Location location, mtr_t *mtr)
NOTE! The following macros should be used instead of buf_page_get_gen, to improve debugging.
Definition: buf0buf.h:441
static void buf_block_modify_clock_inc(buf_block_t *block)
Increment the modify clock.
static bool buf_page_in_file(const buf_page_t *bpage)
Determines if a block is mapped to a tablespace.
page_zip_des_t * buf_block_get_page_zip(buf_block_t *block) noexcept
Gets the compressed page descriptor corresponding to an uncompressed page if applicable.
Definition: buf0buf.h:2762
void buf_pool_invalidate(void)
Invalidates the file pages in the buffer pool when an archive recovery is completed.
Definition: buf0buf.cc:6105
void buf_page_free_descriptor(buf_page_t *bpage)
Free a buf_page_t descriptor.
Definition: buf0buf.cc:1360
buf_block_t * buf_page_create(const page_id_t &page_id, const page_size_t &page_size, rw_lock_type_t rw_latch, mtr_t *mtr)
Initializes a page to the buffer buf_pool.
Definition: buf0buf.cc:5017
double buf_get_modified_ratio_pct(void)
Return the ratio in percents of modified pages in the buffer pool / database pages in the buffer pool...
Definition: buf0buf.cc:6527
static buf_frame_t * buf_frame_align(byte *ptr)
void buf_page_prepare_for_free(buf_page_t *bpage) noexcept
Prepare a page before adding to the free list.
Definition: buf0buf.h:2754
buf_block_t * buf_pool_contains_zip(buf_pool_t *buf_pool, const void *data)
Finds a block in the buffer pool that points to a given compressed page.
Definition: buf0buf.cc:1112
static void buf_block_set_file_page(buf_block_t *block, const page_id_t &page_id)
Map a block to a file page.
void buf_pool_wait_for_no_pending_io()
Waits until there are no pending I/O operations for the buffer pool.
Definition: buf0buf.cc:6893
static buf_page_t * buf_page_alloc_descriptor(void)
Allocates a buf_page_t descriptor.
static enum buf_io_fix buf_block_get_io_fix(const buf_block_t *block)
Gets the io_fix state of a block.
bool buf_block_hash_lock_held_s_or_x(const buf_pool_t *buf_pool, const buf_block_t *block)
Definition: buf0buf.h:2656
bool buf_block_state_valid(buf_block_t *block)
Check if a buf_block_t object is in a valid state.
Definition: buf0buf.h:2009
buf_block_t * buf_block_hash_get_x_locked(buf_pool_t *b, const page_id_t &page_id, rw_lock_t **l)
Definition: buf0buf.h:1054
buf_page_t * buf_page_reset_file_page_was_freed(const page_id_t &page_id)
Sets file_page_was_freed false if the page is found in the buffer pool.
Definition: buf0buf.cc:3187
static bool buf_page_is_old(const buf_page_t *bpage)
Determine if a block has been flagged old.
void buf_block_buf_fix_inc(buf_block_t *b, ut::Location l)
Increments the bufferfix count.
Definition: buf0buf.h:603
static void buf_page_set_sticky(buf_page_t *bpage)
Makes a block sticky.
void buf_get_total_list_size_in_bytes(buf_pools_list_size_t *buf_pools_list_size)
Get total list size in bytes from all buffer pools.
Definition: buf0buf.cc:474
void buf_get_total_list_len(ulint *LRU_len, ulint *free_len, ulint *flush_list_len)
Get total buffer pool statistics.
Definition: buf0buf.cc:454
bool buf_zip_decompress(buf_block_t *block, bool check)
Decompress a block.
Definition: buf0buf.cc:3390
The database buffer buf_pool.
The database buffer pool global types for the directory.
byte buf_frame_t
A buffer frame.
Definition: buf0types.h:62
BPageMutex BufPoolZipMutex
Definition: buf0types.h:201
constexpr uint32_t BUF_BUDDY_LOW
Smallest buddy page size.
Definition: buf0types.h:183
constexpr uint32_t BUF_BUDDY_SIZES_MAX
Maximum number of buddy sizes based on the max page size.
Definition: buf0types.h:189
buf_flush_t
Flags for flush types.
Definition: buf0types.h:68
@ BUF_FLUSH_N_TYPES
Index of last element + 1
Definition: buf0types.h:79
ib_mutex_t BufListMutex
Definition: buf0types.h:199
ib_bpmutex_t BPageMutex
Definition: buf0types.h:198
buf_io_fix
Flags for io_fix types.
Definition: buf0types.h:99
@ BUF_IO_NONE
no pending I/O
Definition: buf0types.h:101
@ BUF_IO_WRITE
write pending
Definition: buf0types.h:107
@ BUF_IO_READ
read pending
Definition: buf0types.h:104
@ BUF_IO_PIN
disallow relocation of block and its removal from the flush_list
Definition: buf0types.h:110
The database buffer pool high-level routines.
Class implementing buf_pool->flush_list hazard pointer.
Definition: buf0buf.h:2087
~FlushHp() override=default
Destructor.
FlushHp(const buf_pool_t *buf_pool, const ib_mutex_t *mutex)
Constructor.
Definition: buf0buf.h:2092
void adjust(const buf_page_t *bpage) override
Adjust the value of hp.
Definition: buf0buf.cc:2869
We use Flush_observer to track flushing of non-redo logged pages in bulk create index(btr0load....
Definition: buf0flu.h:270
A "Hazard Pointer" class used to iterate over page lists inside the buffer pool.
Definition: buf0buf.h:2029
bool is_hp(const buf_page_t *bpage)
Checks if a bpage is the hp.
Definition: buf0buf.cc:2844
void set(buf_page_t *bpage)
Set current value.
Definition: buf0buf.cc:2831
virtual void adjust(const buf_page_t *bpage)=0
Adjust the value of hp.
const ib_mutex_t * m_mutex
mutex that protects access to the m_hp.
