1 | /* |
2 | Copyright (c) 2012,2013 Monty Program Ab |
3 | |
4 | This program is free software; you can redistribute it and/or modify |
5 | it under the terms of the GNU General Public License as published by |
6 | the Free Software Foundation; version 2 of the License. |
7 | |
8 | This program is distributed in the hope that it will be useful, |
9 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
10 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
11 | GNU General Public License for more details. |
12 | |
13 | You should have received a copy of the GNU General Public License |
14 | along with this program; if not, write to the Free Software |
15 | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ |
16 | #pragma once |
17 | |
18 | #ifdef USE_PRAGMA_INTERFACE |
19 | #pragma interface /* gcc class implementation */ |
20 | #endif |
21 | |
22 | /* C++ standard header files */ |
23 | #include <set> |
24 | #include <string> |
25 | #include <unordered_map> |
26 | #include <unordered_set> |
27 | #include <vector> |
28 | |
29 | /* MySQL header files */ |
30 | #include "./handler.h" /* handler */ |
31 | #include "./my_global.h" /* ulonglong */ |
32 | #include "./sql_string.h" |
33 | |
34 | #ifdef _WIN32 |
35 | #undef pthread_key_create |
36 | #undef pthread_key_delete |
37 | #undef pthread_setspecific |
38 | #undef pthread_getspecific |
39 | #endif |
40 | |
41 | |
42 | /* RocksDB header files */ |
43 | #include "rocksdb/cache.h" |
44 | #include "rocksdb/merge_operator.h" |
45 | #include "rocksdb/perf_context.h" |
46 | #include "rocksdb/sst_file_manager.h" |
47 | #include "rocksdb/statistics.h" |
48 | #include "rocksdb/utilities/options_util.h" |
49 | #include "rocksdb/utilities/transaction_db.h" |
50 | #include "rocksdb/utilities/write_batch_with_index.h" |
51 | |
52 | /* MyRocks header files */ |
53 | #include "./rdb_comparator.h" |
54 | #include "./rdb_index_merge.h" |
55 | #include "./rdb_io_watchdog.h" |
56 | #include "./rdb_perf_context.h" |
57 | #include "./rdb_sst_info.h" |
58 | #include "./rdb_utils.h" |
59 | #include "./ut0counter.h" |
60 | |
61 | class Field_blob; |
62 | class Field_varstring; |
63 | /** |
64 | @note MyRocks Coding Conventions: |
65 | MyRocks code follows the baseline MySQL coding conventions, available at |
66 | http://dev.mysql.com/doc/internals/en/coding-guidelines.html, with several |
67 | refinements (@see /storage/rocksdb/README file). |
68 | */ |
69 | |
70 | namespace myrocks { |
71 | |
72 | /* |
73 | * class for exporting transaction information for |
74 | * information_schema.rocksdb_trx |
75 | */ |
76 | struct Rdb_trx_info { |
77 | std::string name; |
78 | ulonglong trx_id; |
79 | ulonglong write_count; |
80 | ulonglong lock_count; |
81 | int timeout_sec; |
82 | std::string state; |
83 | std::string waiting_key; |
84 | ulonglong waiting_cf_id; |
85 | int is_replication; |
86 | int skip_trx_api; |
87 | int read_only; |
88 | int deadlock_detect; |
89 | int num_ongoing_bulk_load; |
90 | ulong thread_id; |
91 | std::string query_str; |
92 | }; |
93 | |
94 | std::vector<Rdb_trx_info> rdb_get_all_trx_info(); |
95 | |
96 | /* |
97 | * class for exporting deadlock transaction information for |
98 | * information_schema.rocksdb_deadlock |
99 | */ |
100 | struct Rdb_deadlock_info { |
101 | struct Rdb_dl_trx_info { |
102 | ulonglong trx_id; |
103 | std::string cf_name; |
104 | std::string waiting_key; |
105 | bool exclusive_lock; |
106 | std::string index_name; |
107 | std::string table_name; |
108 | }; |
109 | std::vector <Rdb_dl_trx_info> path; |
110 | ulonglong victim_trx_id; |
111 | }; |
112 | |
113 | std::vector<Rdb_deadlock_info> rdb_get_deadlock_info(); |
114 | |
115 | /* |
116 | This is |
117 | - the name of the default Column Family (the CF which stores indexes which |
118 | didn't explicitly specify which CF they are in) |
119 | - the name used to set the default column family parameter for per-cf |
120 | arguments. |
121 | */ |
122 | extern const std::string DEFAULT_CF_NAME; |
123 | |
124 | /* |
125 | This is the name of the Column Family used for storing the data dictionary. |
126 | */ |
127 | extern const std::string DEFAULT_SYSTEM_CF_NAME; |
128 | |
129 | /* |
130 | This is the name of the hidden primary key for tables with no pk. |
131 | */ |
132 | const char *const HIDDEN_PK_NAME = "HIDDEN_PK_ID" ; |
133 | |
134 | /* |
135 | Column family name which means "put this index into its own column family". |
136 | DEPRECATED!!! |
137 | */ |
138 | extern const std::string PER_INDEX_CF_NAME; |
139 | |
140 | /* |
141 | Name for the background thread. |
142 | */ |
143 | const char *const BG_THREAD_NAME = "myrocks-bg" ; |
144 | |
145 | /* |
146 | Name for the drop index thread. |
147 | */ |
148 | const char *const INDEX_THREAD_NAME = "myrocks-index" ; |
149 | |
150 | /* |
151 | Separator between partition name and the qualifier. Sample usage: |
152 | |
153 | - p0_cfname=foo |
154 | - p3_tts_col=bar |
155 | */ |
156 | const char RDB_PER_PARTITION_QUALIFIER_NAME_SEP = '_'; |
157 | |
158 | /* |
159 | Separator between qualifier name and value. Sample usage: |
160 | |
161 | - p0_cfname=foo |
162 | - p3_tts_col=bar |
163 | */ |
164 | const char RDB_QUALIFIER_VALUE_SEP = '='; |
165 | |
166 | /* |
167 | Separator between multiple qualifier assignments. Sample usage: |
168 | |
169 | - p0_cfname=foo;p1_cfname=bar;p2_cfname=baz |
170 | */ |
171 | const char RDB_QUALIFIER_SEP = ';'; |
172 | |
173 | /* |
174 | Qualifier name for a custom per partition column family. |
175 | */ |
176 | const char *const RDB_CF_NAME_QUALIFIER = "cfname" ; |
177 | |
178 | /* |
179 | Qualifier name for a custom per partition ttl duration. |
180 | */ |
181 | const char *const RDB_TTL_DURATION_QUALIFIER = "ttl_duration" ; |
182 | |
183 | /* |
184 | Qualifier name for a custom per partition ttl duration. |
185 | */ |
186 | const char *const RDB_TTL_COL_QUALIFIER = "ttl_col" ; |
187 | |
188 | /* |
189 | Default, minimal valid, and maximum valid sampling rate values when collecting |
190 | statistics about table. |
191 | */ |
192 | #define RDB_DEFAULT_TBL_STATS_SAMPLE_PCT 10 |
193 | #define RDB_TBL_STATS_SAMPLE_PCT_MIN 1 |
194 | #define RDB_TBL_STATS_SAMPLE_PCT_MAX 100 |
195 | |
196 | /* |
197 | Default and maximum values for rocksdb-compaction-sequential-deletes and |
198 | rocksdb-compaction-sequential-deletes-window to add basic boundary checking. |
199 | */ |
200 | #define DEFAULT_COMPACTION_SEQUENTIAL_DELETES 0 |
201 | #define MAX_COMPACTION_SEQUENTIAL_DELETES 2000000 |
202 | |
203 | #define DEFAULT_COMPACTION_SEQUENTIAL_DELETES_WINDOW 0 |
204 | #define MAX_COMPACTION_SEQUENTIAL_DELETES_WINDOW 2000000 |
205 | |
206 | /* |
207 | Default and maximum values for various compaction and flushing related |
208 | options. Numbers are based on the hardware we currently use and our internal |
209 | benchmarks which indicate that parallelization helps with the speed of |
210 | compactions. |
211 | |
212 | Ideally of course we'll use heuristic technique to determine the number of |
213 | CPU-s and derive the values from there. This however has its own set of |
214 | problems and we'll choose simplicity for now. |
215 | */ |
216 | #define MAX_BACKGROUND_JOBS 64 |
217 | |
218 | #define DEFAULT_SUBCOMPACTIONS 1 |
219 | #define MAX_SUBCOMPACTIONS 64 |
