Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * vacuum.c
4 : : * The postgres vacuum cleaner.
5 : : *
6 : : * This file includes (a) control and dispatch code for VACUUM and ANALYZE
7 : : * commands, (b) code to compute various vacuum thresholds, and (c) index
8 : : * vacuum code.
9 : : *
10 : : * VACUUM for heap AM is implemented in vacuumlazy.c, parallel vacuum in
11 : : * vacuumparallel.c, ANALYZE in analyze.c, and VACUUM FULL is a variant of
12 : : * REPACK, handled in repack.c.
13 : : *
14 : : *
15 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
16 : : * Portions Copyright (c) 1994, Regents of the University of California
17 : : *
18 : : *
19 : : * IDENTIFICATION
20 : : * src/backend/commands/vacuum.c
21 : : *
22 : : *-------------------------------------------------------------------------
23 : : */
24 : : #include "postgres.h"
25 : :
26 : : #include <math.h>
27 : :
28 : : #include "access/clog.h"
29 : : #include "access/commit_ts.h"
30 : : #include "access/genam.h"
31 : : #include "access/heapam.h"
32 : : #include "access/htup_details.h"
33 : : #include "access/multixact.h"
34 : : #include "access/reloptions.h"
35 : : #include "access/tableam.h"
36 : : #include "access/transam.h"
37 : : #include "access/xact.h"
38 : : #include "catalog/namespace.h"
39 : : #include "catalog/pg_database.h"
40 : : #include "catalog/pg_inherits.h"
41 : : #include "commands/async.h"
42 : : #include "commands/defrem.h"
43 : : #include "commands/progress.h"
44 : : #include "commands/repack.h"
45 : : #include "commands/vacuum.h"
46 : : #include "miscadmin.h"
47 : : #include "nodes/makefuncs.h"
48 : : #include "pgstat.h"
49 : : #include "postmaster/autovacuum.h"
50 : : #include "postmaster/bgworker_internals.h"
51 : : #include "postmaster/interrupt.h"
52 : : #include "storage/bufmgr.h"
53 : : #include "storage/lmgr.h"
54 : : #include "storage/pmsignal.h"
55 : : #include "storage/proc.h"
56 : : #include "storage/procarray.h"
57 : : #include "utils/acl.h"
58 : : #include "utils/fmgroids.h"
59 : : #include "utils/guc.h"
60 : : #include "utils/guc_hooks.h"
61 : : #include "utils/injection_point.h"
62 : : #include "utils/memutils.h"
63 : : #include "utils/snapmgr.h"
64 : : #include "utils/syscache.h"
65 : : #include "utils/wait_event.h"
66 : :
67 : : /*
68 : : * Minimum interval for cost-based vacuum delay reports from a parallel worker.
69 : : * This aims to avoid sending too many messages and waking up the leader too
70 : : * frequently.
71 : : */
72 : : #define PARALLEL_VACUUM_DELAY_REPORT_INTERVAL_NS (NS_PER_S)
73 : :
74 : : /*
75 : : * GUC parameters
76 : : */
77 : : int vacuum_freeze_min_age;
78 : : int vacuum_freeze_table_age;
79 : : int vacuum_multixact_freeze_min_age;
80 : : int vacuum_multixact_freeze_table_age;
81 : : int vacuum_failsafe_age;
82 : : int vacuum_multixact_failsafe_age;
83 : : double vacuum_max_eager_freeze_failure_rate;
84 : : bool track_cost_delay_timing;
85 : : bool vacuum_truncate;
86 : :
87 : : /*
88 : : * Variables for cost-based vacuum delay. The defaults differ between
89 : : * autovacuum and vacuum. They should be set with the appropriate GUC value in
90 : : * vacuum code. They are initialized here to the defaults for client backends
91 : : * executing VACUUM or ANALYZE.
92 : : */
93 : : double vacuum_cost_delay = 0;
94 : : int vacuum_cost_limit = 200;
95 : :
96 : : /* Variable for reporting cost-based vacuum delay from parallel workers. */
97 : : int64 parallel_vacuum_worker_delay_ns = 0;
98 : :
99 : : /*
100 : : * VacuumFailsafeActive is a defined as a global so that we can determine
101 : : * whether or not to re-enable cost-based vacuum delay when vacuuming a table.
102 : : * If failsafe mode has been engaged, we will not re-enable cost-based delay
103 : : * for the table until after vacuuming has completed, regardless of other
104 : : * settings.
105 : : *
106 : : * Only VACUUM code should inspect this variable and only table access methods
107 : : * should set it to true. In Table AM-agnostic VACUUM code, this variable is
108 : : * inspected to determine whether or not to allow cost-based delays. Table AMs
109 : : * are free to set it if they desire this behavior, but it is false by default
110 : : * and reset to false in between vacuuming each relation.
111 : : */
112 : : bool VacuumFailsafeActive = false;
113 : :
114 : : /*
115 : : * Variables for cost-based parallel vacuum. See comments atop
116 : : * compute_parallel_delay to understand how it works.
117 : : */
118 : : pg_atomic_uint32 *VacuumSharedCostBalance = NULL;
119 : : pg_atomic_uint32 *VacuumActiveNWorkers = NULL;
120 : : int VacuumCostBalanceLocal = 0;
121 : :
122 : : /* non-export function prototypes */
123 : : static List *expand_vacuum_rel(VacuumRelation *vrel,
124 : : MemoryContext vac_context, int options);
125 : : static List *get_all_vacuum_rels(MemoryContext vac_context, int options);
126 : : static void vac_truncate_clog(TransactionId frozenXID,
127 : : MultiXactId minMulti,
128 : : TransactionId lastSaneFrozenXid,
129 : : MultiXactId lastSaneMinMulti);
130 : : static bool vacuum_rel(Oid relid, RangeVar *relation, VacuumParams params,
131 : : BufferAccessStrategy bstrategy, bool isTopLevel);
132 : : static double compute_parallel_delay(void);
133 : : static VacOptValue get_vacoptval_from_boolean(DefElem *def);
134 : : static bool vac_tid_reaped(ItemPointer itemptr, void *state);
135 : :
136 : : /*
137 : : * GUC check function to ensure GUC value specified is within the allowable
138 : : * range.
139 : : */
140 : : bool
141 : 1325 : check_vacuum_buffer_usage_limit(int *newval, void **extra,
142 : : GucSource source)
143 : : {
144 : : /* Value upper and lower hard limits are inclusive */
145 [ + - + - ]: 1325 : if (*newval == 0 || (*newval >= MIN_BAS_VAC_RING_SIZE_KB &&
146 [ + - ]: 1325 : *newval <= MAX_BAS_VAC_RING_SIZE_KB))
147 : 1325 : return true;
148 : :
149 : : /* Value does not fall within any allowable range */
150 : 0 : GUC_check_errdetail("\"%s\" must be 0 or between %d kB and %d kB.",
151 : : "vacuum_buffer_usage_limit",
152 : : MIN_BAS_VAC_RING_SIZE_KB, MAX_BAS_VAC_RING_SIZE_KB);
153 : :
154 : 0 : return false;
155 : : }
156 : :
157 : : /*
158 : : * Primary entry point for manual VACUUM and ANALYZE commands
159 : : *
160 : : * This is mainly a preparation wrapper for the real operations that will
161 : : * happen in vacuum().
162 : : */
163 : : void
164 : 9111 : ExecVacuum(ParseState *pstate, VacuumStmt *vacstmt, bool isTopLevel)
165 : : {
166 : : VacuumParams params;
167 : 9111 : BufferAccessStrategy bstrategy = NULL;
168 : 9111 : bool verbose = false;
169 : 9111 : bool skip_locked = false;
170 : 9111 : bool analyze = false;
171 : 9111 : bool freeze = false;
172 : 9111 : bool full = false;
173 : 9111 : bool disable_page_skipping = false;
174 : 9111 : bool process_main = true;
175 : 9111 : bool process_toast = true;
176 : : int ring_size;
177 : 9111 : bool skip_database_stats = false;
178 : 9111 : bool only_database_stats = false;
179 : : MemoryContext vac_context;
180 : : ListCell *lc;
181 : :
182 : : /* index_cleanup and truncate values unspecified for now */
183 : 9111 : params.index_cleanup = VACOPTVALUE_UNSPECIFIED;
184 : 9111 : params.truncate = VACOPTVALUE_UNSPECIFIED;
185 : :
186 : : /* By default parallel vacuum is enabled */
187 : 9111 : params.nworkers = 0;
188 : :
189 : : /* Will be set later if we recurse to a TOAST table. */
190 : 9111 : params.toast_parent = InvalidOid;
191 : 9111 : params.main_relopts = NULL;
192 : :
193 : : /*
194 : : * Set this to an invalid value so it is clear whether or not a
195 : : * BUFFER_USAGE_LIMIT was specified when making the access strategy.
196 : : */
197 : 9111 : ring_size = -1;
198 : :
199 : : /* Parse options list */
200 [ + + + + : 18714 : foreach(lc, vacstmt->options)
+ + ]
201 : : {
202 : 9627 : DefElem *opt = (DefElem *) lfirst(lc);
203 : :
204 : : /* Parse common options for VACUUM and ANALYZE */
205 [ + + ]: 9627 : if (strcmp(opt->defname, "verbose") == 0)
206 : 33 : verbose = defGetBoolean(opt);
207 [ + + ]: 9594 : else if (strcmp(opt->defname, "skip_locked") == 0)
208 : 182 : skip_locked = defGetBoolean(opt);
209 [ + + ]: 9412 : else if (strcmp(opt->defname, "buffer_usage_limit") == 0)
210 : : {
211 : : const char *hintmsg;
212 : : int result;
213 : : char *vac_buffer_size;
214 : :
215 : 36 : vac_buffer_size = defGetString(opt);
216 : :
217 : : /*
218 : : * Check that the specified value is valid and the size falls
219 : : * within the hard upper and lower limits if it is not 0.
220 : : */
221 [ + + ]: 36 : if (!parse_int(vac_buffer_size, &result, GUC_UNIT_KB, &hintmsg) ||
222 [ + + ]: 32 : (result != 0 &&
223 [ + + + + ]: 24 : (result < MIN_BAS_VAC_RING_SIZE_KB || result > MAX_BAS_VAC_RING_SIZE_KB)))
224 : : {
225 [ + - + + ]: 12 : ereport(ERROR,
226 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
227 : : errmsg("%s option must be 0 or between %d kB and %d kB",
228 : : "BUFFER_USAGE_LIMIT",
229 : : MIN_BAS_VAC_RING_SIZE_KB, MAX_BAS_VAC_RING_SIZE_KB),
230 : : hintmsg ? errhint_internal("%s", _(hintmsg)) : 0));
231 : : }
232 : :
233 : 24 : ring_size = result;
234 : : }
235 [ + + ]: 9376 : else if (!vacstmt->is_vacuumcmd)
236 [ + - ]: 4 : ereport(ERROR,
237 : : (errcode(ERRCODE_SYNTAX_ERROR),
238 : : errmsg("unrecognized %s option \"%s\"",
239 : : "ANALYZE", opt->defname),
240 : : parser_errposition(pstate, opt->location)));
241 : :
242 : : /* Parse options available on VACUUM */
243 [ + + ]: 9372 : else if (strcmp(opt->defname, "analyze") == 0)
244 : 1852 : analyze = defGetBoolean(opt);
245 [ + + ]: 7520 : else if (strcmp(opt->defname, "freeze") == 0)
246 : 1899 : freeze = defGetBoolean(opt);
247 [ + + ]: 5621 : else if (strcmp(opt->defname, "full") == 0)
248 : 234 : full = defGetBoolean(opt);
249 [ + + ]: 5387 : else if (strcmp(opt->defname, "disable_page_skipping") == 0)
250 : 124 : disable_page_skipping = defGetBoolean(opt);
251 [ + + ]: 5263 : else if (strcmp(opt->defname, "index_cleanup") == 0)
252 : : {
253 : : /* Interpret no string as the default, which is 'auto' */
254 [ - + ]: 100 : if (!opt->arg)
255 : 0 : params.index_cleanup = VACOPTVALUE_AUTO;
256 : : else
257 : : {
258 : 100 : char *sval = defGetString(opt);
259 : :
260 : : /* Try matching on 'auto' string, or fall back on boolean */
261 [ + + ]: 100 : if (pg_strcasecmp(sval, "auto") == 0)
262 : 4 : params.index_cleanup = VACOPTVALUE_AUTO;
263 : : else
264 : 96 : params.index_cleanup = get_vacoptval_from_boolean(opt);
265 : : }
266 : : }
267 [ + + ]: 5163 : else if (strcmp(opt->defname, "process_main") == 0)
268 : 85 : process_main = defGetBoolean(opt);
269 [ + + ]: 5078 : else if (strcmp(opt->defname, "process_toast") == 0)
270 : 89 : process_toast = defGetBoolean(opt);
271 [ + + ]: 4989 : else if (strcmp(opt->defname, "truncate") == 0)
272 : 88 : params.truncate = get_vacoptval_from_boolean(opt);
273 [ + + ]: 4901 : else if (strcmp(opt->defname, "parallel") == 0)
274 : : {
275 : 199 : int nworkers = defGetInt32(opt);
276 : :
277 [ + + - + ]: 195 : if (nworkers < 0 || nworkers > MAX_PARALLEL_WORKER_LIMIT)
278 [ + - ]: 4 : ereport(ERROR,
279 : : (errcode(ERRCODE_SYNTAX_ERROR),
280 : : errmsg("%s option must be between 0 and %d",
281 : : "PARALLEL",
282 : : MAX_PARALLEL_WORKER_LIMIT),
283 : : parser_errposition(pstate, opt->location)));
284 : :
285 : : /*
286 : : * Disable parallel vacuum, if user has specified parallel degree
287 : : * as zero.
288 : : */
289 [ + + ]: 191 : if (nworkers == 0)
290 : 86 : params.nworkers = -1;
291 : : else
292 : 105 : params.nworkers = nworkers;
293 : : }
294 [ + + ]: 4702 : else if (strcmp(opt->defname, "skip_database_stats") == 0)
295 : 4623 : skip_database_stats = defGetBoolean(opt);
296 [ + - ]: 79 : else if (strcmp(opt->defname, "only_database_stats") == 0)
297 : 79 : only_database_stats = defGetBoolean(opt);
298 : : else
299 [ # # ]: 0 : ereport(ERROR,
300 : : (errcode(ERRCODE_SYNTAX_ERROR),
301 : : errmsg("unrecognized %s option \"%s\"",
302 : : "VACUUM", opt->defname),
303 : : parser_errposition(pstate, opt->location)));
304 : : }
305 : :
306 : : /* Set vacuum options */
307 : 9087 : params.options =
308 [ + + ]: 9087 : (vacstmt->is_vacuumcmd ? VACOPT_VACUUM : VACOPT_ANALYZE) |
309 [ + + ]: 9087 : (verbose ? VACOPT_VERBOSE : 0) |
310 [ + + ]: 9087 : (skip_locked ? VACOPT_SKIP_LOCKED : 0) |
311 [ + + ]: 9087 : (analyze ? VACOPT_ANALYZE : 0) |
312 [ + + ]: 9087 : (freeze ? VACOPT_FREEZE : 0) |
313 [ + + ]: 9087 : (full ? VACOPT_FULL : 0) |
314 [ + + ]: 9087 : (disable_page_skipping ? VACOPT_DISABLE_PAGE_SKIPPING : 0) |
315 [ + + ]: 9087 : (process_main ? VACOPT_PROCESS_MAIN : 0) |
316 [ + + ]: 9087 : (process_toast ? VACOPT_PROCESS_TOAST : 0) |
317 [ + + ]: 9087 : (skip_database_stats ? VACOPT_SKIP_DATABASE_STATS : 0) |
318 [ + + ]: 9087 : (only_database_stats ? VACOPT_ONLY_DATABASE_STATS : 0);
319 : :
320 : : /* sanity checks on options */
321 : : Assert(params.options & (VACOPT_VACUUM | VACOPT_ANALYZE));
322 : : Assert((params.options & VACOPT_VACUUM) ||
323 : : !(params.options & (VACOPT_FULL | VACOPT_FREEZE)));
324 : :
325 [ + + + + ]: 9087 : if ((params.options & VACOPT_FULL) && params.nworkers > 0)
326 [ + - ]: 4 : ereport(ERROR,
327 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
328 : : errmsg("VACUUM FULL cannot be performed in parallel")));
329 : :
330 : : /*
331 : : * BUFFER_USAGE_LIMIT does nothing for VACUUM (FULL) so just raise an
332 : : * ERROR for that case. VACUUM (FULL, ANALYZE) does make use of it, so
333 : : * we'll permit that.
