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 : : * CLUSTER, handled in cluster.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/tableam.h"
35 : : #include "access/transam.h"
36 : : #include "access/xact.h"
37 : : #include "catalog/namespace.h"
38 : : #include "catalog/pg_database.h"
39 : : #include "catalog/pg_inherits.h"
40 : : #include "commands/async.h"
41 : : #include "commands/cluster.h"
42 : : #include "commands/defrem.h"
43 : : #include "commands/progress.h"
44 : : #include "commands/vacuum.h"
45 : : #include "miscadmin.h"
46 : : #include "nodes/makefuncs.h"
47 : : #include "pgstat.h"
48 : : #include "postmaster/autovacuum.h"
49 : : #include "postmaster/bgworker_internals.h"
50 : : #include "postmaster/interrupt.h"
51 : : #include "storage/bufmgr.h"
52 : : #include "storage/lmgr.h"
53 : : #include "storage/pmsignal.h"
54 : : #include "storage/proc.h"
55 : : #include "storage/procarray.h"
56 : : #include "utils/acl.h"
57 : : #include "utils/fmgroids.h"
58 : : #include "utils/guc.h"
59 : : #include "utils/guc_hooks.h"
60 : : #include "utils/injection_point.h"
61 : : #include "utils/memutils.h"
62 : : #include "utils/snapmgr.h"
63 : : #include "utils/syscache.h"
64 : :
65 : : /*
66 : : * Minimum interval for cost-based vacuum delay reports from a parallel worker.
67 : : * This aims to avoid sending too many messages and waking up the leader too
68 : : * frequently.
69 : : */
70 : : #define PARALLEL_VACUUM_DELAY_REPORT_INTERVAL_NS (NS_PER_S)
71 : :
72 : : /*
73 : : * GUC parameters
74 : : */
75 : : int vacuum_freeze_min_age;
76 : : int vacuum_freeze_table_age;
77 : : int vacuum_multixact_freeze_min_age;
78 : : int vacuum_multixact_freeze_table_age;
79 : : int vacuum_failsafe_age;
80 : : int vacuum_multixact_failsafe_age;
81 : : double vacuum_max_eager_freeze_failure_rate;
82 : : bool track_cost_delay_timing;
83 : : bool vacuum_truncate;
84 : :
85 : : /*
86 : : * Variables for cost-based vacuum delay. The defaults differ between
87 : : * autovacuum and vacuum. They should be set with the appropriate GUC value in
88 : : * vacuum code. They are initialized here to the defaults for client backends
89 : : * executing VACUUM or ANALYZE.
90 : : */
91 : : double vacuum_cost_delay = 0;
92 : : int vacuum_cost_limit = 200;
93 : :
94 : : /* Variable for reporting cost-based vacuum delay from parallel workers. */
95 : : int64 parallel_vacuum_worker_delay_ns = 0;
96 : :
97 : : /*
98 : : * VacuumFailsafeActive is a defined as a global so that we can determine
99 : : * whether or not to re-enable cost-based vacuum delay when vacuuming a table.
100 : : * If failsafe mode has been engaged, we will not re-enable cost-based delay
101 : : * for the table until after vacuuming has completed, regardless of other
102 : : * settings.
103 : : *
104 : : * Only VACUUM code should inspect this variable and only table access methods
105 : : * should set it to true. In Table AM-agnostic VACUUM code, this variable is
106 : : * inspected to determine whether or not to allow cost-based delays. Table AMs
107 : : * are free to set it if they desire this behavior, but it is false by default
108 : : * and reset to false in between vacuuming each relation.
109 : : */
110 : : bool VacuumFailsafeActive = false;
111 : :
112 : : /*
113 : : * Variables for cost-based parallel vacuum. See comments atop
114 : : * compute_parallel_delay to understand how it works.
115 : : */
116 : : pg_atomic_uint32 *VacuumSharedCostBalance = NULL;
117 : : pg_atomic_uint32 *VacuumActiveNWorkers = NULL;
118 : : int VacuumCostBalanceLocal = 0;
119 : :
120 : : /* non-export function prototypes */
121 : : static List *expand_vacuum_rel(VacuumRelation *vrel,
122 : : MemoryContext vac_context, int options);
123 : : static List *get_all_vacuum_rels(MemoryContext vac_context, int options);
124 : : static void vac_truncate_clog(TransactionId frozenXID,
125 : : MultiXactId minMulti,
126 : : TransactionId lastSaneFrozenXid,
127 : : MultiXactId lastSaneMinMulti);
128 : : static bool vacuum_rel(Oid relid, RangeVar *relation, VacuumParams params,
129 : : BufferAccessStrategy bstrategy);
130 : : static double compute_parallel_delay(void);
131 : : static VacOptValue get_vacoptval_from_boolean(DefElem *def);
132 : : static bool vac_tid_reaped(ItemPointer itemptr, void *state);
133 : :
134 : : /*
135 : : * GUC check function to ensure GUC value specified is within the allowable
136 : : * range.
137 : : */
138 : : bool
139 : 6 : check_vacuum_buffer_usage_limit(int *newval, void **extra,
140 : : GucSource source)
141 : : {
142 : : /* Value upper and lower hard limits are inclusive */
143 [ + - + - : 6 : if (*newval == 0 || (*newval >= MIN_BAS_VAC_RING_SIZE_KB &&
+ - ]
144 : 6 : *newval <= MAX_BAS_VAC_RING_SIZE_KB))
145 : 6 : return true;
146 : :
147 : : /* Value does not fall within any allowable range */
148 : 0 : GUC_check_errdetail("\"%s\" must be 0 or between %d kB and %d kB.",
149 : : "vacuum_buffer_usage_limit",
150 : : MIN_BAS_VAC_RING_SIZE_KB, MAX_BAS_VAC_RING_SIZE_KB);
151 : :
152 : 0 : return false;
153 : 6 : }
154 : :
155 : : /*
156 : : * Primary entry point for manual VACUUM and ANALYZE commands
157 : : *
158 : : * This is mainly a preparation wrapper for the real operations that will
159 : : * happen in vacuum().
160 : : */
161 : : void
162 : 541 : ExecVacuum(ParseState *pstate, VacuumStmt *vacstmt, bool isTopLevel)
163 : : {
164 : 541 : VacuumParams params;
165 : 541 : BufferAccessStrategy bstrategy = NULL;
166 : 541 : bool verbose = false;
167 : 541 : bool skip_locked = false;
168 : 541 : bool analyze = false;
169 : 541 : bool freeze = false;
170 : 541 : bool full = false;
171 : 541 : bool disable_page_skipping = false;
172 : 541 : bool process_main = true;
173 : 541 : bool process_toast = true;
174 : 541 : int ring_size;
175 : 541 : bool skip_database_stats = false;
176 : 541 : bool only_database_stats = false;
177 : 541 : MemoryContext vac_context;
178 : 541 : ListCell *lc;
179 : :
180 : : /* index_cleanup and truncate values unspecified for now */
181 : 541 : params.index_cleanup = VACOPTVALUE_UNSPECIFIED;
182 : 541 : params.truncate = VACOPTVALUE_UNSPECIFIED;
183 : :
184 : : /* By default parallel vacuum is enabled */
185 : 541 : params.nworkers = 0;
186 : :
187 : : /* Will be set later if we recurse to a TOAST table. */
188 : 541 : params.toast_parent = InvalidOid;
189 : :
190 : : /*
191 : : * Set this to an invalid value so it is clear whether or not a
192 : : * BUFFER_USAGE_LIMIT was specified when making the access strategy.
193 : : */
194 : 541 : ring_size = -1;
195 : :
196 : : /* Parse options list */
197 [ + + + + : 710 : foreach(lc, vacstmt->options)
+ + ]
198 : : {
199 : 174 : DefElem *opt = (DefElem *) lfirst(lc);
200 : :
201 : : /* Parse common options for VACUUM and ANALYZE */
202 [ + + ]: 174 : if (strcmp(opt->defname, "verbose") == 0)
203 : 3 : verbose = defGetBoolean(opt);
204 [ + + ]: 171 : else if (strcmp(opt->defname, "skip_locked") == 0)
205 : 5 : skip_locked = defGetBoolean(opt);
206 [ + + ]: 166 : else if (strcmp(opt->defname, "buffer_usage_limit") == 0)
207 : : {
208 : 9 : const char *hintmsg;
209 : 9 : int result;
210 : 9 : char *vac_buffer_size;
211 : :
212 : 9 : vac_buffer_size = defGetString(opt);
213 : :
214 : : /*
215 : : * Check that the specified value is valid and the size falls
216 : : * within the hard upper and lower limits if it is not 0.