220 | |
221 | /* |
222 | Default value for rocksdb_sst_mgr_rate_bytes_per_sec = 0 (disabled). |
223 | */ |
224 | #define DEFAULT_SST_MGR_RATE_BYTES_PER_SEC 0 |
225 | |
226 | /* |
227 | Defines the field sizes for serializing XID object to a string representation. |
228 | string byte format: [field_size: field_value, ...] |
229 | [ |
230 | 8: XID.formatID, |
231 | 1: XID.gtrid_length, |
232 | 1: XID.bqual_length, |
233 | XID.gtrid_length + XID.bqual_length: XID.data |
234 | ] |
235 | */ |
236 | #define RDB_FORMATID_SZ 8 |
237 | #define RDB_GTRID_SZ 1 |
238 | #define RDB_BQUAL_SZ 1 |
239 | #define RDB_XIDHDR_LEN (RDB_FORMATID_SZ + RDB_GTRID_SZ + RDB_BQUAL_SZ) |
240 | |
241 | /* collations, used in MariaRocks */ |
242 | enum collations_used { |
243 | COLLATION_UTF8MB4_BIN = 46, |
244 | COLLATION_LATIN1_BIN = 47, |
245 | COLLATION_UTF16LE_BIN = 55, |
246 | COLLATION_UTF32_BIN = 61, |
247 | COLLATION_UTF16_BIN = 62, |
248 | COLLATION_BINARY = 63, |
249 | COLLATION_UTF8_BIN = 83 |
250 | }; |
251 | |
252 | /* |
253 | To fix an unhandled exception we specify the upper bound as LONGLONGMAX |
254 | instead of ULONGLONGMAX because the latter is -1 and causes an exception when |
255 | cast to jlong (signed) of JNI |
256 | |
257 | The reason behind the cast issue is the lack of unsigned int support in Java. |
258 | */ |
259 | #define MAX_RATE_LIMITER_BYTES_PER_SEC static_cast<uint64_t>(LONGLONG_MAX) |
260 | |
261 | /* |
262 | Hidden PK column (for tables with no primary key) is a longlong (aka 8 bytes). |
263 | static_assert() in code will validate this assumption. |
264 | */ |
265 | #define ROCKSDB_SIZEOF_HIDDEN_PK_COLUMN sizeof(longlong) |
266 | |
267 | /* |
268 | Bytes used to store TTL, in the beginning of all records for tables with TTL |
269 | enabled. |
270 | */ |
271 | #define ROCKSDB_SIZEOF_TTL_RECORD sizeof(longlong) |
272 | |
273 | #define ROCKSDB_SIZEOF_AUTOINC_VALUE sizeof(longlong) |
274 | |
275 | /* |
276 | Maximum index prefix length in bytes. |
277 | */ |
278 | #define MAX_INDEX_COL_LEN_LARGE 3072 |
279 | #define MAX_INDEX_COL_LEN_SMALL 767 |
280 | |
281 | /* |
282 | MyRocks specific error codes. NB! Please make sure that you will update |
283 | HA_ERR_ROCKSDB_LAST when adding new ones. Also update the strings in |
284 | rdb_error_messages to include any new error messages. |
285 | */ |
286 | #define HA_ERR_ROCKSDB_FIRST (HA_ERR_LAST + 1) |
287 | #define HA_ERR_ROCKSDB_PK_REQUIRED (HA_ERR_ROCKSDB_FIRST + 0) |
288 | #define HA_ERR_ROCKSDB_TABLE_DATA_DIRECTORY_NOT_SUPPORTED \ |
289 | (HA_ERR_ROCKSDB_FIRST + 1) |
290 | #define HA_ERR_ROCKSDB_TABLE_INDEX_DIRECTORY_NOT_SUPPORTED \ |
291 | (HA_ERR_ROCKSDB_FIRST + 2) |
292 | #define HA_ERR_ROCKSDB_COMMIT_FAILED (HA_ERR_ROCKSDB_FIRST + 3) |
293 | #define HA_ERR_ROCKSDB_BULK_LOAD (HA_ERR_ROCKSDB_FIRST + 4) |
294 | #define HA_ERR_ROCKSDB_CORRUPT_DATA (HA_ERR_ROCKSDB_FIRST + 5) |
295 | #define HA_ERR_ROCKSDB_CHECKSUM_MISMATCH (HA_ERR_ROCKSDB_FIRST + 6) |
296 | #define HA_ERR_ROCKSDB_INVALID_TABLE (HA_ERR_ROCKSDB_FIRST + 7) |
297 | #define HA_ERR_ROCKSDB_PROPERTIES (HA_ERR_ROCKSDB_FIRST + 8) |
298 | #define HA_ERR_ROCKSDB_MERGE_FILE_ERR (HA_ERR_ROCKSDB_FIRST + 9) |
299 | /* |
300 | Each error code below maps to a RocksDB status code found in: |
301 | rocksdb/include/rocksdb/status.h |
302 | */ |
303 | #define HA_ERR_ROCKSDB_STATUS_NOT_FOUND (HA_ERR_LAST + 10) |
304 | #define HA_ERR_ROCKSDB_STATUS_CORRUPTION (HA_ERR_LAST + 11) |
305 | #define HA_ERR_ROCKSDB_STATUS_NOT_SUPPORTED (HA_ERR_LAST + 12) |
306 | #define HA_ERR_ROCKSDB_STATUS_INVALID_ARGUMENT (HA_ERR_LAST + 13) |
307 | #define HA_ERR_ROCKSDB_STATUS_IO_ERROR (HA_ERR_LAST + 14) |
308 | #define HA_ERR_ROCKSDB_STATUS_NO_SPACE (HA_ERR_LAST + 15) |
309 | #define HA_ERR_ROCKSDB_STATUS_MERGE_IN_PROGRESS (HA_ERR_LAST + 16) |
310 | #define HA_ERR_ROCKSDB_STATUS_INCOMPLETE (HA_ERR_LAST + 17) |
311 | #define HA_ERR_ROCKSDB_STATUS_SHUTDOWN_IN_PROGRESS (HA_ERR_LAST + 18) |
312 | #define HA_ERR_ROCKSDB_STATUS_TIMED_OUT (HA_ERR_LAST + 19) |
313 | #define HA_ERR_ROCKSDB_STATUS_ABORTED (HA_ERR_LAST + 20) |
314 | #define HA_ERR_ROCKSDB_STATUS_LOCK_LIMIT (HA_ERR_LAST + 21) |
315 | #define HA_ERR_ROCKSDB_STATUS_BUSY (HA_ERR_LAST + 22) |
316 | #define HA_ERR_ROCKSDB_STATUS_DEADLOCK (HA_ERR_LAST + 23) |
317 | #define HA_ERR_ROCKSDB_STATUS_EXPIRED (HA_ERR_LAST + 24) |
318 | #define HA_ERR_ROCKSDB_STATUS_TRY_AGAIN (HA_ERR_LAST + 25) |
319 | #define HA_ERR_ROCKSDB_LAST HA_ERR_ROCKSDB_STATUS_TRY_AGAIN |
320 | |
321 | /** |
322 | @brief |
323 | Rdb_table_handler is a reference-counted structure storing information for |
324 | each open table. All the objects are stored in a global hash map. |
325 | |
326 | //TODO: join this with Rdb_tbl_def ? |
327 | */ |
328 | struct Rdb_table_handler { |
329 | char *m_table_name; |
330 | uint m_table_name_length; |
331 | int m_ref_count; |
332 | atomic_stat<int> m_lock_wait_timeout_counter; |
333 | atomic_stat<int> m_deadlock_counter; |
334 | |
335 | my_core::THR_LOCK m_thr_lock; ///< MySQL latch needed by m_db_lock |
336 | |
337 | /* Stores cumulative table statistics */ |
338 | my_io_perf_atomic_t m_io_perf_read; |
339 | my_io_perf_atomic_t m_io_perf_write; |
340 | Rdb_atomic_perf_counters m_table_perf_context; |
341 | |
342 | /* Stores cached memtable estimate statistics */ |
343 | std::atomic_uint m_mtcache_lock; |
344 | uint64_t m_mtcache_count; |
345 | uint64_t m_mtcache_size; |
346 | uint64_t m_mtcache_last_update; |
347 | }; |
348 | |
349 | class Rdb_key_def; |
350 | class Rdb_tbl_def; |
351 | class Rdb_transaction; |
352 | class Rdb_transaction_impl; |
353 | class Rdb_writebatch_impl; |
354 | class Rdb_field_encoder; |
355 | |
356 | const char *const rocksdb_hton_name = "ROCKSDB" ; |
357 | |
358 | typedef struct _gl_index_id_s { |
359 | uint32_t cf_id; |
360 | uint32_t index_id; |
361 | bool operator==(const struct _gl_index_id_s &other) const { |
362 | return cf_id == other.cf_id && index_id == other.index_id; |
363 | } |
364 | bool operator!=(const struct _gl_index_id_s &other) const { |
365 | return cf_id != other.cf_id || index_id != other.index_id; |
366 | } |
367 | bool operator<(const struct _gl_index_id_s &other) const { |
368 | return cf_id < other.cf_id || |
369 | (cf_id == other.cf_id && index_id < other.index_id); |
370 | } |
371 | bool operator<=(const struct _gl_index_id_s &other) const { |
372 | return cf_id < other.cf_id || |
373 | (cf_id == other.cf_id && index_id <= other.index_id); |
374 | } |
375 | bool operator>(const struct _gl_index_id_s &other) const { |
376 | return cf_id > other.cf_id || |
377 | (cf_id == other.cf_id && index_id > other.index_id); |
378 | } |
379 | bool operator>=(const struct _gl_index_id_s &other) const { |
380 | return cf_id > other.cf_id || |
381 | (cf_id == other.cf_id && index_id >= other.index_id); |
382 | } |
383 | } GL_INDEX_ID; |
384 | |
385 | enum operation_type : int { |
386 | ROWS_DELETED = 0, |
387 | ROWS_INSERTED, |
388 | ROWS_READ, |
389 | ROWS_UPDATED, |
390 | ROWS_DELETED_BLIND, |
391 | ROWS_EXPIRED, |
392 | ROWS_FILTERED, |
393 | ROWS_HIDDEN_NO_SNAPSHOT, |
394 | ROWS_MAX |
395 | }; |
396 | |
397 | enum query_type : int { QUERIES_POINT = 0, QUERIES_RANGE, QUERIES_MAX }; |
398 | |
399 | #if defined(HAVE_SCHED_GETCPU) |