334 : : */
335 [ + + + + ]: 9083 : if (ring_size != -1 && (params.options & VACOPT_FULL) &&
336 [ + - ]: 4 : !(params.options & VACOPT_ANALYZE))
337 [ + - ]: 4 : ereport(ERROR,
338 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
339 : : errmsg("BUFFER_USAGE_LIMIT cannot be specified for VACUUM FULL")));
340 : :
341 : : /*
342 : : * Make sure VACOPT_ANALYZE is specified if any column lists are present.
343 : : */
344 [ + + ]: 9079 : if (!(params.options & VACOPT_ANALYZE))
345 : : {
346 [ + + + + : 7842 : foreach(lc, vacstmt->rels)
+ + ]
347 : : {
348 : 3866 : VacuumRelation *vrel = lfirst_node(VacuumRelation, lc);
349 : :
350 [ + + ]: 3866 : if (vrel->va_cols != NIL)
351 [ + - ]: 4 : ereport(ERROR,
352 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
353 : : errmsg("ANALYZE option must be specified when a column list is provided")));
354 : : }
355 : : }
356 : :
357 : : /*
358 : : * Sanity check DISABLE_PAGE_SKIPPING option.
359 : : */
360 [ + + ]: 9075 : if ((params.options & VACOPT_FULL) != 0 &&
361 [ - + ]: 218 : (params.options & VACOPT_DISABLE_PAGE_SKIPPING) != 0)
362 [ # # ]: 0 : ereport(ERROR,
363 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
364 : : errmsg("VACUUM option DISABLE_PAGE_SKIPPING cannot be used with FULL")));
365 : :
366 : : /* sanity check for PROCESS_TOAST */
367 [ + + ]: 9075 : if ((params.options & VACOPT_FULL) != 0 &&
368 [ + + ]: 218 : (params.options & VACOPT_PROCESS_TOAST) == 0)
369 [ + - ]: 4 : ereport(ERROR,
370 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
371 : : errmsg("PROCESS_TOAST required with VACUUM FULL")));
372 : :
373 : : /* sanity check for ONLY_DATABASE_STATS */
374 [ + + ]: 9071 : if (params.options & VACOPT_ONLY_DATABASE_STATS)
375 : : {
376 : : Assert(params.options & VACOPT_VACUUM);
377 [ + + ]: 79 : if (vacstmt->rels != NIL)
378 [ + - ]: 4 : ereport(ERROR,
379 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
380 : : errmsg("ONLY_DATABASE_STATS cannot be specified with a list of tables")));
381 : : /* don't require people to turn off PROCESS_TOAST/MAIN explicitly */
382 [ - + ]: 75 : if (params.options & ~(VACOPT_VACUUM |
383 : : VACOPT_VERBOSE |
384 : : VACOPT_PROCESS_MAIN |
385 : : VACOPT_PROCESS_TOAST |
386 : : VACOPT_ONLY_DATABASE_STATS))
387 [ # # ]: 0 : ereport(ERROR,
388 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
389 : : errmsg("ONLY_DATABASE_STATS cannot be specified with other VACUUM options")));
390 : : }
391 : :
392 : : /*
393 : : * All freeze ages are zero if the FREEZE option is given; otherwise pass
394 : : * them as -1 which means to use the default values.
395 : : */
396 [ + + ]: 9067 : if (params.options & VACOPT_FREEZE)
397 : : {
398 : 1899 : params.freeze_min_age = 0;
399 : 1899 : params.freeze_table_age = 0;
400 : 1899 : params.multixact_freeze_min_age = 0;
401 : 1899 : params.multixact_freeze_table_age = 0;
402 : : }
403 : : else
404 : : {
405 : 7168 : params.freeze_min_age = -1;
406 : 7168 : params.freeze_table_age = -1;
407 : 7168 : params.multixact_freeze_min_age = -1;
408 : 7168 : params.multixact_freeze_table_age = -1;
409 : : }
410 : :
411 : : /* user-invoked vacuum is never "for wraparound" */
412 : 9067 : params.is_wraparound = false;
413 : :
414 : : /*
415 : : * user-invoked vacuum uses VACOPT_VERBOSE instead of
416 : : * log_vacuum_min_duration and log_analyze_min_duration
417 : : */
418 : 9067 : params.log_vacuum_min_duration = -1;
419 : 9067 : params.log_analyze_min_duration = -1;
420 : :
421 : : /*
422 : : * Later, in vacuum_rel(), we check if a reloption override was specified.
423 : : */
424 : 9067 : params.max_eager_freeze_failure_rate = vacuum_max_eager_freeze_failure_rate;
425 : :
426 : : /*
427 : : * Create special memory context for cross-transaction storage.
428 : : *
429 : : * Since it is a child of PortalContext, it will go away eventually even
430 : : * if we suffer an error; there's no need for special abort cleanup logic.
431 : : */
432 : 9067 : vac_context = AllocSetContextCreate(PortalContext,
433 : : "Vacuum",
434 : : ALLOCSET_DEFAULT_SIZES);
435 : :
436 : : /*
437 : : * Make a buffer strategy object in the cross-transaction memory context.
438 : : * We needn't bother making this for VACUUM (FULL) or VACUUM
439 : : * (ONLY_DATABASE_STATS) as they'll not make use of it. VACUUM (FULL,
440 : : * ANALYZE) is possible, so we'd better ensure that we make a strategy
441 : : * when we see ANALYZE.
442 : : */
443 [ + + ]: 9067 : if ((params.options & (VACOPT_ONLY_DATABASE_STATS |
444 : 289 : VACOPT_FULL)) == 0 ||
445 [ + + ]: 289 : (params.options & VACOPT_ANALYZE) != 0)
446 : : {
447 : :
448 : 8782 : MemoryContext old_context = MemoryContextSwitchTo(vac_context);
449 : :
450 : : Assert(ring_size >= -1);
451 : :
452 : : /*
453 : : * If BUFFER_USAGE_LIMIT was specified by the VACUUM or ANALYZE
454 : : * command, it overrides the value of VacuumBufferUsageLimit. Either
455 : : * value may be 0, in which case GetAccessStrategyWithSize() will
456 : : * return NULL, effectively allowing full use of shared buffers.
457 : : */
458 [ + + ]: 8782 : if (ring_size == -1)
459 : 8762 : ring_size = VacuumBufferUsageLimit;
460 : :
461 : 8782 : bstrategy = GetAccessStrategyWithSize(BAS_VACUUM, ring_size);
462 : :
463 : 8782 : MemoryContextSwitchTo(old_context);
464 : : }
465 : :
466 : : /* Now go through the common routine */
467 : 9067 : vacuum(vacstmt->rels, ¶ms, bstrategy, vac_context, isTopLevel);
468 : :
469 : : /* Finally, clean up the vacuum memory context */
470 : 8977 : MemoryContextDelete(vac_context);
471 : 8977 : }
472 : :
473 : : /*
474 : : * Internal entry point for autovacuum and the VACUUM / ANALYZE commands.
475 : : *
476 : : * relations, if not NIL, is a list of VacuumRelation to process; otherwise,
477 : : * we process all relevant tables in the database. For each VacuumRelation,
478 : : * if a valid OID is supplied, the table with that OID is what to process;
479 : : * otherwise, the VacuumRelation's RangeVar indicates what to process.
480 : : *
481 : : * params contains a set of parameters that can be used to customize the
482 : : * behavior.
483 : : *
484 : : * bstrategy may be passed in as NULL when the caller does not want to
485 : : * restrict the number of shared_buffers that VACUUM / ANALYZE can use,
486 : : * otherwise, the caller must build a BufferAccessStrategy with the number of
487 : : * shared_buffers that VACUUM / ANALYZE should try to limit themselves to
488 : : * using.
489 : : *
490 : : * isTopLevel should be passed down from ProcessUtility.
491 : : *
492 : : * It is the caller's responsibility that all parameters are allocated in a
493 : : * memory context that will not disappear at transaction commit.
494 : : */
495 : : void
496 : 122590 : vacuum(List *relations, const VacuumParams *params, BufferAccessStrategy bstrategy,
497 : : MemoryContext vac_context, bool isTopLevel)
498 : : {
499 : : static bool in_vacuum = false;
500 : :
501 : : const char *stmttype;
502 : : volatile bool in_outer_xact,
503 : : use_own_xacts;
504 : :
505 [ + + ]: 122590 : stmttype = (params->options & VACOPT_VACUUM) ? "VACUUM" : "ANALYZE";
506 : :
507 : : /*
508 : : * We cannot run VACUUM inside a user transaction block; if we were inside
509 : : * a transaction, then our commit- and start-transaction-command calls
510 : : * would not have the intended effect! There are numerous other subtle
511 : : * dependencies on this, too.
512 : : *
513 : : * ANALYZE (without VACUUM) can run either way.
514 : : */
515 [ + + ]: 122590 : if (params->options & VACOPT_VACUUM)
516 : : {
517 : 119082 : PreventInTransactionBlock(isTopLevel, stmttype);
518 : 119069 : in_outer_xact = false;
519 : : }
520 : : else
521 : 3508 : in_outer_xact = IsInTransactionBlock(isTopLevel);
522 : :
523 : : /*
524 : : * Check for and disallow recursive calls. This could happen when VACUUM
525 : : * FULL or ANALYZE calls a hostile index expression that itself calls
526 : : * ANALYZE.
527 : : */
528 [ + + ]: 122577 : if (in_vacuum)
529 [ + - ]: 8 : ereport(ERROR,
530 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
531 : : errmsg("%s cannot be executed from VACUUM or ANALYZE",
532 : : stmttype)));
533 : :
534 : : /*
535 : : * Build list of relation(s) to process, putting any new data in
536 : : * vac_context for safekeeping.
537 : : */
538 [ + + ]: 122569 : if (params->options & VACOPT_ONLY_DATABASE_STATS)
539 : : {
540 : : /* We don't process any tables in this case */
541 : : Assert(relations == NIL);
542 : : }
543 [ + + ]: 122494 : else if (relations != NIL)
544 : : {
545 : 122371 : List *newrels = NIL;
546 : : ListCell *lc;
547 : :
548 [ + - + + : 244827 : foreach(lc, relations)
+ + ]
549 : : {
550 : 122480 : VacuumRelation *vrel = lfirst_node(VacuumRelation, lc);
551 : : List *sublist;
552 : : MemoryContext old_context;
553 : :
554 : 122480 : sublist = expand_vacuum_rel(vrel, vac_context, params->options);
555 : 122456 : old_context = MemoryContextSwitchTo(vac_context);
556 : 122456 : newrels = list_concat(newrels, sublist);
557 : 122456 : MemoryContextSwitchTo(old_context);
558 : : }
559 : 122347 : relations = newrels;
560 : : }
561 : : else
562 : 123 : relations = get_all_vacuum_rels(vac_context, params->options);
563 : :
564 : : /*
565 : : * Decide whether we need to start/commit our own transactions.
566 : : *
567 : : * For VACUUM (with or without ANALYZE): always do so, so that we can
568 : : * release locks as soon as possible. (We could possibly use the outer
569 : : * transaction for a one-table VACUUM, but handling TOAST tables would be
570 : : * problematic.)
571 : : *
572 : : * For ANALYZE (no VACUUM): if inside a transaction block, we cannot
573 : : * start/commit our own transactions. Also, there's no need to do so if
574 : : * only processing one relation. For multiple relations when not within a
575 : : * transaction block, and also in an autovacuum worker, use own
576 : : * transactions so we can release locks sooner.
577 : : */
578 [ + + ]: 122545 : if (params->options & VACOPT_VACUUM)
579 : 119061 : use_own_xacts = true;
580 : : else
581 : : {
582 : : Assert(params->options & VACOPT_ANALYZE);
583 [ + + ]: 3484 : if (AmAutoVacuumWorkerProcess())
584 : 261 : use_own_xacts = true;
585 [ + + ]: 3223 : else if (in_outer_xact)
586 : 168 : use_own_xacts = false;
587 [ + + ]: 3055 : else if (list_length(relations) > 1)
588 : 527 : use_own_xacts = true;
589 : : else
590 : 2528 : use_own_xacts = false;
591 : : }
592 : :
593 : : /*
594 : : * vacuum_rel expects to be entered with no transaction active; it will
595 : : * start and commit its own transaction. But we are called by an SQL
596 : : * command, and so we are executing inside a transaction already. We
597 : : * commit the transaction started in PostgresMain() here, and start
598 : : * another one before exiting to match the commit waiting for us back in
599 : : * PostgresMain().
600 : : */
601 [ + + ]: 122545 : if (use_own_xacts)
602 : : {
603 : : Assert(!in_outer_xact);
604 : :
605 : : /* ActiveSnapshot is not set by autovacuum */
606 [ + + ]: 119849 : if (ActiveSnapshotSet())
607 : 6326 : PopActiveSnapshot();
608 : :
609 : : /* matches the StartTransaction in PostgresMain() */
610 : 119849 : CommitTransactionCommand();
611 : : }
612 : :
613 : : /* Turn vacuum cost accounting on or off, and set/clear in_vacuum */
614 [ + + ]: 122545 : PG_TRY();
615 : : {
616 : : ListCell *cur;
617 : :
618 : 122545 : in_vacuum = true;
619 : 122545 : VacuumFailsafeActive = false;
620 : 122545 : VacuumUpdateCosts();
621 : 122544 : VacuumCostBalance = 0;
622 : 122544 : VacuumCostBalanceLocal = 0;
623 : 122544 : VacuumSharedCostBalance = NULL;
624 : 122544 : VacuumActiveNWorkers = NULL;
625 : :
626 : : /*
627 : : * Loop to process each selected relation.
628 : : */
629 [ + + + + : 255658 : foreach(cur, relations)
+ + ]
630 : : {
631 : 133160 : VacuumRelation *vrel = lfirst_node(VacuumRelation, cur);
632 : :
633 [ + + ]: 133160 : if (params->options & VACOPT_VACUUM)
634 : : {
635 [ + + ]: 124521 : if (!vacuum_rel(vrel->oid, vrel->relation, *params, bstrategy,
636 : : isTopLevel))
637 : 62 : continue;
638 : : }
639 : :
640 [ + + ]: 133092 : if (params->options & VACOPT_ANALYZE)
641 : : {
642 : : /*
643 : : * If using separate xacts, start one for analyze. Otherwise,
644 : : * we can use the outer transaction.
645 : : */
646 [ + + ]: 10738 : if (use_own_xacts)
647 : : {
648 : 8054 : StartTransactionCommand();
649 : : /* functions in indexes may want a snapshot set */
650 : 8054 : PushActiveSnapshot(GetTransactionSnapshot());
651 : : }
652 : :
653 : 10738 : analyze_rel(vrel->oid, vrel->relation, params,
654 : : vrel->va_cols, in_outer_xact, bstrategy);
655 : :
656 [ + + ]: 10698 : if (use_own_xacts)
657 : : {
658 : 8029 : PopActiveSnapshot();
659 : : /* standard_ProcessUtility() does CCI if !use_own_xacts */
660 : 8029 : CommandCounterIncrement();
661 : 8029 : CommitTransactionCommand();
662 : : }
663 : : else
664 : : {
665 : : /*
666 : : * If we're not using separate xacts, better separate the
667 : : * ANALYZE actions with CCIs. This avoids trouble if user
668 : : * says "ANALYZE t, t".
669 : : */
670 : 2669 : CommandCounterIncrement();
671 : : }
672 : : }
673 : :
674 : : /*
675 : : * Ensure VacuumFailsafeActive has been reset before vacuuming the
676 : : * next relation.
677 : : */
678 : 133052 : VacuumFailsafeActive = false;
679 : : }
680 : : }
681 : 45 : PG_FINALLY();
682 : : {
683 : 122543 : in_vacuum = false;
684 : 122543 : VacuumCostActive = false;
685 : 122543 : VacuumFailsafeActive = false;
686 : 122543 : VacuumCostBalance = 0;
687 : : }
688 [ + + ]: 122543 : PG_END_TRY();
689 : :
690 : : /*
691 : : * Finish up processing.
692 : : */
693 [ + + ]: 122498 : if (use_own_xacts)
694 : : {
695 : : /* here, we are not in a transaction */
696 : :
697 : : /*
698 : : * This matches the CommitTransaction waiting for us in
699 : : * PostgresMain().
700 : : */
701 : 119817 : StartTransactionCommand();
702 : : }
703 : :
704 [ + + ]: 122498 : if ((params->options & VACOPT_VACUUM) &&
705 [ + + ]: 119037 : !(params->options & VACOPT_SKIP_DATABASE_STATS))
706 : : {
707 : : /*
708 : : * Update pg_database.datfrozenxid, and truncate pg_xact if possible.
709 : : */
710 : 1155 : vac_update_datfrozenxid();
711 : : }
712 : :
713 : 122498 : }
714 : :
715 : : /*
716 : : * Check if the current user has privileges to vacuum or analyze the relation.