217 : : */
218 [ + + ]: 14 : if (!parse_int(vac_buffer_size, &result, GUC_UNIT_KB, &hintmsg) ||
219 [ + + ]: 6 : (result != 0 &&
220 : 5 : (result < MIN_BAS_VAC_RING_SIZE_KB || result > MAX_BAS_VAC_RING_SIZE_KB)))
221 : : {
222 [ + - + - : 3 : ereport(ERROR,
+ + ]
223 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
224 : : errmsg("%s option must be 0 or between %d kB and %d kB",
225 : : "BUFFER_USAGE_LIMIT",
226 : : MIN_BAS_VAC_RING_SIZE_KB, MAX_BAS_VAC_RING_SIZE_KB),
227 : : hintmsg ? errhint_internal("%s", _(hintmsg)) : 0));
228 : 0 : }
229 : :
230 : 6 : ring_size = result;
231 : 6 : }
232 [ + + ]: 157 : else if (!vacstmt->is_vacuumcmd)
233 [ + - + - ]: 1 : ereport(ERROR,
234 : : (errcode(ERRCODE_SYNTAX_ERROR),
235 : : errmsg("unrecognized %s option \"%s\"",
236 : : "ANALYZE", opt->defname),
237 : : parser_errposition(pstate, opt->location)));
238 : :
239 : : /* Parse options available on VACUUM */
240 [ + + ]: 156 : else if (strcmp(opt->defname, "analyze") == 0)
241 : 76 : analyze = defGetBoolean(opt);
242 [ + + ]: 80 : else if (strcmp(opt->defname, "freeze") == 0)
243 : 12 : freeze = defGetBoolean(opt);
244 [ + + ]: 68 : else if (strcmp(opt->defname, "full") == 0)
245 : 29 : full = defGetBoolean(opt);
246 [ + + ]: 39 : else if (strcmp(opt->defname, "disable_page_skipping") == 0)
247 : 8 : disable_page_skipping = defGetBoolean(opt);
248 [ + + ]: 31 : else if (strcmp(opt->defname, "index_cleanup") == 0)
249 : : {
250 : : /* Interpret no string as the default, which is 'auto' */
251 [ + - ]: 6 : if (!opt->arg)
252 : 0 : params.index_cleanup = VACOPTVALUE_AUTO;
253 : : else
254 : : {
255 : 6 : char *sval = defGetString(opt);
256 : :
257 : : /* Try matching on 'auto' string, or fall back on boolean */
258 [ + + ]: 6 : if (pg_strcasecmp(sval, "auto") == 0)
259 : 1 : params.index_cleanup = VACOPTVALUE_AUTO;
260 : : else
261 : 5 : params.index_cleanup = get_vacoptval_from_boolean(opt);
262 : 6 : }
263 : 6 : }
264 [ + + ]: 25 : else if (strcmp(opt->defname, "process_main") == 0)
265 : 3 : process_main = defGetBoolean(opt);
266 [ + + ]: 22 : else if (strcmp(opt->defname, "process_toast") == 0)
267 : 4 : process_toast = defGetBoolean(opt);
268 [ + + ]: 18 : else if (strcmp(opt->defname, "truncate") == 0)
269 : 3 : params.truncate = get_vacoptval_from_boolean(opt);
270 [ + + ]: 15 : else if (strcmp(opt->defname, "parallel") == 0)
271 : : {
272 : 12 : int nworkers = defGetInt32(opt);
273 : :
274 [ + + ]: 12 : if (nworkers < 0 || nworkers > MAX_PARALLEL_WORKER_LIMIT)
275 [ + - + - ]: 1 : ereport(ERROR,
276 : : (errcode(ERRCODE_SYNTAX_ERROR),
277 : : errmsg("%s option must be between 0 and %d",
278 : : "PARALLEL",
279 : : MAX_PARALLEL_WORKER_LIMIT),
280 : : parser_errposition(pstate, opt->location)));
281 : :
282 : : /*
283 : : * Disable parallel vacuum, if user has specified parallel degree
284 : : * as zero.
285 : : */
286 [ + + ]: 11 : if (nworkers == 0)
287 : 3 : params.nworkers = -1;
288 : : else
289 : 8 : params.nworkers = nworkers;
290 : 11 : }
291 [ + + ]: 3 : else if (strcmp(opt->defname, "skip_database_stats") == 0)
292 : 1 : skip_database_stats = defGetBoolean(opt);
293 [ + - ]: 2 : else if (strcmp(opt->defname, "only_database_stats") == 0)
294 : 2 : only_database_stats = defGetBoolean(opt);
295 : : else
296 [ # # # # ]: 0 : ereport(ERROR,
297 : : (errcode(ERRCODE_SYNTAX_ERROR),
298 : : errmsg("unrecognized %s option \"%s\"",
299 : : "VACUUM", opt->defname),
300 : : parser_errposition(pstate, opt->location)));
301 : 169 : }
302 : :
303 : : /* Set vacuum options */
304 : 536 : params.options =
305 : 1072 : (vacstmt->is_vacuumcmd ? VACOPT_VACUUM : VACOPT_ANALYZE) |
306 : 1072 : (verbose ? VACOPT_VERBOSE : 0) |
307 : 1072 : (skip_locked ? VACOPT_SKIP_LOCKED : 0) |
308 : 1072 : (analyze ? VACOPT_ANALYZE : 0) |
309 : 1072 : (freeze ? VACOPT_FREEZE : 0) |
310 : 1072 : (full ? VACOPT_FULL : 0) |
311 : 1072 : (disable_page_skipping ? VACOPT_DISABLE_PAGE_SKIPPING : 0) |
312 : 1072 : (process_main ? VACOPT_PROCESS_MAIN : 0) |
313 : 1072 : (process_toast ? VACOPT_PROCESS_TOAST : 0) |
314 : 1072 : (skip_database_stats ? VACOPT_SKIP_DATABASE_STATS : 0) |
315 : 536 : (only_database_stats ? VACOPT_ONLY_DATABASE_STATS : 0);
316 : :
317 : : /* sanity checks on options */
318 [ + - ]: 536 : Assert(params.options & (VACOPT_VACUUM | VACOPT_ANALYZE));
319 [ + + + - ]: 536 : Assert((params.options & VACOPT_VACUUM) ||
320 : : !(params.options & (VACOPT_FULL | VACOPT_FREEZE)));
321 : :
322 [ + + + + ]: 536 : if ((params.options & VACOPT_FULL) && params.nworkers > 0)
323 [ + - + - ]: 1 : ereport(ERROR,
324 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
325 : : errmsg("VACUUM FULL cannot be performed in parallel")));
326 : :
327 : : /*
328 : : * BUFFER_USAGE_LIMIT does nothing for VACUUM (FULL) so just raise an
329 : : * ERROR for that case. VACUUM (FULL, ANALYZE) does make use of it, so
330 : : * we'll permit that.
331 : : */
332 [ + + + + : 535 : if (ring_size != -1 && (params.options & VACOPT_FULL) &&
- + ]
333 : 1 : !(params.options & VACOPT_ANALYZE))
334 [ + - + - ]: 1 : ereport(ERROR,
335 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
336 : : errmsg("BUFFER_USAGE_LIMIT cannot be specified for VACUUM FULL")));
337 : :
338 : : /*
339 : : * Make sure VACOPT_ANALYZE is specified if any column lists are present.
340 : : */
341 [ + + ]: 534 : if (!(params.options & VACOPT_ANALYZE))
342 : : {
343 [ + + + + : 247 : foreach(lc, vacstmt->rels)
+ + ]
344 : : {
345 : 125 : VacuumRelation *vrel = lfirst_node(VacuumRelation, lc);
346 : :
347 [ + + ]: 125 : if (vrel->va_cols != NIL)
348 [ + - + - ]: 1 : ereport(ERROR,
349 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
350 : : errmsg("ANALYZE option must be specified when a column list is provided")));
351 : 124 : }
352 : 122 : }
353 : :
354 : :
355 : : /*
356 : : * Sanity check DISABLE_PAGE_SKIPPING option.