Definition: buf0buf.h:2079
void move(const buf_page_t *bpage, buf_page_t *dpage)
Adjust the value of hp for moving.
Definition: buf0buf.cc:2856
buf_page_t * m_hp
hazard pointer.
Definition: buf0buf.h:2083
virtual ~HazardPointer()=default
Destructor.
const buf_pool_t * m_buf_pool
Buffer pool instance.
Definition: buf0buf.h:2075
HazardPointer(const HazardPointer &)
Disable copying.
buf_page_t * get() const
Get current value.
Definition: buf0buf.h:2041
HazardPointer(const buf_pool_t *buf_pool, const ib_mutex_t *mutex)
Constructor.
Definition: buf0buf.h:2034
HazardPointer & operator=(const HazardPointer &)
The IO Context that is passed down to the low level IO code.
Definition: os0file.h:265
Class implementing buf_pool->LRU hazard pointer.
Definition: buf0buf.h:2106
~LRUHp() override=default
Destructor.
void adjust(const buf_page_t *bpage) override
Adjust the value of hp.
Definition: buf0buf.cc:2884
LRUHp(const buf_pool_t *buf_pool, const ib_mutex_t *mutex)
Constructor.
Definition: buf0buf.h:2111
Special purpose iterators to be used when scanning the LRU list.
Definition: buf0buf.h:2128
~LRUItr() override=default
Destructor.
buf_page_t * start()
Selects from where to start a scan.
Definition: buf0buf.cc:2899
LRUItr(const buf_pool_t *buf_pool, const ib_mutex_t *mutex)
Constructor.
Definition: buf0buf.h:2133
The purpose of this class is to hide the knowledge that Buf_io_fix_latching_rules even exists from us...
Definition: buf0buf.cc:5582
Definition: buf0buf.h:1424
bool someone_is_responsible() const
Checks if there is any thread responsible for I/O on this page now.
Definition: buf0buf.h:1432
void release()
Called by the thread responsible for I/O on this page to release its responsibility.
Definition: buf0buf.h:1444
bool current_thread_is_responsible() const
Checks if the current thread is responsible for I/O on this page now.
Definition: buf0buf.h:1438
void take()
Called by the thread which becomes responsible for I/O on this page to indicate that it takes the res...
Definition: buf0buf.h:1451
std::thread::id responsible_thread
The thread responsible for I/O on this page, or an impossible value if no thread is currently respons...
Definition: buf0buf.h:1427
Definition: buf0buf.h:1153
uint32_t freed_page_clock
The value of buf_pool->freed_page_clock when this block was the last time put to the head of the LRU ...
Definition: buf0buf.h:1668
buf_page_t(const buf_page_t &other)
Copy constructor.
Definition: buf0buf.h:1157
void space_id_changed()
Updates new space reference and acquires "reference count latch" and the current version of the space...
Definition: buf0buf.h:1315
void take_io_responsibility()
Called by the thread which becomes responsible for I/O on this page to indicate that it takes the res...
Definition: buf0buf.h:1478
bool is_stale() const
Checks if this space reference saved during last page ID initialization was deleted or truncated sinc...
Definition: buf0buf.h:1232
void init_io_fix()
This is called only when having full ownership of the page object and no other thread can reach it.
Definition: buf0buf.h:1508
void set_io_fix(buf_io_fix io_fix)
Sets io_fix to specified value.
Definition: buf0buf.cc:5656
bool someone_has_io_responsibility() const
Checks if there is any thread responsible for I/O on this page now.
Definition: buf0buf.h:1462
bool has_correct_io_fix_value() const
Checks if io_fix has any of the known enum values.
Definition: buf0buf.h:1411
lsn_t get_newest_lsn() const noexcept
Definition: buf0buf.h:1348
bool current_thread_has_io_responsibility() const
Checks if the current thread is responsible for I/O on this page now.
Definition: buf0buf.h:1468
void set_newest_lsn(lsn_t lsn) noexcept
Set the latest modification LSN.
Definition: buf0buf.h:1359
buf_page_t * hash
Node used in chaining to buf_pool->page_hash or buf_pool->zip_hash.
Definition: buf0buf.h:1598
bool is_io_fix_write() const
Checks if the current value of io_fix is BUF_IO_WRITE.
Definition: buf0buf.cc:5636
Flush_observer * get_flush_observer() noexcept
Definition: buf0buf.h:1332
fil_space_t * get_space() const
Retrieve the tablespace object if one was available during page ID initialization.
Definition: buf0buf.h:1271
bool is_io_fix_read_as_opposed_to_write() const
Assuming that io_fix is either BUF_IO_READ or BUF_IO_WRITE determines which of the two it is.
Definition: buf0buf.cc:5648
space_id_t space() const noexcept
Retrieve the tablespace id.
Definition: buf0buf.h:1221
bool in_zip_hash
true if in buf_pool->zip_hash
Definition: buf0buf.h:1713
bool in_free_list
true if in buf_pool->free; when buf_pool->free_list_mutex is free, the following should hold: in_free...
Definition: buf0buf.h:1704
std::chrono::steady_clock::time_point access_time
Time of first access, or 0 if the block was never accessed in the buffer pool.
Definition: buf0buf.h:1677
Flush_observer * m_flush_observer
Flush observer instance.
Definition: buf0buf.h:1660
buf_fix_count_atomic_t buf_fix_count
Count of how many fold this block is currently bufferfixed.
Definition: buf0buf.h:1382
bool in_flush_list
true if in buf_pool->flush_list; when buf_pool->flush_list_mutex is free, the following should hold: ...
Definition: buf0buf.h:1700
void reset_page_id()
Sets stored value to invalid/empty value.