717 : : * If not, issue a WARNING log message and return false to let the caller
718 : : * decide what to do with this relation. This routine is used to decide if a
719 : : * relation can be processed for VACUUM or ANALYZE.
720 : : *
721 : : * If missing_ok is true, we silently return false if the relation is
722 : : * concurrently dropped. Callers without a lock on the relation must specify
723 : : * missing_ok; all others must hold at least AccessShareLock.
724 : : */
725 : : bool
726 : 161345 : vacuum_is_permitted_for_relation(Oid relid, Form_pg_class reltuple,
727 : : uint32 options, bool missing_ok)
728 : : {
729 : : char *relname;
730 : 161345 : bool is_missing = false;
731 : :
732 : : Assert((options & (VACOPT_VACUUM | VACOPT_ANALYZE)) != 0);
733 : : Assert(missing_ok ||
734 : : CheckRelationOidLockedByMe(relid, AccessShareLock, true));
735 : :
736 : : /*----------
737 : : * A role has privileges to vacuum or analyze the relation if any of the
738 : : * following are true:
739 : : * - the role owns the current database and the relation is not shared
740 : : * - the role has the MAINTAIN privilege on the relation
741 : : *----------
742 : : */
743 [ + + ]: 161345 : if ((object_ownercheck(DatabaseRelationId, MyDatabaseId, GetUserId()) &&
744 [ + + + + ]: 184717 : !reltuple->relisshared) ||
745 [ + + ]: 25864 : pg_class_aclcheck_ext(relid, GetUserId(), ACL_MAINTAIN,
746 : : missing_ok ? &is_missing : NULL) == ACLCHECK_OK)
747 : 159165 : return true;
748 : :
749 : : /*
750 : : * If the relation was concurrently dropped, nothing to do. Note that
751 : : * this is only reachable when the caller specified missing_ok.
752 : : */
753 [ - + ]: 2180 : if (is_missing)
754 : : {
755 : : Assert(missing_ok);
756 : 0 : return false;
757 : : }
758 : :
759 : 2180 : relname = NameStr(reltuple->relname);
760 : :
761 [ + + ]: 2180 : if ((options & VACOPT_VACUUM) != 0)
762 : : {
763 [ + - ]: 148 : ereport(WARNING,
764 : : (errmsg("permission denied to vacuum \"%s\", skipping it",
765 : : relname)));
766 : :
767 : : /*
768 : : * For VACUUM ANALYZE, both logs could show up, but just generate
769 : : * information for VACUUM as that would be the first one to be
770 : : * processed.
771 : : */
772 : 148 : return false;
773 : : }
774 : :
775 [ + - ]: 2032 : if ((options & VACOPT_ANALYZE) != 0)
776 [ + - ]: 2032 : ereport(WARNING,
777 : : (errmsg("permission denied to analyze \"%s\", skipping it",
778 : : relname)));
779 : :
780 : 2032 : return false;
781 : : }
782 : :
783 : :
784 : : /*
785 : : * vacuum_open_relation
786 : : *
787 : : * This routine is used for attempting to open and lock a relation which
788 : : * is going to be vacuumed or analyzed. If the relation cannot be opened
789 : : * or locked, a log is emitted if possible.
790 : : */
791 : : Relation
792 : 141041 : vacuum_open_relation(Oid relid, RangeVar *relation, uint32 options,
793 : : bool verbose, LOCKMODE lmode)
794 : : {
795 : : Relation rel;
796 : 141041 : bool rel_lock = true;
797 : : int elevel;
798 : :
799 : : Assert((options & (VACOPT_VACUUM | VACOPT_ANALYZE)) != 0);
800 : :
801 : : /*
802 : : * Open the relation and get the appropriate lock on it.
803 : : *
804 : : * There's a race condition here: the relation may have gone away since
805 : : * the last time we saw it. If so, we don't need to vacuum or analyze it.
806 : : *
807 : : * If we've been asked not to wait for the relation lock, acquire it first
808 : : * in non-blocking mode, before calling try_relation_open().
809 : : */
810 [ + + ]: 141041 : if (!(options & VACOPT_SKIP_LOCKED))
811 : 140091 : rel = try_relation_open(relid, lmode);
812 [ + + ]: 950 : else if (ConditionalLockRelationOid(relid, lmode))
813 : 940 : rel = try_relation_open(relid, NoLock);
814 : : else
815 : : {
816 : 10 : rel = NULL;
817 : 10 : rel_lock = false;
818 : : }
819 : :
820 : : /* if relation is opened, leave */
821 [ + + ]: 141041 : if (rel)
822 : 141025 : return rel;
823 : :
824 : : /*
825 : : * Relation could not be opened, hence generate if possible a log
826 : : * informing on the situation.
827 : : *
828 : : * If the RangeVar is not defined, we do not have enough information to
829 : : * provide a meaningful log statement. Chances are that the caller has
830 : : * intentionally not provided this information so that this logging is
831 : : * skipped, anyway.
832 : : */
833 [ + + ]: 16 : if (relation == NULL)
834 : 9 : return NULL;
835 : :
836 : : /*
837 : : * Determine the log level.
838 : : *
839 : : * For manual VACUUM or ANALYZE, we emit a WARNING to match the log
840 : : * statements in the permission checks; otherwise, only log if the caller
841 : : * so requested.
842 : : */
843 [ + - ]: 7 : if (!AmAutoVacuumWorkerProcess())
844 : 7 : elevel = WARNING;
845 [ # # ]: 0 : else if (verbose)
846 : 0 : elevel = LOG;
847 : : else
848 : 0 : return NULL;
849 : :
850 [ + + ]: 7 : if ((options & VACOPT_VACUUM) != 0)
851 : : {
852 [ + + ]: 5 : if (!rel_lock)
853 [ + - ]: 3 : ereport(elevel,
854 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
855 : : errmsg("skipping vacuum of \"%s\" --- lock not available",
856 : : relation->relname)));
857 : : else
858 [ + - ]: 2 : ereport(elevel,
859 : : (errcode(ERRCODE_UNDEFINED_TABLE),
860 : : errmsg("skipping vacuum of \"%s\" --- relation no longer exists",
861 : : relation->relname)));
862 : :
863 : : /*
864 : : * For VACUUM ANALYZE, both logs could show up, but just generate
865 : : * information for VACUUM as that would be the first one to be
866 : : * processed.
867 : : */
868 : 5 : return NULL;
869 : : }
870 : :
871 [ + - ]: 2 : if ((options & VACOPT_ANALYZE) != 0)
872 : : {
873 [ + + ]: 2 : if (!rel_lock)
874 [ + - ]: 1 : ereport(elevel,
875 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
876 : : errmsg("skipping analyze of \"%s\" --- lock not available",
877 : : relation->relname)));
878 : : else
879 [ + - ]: 1 : ereport(elevel,
880 : : (errcode(ERRCODE_UNDEFINED_TABLE),
881 : : errmsg("skipping analyze of \"%s\" --- relation no longer exists",
882 : : relation->relname)));
883 : : }
884 : :
885 : 2 : return NULL;
886 : : }
887 : :
888 : :
889 : : /*
890 : : * Given a VacuumRelation, fill in the table OID if it wasn't specified,
891 : : * and optionally add VacuumRelations for partitions or inheritance children.
892 : : *
893 : : * If a VacuumRelation does not have an OID supplied and is a partitioned
894 : : * table, an extra entry will be added to the output for each partition.
895 : : * Presently, only autovacuum supplies OIDs when calling vacuum(), and
896 : : * it does not want us to expand partitioned tables.
897 : : *
898 : : * We take care not to modify the input data structure, but instead build
899 : : * new VacuumRelation(s) to return. (But note that they will reference
900 : : * unmodified parts of the input, eg column lists.) New data structures
901 : : * are made in vac_context.
902 : : */
903 : : static List *
904 : 122480 : expand_vacuum_rel(VacuumRelation *vrel, MemoryContext vac_context,
905 : : int options)
906 : : {
907 : 122480 : List *vacrels = NIL;
908 : : MemoryContext oldcontext;
909 : :
910 : : /* If caller supplied OID, there's nothing we need do here. */
911 [ + + ]: 122480 : if (OidIsValid(vrel->oid))
912 : : {
913 : 113523 : oldcontext = MemoryContextSwitchTo(vac_context);
914 : 113523 : vacrels = lappend(vacrels, vrel);
915 : 113523 : MemoryContextSwitchTo(oldcontext);
916 : : }
917 : : else
918 : : {
919 : : /*
920 : : * Process a specific relation, and possibly partitions or child
921 : : * tables thereof.
922 : : */
923 : : Oid relid;
924 : : HeapTuple tuple;
925 : : Form_pg_class classForm;
926 : : bool include_children;
927 : : bool is_partitioned_table;
928 : : int rvr_opts;
929 : :
930 : : /*
931 : : * Since autovacuum workers supply OIDs when calling vacuum(), no
932 : : * autovacuum worker should reach this code.
933 : : */
934 : : Assert(!AmAutoVacuumWorkerProcess());
935 : :
936 : : /*
937 : : * We transiently take AccessShareLock to protect the syscache lookup
938 : : * below, as well as find_all_inheritors's expectation that the caller
939 : : * holds some lock on the starting relation.
940 : : */
941 : 8957 : rvr_opts = (options & VACOPT_SKIP_LOCKED) ? RVR_SKIP_LOCKED : 0;
942 : 8957 : relid = RangeVarGetRelidExtended(vrel->relation,
943 : : AccessShareLock,
944 : : rvr_opts,
945 : : NULL, NULL);
946 : :
947 : : /*
948 : : * If the lock is unavailable, emit the same log statement that
949 : : * vacuum_rel() and analyze_rel() would.
950 : : */
951 [ + + ]: 8933 : if (!OidIsValid(relid))
952 : : {
953 [ + + ]: 4 : if (options & VACOPT_VACUUM)
954 [ + - ]: 3 : ereport(WARNING,
955 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
956 : : errmsg("skipping vacuum of \"%s\" --- lock not available",
957 : : vrel->relation->relname)));
958 : : else
959 [ + - ]: 1 : ereport(WARNING,
960 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
961 : : errmsg("skipping analyze of \"%s\" --- lock not available",
962 : : vrel->relation->relname)));
963 : 4 : return vacrels;
964 : : }
965 : :
966 : : /*
967 : : * To check whether the relation is a partitioned table and its
968 : : * ownership, fetch its syscache entry.
969 : : */
970 : 8929 : tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(relid));
971 [ - + ]: 8929 : if (!HeapTupleIsValid(tuple))
972 [ # # ]: 0 : elog(ERROR, "cache lookup failed for relation %u", relid);
973 : 8929 : classForm = (Form_pg_class) GETSTRUCT(tuple);
974 : :
975 : : /*
976 : : * Make a returnable VacuumRelation for this rel if the user has the
977 : : * required privileges.
978 : : */
979 [ + + ]: 8929 : if (vacuum_is_permitted_for_relation(relid, classForm, options, false))
980 : : {
981 : 8777 : oldcontext = MemoryContextSwitchTo(vac_context);
982 : 8777 : vacrels = lappend(vacrels, makeVacuumRelation(vrel->relation,
983 : : relid,
984 : : vrel->va_cols));
985 : 8777 : MemoryContextSwitchTo(oldcontext);
986 : : }
987 : :
988 : : /*
989 : : * Vacuuming a partitioned table with ONLY will not do anything since
990 : : * the partitioned table itself is empty. Issue a warning if the user
991 : : * requests this.
992 : : */
993 : 8929 : include_children = vrel->relation->inh;
994 : 8929 : is_partitioned_table = (classForm->relkind == RELKIND_PARTITIONED_TABLE);
995 [ + + + + : 8929 : if ((options & VACOPT_VACUUM) && is_partitioned_table && !include_children)
+ + ]
996 [ + - ]: 4 : ereport(WARNING,
997 : : (errmsg("VACUUM ONLY of partitioned table \"%s\" has no effect",
998 : : vrel->relation->relname)));
999 : :
1000 : 8929 : ReleaseSysCache(tuple);
1001 : :
1002 : : /*
1003 : : * Unless the user has specified ONLY, make relation list entries for
1004 : : * its partitions or inheritance child tables. Note that the list
1005 : : * returned by find_all_inheritors() includes the passed-in OID, so we
1006 : : * have to skip that. There's no point in taking locks on the
1007 : : * individual partitions or child tables yet, and doing so would just
1008 : : * add unnecessary deadlock risk. For this last reason, we do not yet
1009 : : * check the ownership of the partitions/tables, which get added to
1010 : : * the list to process. Ownership will be checked later on anyway.
1011 : : */
1012 [ + + ]: 8929 : if (include_children)
1013 : : {
1014 : 8899 : List *part_oids = find_all_inheritors(relid, NoLock, NULL);
1015 : : ListCell *part_lc;
1016 : :
1017 [ + - + + : 19256 : foreach(part_lc, part_oids)
+ + ]
1018 : : {
1019 : 10357 : Oid part_oid = lfirst_oid(part_lc);
1020 : :
1021 [ + + ]: 10357 : if (part_oid == relid)
1022 : 8899 : continue; /* ignore original table */
1023 : :
1024 : : /*
1025 : : * We omit a RangeVar since it wouldn't be appropriate to
1026 : : * complain about failure to open one of these relations
1027 : : * later.
1028 : : */
1029 : 1458 : oldcontext = MemoryContextSwitchTo(vac_context);
1030 : 1458 : vacrels = lappend(vacrels, makeVacuumRelation(NULL,
1031 : : part_oid,
1032 : : vrel->va_cols));
1033 : 1458 : MemoryContextSwitchTo(oldcontext);
1034 : : }
1035 : : }
1036 : :
1037 : : /*
1038 : : * Release lock again. This means that by the time we actually try to
1039 : : * process the table, it might be gone or renamed. In the former case
1040 : : * we'll silently ignore it; in the latter case we'll process it
1041 : : * anyway, but we must beware that the RangeVar doesn't necessarily
1042 : : * identify it anymore. This isn't ideal, perhaps, but there's little
1043 : : * practical alternative, since we're typically going to commit this
1044 : : * transaction and begin a new one between now and then. Moreover,
1045 : : * holding locks on multiple relations would create significant risk
1046 : : * of deadlock.
1047 : : */
1048 : 8929 : UnlockRelationOid(relid, AccessShareLock);
1049 : : }
1050 : :
1051 : 122452 : return vacrels;
1052 : : }
1053 : :
1054 : : /*
1055 : : * Construct a list of VacuumRelations for all vacuumable rels in
1056 : : * the current database. The list is built in vac_context.
1057 : : */
1058 : : static List *
1059 : 123 : get_all_vacuum_rels(MemoryContext vac_context, int options)
1060 : : {
1061 : 123 : List *vacrels = NIL;
1062 : : Relation pgclass;
1063 : : TableScanDesc scan;
1064 : : HeapTuple tuple;
1065 : :
1066 : 123 : pgclass = table_open(RelationRelationId, AccessShareLock);
1067 : :
1068 : 123 : scan = table_beginscan_catalog(pgclass, 0, NULL);
1069 : :
1070 [ + + ]: 61495 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1071 : : {
1072 : 61372 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
1073 : : MemoryContext oldcontext;
1074 : 61372 : Oid relid = classForm->oid;
1075 : :
1076 : : /*
1077 : : * We include partitioned tables here; depending on which operation is
1078 : : * to be performed, caller will decide whether to process or ignore
1079 : : * them.
1080 : : */
1081 [ + + ]: 61372 : if (classForm->relkind != RELKIND_RELATION &&
1082 [ + + ]: 50125 : classForm->relkind != RELKIND_MATVIEW &&
1083 [ + + ]: 50093 : classForm->relkind != RELKIND_PARTITIONED_TABLE)
1084 : 49980 : continue;
1085 : :
1086 : : /* Skip temp relations belonging to other sessions */
1087 [ + + ]: 11392 : if (classForm->relpersistence == RELPERSISTENCE_TEMP &&
1088 [ + + ]: 9 : !isTempOrTempToastNamespace(classForm->relnamespace))
1089 : 1 : continue;
1090 : :
1091 : : /* check permissions of relation */
1092 [ + + ]: 11391 : if (!vacuum_is_permitted_for_relation(relid, classForm, options, true))
1093 : 1956 : continue;
1094 : :
1095 : : /*
1096 : : * Build VacuumRelation(s) specifying the table OIDs to be processed.
1097 : : * We omit a RangeVar since it wouldn't be appropriate to complain
1098 : : * about failure to open one of these relations later.