357 : : */
358 [ + + + - ]: 533 : if ((params.options & VACOPT_FULL) != 0 &&
359 : 25 : (params.options & VACOPT_DISABLE_PAGE_SKIPPING) != 0)
360 [ # # # # ]: 0 : ereport(ERROR,
361 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
362 : : errmsg("VACUUM option DISABLE_PAGE_SKIPPING cannot be used with FULL")));
363 : :
364 : : /* sanity check for PROCESS_TOAST */
365 [ + + + + ]: 533 : if ((params.options & VACOPT_FULL) != 0 &&
366 : 25 : (params.options & VACOPT_PROCESS_TOAST) == 0)
367 [ + - + - ]: 1 : ereport(ERROR,
368 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
369 : : errmsg("PROCESS_TOAST required with VACUUM FULL")));
370 : :
371 : : /* sanity check for ONLY_DATABASE_STATS */
372 [ + + ]: 532 : if (params.options & VACOPT_ONLY_DATABASE_STATS)
373 : : {
374 [ + - ]: 2 : Assert(params.options & VACOPT_VACUUM);
375 [ + + ]: 2 : if (vacstmt->rels != NIL)
376 [ + - + - ]: 1 : ereport(ERROR,
377 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
378 : : errmsg("ONLY_DATABASE_STATS cannot be specified with a list of tables")));
379 : : /* don't require people to turn off PROCESS_TOAST/MAIN explicitly */
380 [ + - ]: 1 : if (params.options & ~(VACOPT_VACUUM |
381 : : VACOPT_VERBOSE |
382 : : VACOPT_PROCESS_MAIN |
383 : : VACOPT_PROCESS_TOAST |
384 : : VACOPT_ONLY_DATABASE_STATS))
385 [ # # # # ]: 0 : ereport(ERROR,
386 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
387 : : errmsg("ONLY_DATABASE_STATS cannot be specified with other VACUUM options")));
388 : 1 : }
389 : :
390 : : /*
391 : : * All freeze ages are zero if the FREEZE option is given; otherwise pass
392 : : * them as -1 which means to use the default values.
393 : : */
394 [ + + ]: 531 : if (params.options & VACOPT_FREEZE)
395 : : {
396 : 12 : params.freeze_min_age = 0;
397 : 12 : params.freeze_table_age = 0;
398 : 12 : params.multixact_freeze_min_age = 0;
399 : 12 : params.multixact_freeze_table_age = 0;
400 : 12 : }
401 : : else
402 : : {
403 : 519 : params.freeze_min_age = -1;
404 : 519 : params.freeze_table_age = -1;
405 : 519 : params.multixact_freeze_min_age = -1;
406 : 519 : params.multixact_freeze_table_age = -1;
407 : : }
408 : :
409 : : /* user-invoked vacuum is never "for wraparound" */
410 : 531 : params.is_wraparound = false;
411 : :
412 : : /*
413 : : * user-invoked vacuum uses VACOPT_VERBOSE instead of
414 : : * log_vacuum_min_duration and log_analyze_min_duration
415 : : */
416 : 531 : params.log_vacuum_min_duration = -1;
417 : 531 : params.log_analyze_min_duration = -1;
418 : :
419 : : /*
420 : : * Later, in vacuum_rel(), we check if a reloption override was specified.
421 : : */
422 : 531 : params.max_eager_freeze_failure_rate = vacuum_max_eager_freeze_failure_rate;
423 : :
424 : : /*
425 : : * Create special memory context for cross-transaction storage.
426 : : *
427 : : * Since it is a child of PortalContext, it will go away eventually even
428 : : * if we suffer an error; there's no need for special abort cleanup logic.
429 : : */
430 : 531 : vac_context = AllocSetContextCreate(PortalContext,
431 : : "Vacuum",
432 : : ALLOCSET_DEFAULT_SIZES);
433 : :
434 : : /*
435 : : * Make a buffer strategy object in the cross-transaction memory context.
436 : : * We needn't bother making this for VACUUM (FULL) or VACUUM
437 : : * (ONLY_DATABASE_STATS) as they'll not make use of it. VACUUM (FULL,
438 : : * ANALYZE) is possible, so we'd better ensure that we make a strategy
439 : : * when we see ANALYZE.
440 : : */
441 : 531 : if ((params.options & (VACOPT_ONLY_DATABASE_STATS |
442 [ + + + + ]: 531 : VACOPT_FULL)) == 0 ||
443 : 25 : (params.options & VACOPT_ANALYZE) != 0)
444 : : {
445 : :
446 : 507 : MemoryContext old_context = MemoryContextSwitchTo(vac_context);
447 : :
448 [ + - ]: 507 : Assert(ring_size >= -1);
449 : :
450 : : /*
451 : : * If BUFFER_USAGE_LIMIT was specified by the VACUUM or ANALYZE
452 : : * command, it overrides the value of VacuumBufferUsageLimit. Either
453 : : * value may be 0, in which case GetAccessStrategyWithSize() will
454 : : * return NULL, effectively allowing full use of shared buffers.
455 : : */
456 [ + + ]: 507 : if (ring_size == -1)
457 : 502 : ring_size = VacuumBufferUsageLimit;
458 : :
459 : 507 : bstrategy = GetAccessStrategyWithSize(BAS_VACUUM, ring_size);
460 : :
461 : 507 : MemoryContextSwitchTo(old_context);
462 : 507 : }
463 : :
464 : : /* Now go through the common routine */
465 : 531 : vacuum(vacstmt->rels, params, bstrategy, vac_context, isTopLevel);
466 : :
467 : : /* Finally, clean up the vacuum memory context */
468 : 531 : MemoryContextDelete(vac_context);
469 : 531 : }
470 : :
471 : : /*
472 : : * Internal entry point for autovacuum and the VACUUM / ANALYZE commands.
473 : : *
474 : : * relations, if not NIL, is a list of VacuumRelation to process; otherwise,
475 : : * we process all relevant tables in the database. For each VacuumRelation,
476 : : * if a valid OID is supplied, the table with that OID is what to process;
477 : : * otherwise, the VacuumRelation's RangeVar indicates what to process.
478 : : *
479 : : * params contains a set of parameters that can be used to customize the
480 : : * behavior.
481 : : *
482 : : * bstrategy may be passed in as NULL when the caller does not want to
483 : : * restrict the number of shared_buffers that VACUUM / ANALYZE can use,
484 : : * otherwise, the caller must build a BufferAccessStrategy with the number of
485 : : * shared_buffers that VACUUM / ANALYZE should try to limit themselves to
486 : : * using.
487 : : *
488 : : * isTopLevel should be passed down from ProcessUtility.
489 : : *
490 : : * It is the caller's responsibility that all parameters are allocated in a
491 : : * memory context that will not disappear at transaction commit.
492 : : */
493 : : void
494 : 522 : vacuum(List *relations, const VacuumParams params, BufferAccessStrategy bstrategy,
495 : : MemoryContext vac_context, bool isTopLevel)
496 : : {
497 : : static bool in_vacuum = false;
498 : :
499 : 522 : const char *stmttype;
500 : 522 : volatile bool in_outer_xact,
501 : : use_own_xacts;
502 : :
503 : 522 : stmttype = (params.options & VACOPT_VACUUM) ? "VACUUM" : "ANALYZE";
504 : :
505 : : /*
506 : : * We cannot run VACUUM inside a user transaction block; if we were inside
507 : : * a transaction, then our commit- and start-transaction-command calls
508 : : * would not have the intended effect! There are numerous other subtle
509 : : * dependencies on this, too.
510 : : *
511 : : * ANALYZE (without VACUUM) can run either way.
512 : : */
513 [ + + ]: 522 : if (params.options & VACOPT_VACUUM)
514 : : {
515 : 187 : PreventInTransactionBlock(isTopLevel, stmttype);
516 : 187 : in_outer_xact = false;
517 : 187 : }
518 : : else
519 : 335 : in_outer_xact = IsInTransactionBlock(isTopLevel);
520 : :
521 : : /*
522 : : * Check for and disallow recursive calls. This could happen when VACUUM
523 : : * FULL or ANALYZE calls a hostile index expression that itself calls
524 : : * ANALYZE.
525 : : */
526 [ + + ]: 522 : if (in_vacuum)
527 [ + - + - ]: 2 : ereport(ERROR,
528 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
529 : : errmsg("%s cannot be executed from VACUUM or ANALYZE",
530 : : stmttype)));
531 : :
532 : : /*
533 : : * Build list of relation(s) to process, putting any new data in
534 : : * vac_context for safekeeping.