400 | #define RDB_INDEXER get_sched_indexer_t |
401 | #else |
402 | #define RDB_INDEXER thread_id_indexer_t |
403 | #endif |
404 | |
405 | /* Global statistics struct used inside MyRocks */ |
406 | struct st_global_stats { |
407 | ib_counter_t<ulonglong, 64, RDB_INDEXER> rows[ROWS_MAX]; |
408 | |
409 | // system_rows_ stats are only for system |
410 | // tables. They are not counted in rows_* stats. |
411 | ib_counter_t<ulonglong, 64, RDB_INDEXER> system_rows[ROWS_MAX]; |
412 | |
413 | ib_counter_t<ulonglong, 64, RDB_INDEXER> queries[QUERIES_MAX]; |
414 | |
415 | ib_counter_t<ulonglong, 64, RDB_INDEXER> covered_secondary_key_lookups; |
416 | }; |
417 | |
418 | /* Struct used for exporting status to MySQL */ |
419 | struct st_export_stats { |
420 | ulonglong rows_deleted; |
421 | ulonglong rows_inserted; |
422 | ulonglong rows_read; |
423 | ulonglong rows_updated; |
424 | ulonglong rows_deleted_blind; |
425 | ulonglong rows_expired; |
426 | ulonglong rows_filtered; |
427 | ulonglong rows_hidden_no_snapshot; |
428 | |
429 | ulonglong system_rows_deleted; |
430 | ulonglong system_rows_inserted; |
431 | ulonglong system_rows_read; |
432 | ulonglong system_rows_updated; |
433 | |
434 | ulonglong queries_point; |
435 | ulonglong queries_range; |
436 | |
437 | ulonglong covered_secondary_key_lookups; |
438 | }; |
439 | |
440 | /* Struct used for exporting RocksDB memory status */ |
441 | struct st_memory_stats { |
442 | ulonglong memtable_total; |
443 | ulonglong memtable_unflushed; |
444 | }; |
445 | |
446 | /* Struct used for exporting RocksDB IO stalls stats */ |
447 | struct st_io_stall_stats { |
448 | ulonglong level0_slowdown; |
449 | ulonglong level0_slowdown_with_compaction; |
450 | ulonglong level0_numfiles; |
451 | ulonglong level0_numfiles_with_compaction; |
452 | ulonglong stop_for_pending_compaction_bytes; |
453 | ulonglong slowdown_for_pending_compaction_bytes; |
454 | ulonglong memtable_compaction; |
455 | ulonglong memtable_slowdown; |
456 | ulonglong total_stop; |
457 | ulonglong total_slowdown; |
458 | |
459 | st_io_stall_stats() |
460 | : level0_slowdown(0), level0_slowdown_with_compaction(0), |
461 | level0_numfiles(0), level0_numfiles_with_compaction(0), |
462 | stop_for_pending_compaction_bytes(0), |
463 | slowdown_for_pending_compaction_bytes(0), memtable_compaction(0), |
464 | memtable_slowdown(0), total_stop(0), total_slowdown(0) {} |
465 | }; |
466 | |
467 | } // namespace myrocks |
468 | |
469 | #include "./rdb_buff.h" |
470 | |
471 | /* Provide hash function for GL_INDEX_ID so we can include it in sets */ |
472 | namespace std { |
473 | template <> struct hash<myrocks::GL_INDEX_ID> { |
474 | std::size_t operator()(const myrocks::GL_INDEX_ID &gl_index_id) const { |
475 | const uint64_t val = |
476 | ((uint64_t)gl_index_id.cf_id << 32 | (uint64_t)gl_index_id.index_id); |
477 | return std::hash<uint64_t>()(val); |
478 | } |
479 | }; |
480 | } // namespace std |
481 | |
482 | namespace myrocks { |
483 | |
484 | /** |
485 | @brief |
486 | Class definition for ROCKSDB storage engine plugin handler |
487 | */ |
488 | |
489 | class ha_rocksdb : public my_core::handler { |
490 | my_core::THR_LOCK_DATA m_db_lock; ///< MySQL database lock |
491 | |
492 | Rdb_table_handler *m_table_handler; ///< Open table handler |
493 | |
494 | /* Iterator used for range scans and for full table/index scans */ |
495 | rocksdb::Iterator *m_scan_it; |
496 | |
497 | /* Same as handler::end_key but for start. Reverse-ordered scans need it */ |
498 | key_range m_save_start_range; |
499 | const key_range *m_start_range; |
500 | |
501 | /* Whether m_scan_it was created with skip_bloom=true */ |
502 | bool m_scan_it_skips_bloom; |
503 | |
504 | const rocksdb::Snapshot *m_scan_it_snapshot; |
505 | |
506 | Rdb_tbl_def *m_tbl_def; |
507 | |
508 | /* Primary Key encoder from KeyTupleFormat to StorageFormat */ |
509 | std::shared_ptr<Rdb_key_def> m_pk_descr; |
510 | |
511 | /* Array of index descriptors */ |
512 | std::shared_ptr<Rdb_key_def> *m_key_descr_arr; |
513 | |
514 | bool check_keyread_allowed(uint inx, uint part, bool all_parts) const; |
515 | |
516 | /* |
517 | Number of key parts in PK. This is the same as |
518 | table->key_info[table->s->primary_key].keyparts |
519 | */ |
520 | uint m_pk_key_parts; |
521 | |
522 | /* |
523 | TRUE <=> Primary Key columns can be decoded from the index |
524 | */ |
525 | mutable bool m_pk_can_be_decoded; |
526 | |
527 | /* |
528 | TRUE <=> Some fields in the PK may require unpack_info. |
529 | */ |
530 | bool m_maybe_unpack_info; |
531 | |
532 | uchar *m_pk_tuple; /* Buffer for storing PK in KeyTupleFormat */ |
533 | uchar *m_pk_packed_tuple; /* Buffer for storing PK in StorageFormat */ |
534 | // ^^ todo: change it to 'char*'? TODO: ^ can we join this with last_rowkey? |
535 | |
536 | /* |
537 | Temporary buffers for storing the key part of the Key/Value pair |
538 | for secondary indexes. |
539 | */ |
540 | uchar *m_sk_packed_tuple; |
541 | |
542 | /* |
543 | Temporary buffers for storing end key part of the Key/Value pair. |
544 | This is used for range scan only. |
545 | */ |
546 | uchar *m_end_key_packed_tuple; |
547 | |
548 | Rdb_string_writer m_sk_tails; |
549 | Rdb_string_writer m_pk_unpack_info; |
550 | |
551 | /* |
552 | ha_rockdb->index_read_map(.. HA_READ_KEY_EXACT or similar) will save here |
553 | mem-comparable form of the index lookup tuple. |
554 | */ |
555 | uchar *m_sk_match_prefix; |
556 | uint m_sk_match_length; |
557 | |
558 | /* Buffer space for the above */ |
559 | uchar *m_sk_match_prefix_buf; |
560 | |
561 | /* Second buffers, used by UPDATE. */ |
562 | uchar *m_sk_packed_tuple_old; |
563 | Rdb_string_writer m_sk_tails_old; |
564 | |
565 | /* Buffers used for duplicate checking during unique_index_creation */ |
566 | uchar *m_dup_sk_packed_tuple; |
567 | uchar *m_dup_sk_packed_tuple_old; |
568 | |
569 | /* Buffers used for passing upper/bound eq conditions. */ |
570 | uchar *m_eq_cond_lower_bound; |
571 | uchar *m_eq_cond_upper_bound; |
572 | rocksdb::Slice m_eq_cond_lower_bound_slice; |
573 | rocksdb::Slice m_eq_cond_upper_bound_slice; |
574 | |
575 | /* |
576 | Temporary space for packing VARCHARs (we provide it to |
577 | pack_record()/pack_index_tuple() calls). |
578 | */ |
579 | uchar *m_pack_buffer; |
580 | |
581 | /* |
582 | Pointer to the original TTL timestamp value (8 bytes) during UPDATE. |
583 | */ |
584 | char m_ttl_bytes[ROCKSDB_SIZEOF_TTL_RECORD]; |
585 | /* |
586 | The TTL timestamp value can change if the explicit TTL column is |
587 | updated. If we detect this when updating the PK, we indicate it here so |
588 | we know we must always update any SK's. |
589 | */ |
590 | bool m_ttl_bytes_updated; |
591 | |
592 | /* rowkey of the last record we've read, in StorageFormat. */ |
593 | String m_last_rowkey; |
594 | |
595 | /* Buffer used by convert_record_to_storage_format() */ |
596 | String m_storage_record; |
597 | |
598 | /* |
599 | Last retrieved record, in table->record[0] data format. |
600 | |
601 | This is used only when we get the record with rocksdb's Get() call (The |
602 | other option is when we get a rocksdb::Slice from an iterator) |
603 | */ |
604 | rocksdb::PinnableSlice m_retrieved_record; |
605 | |
606 | /* Type of locking to apply to rows */ |