1099 : : */
1100 : 9435 : oldcontext = MemoryContextSwitchTo(vac_context);
1101 : 9435 : vacrels = lappend(vacrels, makeVacuumRelation(NULL,
1102 : : relid,
1103 : : NIL));
1104 : 9435 : MemoryContextSwitchTo(oldcontext);
1105 : : }
1106 : :
1107 : 123 : table_endscan(scan);
1108 : 123 : table_close(pgclass, AccessShareLock);
1109 : :
1110 : 123 : return vacrels;
1111 : : }
1112 : :
1113 : : /*
1114 : : * vacuum_get_cutoffs() -- compute OldestXmin and freeze cutoff points
1115 : : *
1116 : : * The target relation and VACUUM parameters are our inputs.
1117 : : *
1118 : : * Output parameters are the cutoffs that VACUUM caller should use.
1119 : : *
1120 : : * Return value indicates if vacuumlazy.c caller should make its VACUUM
1121 : : * operation aggressive. An aggressive VACUUM must advance relfrozenxid up to
1122 : : * FreezeLimit (at a minimum), and relminmxid up to MultiXactCutoff (at a
1123 : : * minimum).
1124 : : */
1125 : : bool
1126 : 130207 : vacuum_get_cutoffs(Relation rel, const VacuumParams *params,
1127 : : struct VacuumCutoffs *cutoffs)
1128 : : {
1129 : : int freeze_min_age,
1130 : : multixact_freeze_min_age,
1131 : : freeze_table_age,
1132 : : multixact_freeze_table_age,
1133 : : effective_multixact_freeze_max_age;
1134 : : TransactionId nextXID,
1135 : : safeOldestXmin,
1136 : : aggressiveXIDCutoff;
1137 : : MultiXactId nextMXID,
1138 : : safeOldestMxact,
1139 : : aggressiveMXIDCutoff;
1140 : :
1141 : : /* Use mutable copies of freeze age parameters */
1142 : 130207 : freeze_min_age = params->freeze_min_age;
1143 : 130207 : multixact_freeze_min_age = params->multixact_freeze_min_age;
1144 : 130207 : freeze_table_age = params->freeze_table_age;
1145 : 130207 : multixact_freeze_table_age = params->multixact_freeze_table_age;
1146 : :
1147 : : /* Set pg_class fields in cutoffs */
1148 : 130207 : cutoffs->relfrozenxid = rel->rd_rel->relfrozenxid;
1149 : 130207 : cutoffs->relminmxid = rel->rd_rel->relminmxid;
1150 : :
1151 : : /*
1152 : : * Acquire OldestXmin.
1153 : : *
1154 : : * We can always ignore processes running lazy vacuum. This is because we
1155 : : * use these values only for deciding which tuples we must keep in the
1156 : : * tables. Since lazy vacuum doesn't write its XID anywhere (usually no
1157 : : * XID assigned), it's safe to ignore it. In theory it could be
1158 : : * problematic to ignore lazy vacuums in a full vacuum, but keep in mind
1159 : : * that only one vacuum process can be working on a particular table at
1160 : : * any time, and that each vacuum is always an independent transaction.
1161 : : */
1162 : 130207 : cutoffs->OldestXmin = GetOldestNonRemovableTransactionId(rel);
1163 : :
1164 : : Assert(TransactionIdIsNormal(cutoffs->OldestXmin));
1165 : :
1166 : : /* Acquire OldestMxact */
1167 : 130207 : cutoffs->OldestMxact = GetOldestMultiXactId();
1168 : : Assert(MultiXactIdIsValid(cutoffs->OldestMxact));
1169 : :
1170 : : /* Acquire next XID/next MXID values used to apply age-based settings */
1171 : 130207 : nextXID = ReadNextTransactionId();
1172 : 130207 : nextMXID = ReadNextMultiXactId();
1173 : :
1174 : : /*
1175 : : * Also compute the multixact age for which freezing is urgent. This is
1176 : : * normally autovacuum_multixact_freeze_max_age, but may be less if
1177 : : * multixact members are bloated.
1178 : : */
1179 : 130207 : effective_multixact_freeze_max_age = MultiXactMemberFreezeThreshold();
1180 : :
1181 : : /*
1182 : : * Almost ready to set freeze output parameters; check if OldestXmin or
1183 : : * OldestMxact are held back to an unsafe degree before we start on that
1184 : : */
1185 : 130207 : safeOldestXmin = nextXID - autovacuum_freeze_max_age;
1186 [ - + ]: 130207 : if (!TransactionIdIsNormal(safeOldestXmin))
1187 : 0 : safeOldestXmin = FirstNormalTransactionId;
1188 : 130207 : safeOldestMxact = nextMXID - effective_multixact_freeze_max_age;
1189 [ - + ]: 130207 : if (safeOldestMxact < FirstMultiXactId)
1190 : 0 : safeOldestMxact = FirstMultiXactId;
1191 [ + + ]: 130207 : if (TransactionIdPrecedes(cutoffs->OldestXmin, safeOldestXmin))
1192 [ + - ]: 94576 : ereport(WARNING,
1193 : : (errmsg("cutoff for removing and freezing tuples is far in the past"),
1194 : : errhint("Close open transactions soon to avoid wraparound problems.\n"
1195 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1196 [ - + ]: 130207 : if (MultiXactIdPrecedes(cutoffs->OldestMxact, safeOldestMxact))
1197 [ # # ]: 0 : ereport(WARNING,
1198 : : (errmsg("cutoff for freezing multixacts is far in the past"),
1199 : : errhint("Close open transactions soon to avoid wraparound problems.\n"
1200 : : "You might also need to commit or roll back old prepared transactions.")));
1201 : :
1202 : : /*
1203 : : * Determine the minimum freeze age to use: as specified by the caller, or
1204 : : * vacuum_freeze_min_age, but in any case not more than half
1205 : : * autovacuum_freeze_max_age, so that autovacuums to prevent XID
1206 : : * wraparound won't occur too frequently.
1207 : : */
1208 [ + + ]: 130207 : if (freeze_min_age < 0)
1209 : 7061 : freeze_min_age = vacuum_freeze_min_age;
1210 : 130207 : freeze_min_age = Min(freeze_min_age, autovacuum_freeze_max_age / 2);
1211 : : Assert(freeze_min_age >= 0);
1212 : :
1213 : : /* Compute FreezeLimit, being careful to generate a normal XID */
1214 : 130207 : cutoffs->FreezeLimit = nextXID - freeze_min_age;
1215 [ - + ]: 130207 : if (!TransactionIdIsNormal(cutoffs->FreezeLimit))
1216 : 0 : cutoffs->FreezeLimit = FirstNormalTransactionId;
1217 : : /* FreezeLimit must always be <= OldestXmin */
1218 [ + + ]: 130207 : if (TransactionIdPrecedes(cutoffs->OldestXmin, cutoffs->FreezeLimit))
1219 : 107748 : cutoffs->FreezeLimit = cutoffs->OldestXmin;
1220 : :
1221 : : /*
1222 : : * Determine the minimum multixact freeze age to use: as specified by
1223 : : * caller, or vacuum_multixact_freeze_min_age, but in any case not more
1224 : : * than half effective_multixact_freeze_max_age, so that autovacuums to
1225 : : * prevent MultiXact wraparound won't occur too frequently.
1226 : : */
1227 [ + + ]: 130207 : if (multixact_freeze_min_age < 0)
1228 : 7061 : multixact_freeze_min_age = vacuum_multixact_freeze_min_age;
1229 : 130207 : multixact_freeze_min_age = Min(multixact_freeze_min_age,
1230 : : effective_multixact_freeze_max_age / 2);
1231 : : Assert(multixact_freeze_min_age >= 0);
1232 : :
1233 : : /* Compute MultiXactCutoff, being careful to generate a valid value */
1234 : 130207 : cutoffs->MultiXactCutoff = nextMXID - multixact_freeze_min_age;
1235 [ - + ]: 130207 : if (cutoffs->MultiXactCutoff < FirstMultiXactId)
1236 : 0 : cutoffs->MultiXactCutoff = FirstMultiXactId;
1237 : : /* MultiXactCutoff must always be <= OldestMxact */
1238 [ + + ]: 130207 : if (MultiXactIdPrecedes(cutoffs->OldestMxact, cutoffs->MultiXactCutoff))
1239 : 2 : cutoffs->MultiXactCutoff = cutoffs->OldestMxact;
1240 : :
1241 : : /*
1242 : : * Finally, figure out if caller needs to do an aggressive VACUUM or not.
1243 : : *
1244 : : * Determine the table freeze age to use: as specified by the caller, or
1245 : : * the value of the vacuum_freeze_table_age GUC, but in any case not more
1246 : : * than autovacuum_freeze_max_age * 0.95, so that if you have e.g nightly
1247 : : * VACUUM schedule, the nightly VACUUM gets a chance to freeze XIDs before
1248 : : * anti-wraparound autovacuum is launched.
1249 : : */
1250 [ + + ]: 130207 : if (freeze_table_age < 0)
1251 : 7061 : freeze_table_age = vacuum_freeze_table_age;
1252 [ + - ]: 130207 : freeze_table_age = Min(freeze_table_age, autovacuum_freeze_max_age * 0.95);
1253 : : Assert(freeze_table_age >= 0);
1254 : 130207 : aggressiveXIDCutoff = nextXID - freeze_table_age;
1255 [ - + ]: 130207 : if (!TransactionIdIsNormal(aggressiveXIDCutoff))
1256 : 0 : aggressiveXIDCutoff = FirstNormalTransactionId;
1257 [ + + ]: 130207 : if (TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid,
1258 : : aggressiveXIDCutoff))
1259 : 123060 : return true;
1260 : :
1261 : : /*
1262 : : * Similar to the above, determine the table freeze age to use for
1263 : : * multixacts: as specified by the caller, or the value of the
1264 : : * vacuum_multixact_freeze_table_age GUC, but in any case not more than
1265 : : * effective_multixact_freeze_max_age * 0.95, so that if you have e.g.
1266 : : * nightly VACUUM schedule, the nightly VACUUM gets a chance to freeze
1267 : : * multixacts before anti-wraparound autovacuum is launched.
1268 : : */
1269 [ + + ]: 7147 : if (multixact_freeze_table_age < 0)
1270 : 6945 : multixact_freeze_table_age = vacuum_multixact_freeze_table_age;
1271 : 7147 : multixact_freeze_table_age =
1272 [ + - ]: 7147 : Min(multixact_freeze_table_age,
1273 : : effective_multixact_freeze_max_age * 0.95);
1274 : : Assert(multixact_freeze_table_age >= 0);
1275 : 7147 : aggressiveMXIDCutoff = nextMXID - multixact_freeze_table_age;
1276 [ - + ]: 7147 : if (aggressiveMXIDCutoff < FirstMultiXactId)
1277 : 0 : aggressiveMXIDCutoff = FirstMultiXactId;
1278 [ - + ]: 7147 : if (MultiXactIdPrecedesOrEquals(cutoffs->relminmxid,
1279 : : aggressiveMXIDCutoff))
1280 : 0 : return true;
1281 : :
1282 : : /* Non-aggressive VACUUM */
1283 : 7147 : return false;
1284 : : }
1285 : :
1286 : : /*
1287 : : * vacuum_xid_failsafe_check() -- Used by VACUUM's wraparound failsafe
1288 : : * mechanism to determine if its table's relfrozenxid and relminmxid are now
1289 : : * dangerously far in the past.
1290 : : *
1291 : : * When we return true, VACUUM caller triggers the failsafe.
1292 : : */
1293 : : bool
1294 : 132300 : vacuum_xid_failsafe_check(const struct VacuumCutoffs *cutoffs)
1295 : : {
1296 : 132300 : TransactionId relfrozenxid = cutoffs->relfrozenxid;
1297 : 132300 : MultiXactId relminmxid = cutoffs->relminmxid;
1298 : : TransactionId xid_skip_limit;
1299 : : MultiXactId multi_skip_limit;
1300 : : int skip_index_vacuum;
1301 : :
1302 : : Assert(TransactionIdIsNormal(relfrozenxid));
1303 : : Assert(MultiXactIdIsValid(relminmxid));
1304 : :
1305 : : /*
1306 : : * Determine the index skipping age to use. In any case no less than
1307 : : * autovacuum_freeze_max_age * 1.05.
1308 : : */
1309 [ + - ]: 132300 : skip_index_vacuum = Max(vacuum_failsafe_age, autovacuum_freeze_max_age * 1.05);
1310 : :
1311 : 132300 : xid_skip_limit = ReadNextTransactionId() - skip_index_vacuum;
1312 [ - + ]: 132300 : if (!TransactionIdIsNormal(xid_skip_limit))
1313 : 0 : xid_skip_limit = FirstNormalTransactionId;
1314 : :
1315 [ + + ]: 132300 : if (TransactionIdPrecedes(relfrozenxid, xid_skip_limit))
1316 : : {
1317 : : /* The table's relfrozenxid is too old */
1318 : 27978 : return true;
1319 : : }
1320 : :
1321 : : /*
1322 : : * Similar to above, determine the index skipping age to use for
1323 : : * multixact. In any case no less than autovacuum_multixact_freeze_max_age *
1324 : : * 1.05.
1325 : : */
1326 [ + - ]: 104322 : skip_index_vacuum = Max(vacuum_multixact_failsafe_age,
1327 : : autovacuum_multixact_freeze_max_age * 1.05);
1328 : :
1329 : 104322 : multi_skip_limit = ReadNextMultiXactId() - skip_index_vacuum;
1330 [ - + ]: 104322 : if (multi_skip_limit < FirstMultiXactId)
1331 : 0 : multi_skip_limit = FirstMultiXactId;
1332 : :
1333 [ - + ]: 104322 : if (MultiXactIdPrecedes(relminmxid, multi_skip_limit))
1334 : : {
1335 : : /* The table's relminmxid is too old */
1336 : 0 : return true;
1337 : : }
1338 : :
1339 : 104322 : return false;
1340 : : }
1341 : :
1342 : : /*
1343 : : * vac_estimate_reltuples() -- estimate the new value for pg_class.reltuples
1344 : : *
1345 : : * If we scanned the whole relation then we should just use the count of
1346 : : * live tuples seen; but if we did not, we should not blindly extrapolate
1347 : : * from that number, since VACUUM may have scanned a quite nonrandom
1348 : : * subset of the table. When we have only partial information, we take
1349 : : * the old value of pg_class.reltuples/pg_class.relpages as a measurement
1350 : : * of the tuple density in the unscanned pages.
1351 : : *
1352 : : * Note: scanned_tuples should count only *live* tuples, since
1353 : : * pg_class.reltuples is defined that way.
1354 : : */
1355 : : double
1356 : 129800 : vac_estimate_reltuples(Relation relation,
1357 : : BlockNumber total_pages,
1358 : : BlockNumber scanned_pages,
1359 : : double scanned_tuples)
1360 : : {
1361 : 129800 : BlockNumber old_rel_pages = relation->rd_rel->relpages;
1362 : 129800 : double old_rel_tuples = relation->rd_rel->reltuples;
1363 : : double old_density;
1364 : : double unscanned_pages;
1365 : : double total_tuples;
1366 : :
1367 : : /* If we did scan the whole table, just use the count as-is */
1368 [ + + ]: 129800 : if (scanned_pages >= total_pages)
1369 : 125469 : return scanned_tuples;
1370 : :
1371 : : /*
1372 : : * When successive VACUUM commands scan the same few pages again and
1373 : : * again, without anything from the table really changing, there is a risk
1374 : : * that our beliefs about tuple density will gradually become distorted.
1375 : : * This might be caused by vacuumlazy.c implementation details, such as
1376 : : * its tendency to always scan the last heap page. Handle that here.
1377 : : *
1378 : : * If the relation is _exactly_ the same size according to the existing
1379 : : * pg_class entry, and only a few of its pages (less than 2%) were
1380 : : * scanned, keep the existing value of reltuples. Also keep the existing
1381 : : * value when only a subset of rel's pages <= a single page were scanned.
1382 : : *
1383 : : * (Note: we might be returning -1 here.)
1384 : : */
1385 [ + + ]: 4331 : if (old_rel_pages == total_pages &&
1386 [ + + ]: 4306 : scanned_pages < (double) total_pages * 0.02)
1387 : 3116 : return old_rel_tuples;
1388 [ + + ]: 1215 : if (scanned_pages <= 1)
1389 : 1072 : return old_rel_tuples;
1390 : :
1391 : : /*
1392 : : * If old density is unknown, we can't do much except scale up
1393 : : * scanned_tuples to match total_pages.
1394 : : */
1395 [ + + - + ]: 143 : if (old_rel_tuples < 0 || old_rel_pages == 0)
1396 : 5 : return floor((scanned_tuples / scanned_pages) * total_pages + 0.5);
1397 : :
1398 : : /*
1399 : : * Okay, we've covered the corner cases. The normal calculation is to
1400 : : * convert the old measurement to a density (tuples per page), then
1401 : : * estimate the number of tuples in the unscanned pages using that figure,
1402 : : * and finally add on the number of tuples in the scanned pages.