535 : : */
536 [ + + ]: 520 : if (params.options & VACOPT_ONLY_DATABASE_STATS)
537 : : {
538 : : /* We don't process any tables in this case */
539 [ + - ]: 1 : Assert(relations == NIL);
540 : 1 : }
541 [ + + ]: 519 : else if (relations != NIL)
542 : : {
543 : 515 : List *newrels = NIL;
544 : 515 : ListCell *lc;
545 : :
546 [ + - + + : 1056 : foreach(lc, relations)
+ + ]
547 : : {
548 : 541 : VacuumRelation *vrel = lfirst_node(VacuumRelation, lc);
549 : 541 : List *sublist;
550 : 541 : MemoryContext old_context;
551 : :
552 : 541 : sublist = expand_vacuum_rel(vrel, vac_context, params.options);
553 : 541 : old_context = MemoryContextSwitchTo(vac_context);
554 : 541 : newrels = list_concat(newrels, sublist);
555 : 541 : MemoryContextSwitchTo(old_context);
556 : 541 : }
557 : 515 : relations = newrels;
558 : 515 : }
559 : : else
560 : 4 : relations = get_all_vacuum_rels(vac_context, params.options);
561 : :
562 : : /*
563 : : * Decide whether we need to start/commit our own transactions.
564 : : *
565 : : * For VACUUM (with or without ANALYZE): always do so, so that we can
566 : : * release locks as soon as possible. (We could possibly use the outer
567 : : * transaction for a one-table VACUUM, but handling TOAST tables would be
568 : : * problematic.)
569 : : *
570 : : * For ANALYZE (no VACUUM): if inside a transaction block, we cannot
571 : : * start/commit our own transactions. Also, there's no need to do so if
572 : : * only processing one relation. For multiple relations when not within a
573 : : * transaction block, and also in an autovacuum worker, use own
574 : : * transactions so we can release locks sooner.
575 : : */
576 [ + + ]: 520 : if (params.options & VACOPT_VACUUM)
577 : 191 : use_own_xacts = true;
578 : : else
579 : : {
580 [ + - ]: 329 : Assert(params.options & VACOPT_ANALYZE);
581 [ - + ]: 329 : if (AmAutoVacuumWorkerProcess())
582 : 0 : use_own_xacts = true;
583 [ + + ]: 329 : else if (in_outer_xact)
584 : 34 : use_own_xacts = false;
585 [ + + ]: 295 : else if (list_length(relations) > 1)
586 : 113 : use_own_xacts = true;
587 : : else
588 : 182 : use_own_xacts = false;
589 : : }
590 : :
591 : : /*
592 : : * vacuum_rel expects to be entered with no transaction active; it will
593 : : * start and commit its own transaction. But we are called by an SQL
594 : : * command, and so we are executing inside a transaction already. We
595 : : * commit the transaction started in PostgresMain() here, and start
596 : : * another one before exiting to match the commit waiting for us back in
597 : : * PostgresMain().
598 : : */
599 [ + + ]: 520 : if (use_own_xacts)
600 : : {
601 [ + - ]: 304 : Assert(!in_outer_xact);
602 : :
603 : : /* ActiveSnapshot is not set by autovacuum */
604 [ - + ]: 304 : if (ActiveSnapshotSet())
605 : 304 : PopActiveSnapshot();
606 : :
607 : : /* matches the StartTransaction in PostgresMain() */
608 : 304 : CommitTransactionCommand();
609 : 304 : }
610 : :
611 : : /* Turn vacuum cost accounting on or off, and set/clear in_vacuum */
612 [ + + ]: 520 : PG_TRY();
613 : : {
614 : 510 : ListCell *cur;
615 : :
616 : 510 : in_vacuum = true;
617 : 510 : VacuumFailsafeActive = false;
618 : 510 : VacuumUpdateCosts();
619 : 510 : VacuumCostBalance = 0;
620 : 510 : VacuumCostBalanceLocal = 0;
621 : 510 : VacuumSharedCostBalance = NULL;
622 : 510 : VacuumActiveNWorkers = NULL;
623 : :
624 : : /*
625 : : * Loop to process each selected relation.
626 : : */
627 [ + + + + : 1728 : foreach(cur, relations)
+ + ]
628 : : {
629 : 1218 : VacuumRelation *vrel = lfirst_node(VacuumRelation, cur);
630 : :
631 [ + + ]: 1218 : if (params.options & VACOPT_VACUUM)
632 : : {
633 [ + + ]: 517 : if (!vacuum_rel(vrel->oid, vrel->relation, params, bstrategy))
634 : 12 : continue;
635 : 505 : }
636 : :
637 [ + + ]: 1206 : if (params.options & VACOPT_ANALYZE)
638 : : {
639 : : /*
640 : : * If using separate xacts, start one for analyze. Otherwise,
641 : : * we can use the outer transaction.
642 : : */
643 [ + + ]: 782 : if (use_own_xacts)
644 : : {
645 : 581 : StartTransactionCommand();
646 : : /* functions in indexes may want a snapshot set */
647 : 581 : PushActiveSnapshot(GetTransactionSnapshot());
648 : 581 : }
649 : :
650 : 1564 : analyze_rel(vrel->oid, vrel->relation, params,
651 : 782 : vrel->va_cols, in_outer_xact, bstrategy);
652 : :
653 [ + + ]: 782 : if (use_own_xacts)
654 : : {
655 : 575 : PopActiveSnapshot();
656 : : /* standard_ProcessUtility() does CCI if !use_own_xacts */
657 : 575 : CommandCounterIncrement();
658 : 575 : CommitTransactionCommand();
659 : 575 : }
660 : : else
661 : : {
662 : : /*
663 : : * If we're not using separate xacts, better separate the
664 : : * ANALYZE actions with CCIs. This avoids trouble if user
665 : : * says "ANALYZE t, t".
666 : : */
667 : 207 : CommandCounterIncrement();
668 : : }
669 : 782 : }
670 : :
671 : : /*
672 : : * Ensure VacuumFailsafeActive has been reset before vacuuming the
673 : : * next relation.
674 : : */
675 : 1206 : VacuumFailsafeActive = false;
676 [ - + + ]: 1218 : }
677 : 510 : }
678 : 520 : PG_FINALLY();
679 : : {
680 : 520 : in_vacuum = false;
681 : 520 : VacuumCostActive = false;
682 : 520 : VacuumFailsafeActive = false;
683 : 520 : VacuumCostBalance = 0;
684 : : }
685 [ + + ]: 520 : PG_END_TRY();
686 : :
687 : : /*
688 : : * Finish up processing.
689 : : */
690 [ + + ]: 510 : if (use_own_xacts)
691 : : {
692 : : /* here, we are not in a transaction */
693 : :
694 : : /*
695 : : * This matches the CommitTransaction waiting for us in
696 : : * PostgresMain().
697 : : */
698 : 297 : StartTransactionCommand();
699 : 297 : }
700 : :
701 [ + + + + ]: 510 : if ((params.options & VACOPT_VACUUM) &&
702 : 186 : !(params.options & VACOPT_SKIP_DATABASE_STATS))
703 : : {
704 : : /*
705 : : * Update pg_database.datfrozenxid, and truncate pg_xact if possible.
706 : : */
707 : 185 : vac_update_datfrozenxid();
708 : 185 : }
709 : :
710 : 510 : }
711 : :
712 : : /*
713 : : * Check if the current user has privileges to vacuum or analyze the relation.
714 : : * If not, issue a WARNING log message and return false to let the caller
715 : : * decide what to do with this relation. This routine is used to decide if a
716 : : * relation can be processed for VACUUM or ANALYZE.
717 : : */
718 : : bool
719 : 2872 : vacuum_is_permitted_for_relation(Oid relid, Form_pg_class reltuple,
720 : : bits32 options)
721 : : {
722 : 2872 : char *relname;
723 : :
724 [ + - ]: 2872 : Assert((options & (VACOPT_VACUUM | VACOPT_ANALYZE)) != 0);
725 : :
726 : : /*----------
727 : : * A role has privileges to vacuum or analyze the relation if any of the
728 : : * following are true:
729 : : * - the role owns the current database and the relation is not shared
730 : : * - the role has the MAINTAIN privilege on the relation
731 : : *----------
732 : : */
733 [ + + ]: 2872 : if ((object_ownercheck(DatabaseRelationId, MyDatabaseId, GetUserId()) &&
734 [ + + ]: 2872 : !reltuple->relisshared) ||
735 : 2872 : pg_class_aclcheck(relid, GetUserId(), ACL_MAINTAIN) == ACLCHECK_OK)
736 : 2359 : return true;
737 : :
738 : 513 : relname = NameStr(reltuple->relname);
739 : :
740 [ + + ]: 513 : if ((options & VACOPT_VACUUM) != 0)
741 : : {
742 [ - + + - ]: 36 : ereport(WARNING,
743 : : (errmsg("permission denied to vacuum \"%s\", skipping it",
744 : : relname)));
745 : :
746 : : /*
747 : : * For VACUUM ANALYZE, both logs could show up, but just generate
748 : : * information for VACUUM as that would be the first one to be
749 : : * processed.