Definition: buf0buf.h:1308
bool is_dirty() const noexcept
Definition: buf0buf.h:1355
bool old
true if the block is in the old blocks in buf_pool->LRU_old
Definition: buf0buf.h:1686
bool was_io_fix_read() const
Checks if io_fix is BUF_IO_READ without requiring or acquiring any latches.
Definition: buf0buf.h:1568
void reinit_io_fix()
This is called only when having full ownership of the page object and no other thread can reach it.
Definition: buf0buf.h:1520
void reset_page_id(page_id_t new_page_id)
Sets stored page ID to the new value.
Definition: buf0buf.h:1292
void set_oldest_lsn(lsn_t lsn) noexcept
Set the LSN when the page is modified for the first time.
Definition: buf0buf.ic:746
uint16_t get_dblwr_batch_id() const
Definition: buf0buf.h:1217
bool was_stale() const
Checks if this space reference saved during last page ID initialization was deleted or truncated sinc...
Definition: buf0buf.h:1249
void set_page_id(const page_id_t page_id)
Set the stored page id to a new value.
Definition: buf0buf.h:1276
lsn_t newest_modification
The flush LSN, LSN when this page was written to the redo log.
Definition: buf0buf.h:1631
void release_io_responsibility()
Called by the thread responsible for I/O on this page to release its responsibility.
Definition: buf0buf.h:1474
static bool is_memory(const page_t *const ptr) noexcept
Check if the given ptr lies in a memory block of type BUF_BLOCK_MEMORY.
Definition: buf0buf.cc:7005
lsn_t get_oldest_lsn() const noexcept
Definition: buf0buf.h:1352
lsn_t oldest_modification
log sequence number of the youngest modification to this block, zero if not modified.
Definition: buf0buf.h:1635
page_id_t id
Page id.
Definition: buf0buf.h:1376
copyable_atomic_t< buf_io_fix > io_fix
Type of pending I/O operation.
Definition: buf0buf.h:1388
UT_LIST_NODE_T(buf_page_t) list
Based on state, this is a list node, protected by the corresponding list mutex, in one of the followi...
buf_io_fix get_io_fix_snapshot() const
Retrieves a value of io_fix without requiring or acquiring any latches.
Definition: buf0buf.h:1494
page_size_t size
Page size.
Definition: buf0buf.h:1379
io_responsibility_t io_responsibility
Tracks which thread is responsible for I/O on this page.
Definition: buf0buf.h:1457
uint8_t buf_pool_index
Index number of the buffer pool that this block belongs to.
Definition: buf0buf.h:1595
uint16_t m_dblwr_id
Double write instance ordinal value during writes.
Definition: buf0buf.h:1682
bool was_io_fix_none() const
Checks if io_fix is BUF_IO_NONE without requiring or acquiring any latches.
Definition: buf0buf.h:1585
void set_clean() noexcept
Set page to clean state.
Definition: buf0buf.h:1366
buf_io_fix get_io_fix() const
Retrieves the current value of io_fix.
Definition: buf0buf.h:1536
bool is_io_fix_read() const
Checks if the current value of io_fix is BUF_IO_READ.
Definition: buf0buf.cc:5642
uint32_t m_version
Version of fil_space_t when the page was updated.
Definition: buf0buf.h:1673
bool in_LRU_list
true if the page is in the LRU list; used in debugging
Definition: buf0buf.h:1707
void set_flush_observer(Flush_observer *flush_observer) noexcept
Set the flush observer for the page.
Definition: buf0buf.h:1336
buf_page_state state
Block state.
Definition: buf0buf.h:1588
page_no_t page_no() const noexcept
Retrieve the page number.
Definition: buf0buf.h:1225
fil_space_t * m_space
Tablespace instance that this page belongs to.
Definition: buf0buf.h:1663
bool was_io_fixed() const
Checks if io_fix is BUF_IO_FIX or BUF_IO_READ or BUF_IO_WRITE without requiring or acquiring any latc...
Definition: buf0buf.h:1575
void reset_flush_observer() noexcept
Remove the flush observer.
Definition: buf0buf.h:1344
void set_dblwr_batch_id(uint16_t batch_id)
Set the doublewrite buffer ID.
Definition: buf0buf.h:1214
bool file_page_was_freed
This is set to true when fsp frees a page in buffer pool; protected by buf_pool->zip_mutex or buf_blo...
Definition: buf0buf.h:1691
UT_LIST_NODE_T(buf_page_t) LRU
node of the LRU list
friend class Latching_rules_helpers
Definition: buf0buf.h:1416
bool is_memory() const noexcept
Check if the state of this page is BUF_BLOCK_MEMORY.
Definition: buf0buf.h:1207
buf_flush_t flush_type
If this block is currently being flushed to disk, this tells the flush_type.
Definition: buf0buf.h:1592
void set_page_size(const page_size_t &page_size)
Set the page size to a new value.
Definition: buf0buf.h:1284
bool in_page_hash
true if in buf_pool->page_hash
Definition: buf0buf.h:1710
page_zip_des_t zip
compressed page; zip.data (but not the data it points to) is protected by buf_pool->zip_mutex; state ...
Definition: buf0buf.h:1656
static bool is_correct_io_fix_value(buf_io_fix io_fix)
Checks if io_fix has any of the known enum values.
Definition: buf0buf.h:1396
Definition: buf0buf.h:1146
copyable_atomic_t(const copyable_atomic_t< T > &other)
Definition: buf0buf.h:1148
Definition: hash0hash.h:375
Page identifier.
Definition: buf0types.h:207
uint64_t hash() const
Retrieve the hash value.
Definition: buf0types.h:247
page_no_t page_no() const
Retrieve the page number.
Definition: buf0types.h:243
Page size descriptor.
Definition: page0size.h:50
size_t logical() const
Retrieve the logical page size (in-memory).
Definition: page0size.h:129
int page
Definition: ctype-mb.cc:1234
dberr_t
Definition: db0err.h:39
fil_space_t * fil_space_get(space_id_t space_id)
Look up a tablespace.