1403 : : */
1404 : 138 : old_density = old_rel_tuples / old_rel_pages;
1405 : 138 : unscanned_pages = (double) total_pages - (double) scanned_pages;
1406 : 138 : total_tuples = old_density * unscanned_pages + scanned_tuples;
1407 : 138 : return floor(total_tuples + 0.5);
1408 : : }
1409 : :
1410 : :
1411 : : /*
1412 : : * vac_update_relstats() -- update statistics for one relation
1413 : : *
1414 : : * Update the whole-relation statistics that are kept in its pg_class
1415 : : * row. There are additional stats that will be updated if we are
1416 : : * doing ANALYZE, but we always update these stats. This routine works
1417 : : * for both index and heap relation entries in pg_class.
1418 : : *
1419 : : * We violate transaction semantics here by overwriting the rel's
1420 : : * existing pg_class tuple with the new values. This is reasonably
1421 : : * safe as long as we're sure that the new values are correct whether or
1422 : : * not this transaction commits. The reason for doing this is that if
1423 : : * we updated these tuples in the usual way, vacuuming pg_class itself
1424 : : * wouldn't work very well --- by the time we got done with a vacuum
1425 : : * cycle, most of the tuples in pg_class would've been obsoleted. Of
1426 : : * course, this only works for fixed-size not-null columns, but these are.
1427 : : *
1428 : : * Another reason for doing it this way is that when we are in a lazy
1429 : : * VACUUM and have PROC_IN_VACUUM set, we mustn't do any regular updates.
1430 : : * Somebody vacuuming pg_class might think they could delete a tuple
1431 : : * marked with xmin = our xid.
1432 : : *
1433 : : * In addition to fundamentally nontransactional statistics such as
1434 : : * relpages and relallvisible, we try to maintain certain lazily-updated
1435 : : * DDL flags such as relhasindex, by clearing them if no longer correct.
1436 : : * It's safe to do this in VACUUM, which can't run in parallel with
1437 : : * CREATE INDEX/RULE/TRIGGER and can't be part of a transaction block.
1438 : : * However, it's *not* safe to do it in an ANALYZE that's within an
1439 : : * outer transaction, because for example the current transaction might
1440 : : * have dropped the last index; then we'd think relhasindex should be
1441 : : * cleared, but if the transaction later rolls back this would be wrong.
1442 : : * So we refrain from updating the DDL flags if we're inside an outer
1443 : : * transaction. This is OK since postponing the flag maintenance is
1444 : : * always allowable.
1445 : : *
1446 : : * Note: num_tuples should count only *live* tuples, since
1447 : : * pg_class.reltuples is defined that way.
1448 : : *
1449 : : * This routine is shared by VACUUM and ANALYZE.
1450 : : */
1451 : : void
1452 : 156977 : vac_update_relstats(Relation relation,
1453 : : BlockNumber num_pages, double num_tuples,
1454 : : BlockNumber num_all_visible_pages,
1455 : : BlockNumber num_all_frozen_pages,
1456 : : bool hasindex, TransactionId frozenxid,
1457 : : MultiXactId minmulti,
1458 : : bool *frozenxid_updated, bool *minmulti_updated,
1459 : : bool in_outer_xact)
1460 : : {
1461 : 156977 : Oid relid = RelationGetRelid(relation);
1462 : : Relation rd;
1463 : : ScanKeyData key[1];
1464 : : HeapTuple ctup;
1465 : : void *inplace_state;
1466 : : Form_pg_class pgcform;
1467 : : bool dirty,
1468 : : futurexid,
1469 : : futuremxid;
1470 : : TransactionId oldfrozenxid;
1471 : : MultiXactId oldminmulti;
1472 : :
1473 : 156977 : rd = table_open(RelationRelationId, RowExclusiveLock);
1474 : :
1475 : : /* Fetch a copy of the tuple to scribble on */
1476 : 156977 : ScanKeyInit(&key[0],
1477 : : Anum_pg_class_oid,
1478 : : BTEqualStrategyNumber, F_OIDEQ,
1479 : : ObjectIdGetDatum(relid));
1480 : 156977 : systable_inplace_update_begin(rd, ClassOidIndexId, true,
1481 : : NULL, 1, key, &ctup, &inplace_state);
1482 [ - + ]: 156976 : if (!HeapTupleIsValid(ctup))
1483 [ # # ]: 0 : elog(ERROR, "pg_class entry for relid %u vanished during vacuuming",
1484 : : relid);
1485 : 156976 : pgcform = (Form_pg_class) GETSTRUCT(ctup);
1486 : :
1487 : : /* Apply statistical updates, if any, to copied tuple */
1488 : :
1489 : 156976 : dirty = false;
1490 [ + + ]: 156976 : if (pgcform->relpages != (int32) num_pages)
1491 : : {
1492 : 5861 : pgcform->relpages = (int32) num_pages;
1493 : 5861 : dirty = true;
1494 : : }
1495 [ + + ]: 156976 : if (pgcform->reltuples != (float4) num_tuples)
1496 : : {
1497 : 12678 : pgcform->reltuples = (float4) num_tuples;
1498 : 12678 : dirty = true;
1499 : : }
1500 [ + + ]: 156976 : if (pgcform->relallvisible != (int32) num_all_visible_pages)
1501 : : {
1502 : 3913 : pgcform->relallvisible = (int32) num_all_visible_pages;
1503 : 3913 : dirty = true;
1504 : : }
1505 [ + + ]: 156976 : if (pgcform->relallfrozen != (int32) num_all_frozen_pages)
1506 : : {
1507 : 3279 : pgcform->relallfrozen = (int32) num_all_frozen_pages;
1508 : 3279 : dirty = true;
1509 : : }
1510 : :
1511 : : /* Apply DDL updates, but not inside an outer transaction (see above) */
1512 : :
1513 [ + + ]: 156976 : if (!in_outer_xact)
1514 : : {
1515 : : /*
1516 : : * If we didn't find any indexes, reset relhasindex.
1517 : : */
1518 [ + + + + ]: 156682 : if (pgcform->relhasindex && !hasindex)
1519 : : {
1520 : 14 : pgcform->relhasindex = false;
1521 : 14 : dirty = true;
1522 : : }
1523 : :
1524 : : /* We also clear relhasrules and relhastriggers if needed */
1525 [ + + - + ]: 156682 : if (pgcform->relhasrules && relation->rd_rules == NULL)
1526 : : {
1527 : 0 : pgcform->relhasrules = false;
1528 : 0 : dirty = true;
1529 : : }
1530 [ + + + + ]: 156682 : if (pgcform->relhastriggers && relation->trigdesc == NULL)
1531 : : {
1532 : 4 : pgcform->relhastriggers = false;
1533 : 4 : dirty = true;
1534 : : }
1535 : : }
1536 : :
1537 : : /*
1538 : : * Update relfrozenxid, unless caller passed InvalidTransactionId
1539 : : * indicating it has no new data.
1540 : : *
1541 : : * Ordinarily, we don't let relfrozenxid go backwards. However, if the
1542 : : * stored relfrozenxid is "in the future" then it seems best to assume
1543 : : * it's corrupt, and overwrite with the oldest remaining XID in the table.
1544 : : * This should match vac_update_datfrozenxid() concerning what we consider
1545 : : * to be "in the future".
1546 : : */
1547 : 156976 : oldfrozenxid = pgcform->relfrozenxid;
1548 : 156976 : futurexid = false;
1549 [ + + ]: 156976 : if (frozenxid_updated)
1550 : 129796 : *frozenxid_updated = false;
1551 [ + + + + ]: 156976 : if (TransactionIdIsNormal(frozenxid) && oldfrozenxid != frozenxid)
1552 : : {
1553 : 33913 : bool update = false;
1554 : :
1555 [ + + ]: 33913 : if (TransactionIdPrecedes(oldfrozenxid, frozenxid))
1556 : 33847 : update = true;
1557 [ - + ]: 66 : else if (TransactionIdPrecedes(ReadNextTransactionId(), oldfrozenxid))
1558 : 0 : futurexid = update = true;
1559 : :
1560 [ + + ]: 33913 : if (update)
1561 : : {
1562 : 33847 : pgcform->relfrozenxid = frozenxid;
1563 : 33847 : dirty = true;
1564 [ + - ]: 33847 : if (frozenxid_updated)
1565 : 33847 : *frozenxid_updated = true;
1566 : : }
1567 : : }
1568 : :
1569 : : /* Similarly for relminmxid */
1570 : 156976 : oldminmulti = pgcform->relminmxid;
1571 : 156976 : futuremxid = false;
1572 [ + + ]: 156976 : if (minmulti_updated)
1573 : 129796 : *minmulti_updated = false;
1574 [ + + + + ]: 156976 : if (MultiXactIdIsValid(minmulti) && oldminmulti != minmulti)
1575 : : {
1576 : 254 : bool update = false;
1577 : :
1578 [ + - ]: 254 : if (MultiXactIdPrecedes(oldminmulti, minmulti))
1579 : 254 : update = true;
1580 [ # # ]: 0 : else if (MultiXactIdPrecedes(ReadNextMultiXactId(), oldminmulti))
1581 : 0 : futuremxid = update = true;
1582 : :
1583 [ + - ]: 254 : if (update)
1584 : : {
1585 : 254 : pgcform->relminmxid = minmulti;
1586 : 254 : dirty = true;
1587 [ + - ]: 254 : if (minmulti_updated)
1588 : 254 : *minmulti_updated = true;
1589 : : }
1590 : : }
1591 : :
1592 : : /* If anything changed, write out the tuple. */
1593 [ + + ]: 156976 : if (dirty)
1594 : 42881 : systable_inplace_update_finish(inplace_state, ctup);
1595 : : else
1596 : 114095 : systable_inplace_update_cancel(inplace_state);
1597 : :
1598 : 156976 : table_close(rd, RowExclusiveLock);
1599 : :
1600 [ - + ]: 156976 : if (futurexid)
1601 [ # # ]: 0 : ereport(WARNING,
1602 : : (errcode(ERRCODE_DATA_CORRUPTED),
1603 : : errmsg_internal("overwrote invalid relfrozenxid value %u with new value %u for table \"%s\"",
1604 : : oldfrozenxid, frozenxid,
1605 : : RelationGetRelationName(relation))));
1606 [ - + ]: 156976 : if (futuremxid)
1607 [ # # ]: 0 : ereport(WARNING,
1608 : : (errcode(ERRCODE_DATA_CORRUPTED),
1609 : : errmsg_internal("overwrote invalid relminmxid value %u with new value %u for table \"%s\"",
1610 : : oldminmulti, minmulti,
1611 : : RelationGetRelationName(relation))));
1612 : 156976 : }
1613 : :
1614 : :
1615 : : /*
1616 : : * vac_update_datfrozenxid() -- update pg_database.datfrozenxid for our DB
1617 : : *
1618 : : * Update pg_database's datfrozenxid entry for our database to be the
1619 : : * minimum of the pg_class.relfrozenxid values.
1620 : : *
1621 : : * Similarly, update our datminmxid to be the minimum of the
1622 : : * pg_class.relminmxid values.
1623 : : *
1624 : : * If we are able to advance either pg_database value, also try to
1625 : : * truncate pg_xact and pg_multixact.
1626 : : *
1627 : : * We violate transaction semantics here by overwriting the database's
1628 : : * existing pg_database tuple with the new values. This is reasonably
1629 : : * safe since the new values are correct whether or not this transaction
1630 : : * commits. As with vac_update_relstats, this avoids leaving dead tuples
1631 : : * behind after a VACUUM.
1632 : : */
1633 : : void
1634 : 2727 : vac_update_datfrozenxid(void)
1635 : : {
1636 : : HeapTuple tuple;
1637 : : Form_pg_database dbform;
1638 : : Relation relation;
1639 : : SysScanDesc scan;
1640 : : HeapTuple classTup;
1641 : : TransactionId newFrozenXid;
1642 : : MultiXactId newMinMulti;
1643 : : TransactionId lastSaneFrozenXid;
1644 : : MultiXactId lastSaneMinMulti;
1645 : 2727 : bool bogus = false;
1646 : 2727 : bool dirty = false;
1647 : : ScanKeyData key[1];
1648 : : void *inplace_state;
1649 : :
1650 : : /*
1651 : : * Restrict this task to one backend per database. This avoids race
1652 : : * conditions that would move datfrozenxid or datminmxid backward. It
1653 : : * avoids calling vac_truncate_clog() with a datfrozenxid preceding a
1654 : : * datfrozenxid passed to an earlier vac_truncate_clog() call.
1655 : : */
1656 : 2727 : LockDatabaseFrozenIds(ExclusiveLock);
1657 : :
1658 : : /*
1659 : : * Initialize the "min" calculation with
1660 : : * GetOldestNonRemovableTransactionId(), which is a reasonable
1661 : : * approximation to the minimum relfrozenxid for not-yet-committed
1662 : : * pg_class entries for new tables; see AddNewRelationTuple(). So we
1663 : : * cannot produce a wrong minimum by starting with this.
1664 : : */
1665 : 2727 : newFrozenXid = GetOldestNonRemovableTransactionId(NULL);
1666 : :
1667 : : /*
1668 : : * Similarly, initialize the MultiXact "min" with the value that would be
1669 : : * used on pg_class for new tables. See AddNewRelationTuple().
1670 : : */
1671 : 2727 : newMinMulti = GetOldestMultiXactId();
1672 : :
1673 : : /*
1674 : : * Identify the latest relfrozenxid and relminmxid values that we could
1675 : : * validly see during the scan. These are conservative values, but it's
1676 : : * not really worth trying to be more exact.
1677 : : */
1678 : 2727 : lastSaneFrozenXid = ReadNextTransactionId();
1679 : 2727 : lastSaneMinMulti = ReadNextMultiXactId();
1680 : :
1681 : : /*
1682 : : * We must seqscan pg_class to find the minimum Xid, because there is no
1683 : : * index that can help us here.
1684 : : *
1685 : : * See vac_truncate_clog() for the race condition to prevent.
1686 : : */
1687 : 2727 : relation = table_open(RelationRelationId, AccessShareLock);
1688 : :
1689 : 2727 : scan = systable_beginscan(relation, InvalidOid, false,
1690 : : NULL, 0, NULL);
1691 : :
1692 [ + + ]: 1589140 : while ((classTup = systable_getnext(scan)) != NULL)
1693 : : {
1694 : 1586413 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(classTup);
1695 : 1586413 : volatile TransactionId *relfrozenxid_p = &classForm->relfrozenxid;
1696 : 1586413 : volatile TransactionId *relminmxid_p = &classForm->relminmxid;
1697 : 1586413 : TransactionId relfrozenxid = *relfrozenxid_p;
1698 : 1586413 : TransactionId relminmxid = *relminmxid_p;
1699 : :
1700 : : /*
1701 : : * Only consider relations able to hold unfrozen XIDs (anything else
1702 : : * should have InvalidTransactionId in relfrozenxid anyway).
1703 : : */
1704 [ + + ]: 1586413 : if (classForm->relkind != RELKIND_RELATION &&
1705 [ + + ]: 1250995 : classForm->relkind != RELKIND_MATVIEW &&
1706 [ + + ]: 1249341 : classForm->relkind != RELKIND_TOASTVALUE)
1707 : : {
1708 : : Assert(!TransactionIdIsValid(relfrozenxid));
1709 : : Assert(!MultiXactIdIsValid(relminmxid));
1710 : 1079106 : continue;
1711 : : }
1712 : :
1713 : : /*
1714 : : * Some table AMs might not need per-relation xid / multixid horizons.
1715 : : * It therefore seems reasonable to allow relfrozenxid and relminmxid
1716 : : * to not be set (i.e. set to their respective Invalid*Id)
1717 : : * independently. Thus validate and compute horizon for each only if
1718 : : * set.
1719 : : *
1720 : : * If things are working properly, no relation should have a
1721 : : * relfrozenxid or relminmxid that is "in the future". However, such
1722 : : * cases have been known to arise due to bugs in pg_upgrade. If we
1723 : : * see any entries that are "in the future", chicken out and don't do
1724 : : * anything. This ensures we won't truncate clog & multixact SLRUs
1725 : : * before those relations have been scanned and cleaned up.