750 : : */
751 : 36 : return false;
752 : : }
753 : :
754 [ - + ]: 477 : if ((options & VACOPT_ANALYZE) != 0)
755 [ - + + - ]: 477 : ereport(WARNING,
756 : : (errmsg("permission denied to analyze \"%s\", skipping it",
757 : : relname)));
758 : :
759 : 477 : return false;
760 : 2872 : }
761 : :
762 : :
763 : : /*
764 : : * vacuum_open_relation
765 : : *
766 : : * This routine is used for attempting to open and lock a relation which
767 : : * is going to be vacuumed or analyzed. If the relation cannot be opened
768 : : * or locked, a log is emitted if possible.
769 : : */
770 : : Relation
771 : 1509 : vacuum_open_relation(Oid relid, RangeVar *relation, bits32 options,
772 : : bool verbose, LOCKMODE lmode)
773 : : {
774 : 1509 : Relation rel;
775 : 1509 : bool rel_lock = true;
776 : 1509 : int elevel;
777 : :
778 [ + - ]: 1509 : Assert((options & (VACOPT_VACUUM | VACOPT_ANALYZE)) != 0);
779 : :
780 : : /*
781 : : * Open the relation and get the appropriate lock on it.
782 : : *
783 : : * There's a race condition here: the relation may have gone away since
784 : : * the last time we saw it. If so, we don't need to vacuum or analyze it.
785 : : *
786 : : * If we've been asked not to wait for the relation lock, acquire it first
787 : : * in non-blocking mode, before calling try_relation_open().
788 : : */
789 [ + + ]: 1509 : if (!(options & VACOPT_SKIP_LOCKED))
790 : 1506 : rel = try_relation_open(relid, lmode);
791 [ + - ]: 3 : else if (ConditionalLockRelationOid(relid, lmode))
792 : 3 : rel = try_relation_open(relid, NoLock);
793 : : else
794 : : {
795 : 0 : rel = NULL;
796 : 0 : rel_lock = false;
797 : : }
798 : :
799 : : /* if relation is opened, leave */
800 [ + - ]: 1509 : if (rel)
801 : 1509 : return rel;
802 : :
803 : : /*
804 : : * Relation could not be opened, hence generate if possible a log
805 : : * informing on the situation.
806 : : *
807 : : * If the RangeVar is not defined, we do not have enough information to
808 : : * provide a meaningful log statement. Chances are that the caller has
809 : : * intentionally not provided this information so that this logging is
810 : : * skipped, anyway.
811 : : */
812 [ # # ]: 0 : if (relation == NULL)
813 : 0 : return NULL;
814 : :
815 : : /*
816 : : * Determine the log level.
817 : : *
818 : : * For manual VACUUM or ANALYZE, we emit a WARNING to match the log
819 : : * statements in the permission checks; otherwise, only log if the caller
820 : : * so requested.
821 : : */
822 [ # # ]: 0 : if (!AmAutoVacuumWorkerProcess())
823 : 0 : elevel = WARNING;
824 [ # # ]: 0 : else if (verbose)
825 : 0 : elevel = LOG;
826 : : else
827 : 0 : return NULL;
828 : :
829 [ # # ]: 0 : if ((options & VACOPT_VACUUM) != 0)
830 : : {
831 [ # # ]: 0 : if (!rel_lock)
832 [ # # # # : 0 : ereport(elevel,
# # # # #
# ]
833 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
834 : : errmsg("skipping vacuum of \"%s\" --- lock not available",
835 : : relation->relname)));
836 : : else
837 [ # # # # : 0 : ereport(elevel,
# # # # #
# ]
838 : : (errcode(ERRCODE_UNDEFINED_TABLE),
839 : : errmsg("skipping vacuum of \"%s\" --- relation no longer exists",
840 : : relation->relname)));
841 : :
842 : : /*
843 : : * For VACUUM ANALYZE, both logs could show up, but just generate
844 : : * information for VACUUM as that would be the first one to be
845 : : * processed.
846 : : */
847 : 0 : return NULL;
848 : : }
849 : :
850 [ # # ]: 0 : if ((options & VACOPT_ANALYZE) != 0)
851 : : {
852 [ # # ]: 0 : if (!rel_lock)
853 [ # # # # : 0 : ereport(elevel,
# # # # #
# ]
854 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
855 : : errmsg("skipping analyze of \"%s\" --- lock not available",
856 : : relation->relname)));
857 : : else
858 [ # # # # : 0 : ereport(elevel,
# # # # #
# ]
859 : : (errcode(ERRCODE_UNDEFINED_TABLE),
860 : : errmsg("skipping analyze of \"%s\" --- relation no longer exists",
861 : : relation->relname)));
862 : 0 : }
863 : :
864 : 0 : return NULL;
865 : 1509 : }
866 : :
867 : :
868 : : /*
869 : : * Given a VacuumRelation, fill in the table OID if it wasn't specified,
870 : : * and optionally add VacuumRelations for partitions or inheritance children.
871 : : *
872 : : * If a VacuumRelation does not have an OID supplied and is a partitioned
873 : : * table, an extra entry will be added to the output for each partition.
874 : : * Presently, only autovacuum supplies OIDs when calling vacuum(), and
875 : : * it does not want us to expand partitioned tables.
876 : : *
877 : : * We take care not to modify the input data structure, but instead build
878 : : * new VacuumRelation(s) to return. (But note that they will reference
879 : : * unmodified parts of the input, eg column lists.) New data structures
880 : : * are made in vac_context.
881 : : */
882 : : static List *
883 : 535 : expand_vacuum_rel(VacuumRelation *vrel, MemoryContext vac_context,
884 : : int options)
885 : : {
886 : 535 : List *vacrels = NIL;
887 : 535 : MemoryContext oldcontext;
888 : :
889 : : /* If caller supplied OID, there's nothing we need do here. */
890 [ - + ]: 535 : if (OidIsValid(vrel->oid))
891 : : {
892 : 0 : oldcontext = MemoryContextSwitchTo(vac_context);
893 : 0 : vacrels = lappend(vacrels, vrel);
894 : 0 : MemoryContextSwitchTo(oldcontext);
895 : 0 : }
896 : : else
897 : : {
898 : : /*
899 : : * Process a specific relation, and possibly partitions or child
900 : : * tables thereof.
901 : : */
902 : 535 : Oid relid;
903 : 535 : HeapTuple tuple;
904 : 535 : Form_pg_class classForm;
905 : 535 : bool include_children;
906 : 535 : bool is_partitioned_table;
907 : 535 : int rvr_opts;
908 : :
909 : : /*
910 : : * Since autovacuum workers supply OIDs when calling vacuum(), no
911 : : * autovacuum worker should reach this code.
912 : : */
913 [ + - ]: 535 : Assert(!AmAutoVacuumWorkerProcess());
914 : :
915 : : /*
916 : : * We transiently take AccessShareLock to protect the syscache lookup
917 : : * below, as well as find_all_inheritors's expectation that the caller
918 : : * holds some lock on the starting relation.
919 : : */
920 : 535 : rvr_opts = (options & VACOPT_SKIP_LOCKED) ? RVR_SKIP_LOCKED : 0;
921 : 1070 : relid = RangeVarGetRelidExtended(vrel->relation,
922 : : AccessShareLock,
923 : 535 : rvr_opts,
924 : : NULL, NULL);
925 : :
926 : : /*
927 : : * If the lock is unavailable, emit the same log statement that
928 : : * vacuum_rel() and analyze_rel() would.
929 : : */
930 [ + - ]: 535 : if (!OidIsValid(relid))
931 : : {
932 [ # # ]: 0 : if (options & VACOPT_VACUUM)
933 [ # # # # ]: 0 : ereport(WARNING,
934 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
935 : : errmsg("skipping vacuum of \"%s\" --- lock not available",
936 : : vrel->relation->relname)));
937 : : else
938 [ # # # # ]: 0 : ereport(WARNING,
939 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
940 : : errmsg("skipping analyze of \"%s\" --- lock not available",
941 : : vrel->relation->relname)));
942 : 0 : return vacrels;
943 : : }
944 : :
945 : : /*
946 : : * To check whether the relation is a partitioned table and its
947 : : * ownership, fetch its syscache entry.
948 : : */
949 : 535 : tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(relid));
950 [ + - ]: 535 : if (!HeapTupleIsValid(tuple))
951 [ # # # # ]: 0 : elog(ERROR, "cache lookup failed for relation %u", relid);
952 : 535 : classForm = (Form_pg_class) GETSTRUCT(tuple);
953 : :
954 : : /*
955 : : * Make a returnable VacuumRelation for this rel if the user has the
956 : : * required privileges.