607 | enum { RDB_LOCK_NONE, RDB_LOCK_READ, RDB_LOCK_WRITE } m_lock_rows; |
608 | |
609 | /* TRUE means we're doing an index-only read. FALSE means otherwise. */ |
610 | bool m_keyread_only; |
611 | |
612 | bool m_skip_scan_it_next_call; |
613 | |
614 | /* TRUE means we are accessing the first row after a snapshot was created */ |
615 | bool m_rnd_scan_is_new_snapshot; |
616 | |
617 | /* TRUE means the replication slave will use Read Free Replication */ |
618 | bool m_use_read_free_rpl; |
619 | |
620 | /* |
621 | TRUE means we should skip unique key checks for this table if the |
622 | replication lag gets too large |
623 | */ |
624 | bool m_skip_unique_check; |
625 | |
626 | /** |
627 | @brief |
628 | This is a bitmap of indexes (i.e. a set) whose keys (in future, values) may |
629 | be changed by this statement. Indexes that are not in the bitmap do not need |
630 | to be updated. |
631 | @note Valid inside UPDATE statements, IIF(m_update_scope_is_valid == true). |
632 | */ |
633 | my_core::key_map m_update_scope; |
634 | bool m_update_scope_is_valid; |
635 | |
636 | /* SST information used for bulk loading the primary key */ |
637 | std::shared_ptr<Rdb_sst_info> m_sst_info; |
638 | |
639 | /* |
640 | MySQL index number for duplicate key error |
641 | */ |
642 | int m_dupp_errkey; |
643 | |
644 | int create_key_defs(const TABLE *const table_arg, |
645 | Rdb_tbl_def *const tbl_def_arg, |
646 | const TABLE *const old_table_arg = nullptr, |
647 | const Rdb_tbl_def *const old_tbl_def_arg = nullptr) const |
648 | MY_ATTRIBUTE((__nonnull__(2, 3), __warn_unused_result__)); |
649 | int secondary_index_read(const int keyno, uchar *const buf) |
650 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
651 | void setup_iterator_for_rnd_scan(); |
652 | bool is_ascending(const Rdb_key_def &keydef, |
653 | enum ha_rkey_function find_flag) const |
654 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
655 | void setup_iterator_bounds(const Rdb_key_def &kd, |
656 | const rocksdb::Slice &eq_cond, |
657 | uchar *lower_bound_buf, |
658 | uchar *upper_bound_buf, |
659 | rocksdb::Slice *out_lower_bound, |
660 | rocksdb::Slice *out_upper_bound); |
661 | bool can_use_bloom_filter(THD *thd, const Rdb_key_def &kd, |
662 | const rocksdb::Slice &eq_cond, |
663 | const bool use_all_keys); |
664 | bool check_bloom_and_set_bounds(THD *thd, const Rdb_key_def &kd, |
665 | const rocksdb::Slice &eq_cond, |
666 | const bool use_all_keys, |
667 | uchar *lower_bound_buf, |
668 | uchar *upper_bound_buf, |
669 | rocksdb::Slice *out_lower_bound, |
670 | rocksdb::Slice *out_upper_bound); |
671 | void setup_scan_iterator(const Rdb_key_def &kd, rocksdb::Slice *slice, |
672 | const bool use_all_keys, const uint eq_cond_len) |
673 | MY_ATTRIBUTE((__nonnull__)); |
674 | void release_scan_iterator(void); |
675 | |
676 | rocksdb::Status |
677 | get_for_update(Rdb_transaction *const tx, |
678 | rocksdb::ColumnFamilyHandle *const column_family, |
679 | const rocksdb::Slice &key, |
680 | rocksdb::PinnableSlice *value) const; |
681 | |
682 | int get_row_by_rowid(uchar *const buf, const char *const rowid, |
683 | const uint rowid_size, const bool skip_lookup = false, |
684 | const bool skip_ttl_check = true) |
685 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
686 | int get_row_by_rowid(uchar *const buf, const uchar *const rowid, |
687 | const uint rowid_size, const bool skip_lookup = false, |
688 | const bool skip_ttl_check = true) |
689 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)) { |
690 | return get_row_by_rowid(buf, reinterpret_cast<const char *>(rowid), |
691 | rowid_size, skip_lookup, skip_ttl_check); |
692 | } |
693 | |
694 | void load_auto_incr_value(); |
695 | ulonglong load_auto_incr_value_from_index(); |
696 | void update_auto_incr_val(ulonglong val); |
697 | void update_auto_incr_val_from_field(); |
698 | rocksdb::Status get_datadic_auto_incr(Rdb_transaction *const tx, |
699 | const GL_INDEX_ID &gl_index_id, |
700 | ulonglong *new_val) const; |
701 | longlong update_hidden_pk_val(); |
702 | int load_hidden_pk_value() MY_ATTRIBUTE((__warn_unused_result__)); |
703 | int read_hidden_pk_id_from_rowkey(longlong *const hidden_pk_id) |
704 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
705 | bool can_use_single_delete(const uint &index) const |
706 | MY_ATTRIBUTE((__warn_unused_result__)); |
707 | bool is_blind_delete_enabled(); |
708 | bool skip_unique_check() const MY_ATTRIBUTE((__warn_unused_result__)); |
709 | #ifdef MARIAROCKS_NOT_YET // MDEV-10975 |
710 | void set_force_skip_unique_check(bool skip) override; |
711 | #endif |
712 | bool commit_in_the_middle() MY_ATTRIBUTE((__warn_unused_result__)); |
713 | bool do_bulk_commit(Rdb_transaction *const tx) |
714 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
715 | bool has_hidden_pk(const TABLE *const table) const |
716 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
717 | |
718 | void update_row_stats(const operation_type &type); |
719 | |
720 | void set_last_rowkey(const uchar *const old_data); |
721 | |
722 | /* |
723 | Array of table->s->fields elements telling how to store fields in the |
724 | record. |
725 | */ |
726 | Rdb_field_encoder *m_encoder_arr; |
727 | |
728 | /* Describes instructions on how to decode the field */ |
729 | class READ_FIELD { |
730 | public: |
731 | /* Points to Rdb_field_encoder describing the field */ |
732 | Rdb_field_encoder *m_field_enc; |
733 | /* if true, decode the field, otherwise skip it */ |
734 | bool m_decode; |
735 | /* Skip this many bytes before reading (or skipping) this field */ |
736 | int m_skip; |
737 | }; |
738 | |
739 | /* |
740 | This tells which table fields should be decoded (or skipped) when |
741 | decoding table row from (pk, encoded_row) pair. (Secondary keys are |
742 | just always decoded in full currently) |
743 | */ |
744 | std::vector<READ_FIELD> m_decoders_vect; |
745 | |
746 | /* |
747 | This tells if any field which is part of the key needs to be unpacked and |
748 | decoded. |
749 | */ |
750 | bool m_key_requested = false; |
751 | |
752 | /* Setup field_decoders based on type of scan and table->read_set */ |
753 | void setup_read_decoders(); |
754 | |
755 | /* |
756 | For the active index, indicates which columns must be covered for the |
757 | current lookup to be covered. If the bitmap field is null, that means this |
758 | index does not cover the current lookup for any record. |
759 | */ |
760 | MY_BITMAP m_lookup_bitmap = {nullptr, nullptr, nullptr, 0, 0}; |
761 | |
762 | /* |
763 | Number of bytes in on-disk (storage) record format that are used for |
764 | storing SQL NULL flags. |
765 | */ |
766 | uint m_null_bytes_in_rec; |
767 | |
768 | void get_storage_type(Rdb_field_encoder *const encoder, const uint &kp); |
769 | void setup_field_converters(); |
770 | int alloc_key_buffers(const TABLE *const table_arg, |
771 | const Rdb_tbl_def *const tbl_def_arg, |
772 | bool alloc_alter_buffers = false) |
773 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
774 | void free_key_buffers(); |
775 | |
776 | // the buffer size should be at least 2*Rdb_key_def::INDEX_NUMBER_SIZE |
777 | rocksdb::Range get_range(const int &i, uchar buf[]) const; |
778 | |
779 | /* |
780 | Perf timers for data reads |
781 | */ |