Definition: fil0fil.cc:2222
void fil_page_reset_type(const page_id_t &page_id, byte *page, ulint type, mtr_t *mtr)
Reset the page type.
Definition: fil0fil.cc:8229
The low-level file system.
uint16_t page_type_t
Definition: fil0fil.h:1221
constexpr page_type_t FIL_PAGE_INDEX
File page types (values of FIL_PAGE_TYPE)
Definition: fil0fil.h:1225
constexpr page_no_t FIL_NULL
'null' (undefined) page offset in the context of file spaces
Definition: fil0fil.h:1159
page_type_t fil_page_get_type(const byte *page)
Get the file page type.
Definition: fil0fil.h:1338
constexpr uint32_t FIL_PAGE_TYPE
file page type: FIL_PAGE_INDEX,..., 2 bytes.
Definition: fil0types.h:76
constexpr uint32_t FIL_PAGE_OFFSET
page offset inside space
Definition: fil0types.h:46
constexpr uint32_t FIL_PAGE_DATA
start of the data on the page
Definition: fil0types.h:111
constexpr uint32_t FIL_PAGE_SPACE_ID
alias for space id
Definition: fil0types.h:108
constexpr uint32_t FIL_PAGE_NEXT
if there is a 'natural' successor of the page, its offset.
Definition: fil0types.h:61
constexpr uint32_t FIL_PAGE_PREV
if there is a 'natural' predecessor of the page, its offset.
Definition: fil0types.h:51
flush_type
Definition: my_sys.h:297
The simple hash table utility.
static rw_lock_t * hash_lock_s_confirm(rw_lock_t *hash_lock, hash_table_t *table, uint64_t hash_value)
If not appropriate rw_lock for a hash value in a hash table, relock S-lock the another rw_lock until ...
static rw_lock_t * hash_lock_x_confirm(rw_lock_t *hash_lock, hash_table_t *table, uint64_t hash_value)
If not appropriate rw_lock for a hash value in a hash table, relock X-lock the another rw_lock until ...
static rw_lock_t * hash_get_lock(hash_table_t *table, uint64_t hash_value)
Gets the rw_lock for a hash value in a hash table.
static int flags[50]
Definition: hp_test1.cc:40
#define malloc(A)
Definition: lexyy.cc:914
#define free(A)
Definition: lexyy.cc:915
lock_mode
Definition: lock0types.h:54
constexpr lsn_t LSN_MAX
Maximum possible lsn value is slightly higher than the maximum sn value, because lsn sequence enumera...
Definition: log0constants.h:159
uint64_t lsn_t
Type used for all log sequence number storage and arithmetic.
Definition: log0types.h:63
static uint16_t mach_read_from_2(const byte *b)
The following function is used to fetch data from 2 consecutive bytes.
static uint32_t mach_read_from_4(const byte *b)
The following function is used to fetch data from 4 consecutive bytes.
static void mach_write_to_4(byte *b, ulint n)
The following function is used to store data in 4 consecutive bytes.
Mini-transaction buffer global types.
unsigned long long int ulonglong
Definition: my_inttypes.h:56
void copy(Shards< COUNT > &dst, const Shards< COUNT > &src) noexcept
Copy the counters, overwrite destination.
Definition: ut0counter.h:354
void clear(Shards< COUNT > &shards) noexcept
Clear the counter - reset to 0.
Definition: ut0counter.h:344
Definition: buf0block_hint.cc:30
const std::string FILE("FILE")
bool load(THD *, const dd::String_type &fname, dd::String_type *buf)
Read an sdi file from disk and store in a buffer.
Definition: sdi_file.cc:308
Definition: os0file.h:89
Provides atomic access in shared-exclusive modes.
Definition: shared_spin_lock.h:79
size_t size(const char *const c)
Definition: base64.h:46
Definition: gcs_xcom_synode.h:64
mode
Definition: file_handle.h:61
pid_type get_id()
Definition: process.h:48
static uint64_t hash_uint64(uint64_t value)
Hashes a 64-bit integer.
Definition: ut0rnd.h:189
std::list< T, ut::allocator< T > > list
Specialization of list which uses ut_allocator.
Definition: ut0new.h:2879
void os_event_set(os_event_t event)
Sets an event semaphore to the signaled state: lets waiting threads proceed.
Definition: os0event.cc:553
bool os_event_is_set(const os_event_t event)
Check if the event is set.
Definition: os0event.cc:545
int64_t os_event_reset(os_event_t event)
Resets an event semaphore to the non-signaled state.
Definition: os0event.cc:568
The interface to the operating system process control primitives.
Index page routines.
byte page_t
Type of the index page.
Definition: page0types.h:152
required string type
Definition: replication_group_member_actions.proto:34
Shutdowns the Innobase database server.
std::atomic< enum srv_shutdown_t > srv_shutdown_state
At a shutdown this value climbs from SRV_SHUTDOWN_NONE to SRV_SHUTDOWN_EXIT_THREADS.
Definition: srv0start.cc:162
@ SRV_SHUTDOWN_EXIT_THREADS
Exit all threads and free resources.
Definition: srv0shutdown.h:111
The server main program.
Functor to validate the LRU list.
Definition: buf0buf.h:2728
static void validate(const buf_pool_t *buf_pool)
Definition: buf0buf.h:2731
void operator()(const buf_page_t *elem) const
Definition: buf0buf.h:2729
Functor to validate the LRU list.
Definition: buf0buf.h:2718
void operator()(const buf_page_t *elem) const
Definition: buf0buf.h:2719
static void validate(const buf_pool_t *buf_pool)
Definition: buf0buf.h:2721
Definition: buf0buf.h:2737
static void validate(const buf_pool_t *buf_pool)
Definition: buf0buf.h:2743
void operator()(const buf_block_t *elem) const
Definition: buf0buf.h:2738
Structure used by AHI to contain information on record prefixes to be considered in hash index subsys...