957 : : */
958 [ + + ]: 535 : if (vacuum_is_permitted_for_relation(relid, classForm, options))
959 : : {
960 : 499 : oldcontext = MemoryContextSwitchTo(vac_context);
961 : 998 : vacrels = lappend(vacrels, makeVacuumRelation(vrel->relation,
962 : 499 : relid,
963 : 499 : vrel->va_cols));
964 : 499 : MemoryContextSwitchTo(oldcontext);
965 : 499 : }
966 : :
967 : : /*
968 : : * Vacuuming a partitioned table with ONLY will not do anything since
969 : : * the partitioned table itself is empty. Issue a warning if the user
970 : : * requests this.
971 : : */
972 : 535 : include_children = vrel->relation->inh;
973 : 535 : is_partitioned_table = (classForm->relkind == RELKIND_PARTITIONED_TABLE);
974 [ + + + + : 535 : if ((options & VACOPT_VACUUM) && is_partitioned_table && !include_children)
+ + ]
975 [ - + + - ]: 1 : ereport(WARNING,
976 : : (errmsg("VACUUM ONLY of partitioned table \"%s\" has no effect",
977 : : vrel->relation->relname)));
978 : :
979 : 535 : ReleaseSysCache(tuple);
980 : :
981 : : /*
982 : : * Unless the user has specified ONLY, make relation list entries for
983 : : * its partitions or inheritance child tables. Note that the list
984 : : * returned by find_all_inheritors() includes the passed-in OID, so we
985 : : * have to skip that. There's no point in taking locks on the
986 : : * individual partitions or child tables yet, and doing so would just
987 : : * add unnecessary deadlock risk. For this last reason, we do not yet
988 : : * check the ownership of the partitions/tables, which get added to
989 : : * the list to process. Ownership will be checked later on anyway.
990 : : */
991 [ + + ]: 535 : if (include_children)
992 : : {
993 : 530 : List *part_oids = find_all_inheritors(relid, NoLock, NULL);
994 : 530 : ListCell *part_lc;
995 : :
996 [ + - + + : 1418 : foreach(part_lc, part_oids)
+ + ]
997 : : {
998 : 888 : Oid part_oid = lfirst_oid(part_lc);
999 : :
1000 [ + + ]: 888 : if (part_oid == relid)
1001 : 530 : continue; /* ignore original table */
1002 : :
1003 : : /*
1004 : : * We omit a RangeVar since it wouldn't be appropriate to
1005 : : * complain about failure to open one of these relations
1006 : : * later.
1007 : : */
1008 : 358 : oldcontext = MemoryContextSwitchTo(vac_context);
1009 : 716 : vacrels = lappend(vacrels, makeVacuumRelation(NULL,
1010 : 358 : part_oid,
1011 : 358 : vrel->va_cols));
1012 : 358 : MemoryContextSwitchTo(oldcontext);
1013 [ - + + ]: 888 : }
1014 : 530 : }
1015 : :
1016 : : /*
1017 : : * Release lock again. This means that by the time we actually try to
1018 : : * process the table, it might be gone or renamed. In the former case
1019 : : * we'll silently ignore it; in the latter case we'll process it
1020 : : * anyway, but we must beware that the RangeVar doesn't necessarily
1021 : : * identify it anymore. This isn't ideal, perhaps, but there's little
1022 : : * practical alternative, since we're typically going to commit this
1023 : : * transaction and begin a new one between now and then. Moreover,
1024 : : * holding locks on multiple relations would create significant risk
1025 : : * of deadlock.
1026 : : */
1027 : 535 : UnlockRelationOid(relid, AccessShareLock);
1028 [ - + ]: 535 : }
1029 : :
1030 : 535 : return vacrels;
1031 : 535 : }
1032 : :
1033 : : /*
1034 : : * Construct a list of VacuumRelations for all vacuumable rels in
1035 : : * the current database. The list is built in vac_context.
1036 : : */
1037 : : static List *
1038 : 4 : get_all_vacuum_rels(MemoryContext vac_context, int options)
1039 : : {
1040 : 4 : List *vacrels = NIL;
1041 : 4 : Relation pgclass;
1042 : 4 : TableScanDesc scan;
1043 : 4 : HeapTuple tuple;
1044 : :
1045 : 4 : pgclass = table_open(RelationRelationId, AccessShareLock);
1046 : :
1047 : 4 : scan = table_beginscan_catalog(pgclass, 0, NULL);
1048 : :
1049 [ + + ]: 2936 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1050 : : {
1051 : 2932 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
1052 : 2932 : MemoryContext oldcontext;
1053 : 2932 : Oid relid = classForm->oid;
1054 : :
1055 : : /*
1056 : : * We include partitioned tables here; depending on which operation is
1057 : : * to be performed, caller will decide whether to process or ignore
1058 : : * them.
1059 : : */
1060 [ + + ]: 2932 : if (classForm->relkind != RELKIND_RELATION &&
1061 [ + + + + ]: 2139 : classForm->relkind != RELKIND_MATVIEW &&
1062 : 2131 : classForm->relkind != RELKIND_PARTITIONED_TABLE)
1063 : 2104 : continue;
1064 : :
1065 : : /* check permissions of relation */
1066 [ + + ]: 828 : if (!vacuum_is_permitted_for_relation(relid, classForm, options))
1067 : 459 : continue;
1068 : :
1069 : : /*
1070 : : * Build VacuumRelation(s) specifying the table OIDs to be processed.
1071 : : * We omit a RangeVar since it wouldn't be appropriate to complain
1072 : : * about failure to open one of these relations later.
1073 : : */
1074 : 369 : oldcontext = MemoryContextSwitchTo(vac_context);
1075 : 738 : vacrels = lappend(vacrels, makeVacuumRelation(NULL,
1076 : 369 : relid,
1077 : : NIL));
1078 : 369 : MemoryContextSwitchTo(oldcontext);
1079 [ - + + ]: 2932 : }
1080 : :
1081 : 4 : table_endscan(scan);
1082 : 4 : table_close(pgclass, AccessShareLock);
1083 : :
1084 : 8 : return vacrels;
1085 : 4 : }
1086 : :
1087 : : /*
1088 : : * vacuum_get_cutoffs() -- compute OldestXmin and freeze cutoff points
1089 : : *
1090 : : * The target relation and VACUUM parameters are our inputs.
1091 : : *
1092 : : * Output parameters are the cutoffs that VACUUM caller should use.
1093 : : *
1094 : : * Return value indicates if vacuumlazy.c caller should make its VACUUM
1095 : : * operation aggressive. An aggressive VACUUM must advance relfrozenxid up to
1096 : : * FreezeLimit (at a minimum), and relminmxid up to MultiXactCutoff (at a
1097 : : * minimum).
1098 : : */
1099 : : bool
1100 : 699 : vacuum_get_cutoffs(Relation rel, const VacuumParams params,
1101 : : struct VacuumCutoffs *cutoffs)
1102 : : {
1103 : 699 : int freeze_min_age,
1104 : : multixact_freeze_min_age,
1105 : : freeze_table_age,
1106 : : multixact_freeze_table_age,
1107 : : effective_multixact_freeze_max_age;
1108 : 699 : TransactionId nextXID,
1109 : : safeOldestXmin,
1110 : : aggressiveXIDCutoff;
1111 : 699 : MultiXactId nextMXID,
1112 : : safeOldestMxact,
1113 : : aggressiveMXIDCutoff;
1114 : :
1115 : : /* Use mutable copies of freeze age parameters */
1116 : 699 : freeze_min_age = params.freeze_min_age;
1117 : 699 : multixact_freeze_min_age = params.multixact_freeze_min_age;
1118 : 699 : freeze_table_age = params.freeze_table_age;
1119 : 699 : multixact_freeze_table_age = params.multixact_freeze_table_age;
1120 : :
1121 : : /* Set pg_class fields in cutoffs */
1122 : 699 : cutoffs->relfrozenxid = rel->rd_rel->relfrozenxid;
1123 : 699 : cutoffs->relminmxid = rel->rd_rel->relminmxid;
1124 : :
1125 : : /*
1126 : : * Acquire OldestXmin.
1127 : : *
1128 : : * We can always ignore processes running lazy vacuum. This is because we
1129 : : * use these values only for deciding which tuples we must keep in the
1130 : : * tables. Since lazy vacuum doesn't write its XID anywhere (usually no
1131 : : * XID assigned), it's safe to ignore it. In theory it could be
1132 : : * problematic to ignore lazy vacuums in a full vacuum, but keep in mind
1133 : : * that only one vacuum process can be working on a particular table at
1134 : : * any time, and that each vacuum is always an independent transaction.