782 | Rdb_io_perf m_io_perf; |
783 | |
784 | /* |
785 | A counter of how many row checksums were checked for this table. Note that |
786 | this does not include checksums for secondary index entries. |
787 | */ |
788 | my_core::ha_rows m_row_checksums_checked; |
789 | |
790 | /* |
791 | Update stats |
792 | */ |
793 | void update_stats(void); |
794 | |
795 | public: |
796 | /* |
797 | The following two are currently only used for getting the range bounds |
798 | from QUICK_SELECT_DESC. |
799 | We don't need to implement prepare_index_key_scan[_map] because it is |
800 | only used with HA_READ_KEY_EXACT and HA_READ_PREFIX_LAST where one |
801 | can infer the bounds of the range being scanned, anyway. |
802 | */ |
803 | int prepare_index_scan() override; |
804 | int prepare_range_scan(const key_range *start_key, |
805 | const key_range *end_key) override; |
806 | |
807 | /* |
808 | Controls whether writes include checksums. This is updated from the session |
809 | variable |
810 | at the start of each query. |
811 | */ |
812 | bool m_store_row_debug_checksums; |
813 | |
814 | /* Same as above but for verifying checksums when reading */ |
815 | bool m_verify_row_debug_checksums; |
816 | int m_checksums_pct; |
817 | |
818 | ha_rocksdb(my_core::handlerton *const hton, |
819 | my_core::TABLE_SHARE *const table_arg); |
820 | ~ha_rocksdb() { |
821 | int err MY_ATTRIBUTE((__unused__)); |
822 | err = finalize_bulk_load(false); |
823 | if (err != 0) { |
824 | sql_print_error("RocksDB: Error %d finalizing bulk load while closing " |
825 | "handler." , |
826 | err); |
827 | } |
828 | } |
829 | |
830 | /** @brief |
831 | The name that will be used for display purposes. |
832 | */ |
833 | const char *table_type() const /*override*/ { |
834 | DBUG_ENTER_FUNC(); |
835 | // MariaDB: this function is not virtual, however ha_innodb |
836 | // declares it (and then never uses!) psergey-merge-todo:. |
837 | DBUG_RETURN(rocksdb_hton_name); |
838 | } |
839 | |
840 | /* The following is only used by SHOW KEYS: */ |
841 | const char *index_type(uint inx) override { |
842 | DBUG_ENTER_FUNC(); |
843 | |
844 | DBUG_RETURN("LSMTREE" ); |
845 | } |
846 | |
847 | /* |
848 | Not present in MariaDB: |
849 | const char **bas_ext() const override; |
850 | */ |
851 | |
852 | /* |
853 | Returns the name of the table's base name |
854 | */ |
855 | const std::string &get_table_basename() const; |
856 | |
857 | /** @brief |
858 | This is a list of flags that indicate what functionality the storage engine |
859 | implements. The current table flags are documented in handler.h |
860 | */ |
861 | ulonglong table_flags() const override { |
862 | DBUG_ENTER_FUNC(); |
863 | |
864 | /* |
865 | HA_BINLOG_STMT_CAPABLE |
866 | We are saying that this engine is just statement capable to have |
867 | an engine that can only handle statement-based logging. This is |
868 | used in testing. |
869 | HA_REC_NOT_IN_SEQ |
870 | If we don't set it, filesort crashes, because it assumes rowids are |
871 | 1..8 byte numbers |
872 | HA_PRIMARY_KEY_IN_READ_INDEX |
873 | This flag is always set, even for tables that: |
874 | - have no PK |
875 | - have some (or all) of PK that can't be decoded from the secondary |
876 | index. |
877 | */ |
878 | DBUG_RETURN(HA_BINLOG_ROW_CAPABLE | HA_BINLOG_STMT_CAPABLE | |
879 | HA_REC_NOT_IN_SEQ | HA_CAN_INDEX_BLOBS | |
880 | HA_PRIMARY_KEY_IN_READ_INDEX | |
881 | HA_PRIMARY_KEY_REQUIRED_FOR_POSITION | HA_NULL_IN_KEY | |
882 | HA_PARTIAL_COLUMN_READ | |
883 | HA_TABLE_SCAN_ON_INDEX); |
884 | } |
885 | |
886 | private: |
887 | bool init_with_fields(); /* no 'override' in MariaDB */ |
888 | public: |
889 | /** @brief |
890 | This is a bitmap of flags that indicates how the storage engine |
891 | implements indexes. The current index flags are documented in |
892 | handler.h. If you do not implement indexes, just return zero here. |
893 | |
894 | @details |
895 | part is the key part to check. First key part is 0. |
896 | If all_parts is set, MySQL wants to know the flags for the combined |
897 | index, up to and including 'part'. |
898 | */ |
899 | ulong index_flags(uint inx, uint part, bool all_parts) const override; |
900 | |
901 | const key_map *keys_to_use_for_scanning() override { |
902 | DBUG_ENTER_FUNC(); |
903 | |
904 | DBUG_RETURN(&key_map_full); |
905 | } |
906 | |
907 | bool primary_key_is_clustered() override { |
908 | DBUG_ENTER_FUNC(); |
909 | |
910 | DBUG_RETURN(true); |
911 | } |
912 | |
913 | bool should_store_row_debug_checksums() const { |
914 | return m_store_row_debug_checksums && (rand() % 100 < m_checksums_pct); |
915 | } |
916 | |
917 | int rename_table(const char *const from, const char *const to) override |
918 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
919 | |
920 | int convert_blob_from_storage_format(my_core::Field_blob *const blob, |
921 | Rdb_string_reader *const reader, |
922 | bool decode) |
923 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
924 | |
925 | int convert_varchar_from_storage_format( |
926 | my_core::Field_varstring *const field_var, |
927 | Rdb_string_reader *const reader, bool decode) |
928 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
929 | |
930 | int convert_field_from_storage_format(my_core::Field *const field, |
931 | Rdb_string_reader *const reader, |
932 | bool decode, uint len) |
933 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
934 | |
935 | int convert_record_from_storage_format(const rocksdb::Slice *const key, |
936 | const rocksdb::Slice *const value, |
937 | uchar *const buf) |
938 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
939 | |
940 | int convert_record_from_storage_format(const rocksdb::Slice *const key, |
941 | uchar *const buf) |
942 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
943 | |
944 | static const std::vector<std::string> parse_into_tokens(const std::string &s, |
945 | const char delim); |
946 | |
947 | static const std::string generate_cf_name(const uint index, |
948 | const TABLE *const table_arg, |
949 | const Rdb_tbl_def *const tbl_def_arg, |
950 | bool *per_part_match_found); |
951 | |
952 | static const char *get_key_name(const uint index, |
953 | const TABLE *const table_arg, |
954 | const Rdb_tbl_def *const tbl_def_arg) |
955 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
956 | |
957 | static const char *(const uint index, |
958 | const TABLE *const table_arg, |
959 | const Rdb_tbl_def *const tbl_def_arg) |
960 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
961 | |
962 | static const std::string (const TABLE *const table_arg) |
963 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
964 | |
965 | static bool is_hidden_pk(const uint index, const TABLE *const table_arg, |
966 | const Rdb_tbl_def *const tbl_def_arg) |
967 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
968 | |
969 | static uint pk_index(const TABLE *const table_arg, |
970 | const Rdb_tbl_def *const tbl_def_arg) |
971 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
972 | |
973 | static bool is_pk(const uint index, const TABLE *table_arg, |
974 | const Rdb_tbl_def *tbl_def_arg) |
975 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
976 | /** @brief |
977 | unireg.cc will call max_supported_record_length(), max_supported_keys(), |