Definition: buf0buf.h:1722
bool operator==(const btr_search_prefix_info_t &other) const
Definition: buf0buf.h:1736
bool operator!=(const btr_search_prefix_info_t &other) const
Definition: buf0buf.h:1741
bool equals_without_left_side(const btr_search_prefix_info_t &other) const
Definition: buf0buf.h:1732
uint32_t n_bytes
recommended prefix: number of bytes in an incomplete field
Definition: buf0buf.h:1725
uint16_t n_fields
recommended prefix length for hash search: number of full fields
Definition: buf0buf.h:1727
bool left_side
true or false, depending on whether the leftmost record of several records with the same prefix shoul...
Definition: buf0buf.h:1730
Structure that holds most AHI-related fields.
Definition: buf0buf.h:1790
void assert_empty_on_init() const
Definition: buf0buf.h:1856
std::atomic< dict_index_t * > index
Index for which the adaptive hash index has been created, or nullptr if the page does not exist in th...
Definition: buf0buf.h:1809
void assert_empty() const
Definition: buf0buf.h:1854
void validate() const
Definition: buf0buf.h:1832
std::atomic< btr_search_prefix_info_t > prefix_info
Prefix info that was used for building hash index.
Definition: buf0buf.h:1808
std::atomic< uint16_t > n_pointers
Used in debugging.
Definition: buf0buf.h:1830
std::atomic< btr_search_prefix_info_t > recommended_prefix_info
Recommended prefix info for hash search.
Definition: buf0buf.h:1797
The buffer control block structure.
Definition: buf0buf.h:1747
rw_lock_t debug_latch
In the debug version, each thread which bufferfixes the block acquires an s-latch here; so we can use...
Definition: buf0buf.h:1891
uint64_t modify_clock
This clock is incremented every time a pointer to a record on the page may become obsolete; this is u...
Definition: buf0buf.h:1905
BPageLock lock
read-write lock of the buffer frame
Definition: buf0buf.h:1757
std::atomic< uint32_t > n_hash_helps
Counter which controls how many times the current prefix recommendation would help in searches.
Definition: buf0buf.h:1879
page_type_t get_page_type() const
Get the page type of the current buffer block.
Definition: buf0buf.h:1951
struct buf_block_t::ahi_t ahi
buf_page_t page
page information; this must be the first field, so that buf_pool->page_hash can point to buf_page_t o...
Definition: buf0buf.h:1753
page_no_t get_next_page_no() const
Get the next page number of the current buffer block.
Definition: buf0buf.h:1939
page_zip_des_t const * get_page_zip() const noexcept
Const version.
Definition: buf0buf.h:1989
const char * get_page_type_str() const noexcept
Get the page type of the current buffer block as string.
Definition: buf0buf.cc:6981
bool in_withdraw_list
Definition: buf0buf.h:1784
UT_LIST_NODE_T(buf_block_t) unzip_LRU
node of the decompressed LRU list; a block is in the unzip_LRU list if page.state == BUF_BLOCK_FILE_P...
page_no_t get_prev_page_no() const
Get the prev page number of the current buffer block.
Definition: buf0buf.h:1945
bool is_leaf() const
Definition: buf0buf.h:2000
void mark_for_read_io()
Mark the frame with jumbled page_id, while initiating i/o read (BUF_IO_READ).
Definition: buf0buf.h:1958
uint64_t get_modify_clock(bool single_threaded) const noexcept
Get the modified clock (version) value.
Definition: buf0buf.h:1918
bool in_unzip_LRU_list
true if the page is in the decompressed LRU list; used in debugging
Definition: buf0buf.h:1782
bool is_empty() const
Check if this index page is empty.
Definition: buf0buf.cc:7018
const page_id_t & get_page_id() const
Get the page number and space id of the current buffer block.
Definition: buf0buf.h:1931
BPageMutex mutex
mutex protecting this block: state (also protected by the buffer pool mutex), io_fix,...
Definition: buf0buf.h:1912
byte * frame
pointer to buffer frame which is of size UNIV_PAGE_SIZE, and aligned to an address divisible by UNIV_...
Definition: buf0buf.h:1769
page_no_t get_page_no() const
Get the page number of the current buffer block.
Definition: buf0buf.h:1935
bool is_index_page() const
Definition: buf0buf.h:2002
bool is_root() const
Definition: buf0buf.h:1996
bool was_freed() const
Check if the buffer block was freed.
Definition: buf0buf.h:1762
bool made_dirty_with_no_latch
true if block has been made dirty without acquiring X/SX latch as the block belongs to temporary tabl...
Definition: buf0buf.h:1883
page_zip_des_t * get_page_zip() noexcept
Gets the compressed page descriptor corresponding to an uncompressed page if applicable.
Definition: buf0buf.h:1983
bool is_memory() const noexcept
Definition: buf0buf.h:1993
uint16_t get_page_level() const
Definition: buf0buf.cc:7010
bool is_compact() const
Determine whether the page is in new-style compact format.
Definition: buf0buf.cc:7022
Struct that is embedded in the free zip blocks.
Definition: buf0buf.h:2147
UT_LIST_NODE_T(buf_buddy_free_t) list
Node of zip_free list.
ulint size
size of the block
Definition: buf0buf.h:2149
byte bytes[FIL_PAGE_DATA]
stamp[FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID] == BUF_BUDDY_FREE_STAMP denotes a free block.
Definition: buf0buf.h:2150
union buf_buddy_free_t::@194 stamp
buf_page_t bpage
Embedded bpage descriptor.
Definition: buf0buf.h:2161
Definition: buf0buf.h:2259
uint64_t relocated
Definition: buf0buf.h:2261
ulint used
Definition: buf0buf.h:2260
std::chrono::steady_clock::duration relocated_duration
Definition: buf0buf.h:2262
Statistics of buddy blocks of a given size.