1135 : : */
1136 : 699 : cutoffs->OldestXmin = GetOldestNonRemovableTransactionId(rel);
1137 : :
1138 [ + - ]: 699 : Assert(TransactionIdIsNormal(cutoffs->OldestXmin));
1139 : :
1140 : : /* Acquire OldestMxact */
1141 : 699 : cutoffs->OldestMxact = GetOldestMultiXactId();
1142 [ + - ]: 699 : Assert(MultiXactIdIsValid(cutoffs->OldestMxact));
1143 : :
1144 : : /* Acquire next XID/next MXID values used to apply age-based settings */
1145 : 699 : nextXID = ReadNextTransactionId();
1146 : 699 : nextMXID = ReadNextMultiXactId();
1147 : :
1148 : : /*
1149 : : * Also compute the multixact age for which freezing is urgent. This is
1150 : : * normally autovacuum_multixact_freeze_max_age, but may be less if
1151 : : * multixact members are bloated.
1152 : : */
1153 : 699 : effective_multixact_freeze_max_age = MultiXactMemberFreezeThreshold();
1154 : :
1155 : : /*
1156 : : * Almost ready to set freeze output parameters; check if OldestXmin or
1157 : : * OldestMxact are held back to an unsafe degree before we start on that
1158 : : */
1159 : 699 : safeOldestXmin = nextXID - autovacuum_freeze_max_age;
1160 [ + - ]: 699 : if (!TransactionIdIsNormal(safeOldestXmin))
1161 : 0 : safeOldestXmin = FirstNormalTransactionId;
1162 : 699 : safeOldestMxact = nextMXID - effective_multixact_freeze_max_age;
1163 [ + - ]: 699 : if (safeOldestMxact < FirstMultiXactId)
1164 : 0 : safeOldestMxact = FirstMultiXactId;
1165 [ + - ]: 699 : if (TransactionIdPrecedes(cutoffs->OldestXmin, safeOldestXmin))
1166 [ # # # # ]: 0 : ereport(WARNING,
1167 : : (errmsg("cutoff for removing and freezing tuples is far in the past"),
1168 : : errhint("Close open transactions soon to avoid wraparound problems.\n"
1169 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1170 [ + - ]: 699 : if (MultiXactIdPrecedes(cutoffs->OldestMxact, safeOldestMxact))
1171 [ # # # # ]: 0 : ereport(WARNING,
1172 : : (errmsg("cutoff for freezing multixacts is far in the past"),
1173 : : errhint("Close open transactions soon to avoid wraparound problems.\n"
1174 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1175 : :
1176 : : /*
1177 : : * Determine the minimum freeze age to use: as specified by the caller, or
1178 : : * vacuum_freeze_min_age, but in any case not more than half
1179 : : * autovacuum_freeze_max_age, so that autovacuums to prevent XID
1180 : : * wraparound won't occur too frequently.
1181 : : */
1182 [ + + ]: 699 : if (freeze_min_age < 0)
1183 : 521 : freeze_min_age = vacuum_freeze_min_age;
1184 [ + - ]: 699 : freeze_min_age = Min(freeze_min_age, autovacuum_freeze_max_age / 2);
1185 [ + - ]: 699 : Assert(freeze_min_age >= 0);
1186 : :
1187 : : /* Compute FreezeLimit, being careful to generate a normal XID */
1188 : 699 : cutoffs->FreezeLimit = nextXID - freeze_min_age;
1189 [ + - ]: 699 : if (!TransactionIdIsNormal(cutoffs->FreezeLimit))
1190 : 0 : cutoffs->FreezeLimit = FirstNormalTransactionId;
1191 : : /* FreezeLimit must always be <= OldestXmin */
1192 [ + + ]: 699 : if (TransactionIdPrecedes(cutoffs->OldestXmin, cutoffs->FreezeLimit))
1193 : 70 : cutoffs->FreezeLimit = cutoffs->OldestXmin;
1194 : :
1195 : : /*
1196 : : * Determine the minimum multixact freeze age to use: as specified by
1197 : : * caller, or vacuum_multixact_freeze_min_age, but in any case not more
1198 : : * than half effective_multixact_freeze_max_age, so that autovacuums to
1199 : : * prevent MultiXact wraparound won't occur too frequently.
1200 : : */
1201 [ + + ]: 699 : if (multixact_freeze_min_age < 0)
1202 : 521 : multixact_freeze_min_age = vacuum_multixact_freeze_min_age;
1203 [ + - ]: 699 : multixact_freeze_min_age = Min(multixact_freeze_min_age,
1204 : : effective_multixact_freeze_max_age / 2);
1205 [ + - ]: 699 : Assert(multixact_freeze_min_age >= 0);
1206 : :
1207 : : /* Compute MultiXactCutoff, being careful to generate a valid value */
1208 : 699 : cutoffs->MultiXactCutoff = nextMXID - multixact_freeze_min_age;
1209 [ + - ]: 699 : if (cutoffs->MultiXactCutoff < FirstMultiXactId)
1210 : 0 : cutoffs->MultiXactCutoff = FirstMultiXactId;
1211 : : /* MultiXactCutoff must always be <= OldestMxact */
1212 [ + + ]: 699 : if (MultiXactIdPrecedes(cutoffs->OldestMxact, cutoffs->MultiXactCutoff))
1213 : 4 : cutoffs->MultiXactCutoff = cutoffs->OldestMxact;
1214 : :
1215 : : /*
1216 : : * Finally, figure out if caller needs to do an aggressive VACUUM or not.
1217 : : *
1218 : : * Determine the table freeze age to use: as specified by the caller, or
1219 : : * the value of the vacuum_freeze_table_age GUC, but in any case not more
1220 : : * than autovacuum_freeze_max_age * 0.95, so that if you have e.g nightly
1221 : : * VACUUM schedule, the nightly VACUUM gets a chance to freeze XIDs before
1222 : : * anti-wraparound autovacuum is launched.
1223 : : */
1224 [ + + ]: 699 : if (freeze_table_age < 0)
1225 : 521 : freeze_table_age = vacuum_freeze_table_age;
1226 [ + - ]: 699 : freeze_table_age = Min(freeze_table_age, autovacuum_freeze_max_age * 0.95);
1227 [ + - ]: 699 : Assert(freeze_table_age >= 0);
1228 : 699 : aggressiveXIDCutoff = nextXID - freeze_table_age;
1229 [ + - ]: 699 : if (!TransactionIdIsNormal(aggressiveXIDCutoff))
1230 : 0 : aggressiveXIDCutoff = FirstNormalTransactionId;
1231 [ + + + + ]: 1398 : if (TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid,
1232 : 699 : aggressiveXIDCutoff))
1233 : 178 : return true;
1234 : :
1235 : : /*
1236 : : * Similar to the above, determine the table freeze age to use for
1237 : : * multixacts: as specified by the caller, or the value of the
1238 : : * vacuum_multixact_freeze_table_age GUC, but in any case not more than
1239 : : * effective_multixact_freeze_max_age * 0.95, so that if you have e.g.
1240 : : * nightly VACUUM schedule, the nightly VACUUM gets a chance to freeze
1241 : : * multixacts before anti-wraparound autovacuum is launched.
1242 : : */
1243 [ - + ]: 521 : if (multixact_freeze_table_age < 0)
1244 : 521 : multixact_freeze_table_age = vacuum_multixact_freeze_table_age;
1245 : 521 : multixact_freeze_table_age =
1246 [ + - ]: 521 : Min(multixact_freeze_table_age,
1247 : : effective_multixact_freeze_max_age * 0.95);
1248 [ + - ]: 521 : Assert(multixact_freeze_table_age >= 0);
1249 : 521 : aggressiveMXIDCutoff = nextMXID - multixact_freeze_table_age;
1250 [ + - ]: 521 : if (aggressiveMXIDCutoff < FirstMultiXactId)
1251 : 0 : aggressiveMXIDCutoff = FirstMultiXactId;
1252 [ - + - + ]: 1042 : if (MultiXactIdPrecedesOrEquals(cutoffs->relminmxid,
1253 : 521 : aggressiveMXIDCutoff))
1254 : 0 : return true;
1255 : :
1256 : : /* Non-aggressive VACUUM */
1257 : 521 : return false;
1258 : 699 : }
1259 : :
1260 : : /*
1261 : : * vacuum_xid_failsafe_check() -- Used by VACUUM's wraparound failsafe
1262 : : * mechanism to determine if its table's relfrozenxid and relminmxid are now
1263 : : * dangerously far in the past.
1264 : : *
1265 : : * When we return true, VACUUM caller triggers the failsafe.