1726 : : */
1727 : :
1728 [ + - ]: 507307 : if (TransactionIdIsValid(relfrozenxid))
1729 : : {
1730 : : Assert(TransactionIdIsNormal(relfrozenxid));
1731 : :
1732 : : /* check for values in the future */
1733 [ - + ]: 507307 : if (TransactionIdPrecedes(lastSaneFrozenXid, relfrozenxid))
1734 : : {
1735 : 0 : bogus = true;
1736 : 0 : break;
1737 : : }
1738 : :
1739 : : /* determine new horizon */
1740 [ + + ]: 507307 : if (TransactionIdPrecedes(relfrozenxid, newFrozenXid))
1741 : 2499 : newFrozenXid = relfrozenxid;
1742 : : }
1743 : :
1744 [ + - ]: 507307 : if (MultiXactIdIsValid(relminmxid))
1745 : : {
1746 : : /* check for values in the future */
1747 [ - + ]: 507307 : if (MultiXactIdPrecedes(lastSaneMinMulti, relminmxid))
1748 : : {
1749 : 0 : bogus = true;
1750 : 0 : break;
1751 : : }
1752 : :
1753 : : /* determine new horizon */
1754 [ + + ]: 507307 : if (MultiXactIdPrecedes(relminmxid, newMinMulti))
1755 : 223 : newMinMulti = relminmxid;
1756 : : }
1757 : : }
1758 : :
1759 : : /* we're done with pg_class */
1760 : 2727 : systable_endscan(scan);
1761 : 2727 : table_close(relation, AccessShareLock);
1762 : :
1763 : : /* chicken out if bogus data found */
1764 [ - + ]: 2727 : if (bogus)
1765 : 0 : return;
1766 : :
1767 : : Assert(TransactionIdIsNormal(newFrozenXid));
1768 : : Assert(MultiXactIdIsValid(newMinMulti));
1769 : :
1770 : : /* Now fetch the pg_database tuple we need to update. */
1771 : 2727 : relation = table_open(DatabaseRelationId, RowExclusiveLock);
1772 : :
1773 : : /*
1774 : : * Fetch a copy of the tuple to scribble on. We could check the syscache
1775 : : * tuple first. If that concluded !dirty, we'd avoid waiting on
1776 : : * concurrent heap_update() and would avoid exclusive-locking the buffer.
1777 : : * For now, don't optimize that.
1778 : : */
1779 : 2727 : ScanKeyInit(&key[0],
1780 : : Anum_pg_database_oid,
1781 : : BTEqualStrategyNumber, F_OIDEQ,
1782 : : ObjectIdGetDatum(MyDatabaseId));
1783 : :
1784 : 2727 : systable_inplace_update_begin(relation, DatabaseOidIndexId, true,
1785 : : NULL, 1, key, &tuple, &inplace_state);
1786 : :
1787 [ - + ]: 2727 : if (!HeapTupleIsValid(tuple))
1788 [ # # ]: 0 : elog(ERROR, "could not find tuple for database %u", MyDatabaseId);
1789 : :
1790 : 2727 : dbform = (Form_pg_database) GETSTRUCT(tuple);
1791 : :
1792 : : /*
1793 : : * As in vac_update_relstats(), we ordinarily don't want to let
1794 : : * datfrozenxid go backward; but if it's "in the future" then it must be
1795 : : * corrupt and it seems best to overwrite it.
1796 : : */
1797 [ + + - + ]: 3057 : if (dbform->datfrozenxid != newFrozenXid &&
1798 [ - - ]: 330 : (TransactionIdPrecedes(dbform->datfrozenxid, newFrozenXid) ||
1799 : 0 : TransactionIdPrecedes(lastSaneFrozenXid, dbform->datfrozenxid)))
1800 : : {
1801 : 330 : dbform->datfrozenxid = newFrozenXid;
1802 : 330 : dirty = true;
1803 : : }
1804 : : else
1805 : 2397 : newFrozenXid = dbform->datfrozenxid;
1806 : :
1807 : : /* Ditto for datminmxid */
1808 [ + + - + ]: 2728 : if (dbform->datminmxid != newMinMulti &&
1809 [ - - ]: 1 : (MultiXactIdPrecedes(dbform->datminmxid, newMinMulti) ||
1810 : 0 : MultiXactIdPrecedes(lastSaneMinMulti, dbform->datminmxid)))
1811 : : {
1812 : 1 : dbform->datminmxid = newMinMulti;
1813 : 1 : dirty = true;
1814 : : }
1815 : : else
1816 : 2726 : newMinMulti = dbform->datminmxid;
1817 : :
1818 [ + + ]: 2727 : if (dirty)
1819 : 330 : systable_inplace_update_finish(inplace_state, tuple);
1820 : : else
1821 : 2397 : systable_inplace_update_cancel(inplace_state);
1822 : :
1823 : 2727 : heap_freetuple(tuple);
1824 : 2727 : table_close(relation, RowExclusiveLock);
1825 : :
1826 : : /*
1827 : : * If we were able to advance datfrozenxid or datminmxid, see if we can
1828 : : * truncate pg_xact and/or pg_multixact. Also do it if the shared
1829 : : * XID-wrap-limit info is stale, since this action will update that too.
1830 : : */
1831 [ + + + + ]: 2727 : if (dirty || ForceTransactionIdLimitUpdate())
1832 : 1181 : vac_truncate_clog(newFrozenXid, newMinMulti,
1833 : : lastSaneFrozenXid, lastSaneMinMulti);
1834 : : }
1835 : :
1836 : :
1837 : : /*
1838 : : * vac_truncate_clog() -- attempt to truncate the commit log
1839 : : *
1840 : : * Scan pg_database to determine the system-wide oldest datfrozenxid,
1841 : : * and use it to truncate the transaction commit log (pg_xact).
1842 : : * Also update the XID wrap limit info maintained by varsup.c.
1843 : : * Likewise for datminmxid.
1844 : : *
1845 : : * The passed frozenXID and minMulti are the updated values for my own
1846 : : * pg_database entry. They're used to initialize the "min" calculations.
1847 : : * The caller also passes the "last sane" XID and MXID, since it has
1848 : : * those at hand already.
1849 : : *
1850 : : * This routine is only invoked when we've managed to change our
1851 : : * DB's datfrozenxid/datminmxid values, or we found that the shared
1852 : : * XID-wrap-limit info is stale.
1853 : : */
1854 : : static void
1855 : 1181 : vac_truncate_clog(TransactionId frozenXID,
1856 : : MultiXactId minMulti,
1857 : : TransactionId lastSaneFrozenXid,
1858 : : MultiXactId lastSaneMinMulti)
1859 : : {
1860 : 1181 : TransactionId nextXID = ReadNextTransactionId();
1861 : : Relation relation;
1862 : : TableScanDesc scan;
1863 : : HeapTuple tuple;
1864 : : Oid oldestxid_datoid;
1865 : : Oid minmulti_datoid;
1866 : 1181 : bool bogus = false;
1867 : 1181 : bool frozenAlreadyWrapped = false;
1868 : :
1869 : : /* Restrict task to one backend per cluster; see SimpleLruTruncate(). */
1870 : 1181 : LWLockAcquire(WrapLimitsVacuumLock, LW_EXCLUSIVE);
1871 : :
1872 : : /* init oldest datoids to sync with my frozenXID/minMulti values */
1873 : 1181 : oldestxid_datoid = MyDatabaseId;
1874 : 1181 : minmulti_datoid = MyDatabaseId;
1875 : :
1876 : : /*
1877 : : * Scan pg_database to compute the minimum datfrozenxid/datminmxid
1878 : : *
1879 : : * Since vac_update_datfrozenxid updates datfrozenxid/datminmxid in-place,
1880 : : * the values could change while we look at them. Fetch each one just
1881 : : * once to ensure sane behavior of the comparison logic. (Here, as in
1882 : : * many other places, we assume that fetching or updating an XID in shared
1883 : : * storage is atomic.)
1884 : : *
1885 : : * Note: we need not worry about a race condition with new entries being
1886 : : * inserted by CREATE DATABASE. Any such entry will have a copy of some
1887 : : * existing DB's datfrozenxid, and that source DB cannot be ours because
1888 : : * of the interlock against copying a DB containing an active backend.
1889 : : * Hence the new entry will not reduce the minimum. Also, if two VACUUMs
1890 : : * concurrently modify the datfrozenxid's of different databases, the
1891 : : * worst possible outcome is that pg_xact is not truncated as aggressively
1892 : : * as it could be.
1893 : : */
1894 : 1181 : relation = table_open(DatabaseRelationId, AccessShareLock);
1895 : :
1896 : 1181 : scan = table_beginscan_catalog(relation, 0, NULL);
1897 : :
1898 [ + + ]: 4628 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1899 : : {
1900 : 3447 : Form_pg_database dbform = (Form_pg_database) GETSTRUCT(tuple);
1901 : 3447 : volatile TransactionId *datfrozenxid_p = &dbform->datfrozenxid;
1902 : 3447 : volatile TransactionId *datminmxid_p = &dbform->datminmxid;
1903 : 3447 : TransactionId datfrozenxid = *datfrozenxid_p;
1904 : 3447 : TransactionId datminmxid = *datminmxid_p;
1905 : :
1906 : : Assert(TransactionIdIsNormal(datfrozenxid));
1907 : : Assert(MultiXactIdIsValid(datminmxid));
1908 : :
1909 : : /*
1910 : : * If database is in the process of getting dropped, or has been
1911 : : * interrupted while doing so, no connections to it are possible
1912 : : * anymore. Therefore we don't need to take it into account here.
1913 : : * Which is good, because it can't be processed by autovacuum either.
1914 : : */
1915 [ + + ]: 3447 : if (database_is_invalid_form((Form_pg_database) dbform))
1916 : : {
1917 [ - + ]: 3 : elog(DEBUG2,
1918 : : "skipping invalid database \"%s\" while computing relfrozenxid",
1919 : : NameStr(dbform->datname));
1920 : 3 : continue;
1921 : : }
1922 : :
1923 : : /*
1924 : : * If things are working properly, no database should have a
1925 : : * datfrozenxid or datminmxid that is "in the future". However, such
1926 : : * cases have been known to arise due to bugs in pg_upgrade. If we
1927 : : * see any entries that are "in the future", chicken out and don't do
1928 : : * anything. This ensures we won't truncate clog before those
1929 : : * databases have been scanned and cleaned up. (We will issue the
1930 : : * "already wrapped" warning if appropriate, though.)
1931 : : */
1932 [ + - - + ]: 6888 : if (TransactionIdPrecedes(lastSaneFrozenXid, datfrozenxid) ||
1933 : 3444 : MultiXactIdPrecedes(lastSaneMinMulti, datminmxid))
1934 : 0 : bogus = true;
1935 : :
1936 [ - + ]: 3444 : if (TransactionIdPrecedes(nextXID, datfrozenxid))
1937 : 0 : frozenAlreadyWrapped = true;
1938 [ + + ]: 3444 : else if (TransactionIdPrecedes(datfrozenxid, frozenXID))
1939 : : {
1940 : 282 : frozenXID = datfrozenxid;
1941 : 282 : oldestxid_datoid = dbform->oid;
1942 : : }
1943 : :
1944 [ + + ]: 3444 : if (MultiXactIdPrecedes(datminmxid, minMulti))
1945 : : {
1946 : 2 : minMulti = datminmxid;
1947 : 2 : minmulti_datoid = dbform->oid;
1948 : : }
1949 : : }
1950 : :
1951 : 1181 : table_endscan(scan);
1952 : :
1953 : 1181 : table_close(relation, AccessShareLock);
1954 : :
1955 : : /*
1956 : : * Do not truncate CLOG if we seem to have suffered wraparound already;
1957 : : * the computed minimum XID might be bogus. This case should now be
1958 : : * impossible due to the defenses in GetNewTransactionId, but we keep the
1959 : : * test anyway.
1960 : : */
1961 [ - + ]: 1181 : if (frozenAlreadyWrapped)
1962 : : {
1963 [ # # ]: 0 : ereport(WARNING,
1964 : : (errmsg("some databases have not been vacuumed in over 2 billion transactions"),
1965 : : errdetail("You might have already suffered transaction-wraparound data loss.")));
1966 : 0 : LWLockRelease(WrapLimitsVacuumLock);
1967 : 0 : return;
1968 : : }
1969 : :
1970 : : /* chicken out if data is bogus in any other way */
1971 [ - + ]: 1181 : if (bogus)
1972 : : {
1973 : 0 : LWLockRelease(WrapLimitsVacuumLock);
1974 : 0 : return;
1975 : : }
1976 : :
1977 : : /*
1978 : : * Freeze any old transaction IDs in the async notification queue before
1979 : : * CLOG truncation.
1980 : : */
1981 : 1181 : AsyncNotifyFreezeXids(frozenXID);
1982 : :
1983 : : /*
1984 : : * Advance the oldest value for commit timestamps before truncating, so
1985 : : * that if a user requests a timestamp for a transaction we're truncating
1986 : : * away right after this point, they get NULL instead of an ugly "file not
1987 : : * found" error from slru.c. This doesn't matter for xact/multixact
1988 : : * because they are not subject to arbitrary lookups from users.
1989 : : */
1990 : 1181 : AdvanceOldestCommitTsXid(frozenXID);
1991 : :
1992 : : /*
1993 : : * Truncate CLOG, multixact and CommitTs to the oldest computed value.
1994 : : */
1995 : 1181 : TruncateCLOG(frozenXID, oldestxid_datoid);
1996 : 1181 : TruncateCommitTs(frozenXID);
1997 : 1181 : TruncateMultiXact(minMulti, minmulti_datoid);
1998 : :
1999 : : /*
2000 : : * Update the wrap limit for GetNewTransactionId and creation of new
2001 : : * MultiXactIds. Note: these functions will also signal the postmaster
2002 : : * for an(other) autovac cycle if needed. XXX should we avoid possibly
2003 : : * signaling twice?
2004 : : */
2005 : 1181 : SetTransactionIdLimit(frozenXID, oldestxid_datoid);
2006 : 1181 : SetMultiXactIdLimit(minMulti, minmulti_datoid);
2007 : :
2008 : 1181 : LWLockRelease(WrapLimitsVacuumLock);
2009 : : }
2010 : :
2011 : :
2012 : : /*
2013 : : * vacuum_rel() -- vacuum one heap relation
2014 : : *
2015 : : * relid identifies the relation to vacuum. If relation is supplied,
2016 : : * use the name therein for reporting any failure to open/lock the rel;
2017 : : * do not use it once we've successfully opened the rel, since it might
2018 : : * be stale.
2019 : : *
2020 : : * Returns true if it's okay to proceed with a requested ANALYZE
2021 : : * operation on this table.
2022 : : *
2023 : : * Doing one heap at a time incurs extra overhead, since we need to
2024 : : * check that the heap exists again just before we vacuum it. The
2025 : : * reason that we do this is so that vacuuming can be spread across
2026 : : * many small transactions. Otherwise, two-phase locking would require
2027 : : * us to lock the entire database during one pass of the vacuum cleaner.
2028 : : *
2029 : : * At entry and exit, we are not inside a transaction.
2030 : : */
2031 : : static bool
2032 : 130295 : vacuum_rel(Oid relid, RangeVar *relation, VacuumParams params,
2033 : : BufferAccessStrategy bstrategy, bool isTopLevel)
2034 : : {
2035 : : LOCKMODE lmode;
2036 : : Relation rel;
2037 : : LockRelId lockrelid;
2038 : : Oid priv_relid;
2039 : : Oid toast_relid;
2040 : : Oid save_userid;
2041 : : int save_sec_context;
2042 : : int save_nestlevel;
2043 : : VacuumParams toast_vacuum_params;
2044 : : StdRdOptions *relopts;
2045 : : StdRdOptions relopts_copy;
2046 : :
2047 : : /*
2048 : : * This function scribbles on the parameters, so make a copy early to
2049 : : * avoid affecting the TOAST table (if we do end up recursing to it).
2050 : : */
2051 : 130295 : memcpy(&toast_vacuum_params, ¶ms, sizeof(VacuumParams));
2052 : :
2053 : : /* Begin a transaction for vacuuming this relation */
2054 : 130295 : StartTransactionCommand();
2055 : :
2056 [ + + ]: 130295 : if (!(params.options & VACOPT_FULL))
2057 : : {
2058 : : /*
2059 : : * In lazy vacuum, we can set the PROC_IN_VACUUM flag, which lets
2060 : : * other concurrent VACUUMs know that they can ignore this one while
2061 : : * determining their OldestXmin. (The reason we don't set it during a
2062 : : * full VACUUM is exactly that we may have to run user-defined
2063 : : * functions for functional indexes, and we want to make sure that if
2064 : : * they use the snapshot set above, any tuples it requires can't get
2065 : : * removed from other tables. An index function that depends on the
2066 : : * contents of other tables is arguably broken, but we won't break it
2067 : : * here by violating transaction semantics.)
2068 : : *
2069 : : * We also set the VACUUM_FOR_WRAPAROUND flag, which is passed down by
2070 : : * autovacuum; it's used to avoid canceling a vacuum that was invoked
2071 : : * in an emergency.