978 | max_supported_key_parts(), uint max_supported_key_length() |
979 | to make sure that the storage engine can handle the data it is about to |
980 | send. Return *real* limits of your storage engine here; MySQL will do |
981 | min(your_limits, MySQL_limits) automatically. |
982 | */ |
983 | uint max_supported_record_length() const override { |
984 | DBUG_ENTER_FUNC(); |
985 | |
986 | DBUG_RETURN(HA_MAX_REC_LENGTH); |
987 | } |
988 | |
989 | uint max_supported_keys() const override { |
990 | DBUG_ENTER_FUNC(); |
991 | |
992 | DBUG_RETURN(MAX_INDEXES); |
993 | } |
994 | |
995 | uint max_supported_key_parts() const override { |
996 | DBUG_ENTER_FUNC(); |
997 | |
998 | DBUG_RETURN(MAX_REF_PARTS); |
999 | } |
1000 | |
1001 | uint max_supported_key_part_length() const override; |
1002 | |
1003 | /** @brief |
1004 | unireg.cc will call this to make sure that the storage engine can handle |
1005 | the data it is about to send. Return *real* limits of your storage engine |
1006 | here; MySQL will do min(your_limits, MySQL_limits) automatically. |
1007 | |
1008 | @details |
1009 | There is no need to implement ..._key_... methods if your engine doesn't |
1010 | support indexes. |
1011 | */ |
1012 | uint max_supported_key_length() const override { |
1013 | DBUG_ENTER_FUNC(); |
1014 | |
1015 | DBUG_RETURN(16 * 1024); /* just to return something*/ |
1016 | } |
1017 | |
1018 | /** |
1019 | TODO: return actual upper bound of number of records in the table. |
1020 | (e.g. save number of records seen on full table scan and/or use file size |
1021 | as upper bound) |
1022 | */ |
1023 | ha_rows estimate_rows_upper_bound() override { |
1024 | DBUG_ENTER_FUNC(); |
1025 | |
1026 | DBUG_RETURN(HA_POS_ERROR); |
1027 | } |
1028 | |
1029 | /* At the moment, we're ok with default handler::index_init() implementation. |
1030 | */ |
1031 | int index_read_map(uchar *const buf, const uchar *const key, |
1032 | key_part_map keypart_map, |
1033 | enum ha_rkey_function find_flag) override |
1034 | MY_ATTRIBUTE((__warn_unused_result__)); |
1035 | |
1036 | int index_read_map_impl(uchar *const buf, const uchar *const key, |
1037 | key_part_map keypart_map, |
1038 | enum ha_rkey_function find_flag, |
1039 | const key_range *end_key) |
1040 | MY_ATTRIBUTE((__warn_unused_result__)); |
1041 | |
1042 | bool is_using_full_key(key_part_map keypart_map, uint actual_key_parts); |
1043 | int read_range_first(const key_range *const start_key, |
1044 | const key_range *const end_key, bool eq_range, |
1045 | bool sorted) override |
1046 | MY_ATTRIBUTE((__warn_unused_result__)); |
1047 | |
1048 | virtual double scan_time() override { |
1049 | DBUG_ENTER_FUNC(); |
1050 | |
1051 | DBUG_RETURN( |
1052 | static_cast<double>((stats.records + stats.deleted) / 20.0 + 10)); |
1053 | } |
1054 | |
1055 | virtual double read_time(uint, uint, ha_rows rows) override; |
1056 | |
1057 | int open(const char *const name, int mode, uint test_if_locked) override |
1058 | MY_ATTRIBUTE((__warn_unused_result__)); |
1059 | int close(void) override MY_ATTRIBUTE((__warn_unused_result__)); |
1060 | |
1061 | int write_row(uchar *const buf) override |
1062 | MY_ATTRIBUTE((__warn_unused_result__)); |
1063 | int update_row(const uchar *const old_data, const uchar *const new_data) override |
1064 | MY_ATTRIBUTE((__warn_unused_result__)); |
1065 | int delete_row(const uchar *const buf) override |
1066 | MY_ATTRIBUTE((__warn_unused_result__)); |
1067 | rocksdb::Status delete_or_singledelete(uint index, Rdb_transaction *const tx, |
1068 | rocksdb::ColumnFamilyHandle *const cf, |
1069 | const rocksdb::Slice &key) |
1070 | MY_ATTRIBUTE((__warn_unused_result__)); |
1071 | |
1072 | int index_next(uchar *const buf) override |
1073 | MY_ATTRIBUTE((__warn_unused_result__)); |
1074 | int index_next_with_direction(uchar *const buf, bool move_forward) |
1075 | MY_ATTRIBUTE((__warn_unused_result__)); |
1076 | int index_prev(uchar *const buf) override |
1077 | MY_ATTRIBUTE((__warn_unused_result__)); |
1078 | |
1079 | int index_first(uchar *const buf) override |
1080 | MY_ATTRIBUTE((__warn_unused_result__)); |
1081 | int index_last(uchar *const buf) override |
1082 | MY_ATTRIBUTE((__warn_unused_result__)); |
1083 | |
1084 | class Item *idx_cond_push(uint keyno, class Item *const idx_cond) override; |
1085 | /* |
1086 | Default implementation from cancel_pushed_idx_cond() suits us |
1087 | */ |
1088 | private: |
1089 | struct key_def_cf_info { |
1090 | rocksdb::ColumnFamilyHandle *cf_handle; |
1091 | bool is_reverse_cf; |
1092 | bool is_per_partition_cf; |
1093 | }; |
1094 | |
1095 | struct update_row_info { |
1096 | Rdb_transaction *tx; |
1097 | const uchar *new_data; |
1098 | const uchar *old_data; |
1099 | rocksdb::Slice new_pk_slice; |
1100 | rocksdb::Slice old_pk_slice; |
1101 | rocksdb::Slice old_pk_rec; |
1102 | |
1103 | // "unpack_info" data for the new PK value |
1104 | Rdb_string_writer *new_pk_unpack_info; |
1105 | |
1106 | longlong hidden_pk_id; |
1107 | bool skip_unique_check; |
1108 | |
1109 | // In certain cases, TTL is enabled on a table, as well as an explicit TTL |
1110 | // column. The TTL column can be part of either the key or the value part |
1111 | // of the record. If it is part of the key, we store the offset here. |
1112 | // |
1113 | // Later on, we use this offset to store the TTL in the value part of the |
1114 | // record, which we can then access in the compaction filter. |
1115 | // |
1116 | // Set to UINT_MAX by default to indicate that the TTL is not in key. |
1117 | uint ttl_pk_offset = UINT_MAX; |
1118 | }; |
1119 | |
1120 | /* |
1121 | Used to check for duplicate entries during fast unique secondary index |
1122 | creation. |
1123 | */ |
1124 | struct unique_sk_buf_info { |
1125 | bool sk_buf_switch = false; |
1126 | rocksdb::Slice sk_memcmp_key; |
1127 | rocksdb::Slice sk_memcmp_key_old; |
1128 | uchar *dup_sk_buf; |
1129 | uchar *dup_sk_buf_old; |
1130 | |
1131 | /* |
1132 | This method is meant to be called back to back during inplace creation |
1133 | of unique indexes. It will switch between two buffers, which |
1134 | will each store the memcmp form of secondary keys, which are then |
1135 | converted to slices in sk_memcmp_key or sk_memcmp_key_old. |
1136 | |
1137 | Switching buffers on each iteration allows us to retain the |
1138 | sk_memcmp_key_old value for duplicate comparison. |
1139 | */ |
1140 | inline uchar *swap_and_get_sk_buf() { |
1141 | sk_buf_switch = !sk_buf_switch; |
1142 | return sk_buf_switch ? dup_sk_buf : dup_sk_buf_old; |
1143 | } |
1144 | }; |
1145 | |
1146 | int create_cfs(const TABLE *const table_arg, Rdb_tbl_def *const tbl_def_arg, |
1147 | std::array<struct key_def_cf_info, MAX_INDEXES + 1> *const cfs) |
1148 | const MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1149 | |
1150 | int create_key_def(const TABLE *const table_arg, const uint &i, |
1151 | const Rdb_tbl_def *const tbl_def_arg, |
1152 | std::shared_ptr<Rdb_key_def> *const new_key_def, |
1153 | const struct key_def_cf_info &cf_info) const |
1154 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1155 | |
1156 | int create_inplace_key_defs( |
1157 | const TABLE *const table_arg, Rdb_tbl_def *vtbl_def_arg, |
1158 | const TABLE *const old_table_arg, |
1159 | const Rdb_tbl_def *const old_tbl_def_arg, |