Definition: buf0buf.h:2249
uint64_t relocated
Number of blocks relocated by the buddy system.
Definition: buf0buf.h:2254
snapshot_t take_snapshot()
Definition: buf0buf.h:2265
std::atomic< ulint > used
Number of blocks allocated from the buddy system.
Definition: buf0buf.h:2251
std::chrono::steady_clock::duration relocated_duration
Total duration of block relocations.
Definition: buf0buf.h:2257
A chunk of buffers.
Definition: buf0buf.ic:53
This structure defines information we will fetch from each buffer pool.
Definition: buf0buf.h:158
ulint young_making_delta
num of pages made young since last printout
Definition: buf0buf.h:211
std::array< size_t, BUF_FLUSH_N_TYPES > n_pending_flush
Number of pages pending flush of given type.
Definition: buf0buf.h:178
double pages_created_rate
num of pages create per second
Definition: buf0buf.h:207
ulint n_pages_read
buf_pool->n_pages_read
Definition: buf0buf.h:184
ulint pool_unique_id
Buffer Pool ID.
Definition: buf0buf.h:162
ulint n_pend_reads
buf_pool->n_pend_reads, pages pending read
Definition: buf0buf.h:176
ulint n_ra_pages_read_rnd
buf_pool->n_ra_pages_read_rnd, number of pages readahead
Definition: buf0buf.h:192
ulint lru_len
Length of buf_pool->LRU.
Definition: buf0buf.h:166
ulint unzip_cur
buf_LRU_stat_cur.unzip, num pages decompressed in current interval
Definition: buf0buf.h:232
ulint page_read_delta
num of pages read since last printout
Definition: buf0buf.h:209
double pages_written_rate
num of pages written per second
Definition: buf0buf.h:208
double page_made_young_rate
page made young rate in pages per second
Definition: buf0buf.h:202
ulint io_sum
buf_LRU_stat_sum.io
Definition: buf0buf.h:228
ulint io_cur
buf_LRU_stat_cur.io, num of IO for current interval
Definition: buf0buf.h:229
ulint old_lru_len
buf_pool->LRU_old_len
Definition: buf0buf.h:168
double pages_readahead_rnd_rate
random readahead rate in pages per second
Definition: buf0buf.h:217
ulint unzip_sum
buf_LRU_stat_sum.unzip
Definition: buf0buf.h:231
ulint n_pages_made_young
number of pages made young
Definition: buf0buf.h:180
ulint n_ra_pages_read
buf_pool->n_ra_pages_read, number of pages readahead
Definition: buf0buf.h:194
ulint n_pages_not_made_young
number of pages not made young
Definition: buf0buf.h:182
ulint n_ra_pages_evicted
buf_pool->n_ra_pages_evicted, number of readahead pages evicted without access
Definition: buf0buf.h:197
ulint n_pages_created
buf_pool->n_pages_created
Definition: buf0buf.h:186
ulint n_page_get_delta
num of buffer pool page gets since last printout
Definition: buf0buf.h:199
ulint free_list_len
Length of buf_pool->free list.
Definition: buf0buf.h:170
double pages_evicted_rate
rate of readahead page evicted without access, in pages per second
Definition: buf0buf.h:221
ulint not_young_making_delta
num of pages not make young since last printout
Definition: buf0buf.h:213
ulint n_pend_unzip
buf_pool->n_pend_unzip, pages pending decompress
Definition: buf0buf.h:174
double pages_readahead_rate
readahead rate in pages per second
Definition: buf0buf.h:219
ulint unzip_lru_len
length of buf_pool->unzip_LRU list
Definition: buf0buf.h:225
ulint pool_size
Buffer Pool size in pages.
Definition: buf0buf.h:164
ulint flush_list_len
Length of buf_pool->flush_list.
Definition: buf0buf.h:172
double page_not_made_young_rate
page not made young rate in pages per second
Definition: buf0buf.h:204
ulint n_page_gets
buf_pool->n_page_gets
Definition: buf0buf.h:190
double pages_read_rate
num of pages read per second
Definition: buf0buf.h:206
ulint n_pages_written
buf_pool->n_pages_written
Definition: buf0buf.h:188
The buffer pool statistics structure.
Definition: buf0buf.h:2167
uint64_t n_pages_not_made_young
Number of pages not made young because the first access was not long enough ago, in buf_page_peek_if_...
Definition: buf0buf.h:2200
std::atomic< uint64_t > n_pages_read
Number of read operations.
Definition: buf0buf.h:2176
std::atomic< uint64_t > n_pages_created
number of pages created in the pool with no read.
Definition: buf0buf.h:2182
uint64_t n_pages_made_young
Number of pages made young, in calls to buf_LRU_make_block_young().
Definition: buf0buf.h:2196
uint64_t n_ra_pages_evicted
Number of read ahead pages that are evicted without being accessed.
Definition: buf0buf.h:2192
std::atomic< uint64_t > n_ra_pages_read_rnd
Number of pages read in as part of random read ahead.
Definition: buf0buf.h:2185
uint64_t flush_list_bytes
Flush_list size in bytes.
Definition: buf0buf.h:2206
std::atomic< uint64_t > n_ra_pages_read
Number of pages read in as part of read ahead.
Definition: buf0buf.h:2188
Shards m_n_page_gets
Number of page gets performed; also successful searches through the adaptive hash index are counted a...
Definition: buf0buf.h:2173
uint64_t LRU_bytes
LRU size in bytes.
Definition: buf0buf.h:2203
void reset()
Definition: buf0buf.h:2232
static void copy(buf_pool_stat_t &dst, const buf_pool_stat_t &src) noexcept
Definition: buf0buf.h:2208
std::atomic< uint64_t > n_pages_written
Number of write operations.