2072 : : *
2073 : : * Note: these flags remain set until CommitTransaction or
2074 : : * AbortTransaction. We don't want to clear them until we reset
2075 : : * MyProc->xid/xmin, otherwise GetOldestNonRemovableTransactionId()
2076 : : * might appear to go backwards, which is probably Not Good. (We also
2077 : : * set PROC_IN_VACUUM *before* taking our own snapshot, so that our
2078 : : * xmin doesn't become visible ahead of setting the flag.)
2079 : : */
2080 : 130048 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
2081 : 130048 : MyProc->statusFlags |= PROC_IN_VACUUM;
2082 [ + + ]: 130048 : if (params.is_wraparound)
2083 : 113027 : MyProc->statusFlags |= PROC_VACUUM_FOR_WRAPAROUND;
2084 : 130048 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
2085 : 130048 : LWLockRelease(ProcArrayLock);
2086 : : }
2087 : :
2088 : : /*
2089 : : * Need to acquire a snapshot to prevent pg_subtrans from being truncated,
2090 : : * cutoff xids in local memory wrapping around, and to have updated xmin
2091 : : * horizons.
2092 : : */
2093 : 130295 : PushActiveSnapshot(GetTransactionSnapshot());
2094 : :
2095 : : /*
2096 : : * Check for user-requested abort. Note we want this to be inside a
2097 : : * transaction, so xact.c doesn't issue useless WARNING.
2098 : : */
2099 [ - + ]: 130295 : CHECK_FOR_INTERRUPTS();
2100 : :
2101 : : /*
2102 : : * Determine the type of lock we want --- hard exclusive lock for a FULL
2103 : : * vacuum, but just ShareUpdateExclusiveLock for concurrent vacuum. Either
2104 : : * way, we can be sure that no other backend is vacuuming the same table.
2105 : : */
2106 : 260590 : lmode = (params.options & VACOPT_FULL) ?
2107 [ + + ]: 130295 : AccessExclusiveLock : ShareUpdateExclusiveLock;
2108 : :
2109 : : /* open the relation and get the appropriate lock on it */
2110 : 130295 : rel = vacuum_open_relation(relid, relation, params.options,
2111 : 130295 : params.log_vacuum_min_duration >= 0, lmode);
2112 : :
2113 : : /* leave if relation could not be opened or locked */
2114 [ + + ]: 130295 : if (!rel)
2115 : : {
2116 : 12 : PopActiveSnapshot();
2117 : 12 : CommitTransactionCommand();
2118 : 12 : return false;
2119 : : }
2120 : :
2121 : : /*
2122 : : * When recursing to a TOAST table, check privileges on the parent. NB:
2123 : : * This is only safe to do because we hold a session lock on the main
2124 : : * relation that prevents concurrent deletion.
2125 : : */
2126 [ + + ]: 130283 : if (OidIsValid(params.toast_parent))
2127 : 5774 : priv_relid = params.toast_parent;
2128 : : else
2129 : 124509 : priv_relid = RelationGetRelid(rel);
2130 : :
2131 : : /*
2132 : : * Check if relation needs to be skipped based on privileges. This check
2133 : : * happens also when building the relation list to vacuum for a manual
2134 : : * operation, and needs to be done additionally here as VACUUM could
2135 : : * happen across multiple transactions where privileges could have changed
2136 : : * in-between. Make sure to only generate logs for VACUUM in this case.
2137 : : */
2138 [ + + ]: 130283 : if (!vacuum_is_permitted_for_relation(priv_relid,
2139 : : rel->rd_rel,
2140 : 130283 : params.options & ~VACOPT_ANALYZE,
2141 : : false))
2142 : : {
2143 : 48 : relation_close(rel, lmode);
2144 : 48 : PopActiveSnapshot();
2145 : 48 : CommitTransactionCommand();
2146 : 48 : return false;
2147 : : }
2148 : :
2149 : : /*
2150 : : * Check that it's of a vacuumable relkind.
2151 : : */
2152 [ + + ]: 130235 : if (rel->rd_rel->relkind != RELKIND_RELATION &&
2153 [ + + ]: 46752 : rel->rd_rel->relkind != RELKIND_MATVIEW &&
2154 [ + + ]: 46747 : rel->rd_rel->relkind != RELKIND_TOASTVALUE &&
2155 [ + + ]: 125 : rel->rd_rel->relkind != RELKIND_PARTITIONED_TABLE)
2156 : : {
2157 [ + - ]: 1 : ereport(WARNING,
2158 : : (errmsg("skipping \"%s\" --- cannot vacuum non-tables or special system tables",
2159 : : RelationGetRelationName(rel))));
2160 : 1 : relation_close(rel, lmode);
2161 : 1 : PopActiveSnapshot();
2162 : 1 : CommitTransactionCommand();
2163 : 1 : return false;
2164 : : }
2165 : :
2166 : : /*
2167 : : * Silently ignore tables that are temp tables of other backends ---
2168 : : * trying to vacuum these will lead to great unhappiness, since their
2169 : : * contents are probably not up-to-date on disk. (We don't throw a
2170 : : * warning here; it would just lead to chatter during a database-wide
2171 : : * VACUUM.)
2172 : : */
2173 [ + + + + ]: 130234 : if (RELATION_IS_OTHER_TEMP(rel))
2174 : : {
2175 : 1 : relation_close(rel, lmode);
2176 : 1 : PopActiveSnapshot();
2177 : 1 : CommitTransactionCommand();
2178 : 1 : return false;
2179 : : }
2180 : :
2181 : : /*
2182 : : * Silently ignore partitioned tables as there is no work to be done. The
2183 : : * useful work is on their child partitions, which have been queued up for
2184 : : * us separately.
2185 : : */
2186 [ + + ]: 130233 : if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
2187 : : {
2188 : 124 : relation_close(rel, lmode);
2189 : 124 : PopActiveSnapshot();
2190 : 124 : CommitTransactionCommand();
2191 : : /* It's OK to proceed with ANALYZE on this table */
2192 : 124 : return true;
2193 : : }
2194 : :
2195 : : /*
2196 : : * Get a session-level lock too. This will protect our access to the
2197 : : * relation across multiple transactions, so that we can vacuum the
2198 : : * relation's TOAST table (if any) secure in the knowledge that no one is
2199 : : * deleting the parent relation.
2200 : : *
2201 : : * NOTE: this cannot block, even if someone else is waiting for access,
2202 : : * because the lock manager knows that both lock requests are from the
2203 : : * same process.
2204 : : */
2205 : 130109 : lockrelid = rel->rd_lockInfo.lockRelId;
2206 : 130109 : LockRelationIdForSession(&lockrelid, lmode);
2207 : :
2208 : : /*
2209 : : * Determine the storage parameters to use. A TOAST table takes any
2210 : : * storage parameter it accepts but does not set from its main table,
2211 : : * whose parameters the caller handed down for that purpose. For anything
2212 : : * else, params.main_relopts is NULL, and this just copies our own.
2213 : : */
2214 : 130109 : relopts = merge_toast_reloptions((StdRdOptions *) rel->rd_options,
2215 : : params.main_relopts);
2216 : :
2217 : : /*
2218 : : * Set index_cleanup option based on index_cleanup reloption if it wasn't
2219 : : * specified in VACUUM command, or when running in an autovacuum worker
2220 : : */
2221 [ + + ]: 130109 : if (params.index_cleanup == VACOPTVALUE_UNSPECIFIED)
2222 : : {
2223 : : StdRdOptIndexCleanup vacuum_index_cleanup;
2224 : :
2225 [ + + ]: 129962 : if (relopts == NULL)
2226 : 127295 : vacuum_index_cleanup = STDRD_OPTION_VACUUM_INDEX_CLEANUP_NOT_SET;
2227 : : else
2228 : 2667 : vacuum_index_cleanup = relopts->vacuum_index_cleanup;
2229 : :
2230 [ + + + + : 129962 : switch (vacuum_index_cleanup)
- ]
2231 : : {
2232 : 18 : case STDRD_OPTION_VACUUM_INDEX_CLEANUP_ON:
2233 : 18 : params.index_cleanup = VACOPTVALUE_ENABLED;
2234 : 18 : break;
2235 : 20 : case STDRD_OPTION_VACUUM_INDEX_CLEANUP_OFF:
2236 : 20 : params.index_cleanup = VACOPTVALUE_DISABLED;
2237 : 20 : break;
2238 : 12 : case STDRD_OPTION_VACUUM_INDEX_CLEANUP_AUTO:
2239 : 12 : params.index_cleanup = VACOPTVALUE_AUTO;
2240 : 12 : break;
2241 : 129912 : case STDRD_OPTION_VACUUM_INDEX_CLEANUP_NOT_SET:
2242 : 129912 : params.index_cleanup = VACOPTVALUE_AUTO;
2243 : 129912 : break;
2244 : : }
2245 : : }
2246 : :
2247 : : #ifdef USE_INJECTION_POINTS
2248 [ + + ]: 130109 : if (params.index_cleanup == VACOPTVALUE_AUTO)
2249 : 129928 : INJECTION_POINT("vacuum-index-cleanup-auto", NULL);
2250 [ + + ]: 181 : else if (params.index_cleanup == VACOPTVALUE_DISABLED)
2251 : 150 : INJECTION_POINT("vacuum-index-cleanup-disabled", NULL);
2252 [ + - ]: 31 : else if (params.index_cleanup == VACOPTVALUE_ENABLED)
2253 : 31 : INJECTION_POINT("vacuum-index-cleanup-enabled", NULL);
2254 : : #endif
2255 : :
2256 : : /*
2257 : : * Check if the vacuum_max_eager_freeze_failure_rate table storage
2258 : : * parameter was specified. This overrides the GUC value.
2259 : : */
2260 [ + + - + ]: 130109 : if (relopts != NULL && relopts->vacuum_max_eager_freeze_failure_rate >= 0)
2261 : 0 : params.max_eager_freeze_failure_rate = relopts->vacuum_max_eager_freeze_failure_rate;
2262 : :
2263 : : /*
2264 : : * Set truncate option based on truncate reloption or GUC if it wasn't
2265 : : * specified in VACUUM command, or when running in an autovacuum worker
2266 : : */
2267 [ + + ]: 130109 : if (params.truncate == VACOPTVALUE_UNSPECIFIED)
2268 : : {
2269 [ + + + + ]: 129970 : if (relopts && relopts->vacuum_truncate != PG_TERNARY_UNSET)
2270 : : {
2271 [ + + ]: 27 : if (relopts->vacuum_truncate == PG_TERNARY_TRUE)
2272 : 11 : params.truncate = VACOPTVALUE_ENABLED;
2273 : : else
2274 : 16 : params.truncate = VACOPTVALUE_DISABLED;
2275 : : }
2276 [ + + ]: 129943 : else if (vacuum_truncate)
2277 : 129933 : params.truncate = VACOPTVALUE_ENABLED;
2278 : : else
2279 : 10 : params.truncate = VACOPTVALUE_DISABLED;
2280 : : }
2281 : :
2282 : : #ifdef USE_INJECTION_POINTS
2283 [ - + ]: 130109 : if (params.truncate == VACOPTVALUE_AUTO)
2284 : 0 : INJECTION_POINT("vacuum-truncate-auto", NULL);
2285 [ + + ]: 130109 : else if (params.truncate == VACOPTVALUE_DISABLED)
2286 : 164 : INJECTION_POINT("vacuum-truncate-disabled", NULL);
2287 [ + - ]: 129945 : else if (params.truncate == VACOPTVALUE_ENABLED)
2288 : 129945 : INJECTION_POINT("vacuum-truncate-enabled", NULL);
2289 : : #endif
2290 : :
2291 : : /*
2292 : : * Remember the relation's TOAST relation for later, if the caller asked
2293 : : * us to process it. In VACUUM FULL, though, the toast table is
2294 : : * automatically rebuilt by cluster_rel so we shouldn't recurse to it,
2295 : : * unless PROCESS_MAIN is disabled.
2296 : : */
2297 [ + + ]: 130109 : if ((params.options & VACOPT_PROCESS_TOAST) != 0 &&
2298 [ + + ]: 16767 : ((params.options & VACOPT_FULL) == 0 ||
2299 [ + + ]: 230 : (params.options & VACOPT_PROCESS_MAIN) == 0))
2300 : 16541 : toast_relid = rel->rd_rel->reltoastrelid;
2301 : : else
2302 : 113568 : toast_relid = InvalidOid;
2303 : :
2304 : : /*
2305 : : * Hand our storage parameters down for the TOAST table to inherit. Take
2306 : : * a copy while we still have the relation open; the relcache entry can go
2307 : : * away once we close it.
2308 : : */
2309 [ + + + + ]: 130109 : if (OidIsValid(toast_relid) && rel->rd_options)
2310 : : {
2311 : 125 : memcpy(&relopts_copy, rel->rd_options, sizeof(StdRdOptions));
2312 : 125 : toast_vacuum_params.main_relopts = &relopts_copy;
2313 : : }
2314 : :
2315 : : /*
2316 : : * Switch to the table owner's userid, so that any index functions are run
2317 : : * as that user. Also lock down security-restricted operations and
2318 : : * arrange to make GUC variable changes local to this command. (This is
2319 : : * unnecessary, but harmless, for lazy VACUUM.)
2320 : : */
2321 : 130109 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
2322 : 130109 : SetUserIdAndSecContext(rel->rd_rel->relowner,
2323 : : save_sec_context | SECURITY_RESTRICTED_OPERATION);
2324 : 130109 : save_nestlevel = NewGUCNestLevel();
2325 : 130109 : RestrictSearchPath();
2326 : :
2327 : : /*
2328 : : * If PROCESS_MAIN is set (the default), it's time to vacuum the main
2329 : : * relation. Otherwise, we can skip this part. If processing the TOAST
2330 : : * table is required (e.g., PROCESS_TOAST is set), we force PROCESS_MAIN
2331 : : * to be set when we recurse to the TOAST table.
2332 : : */
2333 [ + + ]: 130109 : if (params.options & VACOPT_PROCESS_MAIN)
2334 : : {
2335 : : /*
2336 : : * Do the actual work --- either FULL or "lazy" vacuum
2337 : : */
2338 [ + + ]: 130024 : if (params.options & VACOPT_FULL)
2339 : : {
2340 : 226 : ClusterParams cluster_params = {0};
2341 : :
2342 [ + + ]: 226 : if ((params.options & VACOPT_VERBOSE) != 0)
2343 : 1 : cluster_params.options |= CLUOPT_VERBOSE;
2344 : :
2345 : : /* VACUUM FULL is a variant of REPACK; see repack.c */
2346 : 226 : cluster_rel(REPACK_COMMAND_VACUUMFULL, rel, InvalidOid,
2347 : : &cluster_params, isTopLevel);
2348 : : /* cluster_rel closes the relation, but keeps lock */
2349 : :
2350 : 222 : rel = NULL;
2351 : : }
2352 : : else
2353 : 129798 : table_relation_vacuum(rel, ¶ms, bstrategy);
2354 : : }
2355 : :
2356 : : /* Roll back any GUC changes executed by index functions */
2357 : 130103 : AtEOXact_GUC(false, save_nestlevel);
2358 : :
2359 : : /* Restore userid and security context */
2360 : 130103 : SetUserIdAndSecContext(save_userid, save_sec_context);
2361 : :
2362 : : /* all done with this class, but hold lock until commit */
2363 [ + + ]: 130103 : if (rel)
2364 : 129881 : relation_close(rel, NoLock);
2365 : :
2366 : : /*
2367 : : * Complete the transaction and free all temporary memory used.
2368 : : */
2369 : 130103 : PopActiveSnapshot();
2370 : 130103 : CommitTransactionCommand();
2371 : :
2372 : : /*
2373 : : * If the relation has a secondary toast rel, vacuum that too while we
2374 : : * still hold the session lock on the main table. Note however that
2375 : : * "analyze" will not get done on the toast table. This is good, because
2376 : : * the toaster always uses hardcoded index access and statistics are
2377 : : * totally unimportant for toast relations.
2378 : : */
2379 [ + + ]: 130103 : if (toast_relid != InvalidOid)
2380 : : {
2381 : : /*
2382 : : * Force VACOPT_PROCESS_MAIN so vacuum_rel() processes it. Likewise,
2383 : : * set toast_parent so that the privilege checks are done on the main
2384 : : * relation. NB: This is only safe to do because we hold a session
2385 : : * lock on the main relation that prevents concurrent deletion.
2386 : : */
2387 : 5774 : toast_vacuum_params.options |= VACOPT_PROCESS_MAIN;
2388 : 5774 : toast_vacuum_params.toast_parent = relid;
2389 : :
2390 : 5774 : vacuum_rel(toast_relid, NULL, toast_vacuum_params, bstrategy,
2391 : : isTopLevel);
2392 : : }
2393 : :
2394 : : /*
2395 : : * Now release the session-level lock on the main table.