1160 | const std::array<key_def_cf_info, MAX_INDEXES + 1> &cfs) const |
1161 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1162 | |
1163 | std::unordered_map<std::string, uint> |
1164 | get_old_key_positions(const TABLE *table_arg, const Rdb_tbl_def *tbl_def_arg, |
1165 | const TABLE *old_table_arg, |
1166 | const Rdb_tbl_def *old_tbl_def_arg) const |
1167 | MY_ATTRIBUTE((__nonnull__)); |
1168 | |
1169 | int compare_key_parts(const KEY *const old_key, |
1170 | const KEY *const new_key) const; |
1171 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1172 | |
1173 | int compare_keys(const KEY *const old_key, const KEY *const new_key) const |
1174 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1175 | |
1176 | int convert_record_to_storage_format(const struct update_row_info &row_info, |
1177 | rocksdb::Slice *const packed_rec) |
1178 | MY_ATTRIBUTE((__nonnull__)); |
1179 | |
1180 | bool should_hide_ttl_rec(const Rdb_key_def &kd, |
1181 | const rocksdb::Slice &ttl_rec_val, |
1182 | const int64_t curr_ts) |
1183 | MY_ATTRIBUTE((__warn_unused_result__)); |
1184 | void rocksdb_skip_expired_records(const Rdb_key_def &kd, |
1185 | rocksdb::Iterator *const iter, |
1186 | bool seek_backward); |
1187 | |
1188 | int index_first_intern(uchar *buf) |
1189 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1190 | int index_last_intern(uchar *buf) |
1191 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1192 | |
1193 | int find_icp_matching_index_rec(const bool &move_forward, uchar *const buf) |
1194 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1195 | |
1196 | void calc_updated_indexes(); |
1197 | int update_write_row(const uchar *const old_data, const uchar *const new_data, |
1198 | const bool skip_unique_check) |
1199 | MY_ATTRIBUTE((__warn_unused_result__)); |
1200 | int get_pk_for_update(struct update_row_info *const row_info); |
1201 | int check_and_lock_unique_pk(const uint &key_id, |
1202 | const struct update_row_info &row_info, |
1203 | bool *const found, bool *const pk_changed) |
1204 | MY_ATTRIBUTE((__warn_unused_result__)); |
1205 | int check_and_lock_sk(const uint &key_id, |
1206 | const struct update_row_info &row_info, |
1207 | bool *const found) |
1208 | MY_ATTRIBUTE((__warn_unused_result__)); |
1209 | int check_uniqueness_and_lock(const struct update_row_info &row_info, |
1210 | bool *const pk_changed) |
1211 | MY_ATTRIBUTE((__warn_unused_result__)); |
1212 | bool over_bulk_load_threshold(int *err) |
1213 | MY_ATTRIBUTE((__warn_unused_result__)); |
1214 | int check_duplicate_sk(const TABLE *table_arg, const Rdb_key_def &index, |
1215 | const rocksdb::Slice *key, |
1216 | struct unique_sk_buf_info *sk_info) |
1217 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1218 | int bulk_load_key(Rdb_transaction *const tx, const Rdb_key_def &kd, |
1219 | const rocksdb::Slice &key, const rocksdb::Slice &value, |
1220 | bool sort) |
1221 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1222 | void update_bytes_written(ulonglong bytes_written); |
1223 | int update_pk(const Rdb_key_def &kd, const struct update_row_info &row_info, |
1224 | const bool &pk_changed) MY_ATTRIBUTE((__warn_unused_result__)); |
1225 | int update_sk(const TABLE *const table_arg, const Rdb_key_def &kd, |
1226 | const struct update_row_info &row_info) |
1227 | MY_ATTRIBUTE((__warn_unused_result__)); |
1228 | int update_indexes(const struct update_row_info &row_info, |
1229 | const bool &pk_changed) |
1230 | MY_ATTRIBUTE((__warn_unused_result__)); |
1231 | |
1232 | int read_key_exact(const Rdb_key_def &kd, rocksdb::Iterator *const iter, |
1233 | const bool &using_full_key, |
1234 | const rocksdb::Slice &key_slice, |
1235 | const int64_t ttl_filter_ts) |
1236 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1237 | int read_before_key(const Rdb_key_def &kd, const bool &using_full_key, |
1238 | const rocksdb::Slice &key_slice, |
1239 | const int64_t ttl_filter_ts) |
1240 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1241 | int read_after_key(const Rdb_key_def &kd, const rocksdb::Slice &key_slice, |
1242 | const int64_t ttl_filter_ts) |
1243 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1244 | int position_to_correct_key( |
1245 | const Rdb_key_def &kd, const enum ha_rkey_function &find_flag, |
1246 | const bool &full_key_match, const uchar *const key, |
1247 | const key_part_map &keypart_map, const rocksdb::Slice &key_slice, |
1248 | bool *const move_forward, const int64_t ttl_filter_ts) |
1249 | MY_ATTRIBUTE((__warn_unused_result__)); |
1250 | |
1251 | int read_row_from_primary_key(uchar *const buf) |
1252 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1253 | int read_row_from_secondary_key(uchar *const buf, const Rdb_key_def &kd, |
1254 | bool move_forward) |
1255 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1256 | |
1257 | int calc_eq_cond_len(const Rdb_key_def &kd, |
1258 | const enum ha_rkey_function &find_flag, |
1259 | const rocksdb::Slice &slice, |
1260 | const int &bytes_changed_by_succ, |
1261 | const key_range *const end_key, |
1262 | uint *const end_key_packed_size) |
1263 | MY_ATTRIBUTE((__warn_unused_result__)); |
1264 | |
1265 | Rdb_tbl_def *get_table_if_exists(const char *const tablename) |
1266 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1267 | void read_thd_vars(THD *const thd) MY_ATTRIBUTE((__nonnull__)); |
1268 | |
1269 | bool contains_foreign_key(THD *const thd) |
1270 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1271 | |
1272 | int inplace_populate_sk( |
1273 | TABLE *const table_arg, |
1274 | const std::unordered_set<std::shared_ptr<Rdb_key_def>> &indexes) |
1275 | MY_ATTRIBUTE((__nonnull__, __warn_unused_result__)); |
1276 | |
1277 | int finalize_bulk_load(bool print_client_error = true) |
1278 | MY_ATTRIBUTE((__warn_unused_result__)); |
1279 | |
1280 | public: |
1281 | int index_init(uint idx, bool sorted) override |
1282 | MY_ATTRIBUTE((__warn_unused_result__)); |
1283 | int index_end() override MY_ATTRIBUTE((__warn_unused_result__)); |
1284 | |
1285 | void unlock_row() override; |
1286 | |
1287 | /** @brief |
1288 | Unlike index_init(), rnd_init() can be called two consecutive times |
1289 | without rnd_end() in between (it only makes sense if scan=1). In this |
1290 | case, the second call should prepare for the new table scan (e.g if |
1291 | rnd_init() allocates the cursor, the second call should position the |
1292 | cursor to the start of the table; no need to deallocate and allocate |
1293 | it again. This is a required method. |
1294 | */ |
1295 | int rnd_init(bool scan) override MY_ATTRIBUTE((__warn_unused_result__)); |
1296 | int rnd_end() override MY_ATTRIBUTE((__warn_unused_result__)); |
1297 | |
1298 | int rnd_next(uchar *const buf) override |
1299 | MY_ATTRIBUTE((__warn_unused_result__)); |
1300 | int rnd_next_with_direction(uchar *const buf, bool move_forward) |
1301 | MY_ATTRIBUTE((__warn_unused_result__)); |
1302 | |
1303 | int rnd_pos(uchar *const buf, uchar *const pos) override |
1304 | MY_ATTRIBUTE((__warn_unused_result__)); |
1305 | void position(const uchar *const record) override; |
1306 | int info(uint) override; |
1307 | |
1308 | /* This function will always return success, therefore no annotation related |