Definition: buf0buf.h:2179
The buffer pool structure.
Definition: buf0buf.h:2275
UT_LIST_BASE_NODE_T(buf_buddy_free_t, list) zip_free[BUF_BUDDY_SIZES_MAX]
Buddy free lists.
UT_LIST_BASE_NODE_T(buf_page_t, list) free
Base node of the free block list.
buf_chunk_t * chunks_old
old buffer pool chunks to be freed after resizing buffer pool
Definition: buf0buf.h:2329
void change_flush_state(buf_flush_t flush_type, F &&change)
Executes change() which modifies fields protected by flush_state_mutex.
Definition: buf0buf.h:2548
hash_table_t * zip_hash
Hash table of buf_block_t blocks whose frames are allocated to the zip buddy system,...
Definition: buf0buf.h:2348
FlushHp flush_hp
"Hazard pointer" used during scan of flush_list while doing flush list batch.
Definition: buf0buf.h:2382
buf_buddy_stat_t buddy_stat[BUF_BUDDY_SIZES_MAX+1]
Statistics of buddy system, indexed by block size.
Definition: buf0buf.h:2361
LRUItr single_scan_itr
Iterator used to scan the LRU list when searching for single page flushing victim.
Definition: buf0buf.h:2459
buf_pool_stat_t old_stat
Old statistics.
Definition: buf0buf.h:2367
void deallocate_chunk(buf_chunk_t *chunk)
A wrapper for buf_pool_t::allocator.deallocate_large which also advices the OS that this chunk can be...
Definition: buf0buf.cc:918
bool try_LRU_scan
Set to false when an LRU scan for free block fails.
Definition: buf0buf.h:2421
ulint buddy_n_frames
Number of frames allocated from the buffer pool to the buddy system.
Definition: buf0buf.h:2316
std::atomic< ulint > n_pend_reads
Number of pending read operations.
Definition: buf0buf.h:2351
UT_LIST_BASE_NODE_T(buf_page_t, list) zip_clean
Unmodified compressed pages.
UT_LIST_BASE_NODE_T(buf_page_t, list) flush_list
Base node of the modified block list.
LRUHp lru_hp
"hazard pointer" used during scan of LRU while doing LRU list batch.
Definition: buf0buf.h:2451
bool allocate_chunk(ulonglong mem_size, buf_chunk_t *chunk)
A wrapper for buf_pool_t::allocator.alocate_large which also advices the OS that this chunk should no...
Definition: buf0buf.cc:884
bool is_tracking()
Check if the page modifications are tracked.
Definition: buf0buf.h:2428
UT_LIST_BASE_NODE_T(buf_block_t, unzip_LRU) unzip_LRU
Base node of the unzip_LRU list.
std::chrono::steady_clock::time_point last_printout_time
when buf_print_io was last time called.
Definition: buf0buf.h:2357
ulint LRU_old_ratio
Reserve this much of the buffer pool for "old" blocks.
Definition: buf0buf.h:2312
BufListMutex chunks_mutex
protects (de)allocation of chunks:
Definition: buf0buf.h:2284
lsn_t track_page_lsn
Page Tracking start LSN.
Definition: buf0buf.h:2424
BufListMutex zip_free_mutex
buddy allocator mutex
Definition: buf0buf.h:2293
volatile ulint n_chunks
Number of buffer pool chunks.
Definition: buf0buf.h:2320
LRUItr lru_scan_itr
Iterator used to scan the LRU list when searching for replaceable victim.
Definition: buf0buf.h:2455
bool madvise_dump()
Advices the OS that all chunks in this buffer pool instance can be dumped to a core file.
Definition: buf0buf.cc:929
ulint old_size
Previous pool size in pages.
Definition: buf0buf.h:2335
UT_LIST_BASE_NODE_T(buf_page_t, LRU) LRU
Base node of the LRU list.
FlushHp oldest_hp
Entry pointer to scan the oldest page except for system temporary.
Definition: buf0buf.h:2385
buf_page_t * LRU_old
Pointer to the about LRU_old_ratio/BUF_LRU_OLD_RATIO_DIV oldest blocks in the LRU list; NULL if LRU l...
Definition: buf0buf.h:2467
buf_page_t * watch
Sentinel records for buffer pool watches.
Definition: buf0buf.h:2496
BufListMutex flush_list_mutex
Mutex protecting the flush list access.
Definition: buf0buf.h:2378
ib_mutex_t flush_state_mutex
Flush state protection mutex.
Definition: buf0buf.h:2299
ib_rbt_t * flush_rbt
A red-black tree is used exclusively during recovery to speed up insertions in the flush_list.
Definition: buf0buf.h:2408
buf_chunk_t * chunks
buffer pool chunks
Definition: buf0buf.h:2326
BufListMutex LRU_list_mutex
LRU list mutex.
Definition: buf0buf.h:2287
hash_table_t * page_hash
Hash table of buf_page_t or buf_block_t file pages, buf_page_in_file() == true, indexed by (space_id,...
Definition: buf0buf.h:2344
ulint withdraw_target
Target length of withdraw block list, when withdrawing.
Definition: buf0buf.h:2447
os_event_t no_flush[BUF_FLUSH_N_TYPES]
This is in the set state when there is no flush batch of the given type running.
Definition: buf0buf.h:2400
buf_pool_stat_t stat
Current statistics.
Definition: buf0buf.h:2364
page_no_t read_ahead_area
Size in pages of the area which the read-ahead algorithms read if invoked.
Definition: buf0buf.h:2339
ulint curr_pool_size
Current pool size in bytes.
Definition: buf0buf.h:2309
bool init_flush[BUF_FLUSH_N_TYPES]
This is true when a flush of the given type is being initialized.
Definition: buf0buf.h:2392
ulint curr_size
Current pool size in pages.