2396 : : */
2397 : 130103 : UnlockRelationIdForSession(&lockrelid, lmode);
2398 : :
2399 : : /* Report that we really did it. */
2400 : 130103 : return true;
2401 : : }
2402 : :
2403 : :
2404 : : /*
2405 : : * Open all the vacuumable indexes of the given relation, obtaining the
2406 : : * specified kind of lock on each. Return an array of Relation pointers for
2407 : : * the indexes into *Irel, and the number of indexes into *nindexes.
2408 : : *
2409 : : * We consider an index vacuumable if it is marked insertable (indisready).
2410 : : * If it isn't, probably a CREATE INDEX CONCURRENTLY command failed early in
2411 : : * execution, and what we have is too corrupt to be processable. We will
2412 : : * vacuum even if the index isn't indisvalid; this is important because in a
2413 : : * unique index, uniqueness checks will be performed anyway and had better not
2414 : : * hit dangling index pointers.
2415 : : */
2416 : : void
2417 : 139908 : vac_open_indexes(Relation relation, LOCKMODE lockmode,
2418 : : int *nindexes, Relation **Irel)
2419 : : {
2420 : : List *indexoidlist;
2421 : : ListCell *indexoidscan;
2422 : : int i;
2423 : :
2424 : : Assert(lockmode != NoLock);
2425 : :
2426 : 139908 : indexoidlist = RelationGetIndexList(relation);
2427 : :
2428 : : /* allocate enough memory for all indexes */
2429 : 139908 : i = list_length(indexoidlist);
2430 : :
2431 [ + + ]: 139908 : if (i > 0)
2432 : 131623 : *Irel = palloc_array(Relation, i);
2433 : : else
2434 : 8285 : *Irel = NULL;
2435 : :
2436 : : /* collect just the ready indexes */
2437 : 139908 : i = 0;
2438 [ + + + + : 351667 : foreach(indexoidscan, indexoidlist)
+ + ]
2439 : : {
2440 : 211759 : Oid indexoid = lfirst_oid(indexoidscan);
2441 : : Relation indrel;
2442 : :
2443 : 211759 : indrel = index_open(indexoid, lockmode);
2444 [ + - ]: 211759 : if (indrel->rd_index->indisready)
2445 : 211759 : (*Irel)[i++] = indrel;
2446 : : else
2447 : 0 : index_close(indrel, lockmode);
2448 : : }
2449 : :
2450 : 139908 : *nindexes = i;
2451 : :
2452 : 139908 : list_free(indexoidlist);
2453 : 139908 : }
2454 : :
2455 : : /*
2456 : : * Release the resources acquired by vac_open_indexes. Optionally release
2457 : : * the locks (say NoLock to keep 'em).
2458 : : */
2459 : : void
2460 : 140506 : vac_close_indexes(int nindexes, Relation *Irel, LOCKMODE lockmode)
2461 : : {
2462 [ + + ]: 140506 : if (Irel == NULL)
2463 : 8889 : return;
2464 : :
2465 [ + + ]: 343363 : while (nindexes--)
2466 : : {
2467 : 211746 : Relation ind = Irel[nindexes];
2468 : :
2469 : 211746 : index_close(ind, lockmode);
2470 : : }
2471 : 131617 : pfree(Irel);
2472 : : }
2473 : :
2474 : : /*
2475 : : * vacuum_delay_point --- check for interrupts and cost-based delay.
2476 : : *
2477 : : * This should be called in each major loop of VACUUM processing,
2478 : : * typically once per page processed.
2479 : : */
2480 : : void
2481 : 56338067 : vacuum_delay_point(bool is_analyze)
2482 : : {
2483 : 56338067 : double msec = 0;
2484 : :
2485 : : /* Always check for interrupts */
2486 [ + + ]: 56338067 : CHECK_FOR_INTERRUPTS();
2487 : :
2488 [ - + ]: 56338067 : if (InterruptPending)
2489 : 0 : return;
2490 : :
2491 [ + + ]: 56338067 : if (IsParallelWorker())
2492 : : {
2493 : : /*
2494 : : * Update cost-based vacuum delay parameters for a parallel autovacuum
2495 : : * worker if any changes are detected. It might enable cost-based
2496 : : * delay so it needs to be called before VacuumCostActive check.
2497 : : */
2498 : 313 : parallel_vacuum_update_shared_delay_params();
2499 : : }
2500 : :
2501 [ + + + - ]: 56338067 : if (!VacuumCostActive && !ConfigReloadPending)
2502 : 47464783 : return;
2503 : :
2504 : : /*
2505 : : * Autovacuum workers should reload the configuration file if requested.
2506 : : * This allows changes to [autovacuum_]vacuum_cost_limit and
2507 : : * [autovacuum_]vacuum_cost_delay to take effect while a table is being
2508 : : * vacuumed or analyzed.
2509 : : */
2510 [ + + + - ]: 8873284 : if (ConfigReloadPending && AmAutoVacuumWorkerProcess())
2511 : : {
2512 : 1 : ConfigReloadPending = false;
2513 : 1 : ProcessConfigFile(PGC_SIGHUP);
2514 : 1 : VacuumUpdateCosts();
2515 : :
2516 : : /*
2517 : : * Propagate cost-based vacuum delay parameters to shared memory if
2518 : : * any of them have changed during the config reload.
2519 : : */
2520 : 1 : parallel_vacuum_propagate_shared_delay_params();
2521 : : }
2522 : :
2523 : : /*
2524 : : * If we disabled cost-based delays after reloading the config file,
2525 : : * return.
2526 : : */
2527 [ - + ]: 8873284 : if (!VacuumCostActive)
2528 : 0 : return;
2529 : :
2530 : : /*
2531 : : * For parallel vacuum, the delay is computed based on the shared cost
2532 : : * balance. See compute_parallel_delay.
2533 : : */
2534 [ + + ]: 8873284 : if (VacuumSharedCostBalance != NULL)
2535 : 644 : msec = compute_parallel_delay();
2536 [ + + ]: 8872640 : else if (VacuumCostBalance >= vacuum_cost_limit)
2537 : 3660 : msec = vacuum_cost_delay * VacuumCostBalance / vacuum_cost_limit;
2538 : :
2539 : : /* Nap if appropriate */
2540 [ + + ]: 8873284 : if (msec > 0)
2541 : : {
2542 : : instr_time delay_start;
2543 : :
2544 [ + + ]: 3702 : if (msec > vacuum_cost_delay * 4)
2545 : 9 : msec = vacuum_cost_delay * 4;
2546 : :
2547 [ - + ]: 3702 : if (track_cost_delay_timing)
2548 : 0 : INSTR_TIME_SET_CURRENT(delay_start);
2549 : :
2550 : 3702 : pgstat_report_wait_start(WAIT_EVENT_VACUUM_DELAY);
2551 : 3702 : pg_usleep(msec * 1000);
2552 : 3702 : pgstat_report_wait_end();
2553 : :
2554 [ - + ]: 3702 : if (track_cost_delay_timing)
2555 : : {
2556 : : instr_time delay_end;
2557 : : instr_time delay;
2558 : :
2559 : 0 : INSTR_TIME_SET_CURRENT(delay_end);
2560 : 0 : INSTR_TIME_SET_ZERO(delay);
2561 : 0 : INSTR_TIME_ACCUM_DIFF(delay, delay_end, delay_start);
2562 : :
2563 : : /*
2564 : : * For parallel workers, we only report the delay time every once
2565 : : * in a while to avoid overloading the leader with messages and
2566 : : * interrupts.
2567 : : */
2568 [ # # ]: 0 : if (IsParallelWorker())
2569 : : {
2570 : : static instr_time last_report_time;
2571 : : instr_time time_since_last_report;
2572 : :
2573 : : Assert(!is_analyze);
2574 : :
2575 : : /* Accumulate the delay time */
2576 : 0 : parallel_vacuum_worker_delay_ns += INSTR_TIME_GET_NANOSEC(delay);
2577 : :
2578 : : /* Calculate interval since last report */
2579 : 0 : INSTR_TIME_SET_ZERO(time_since_last_report);
2580 : 0 : INSTR_TIME_ACCUM_DIFF(time_since_last_report, delay_end, last_report_time);
2581 : :
2582 : : /* If we haven't reported in a while, do so now */
2583 [ # # ]: 0 : if (INSTR_TIME_GET_NANOSEC(time_since_last_report) >=
2584 : : PARALLEL_VACUUM_DELAY_REPORT_INTERVAL_NS)
2585 : : {
2586 : 0 : pgstat_progress_parallel_incr_param(PROGRESS_VACUUM_DELAY_TIME,
2587 : : parallel_vacuum_worker_delay_ns);
2588 : :
2589 : : /* Reset variables */
2590 : 0 : last_report_time = delay_end;
2591 : 0 : parallel_vacuum_worker_delay_ns = 0;
2592 : : }
2593 : : }
2594 [ # # ]: 0 : else if (is_analyze)
2595 : 0 : pgstat_progress_incr_param(PROGRESS_ANALYZE_DELAY_TIME,
2596 : : INSTR_TIME_GET_NANOSEC(delay));
2597 : : else
2598 : 0 : pgstat_progress_incr_param(PROGRESS_VACUUM_DELAY_TIME,
2599 : : INSTR_TIME_GET_NANOSEC(delay));
2600 : : }
2601 : :
2602 : : /*
2603 : : * We don't want to ignore postmaster death during very long vacuums
2604 : : * with vacuum_cost_delay configured. We can't use the usual
2605 : : * WaitLatch() approach here because we want microsecond-based sleep
2606 : : * durations above.
2607 : : */
2608 [ + - - + ]: 3702 : if (IsUnderPostmaster && !PostmasterIsAlive())
2609 : 0 : exit(1);
2610 : :
2611 : 3702 : VacuumCostBalance = 0;
2612 : :
2613 : : /*
2614 : : * Balance and update limit values for autovacuum workers. We must do
2615 : : * this periodically, as the number of workers across which we are
2616 : : * balancing the limit may have changed.
2617 : : *
2618 : : * TODO: There may be better criteria for determining when to do this
2619 : : * besides "check after napping".
2620 : : */
2621 : 3702 : AutoVacuumUpdateCostLimit();
2622 : :
2623 [ + + ]: 3702 : if (AmAutoVacuumWorkerProcess())
2624 : 3678 : parallel_vacuum_propagate_shared_delay_params();
2625 : :
2626 : : /* Might have gotten an interrupt while sleeping */
2627 [ + + ]: 3702 : CHECK_FOR_INTERRUPTS();
2628 : : }
2629 : : }
2630 : :
2631 : : /*
2632 : : * Computes the vacuum delay for parallel workers.
2633 : : *
2634 : : * The basic idea of a cost-based delay for parallel vacuum is to allow each
2635 : : * worker to sleep in proportion to the share of work it's done. We achieve this
2636 : : * by allowing all parallel vacuum workers including the leader process to
2637 : : * have a shared view of cost related parameters (mainly VacuumCostBalance).
2638 : : * We allow each worker to update it as and when it has incurred any cost and
2639 : : * then based on that decide whether it needs to sleep. We compute the time
2640 : : * to sleep for a worker based on the cost it has incurred
2641 : : * (VacuumCostBalanceLocal) and then reduce the VacuumSharedCostBalance by
2642 : : * that amount. This avoids putting to sleep those workers which have done less
2643 : : * I/O than other workers and therefore ensure that workers
2644 : : * which are doing more I/O got throttled more.
2645 : : *
2646 : : * We allow a worker to sleep only if it has performed I/O above a certain
2647 : : * threshold, which is calculated based on the number of active workers
2648 : : * (VacuumActiveNWorkers), and the overall cost balance is more than
2649 : : * VacuumCostLimit set by the system. Testing reveals that we achieve
2650 : : * the required throttling if we force a worker that has done more than 50%
2651 : : * of its share of work to sleep.
2652 : : */
2653 : : static double
2654 : 644 : compute_parallel_delay(void)
2655 : : {
2656 : 644 : double msec = 0;
2657 : : uint32 shared_balance;
2658 : : int nworkers;
2659 : :
2660 : : /* Parallel vacuum must be active */
2661 : : Assert(VacuumSharedCostBalance);
2662 : :
2663 : 644 : nworkers = pg_atomic_read_u32(VacuumActiveNWorkers);
2664 : :
2665 : : /* At least count itself */
2666 : : Assert(nworkers >= 1);
2667 : :
2668 : : /* Update the shared cost balance value atomically */
2669 : 644 : shared_balance = pg_atomic_add_fetch_u32(VacuumSharedCostBalance, VacuumCostBalance);
2670 : :
2671 : : /* Compute the total local balance for the current worker */
2672 : 644 : VacuumCostBalanceLocal += VacuumCostBalance;
2673 : :
2674 [ + + ]: 644 : if ((shared_balance >= vacuum_cost_limit) &&
2675 [ + + ]: 120 : (VacuumCostBalanceLocal > 0.5 * ((double) vacuum_cost_limit / nworkers)))
2676 : : {
2677 : : /* Compute sleep time based on the local cost balance */
2678 : 42 : msec = vacuum_cost_delay * VacuumCostBalanceLocal / vacuum_cost_limit;
2679 : 42 : pg_atomic_sub_fetch_u32(VacuumSharedCostBalance, VacuumCostBalanceLocal);
2680 : 42 : VacuumCostBalanceLocal = 0;
2681 : : }
2682 : :
2683 : : /*
2684 : : * Reset the local balance as we accumulated it into the shared value.
2685 : : */
2686 : 644 : VacuumCostBalance = 0;
2687 : :
2688 : 644 : return msec;
2689 : : }
2690 : :
2691 : : /*
2692 : : * A wrapper function of defGetBoolean().
2693 : : *
2694 : : * This function returns VACOPTVALUE_ENABLED and VACOPTVALUE_DISABLED instead
2695 : : * of true and false.
2696 : : */
2697 : : static VacOptValue
2698 : 184 : get_vacoptval_from_boolean(DefElem *def)
2699 : : {
2700 [ + + ]: 184 : return defGetBoolean(def) ? VACOPTVALUE_ENABLED : VACOPTVALUE_DISABLED;
2701 : : }
2702 : :
2703 : : /*
2704 : : * vac_bulkdel_one_index() -- bulk-deletion for index relation.
2705 : : *
2706 : : * Returns bulk delete stats derived from input stats
2707 : : */
2708 : : IndexBulkDeleteResult *
2709 : 1671 : vac_bulkdel_one_index(IndexVacuumInfo *ivinfo, IndexBulkDeleteResult *istat,
2710 : : TidStore *dead_items, VacDeadItemsInfo *dead_items_info)
2711 : : {
2712 : : /* Do bulk deletion */
2713 : 1671 : istat = index_bulk_delete(ivinfo, istat, vac_tid_reaped,
2714 : : dead_items);
2715 : :
2716 [ + + ]: 1670 : ereport(ivinfo->message_level,
2717 : : (errmsg("scanned index \"%s\" to remove %" PRId64 " row versions",
2718 : : RelationGetRelationName(ivinfo->index),
2719 : : dead_items_info->num_items)));
2720 : :
2721 : 1670 : return istat;
2722 : : }
2723 : :
2724 : : /*
2725 : : * vac_cleanup_one_index() -- do post-vacuum cleanup for index relation.
2726 : : *
2727 : : * Returns bulk delete stats derived from input stats
2728 : : */
2729 : : IndexBulkDeleteResult *
2730 : 153977 : vac_cleanup_one_index(IndexVacuumInfo *ivinfo, IndexBulkDeleteResult *istat)
2731 : : {
2732 : 153977 : istat = index_vacuum_cleanup(ivinfo, istat);
2733 : :
2734 [ + + ]: 153977 : if (istat)
2735 [ + + ]: 1856 : ereport(ivinfo->message_level,
2736 : : (errmsg("index \"%s\" now contains %.0f row versions in %u pages",
2737 : : RelationGetRelationName(ivinfo->index),
2738 : : istat->num_index_tuples,
2739 : : istat->num_pages),
2740 : : errdetail("%.0f index row versions were removed.\n"
2741 : : "%u index pages were newly deleted.\n"
2742 : : "%u index pages are currently deleted, of which %u are currently reusable.",
2743 : : istat->tuples_removed,
2744 : : istat->pages_newly_deleted,
2745 : : istat->pages_deleted, istat->pages_free)));
2746 : :
2747 : 153977 : return istat;
2748 : : }
2749 : :
2750 : : /*
2751 : : * vac_tid_reaped() -- is a particular tid deletable?
2752 : : *
2753 : : * This has the right signature to be an IndexBulkDeleteCallback.
2754 : : */
2755 : : static bool
2756 : 5085711 : vac_tid_reaped(ItemPointer itemptr, void *state)
2757 : : {
2758 : 5085711 : TidStore *dead_items = (TidStore *) state;
2759 : :
2760 : 5085711 : return TidStoreIsMember(dead_items, itemptr);
2761 : : }
|