1309 | * to checking the return value. Can't change the signature because it's |
1310 | * required by the interface. */ |
1311 | int (enum ha_extra_function operation) override; |
1312 | |
1313 | int start_stmt(THD *const thd, thr_lock_type lock_type) override |
1314 | MY_ATTRIBUTE((__warn_unused_result__)); |
1315 | int external_lock(THD *const thd, int lock_type) override |
1316 | MY_ATTRIBUTE((__warn_unused_result__)); |
1317 | int truncate() override MY_ATTRIBUTE((__warn_unused_result__)); |
1318 | |
1319 | int reset() override { |
1320 | DBUG_ENTER_FUNC(); |
1321 | |
1322 | /* Free blob data */ |
1323 | m_retrieved_record.Reset(); |
1324 | |
1325 | DBUG_RETURN(HA_EXIT_SUCCESS); |
1326 | } |
1327 | |
1328 | int check(THD *const thd, HA_CHECK_OPT *const check_opt) override |
1329 | MY_ATTRIBUTE((__warn_unused_result__)); |
1330 | int remove_rows(Rdb_tbl_def *const tbl); |
1331 | ha_rows records_in_range(uint inx, key_range *const min_key, |
1332 | key_range *const max_key) override |
1333 | MY_ATTRIBUTE((__warn_unused_result__)); |
1334 | int delete_table(const char *const from) override |
1335 | MY_ATTRIBUTE((__warn_unused_result__)); |
1336 | int create(const char *const name, TABLE *const form, |
1337 | HA_CREATE_INFO *const create_info) override |
1338 | MY_ATTRIBUTE((__warn_unused_result__)); |
1339 | bool check_if_incompatible_data(HA_CREATE_INFO *const info, |
1340 | uint table_changes) override |
1341 | MY_ATTRIBUTE((__warn_unused_result__)); |
1342 | |
1343 | THR_LOCK_DATA **store_lock(THD *const thd, THR_LOCK_DATA **to, |
1344 | enum thr_lock_type lock_type) override |
1345 | MY_ATTRIBUTE((__warn_unused_result__)); |
1346 | |
1347 | my_bool register_query_cache_table(THD *const thd, const char *table_key, |
1348 | uint key_length, |
1349 | qc_engine_callback *const engine_callback, |
1350 | ulonglong *const engine_data) override { |
1351 | DBUG_ENTER_FUNC(); |
1352 | |
1353 | /* Currently, we don't support query cache */ |
1354 | DBUG_RETURN(FALSE); |
1355 | } |
1356 | |
1357 | bool get_error_message(const int error, String *const buf) override |
1358 | MY_ATTRIBUTE((__nonnull__)); |
1359 | |
1360 | static int rdb_error_to_mysql(const rocksdb::Status &s, |
1361 | const char *msg = nullptr) |
1362 | MY_ATTRIBUTE((__warn_unused_result__)); |
1363 | |
1364 | void get_auto_increment(ulonglong offset, ulonglong increment, |
1365 | ulonglong nb_desired_values, |
1366 | ulonglong *const first_value, |
1367 | ulonglong *const nb_reserved_values) override; |
1368 | void update_create_info(HA_CREATE_INFO *const create_info) override; |
1369 | int optimize(THD *const thd, HA_CHECK_OPT *const check_opt) override |
1370 | MY_ATTRIBUTE((__warn_unused_result__)); |
1371 | int analyze(THD *const thd, HA_CHECK_OPT *const check_opt) override |
1372 | MY_ATTRIBUTE((__warn_unused_result__)); |
1373 | int calculate_stats(const TABLE *const table_arg, THD *const thd, |
1374 | HA_CHECK_OPT *const check_opt) |
1375 | MY_ATTRIBUTE((__warn_unused_result__)); |
1376 | |
1377 | enum_alter_inplace_result check_if_supported_inplace_alter( |
1378 | TABLE *altered_table, |
1379 | my_core::Alter_inplace_info *const ha_alter_info) override; |
1380 | |
1381 | bool prepare_inplace_alter_table( |
1382 | TABLE *const altered_table, |
1383 | my_core::Alter_inplace_info *const ha_alter_info) override; |
1384 | |
1385 | bool inplace_alter_table( |
1386 | TABLE *const altered_table, |
1387 | my_core::Alter_inplace_info *const ha_alter_info) override; |
1388 | |
1389 | bool |
1390 | commit_inplace_alter_table(TABLE *const altered_table, |
1391 | my_core::Alter_inplace_info *const ha_alter_info, |
1392 | bool commit) override; |
1393 | |
1394 | #ifdef MARIAROCKS_NOT_YET // MDEV-10976 |
1395 | void set_use_read_free_rpl(const char *const whitelist); |
1396 | #endif |
1397 | void set_skip_unique_check_tables(const char *const whitelist); |
1398 | |
1399 | #ifdef MARIAROCKS_NOT_YET // MDEV-10976 |
1400 | public: |
1401 | virtual void rpl_before_delete_rows() override; |
1402 | virtual void rpl_after_delete_rows() override; |
1403 | virtual void rpl_before_update_rows() override; |
1404 | virtual void rpl_after_update_rows() override; |
1405 | virtual bool use_read_free_rpl(); |
1406 | #endif // MARIAROCKS_NOT_YET |
1407 | |
1408 | private: |
1409 | /* Flags tracking if we are inside different replication operation */ |
1410 | bool m_in_rpl_delete_rows; |
1411 | bool m_in_rpl_update_rows; |
1412 | |
1413 | bool m_force_skip_unique_check; |
1414 | }; |
1415 | |
1416 | /* |
1417 | Helper class for in-place alter, for storing handler context between inplace |
1418 | alter calls |
1419 | */ |
1420 | struct Rdb_inplace_alter_ctx : public my_core::inplace_alter_handler_ctx { |
1421 | /* The new table definition */ |
1422 | Rdb_tbl_def *const m_new_tdef; |
1423 | |
1424 | /* Stores the original key definitions */ |
1425 | std::shared_ptr<Rdb_key_def> *const m_old_key_descr; |
1426 | |
1427 | /* Stores the new key definitions */ |
1428 | std::shared_ptr<Rdb_key_def> *m_new_key_descr; |
1429 | |
1430 | /* Stores the old number of key definitions */ |
1431 | const uint m_old_n_keys; |
1432 | |
1433 | /* Stores the new number of key definitions */ |
1434 | const uint m_new_n_keys; |
1435 | |
1436 | /* Stores the added key glids */ |
1437 | const std::unordered_set<std::shared_ptr<Rdb_key_def>> m_added_indexes; |
1438 | |
1439 | /* Stores the dropped key glids */ |
1440 | const std::unordered_set<GL_INDEX_ID> m_dropped_index_ids; |
1441 | |
1442 | /* Stores number of keys to add */ |
1443 | const uint m_n_added_keys; |
1444 | |
1445 | /* Stores number of keys to drop */ |
1446 | const uint m_n_dropped_keys; |
1447 | |
1448 | /* Stores the largest current auto increment value in the index */ |
1449 | const ulonglong m_max_auto_incr; |
1450 | |
1451 | Rdb_inplace_alter_ctx( |
1452 | Rdb_tbl_def *new_tdef, std::shared_ptr<Rdb_key_def> *old_key_descr, |
1453 | std::shared_ptr<Rdb_key_def> *new_key_descr, uint old_n_keys, |
1454 | uint new_n_keys, |
1455 | std::unordered_set<std::shared_ptr<Rdb_key_def>> added_indexes, |
1456 | std::unordered_set<GL_INDEX_ID> dropped_index_ids, uint n_added_keys, |
1457 | uint n_dropped_keys, ulonglong max_auto_incr) |
1458 | : my_core::inplace_alter_handler_ctx(), m_new_tdef(new_tdef), |
1459 | m_old_key_descr(old_key_descr), m_new_key_descr(new_key_descr), |
1460 | m_old_n_keys(old_n_keys), m_new_n_keys(new_n_keys), |
1461 | m_added_indexes(added_indexes), m_dropped_index_ids(dropped_index_ids), |
1462 | m_n_added_keys(n_added_keys), m_n_dropped_keys(n_dropped_keys), |
1463 | m_max_auto_incr(max_auto_incr) {} |
1464 | |
1465 | ~Rdb_inplace_alter_ctx() {} |
1466 | |
1467 | private: |
1468 | /* Disable Copying */ |
1469 | Rdb_inplace_alter_ctx(const Rdb_inplace_alter_ctx &); |
1470 | Rdb_inplace_alter_ctx &operator=(const Rdb_inplace_alter_ctx &); |
1471 | }; |
1472 | |
1473 | // file name indicating RocksDB data corruption |
1474 | std::string rdb_corruption_marker_file_name(); |
1475 | |
1476 | const int MYROCKS_MARIADB_PLUGIN_MATURITY_LEVEL= MariaDB_PLUGIN_MATURITY_STABLE; |
1477 | |
1478 | extern bool prevent_myrocks_loading; |
1479 | |
1480 | void sql_print_verbose_info(const char *format, ...); |
1481 | |
1482 | } // namespace myrocks |
1483 | |
1484 | |