Definition: buf0buf.h:2332
ulint instance_no
Array index of this buffer pool instance.
Definition: buf0buf.h:2306
ulint LRU_old_len
Length of the LRU list from the block to which LRU_old points onward, including that block; see buf0l...
Definition: buf0buf.h:2473
std::array< size_t, BUF_FLUSH_N_TYPES > n_flush
This is the number of pending writes in the given flush type.
Definition: buf0buf.h:2396
std::atomic< ulint > n_pend_unzip
number of pending decompressions.
Definition: buf0buf.h:2354
UT_LIST_BASE_NODE_T(buf_page_t, list) withdraw
base node of the withdraw block list.
lsn_t max_lsn_io
Maximum LSN for which write io has already started.
Definition: buf0buf.h:2431
ulint freed_page_clock
A sequence number used to count the number of buffer blocks removed from the end of the LRU list; NOT...
Definition: buf0buf.h:2415
BufListMutex free_list_mutex
free and withdraw list mutex
Definition: buf0buf.h:2290
BufPoolZipMutex zip_mutex
Zip mutex of this buffer pool instance, protects compressed only pages (of type buf_page_t,...
Definition: buf0buf.h:2303
bool madvise_dont_dump()
Advices the OS that all chunks in this buffer pool instance should not be dumped to a core file.
Definition: buf0buf.cc:939
volatile ulint n_chunks_new
New number of buffer pool chunks.
Definition: buf0buf.h:2323
BufListMutex zip_hash_mutex
zip_hash mutex
Definition: buf0buf.h:2296
bool is_flushing(buf_flush_t flush_type) const
Checks if the batch is running, which is basically equivalent to !os_event_is_set(no_flush[type]) if ...
Definition: buf0buf.h:2535
The occupied bytes of lists in all buffer pools.
Definition: buf0buf.h:238
ulint LRU_bytes
LRU size in bytes.
Definition: buf0buf.h:239
ulint unzip_LRU_bytes
unzip_LRU size in bytes
Definition: buf0buf.h:240
ulint flush_list_bytes
flush_list size in bytes
Definition: buf0buf.h:241
File space address.
Definition: fil0fil.h:1176
File node of a tablespace or the log data space.
Definition: fil0fil.h:155
Tablespace or log data space.
Definition: fil0fil.h:238
space_id_t id
Tablespace ID.
Definition: fil0fil.h:334
void dec_ref() noexcept
Decrement the page reference count.
Definition: fil0fil.h:296
uint32_t get_current_version() const
Returns current version of the space object.
Definition: fil0fil.cc:11767
uint32_t get_recent_version() const
Returns current version of the space object.
Definition: fil0fil.cc:11771
bool was_not_deleted() const
Definition: fil0fil.cc:11759
void inc_ref() noexcept
Increment the page reference count.
Definition: fil0fil.h:285
bool is_deleted() const
Definition: fil0fil.cc:11754
Red black tree instance.
Definition: ut0rbt.h:72
Mini-transaction handle and buffer.
Definition: mtr0mtr.h:177
InnoDB condition variable.
Definition: os0event.cc:63
Compressed page descriptor.
Definition: page0types.h:201
page_zip_t * data
Compressed page data.
Definition: page0types.h:203
The structure used in the spin lock implementation of a read-write lock.
Definition: sync0rw.h:363
Definition: ut0core.h:36
bool rw_lock_own(const rw_lock_t *lock, ulint lock_type)
Checks if the thread has locked the rw-lock in the specified mode, with the pass value == 0.
Definition: sync0rw.cc:858
rw_lock_type_t
Definition: sync0rw.h:97
@ RW_NO_LATCH
Definition: sync0rw.h:101
bool rw_lock_own_flagged(const rw_lock_t *lock, rw_lock_flags_t flags)
Checks if the thread has locked the rw-lock in the specified mode, with the pass value == 0.
Definition: sync0rw.cc:898
latch_level_t
Latching order levels.
Definition: sync0types.h:201
@ RW_LOCK_S
Definition: sync0types.h:208
@ RW_LOCK_X
Definition: sync0types.h:209
@ RW_LOCK_FLAG_X
Definition: sync0types.h:1216
@ RW_LOCK_FLAG_SX
Definition: sync0types.h:1217
@ RW_LOCK_FLAG_S
Definition: sync0types.h:1215
Version control for database, common definitions, and include files.
constexpr uint32_t UINT32_UNDEFINED
The 'undefined' value for a 32-bit unsigned integer.
Definition: univ.i:428
#define UNIV_PAGE_SIZE_SHIFT
The 2-logarithm of UNIV_PAGE_SIZE:
Definition: univ.i:291
#define IF_DEBUG(...)
Definition: univ.i:674
unsigned long int ulint
Definition: univ.i:406
#define UNIV_MEM_VALID(addr, size)
Definition: univ.i:591
constexpr uint32_t UNIV_ZIP_SIZE_MIN
Smallest compressed page size.
Definition: univ.i:330
Utilities for byte operations.
#define ut_ad(EXPR)
Debug assertion.
Definition: ut0dbg.h:105
#define ut_d(EXPR)
Debug statement.
Definition: ut0dbg.h:107
#define ut_a(EXPR)
Abort execution if EXPR does not evaluate to nonzero.
Definition: ut0dbg.h:93
void ut_list_validate(const List &list, Functor &functor)
Checks the consistency of a two-way list.
Definition: ut0lst.h:493
#define mutex_own(M)
Checks that the current thread owns the mutex.
Definition: ut0mutex.h:165
#define mutex_exit(M)
Definition: ut0mutex.h:123
#define mutex_enter(M)
Definition: ut0mutex.h:117
Various utilities.
static uint64_t lsn
Definition: xcom_base.cc:446
unsigned long id[MAX_DEAD]
Definition: xcom_base.cc:510
int n
Definition: xcom_base.cc:509