1266 : : */
1267 : : bool
1268 : 823 : vacuum_xid_failsafe_check(const struct VacuumCutoffs *cutoffs)
1269 : : {
1270 : 823 : TransactionId relfrozenxid = cutoffs->relfrozenxid;
1271 : 823 : MultiXactId relminmxid = cutoffs->relminmxid;
1272 : 823 : TransactionId xid_skip_limit;
1273 : 823 : MultiXactId multi_skip_limit;
1274 : 823 : int skip_index_vacuum;
1275 : :
1276 [ + - ]: 823 : Assert(TransactionIdIsNormal(relfrozenxid));
1277 [ + - ]: 823 : Assert(MultiXactIdIsValid(relminmxid));
1278 : :
1279 : : /*
1280 : : * Determine the index skipping age to use. In any case no less than
1281 : : * autovacuum_freeze_max_age * 1.05.
1282 : : */
1283 [ + - ]: 823 : skip_index_vacuum = Max(vacuum_failsafe_age, autovacuum_freeze_max_age * 1.05);
1284 : :
1285 : 823 : xid_skip_limit = ReadNextTransactionId() - skip_index_vacuum;
1286 [ + - ]: 823 : if (!TransactionIdIsNormal(xid_skip_limit))
1287 : 0 : xid_skip_limit = FirstNormalTransactionId;
1288 : :
1289 [ - + ]: 823 : if (TransactionIdPrecedes(relfrozenxid, xid_skip_limit))
1290 : : {
1291 : : /* The table's relfrozenxid is too old */
1292 : 0 : return true;
1293 : : }
1294 : :
1295 : : /*
1296 : : * Similar to above, determine the index skipping age to use for
1297 : : * multixact. In any case no less than autovacuum_multixact_freeze_max_age *
1298 : : * 1.05.
1299 : : */
1300 [ + - ]: 823 : skip_index_vacuum = Max(vacuum_multixact_failsafe_age,
1301 : : autovacuum_multixact_freeze_max_age * 1.05);
1302 : :
1303 : 823 : multi_skip_limit = ReadNextMultiXactId() - skip_index_vacuum;
1304 [ + - ]: 823 : if (multi_skip_limit < FirstMultiXactId)
1305 : 0 : multi_skip_limit = FirstMultiXactId;
1306 : :
1307 [ - + ]: 823 : if (MultiXactIdPrecedes(relminmxid, multi_skip_limit))
1308 : : {
1309 : : /* The table's relminmxid is too old */
1310 : 0 : return true;
1311 : : }
1312 : :
1313 : 823 : return false;
1314 : 823 : }
1315 : :
1316 : : /*
1317 : : * vac_estimate_reltuples() -- estimate the new value for pg_class.reltuples
1318 : : *
1319 : : * If we scanned the whole relation then we should just use the count of
1320 : : * live tuples seen; but if we did not, we should not blindly extrapolate
1321 : : * from that number, since VACUUM may have scanned a quite nonrandom
1322 : : * subset of the table. When we have only partial information, we take
1323 : : * the old value of pg_class.reltuples/pg_class.relpages as a measurement
1324 : : * of the tuple density in the unscanned pages.
1325 : : *
1326 : : * Note: scanned_tuples should count only *live* tuples, since
1327 : : * pg_class.reltuples is defined that way.
1328 : : */
1329 : : double
1330 : 645 : vac_estimate_reltuples(Relation relation,
1331 : : BlockNumber total_pages,
1332 : : BlockNumber scanned_pages,
1333 : : double scanned_tuples)
1334 : : {
1335 : 645 : BlockNumber old_rel_pages = relation->rd_rel->relpages;
1336 : 645 : double old_rel_tuples = relation->rd_rel->reltuples;
1337 : 645 : double old_density;
1338 : 645 : double unscanned_pages;
1339 : 645 : double total_tuples;
1340 : :
1341 : : /* If we did scan the whole table, just use the count as-is */
1342 [ + + ]: 645 : if (scanned_pages >= total_pages)
1343 : 637 : return scanned_tuples;
1344 : :
1345 : : /*
1346 : : * When successive VACUUM commands scan the same few pages again and
1347 : : * again, without anything from the table really changing, there is a risk
1348 : : * that our beliefs about tuple density will gradually become distorted.
1349 : : * This might be caused by vacuumlazy.c implementation details, such as
1350 : : * its tendency to always scan the last heap page. Handle that here.
1351 : : *
1352 : : * If the relation is _exactly_ the same size according to the existing
1353 : : * pg_class entry, and only a few of its pages (less than 2%) were
1354 : : * scanned, keep the existing value of reltuples. Also keep the existing
1355 : : * value when only a subset of rel's pages <= a single page were scanned.
1356 : : *
1357 : : * (Note: we might be returning -1 here.)
1358 : : */
1359 [ + + + + ]: 8 : if (old_rel_pages == total_pages &&
1360 : 6 : scanned_pages < (double) total_pages * 0.02)
1361 : 3 : return old_rel_tuples;
1362 [ + + ]: 5 : if (scanned_pages <= 1)
1363 : 3 : return old_rel_tuples;
1364 : :
1365 : : /*
1366 : : * If old density is unknown, we can't do much except scale up
1367 : : * scanned_tuples to match total_pages.
1368 : : */
1369 [ + - - + ]: 2 : if (old_rel_tuples < 0 || old_rel_pages == 0)
1370 : 0 : return floor((scanned_tuples / scanned_pages) * total_pages + 0.5);
1371 : :
1372 : : /*
1373 : : * Okay, we've covered the corner cases. The normal calculation is to
1374 : : * convert the old measurement to a density (tuples per page), then
1375 : : * estimate the number of tuples in the unscanned pages using that figure,
1376 : : * and finally add on the number of tuples in the scanned pages.
1377 : : */
1378 : 2 : old_density = old_rel_tuples / old_rel_pages;
1379 : 2 : unscanned_pages = (double) total_pages - (double) scanned_pages;
1380 : 2 : total_tuples = old_density * unscanned_pages + scanned_tuples;
1381 : 2 : return floor(total_tuples + 0.5);
1382 : 645 : }
1383 : :
1384 : :
1385 : : /*
1386 : : * vac_update_relstats() -- update statistics for one relation
1387 : : *
1388 : : * Update the whole-relation statistics that are kept in its pg_class
1389 : : * row. There are additional stats that will be updated if we are
1390 : : * doing ANALYZE, but we always update these stats. This routine works
1391 : : * for both index and heap relation entries in pg_class.
1392 : : *
1393 : : * We violate transaction semantics here by overwriting the rel's
1394 : : * existing pg_class tuple with the new values. This is reasonably
1395 : : * safe as long as we're sure that the new values are correct whether or
1396 : : * not this transaction commits. The reason for doing this is that if
1397 : : * we updated these tuples in the usual way, vacuuming pg_class itself
1398 : : * wouldn't work very well --- by the time we got done with a vacuum
1399 : : * cycle, most of the tuples in pg_class would've been obsoleted. Of
1400 : : * course, this only works for fixed-size not-null columns, but these are.
1401 : : *
1402 : : * Another reason for doing it this way is that when we are in a lazy
1403 : : * VACUUM and have PROC_IN_VACUUM set, we mustn't do any regular updates.
1404 : : * Somebody vacuuming pg_class might think they could delete a tuple
1405 : : * marked with xmin = our xid.
1406 : : *
1407 : : * In addition to fundamentally nontransactional statistics such as
1408 : : * relpages and relallvisible, we try to maintain certain lazily-updated
1409 : : * DDL flags such as relhasindex, by clearing them if no longer correct.
1410 : : * It's safe to do this in VACUUM, which can't run in parallel with
1411 : : * CREATE INDEX/RULE/TRIGGER and can't be part of a transaction block.
1412 : : * However, it's *not* safe to do it in an ANALYZE that's within an
1413 : : * outer transaction, because for example the current transaction might
1414 : : * have dropped the last index; then we'd think relhasindex should be
1415 : : * cleared, but if the transaction later rolls back this would be wrong.
1416 : : * So we refrain from updating the DDL flags if we're inside an outer
1417 : : * transaction. This is OK since postponing the flag maintenance is
1418 : : * always allowable.
1419 : : *
1420 : : * Note: num_tuples should count only *live* tuples, since
1421 : : * pg_class.reltuples is defined that way.
1422 : : *
1423 : : * This routine is shared by VACUUM and ANALYZE.
1424 : : */
1425 : : void
1426 : 1921 : vac_update_relstats(Relation relation,
1427 : : BlockNumber num_pages, double num_tuples,
1428 : : BlockNumber num_all_visible_pages,
1429 : : BlockNumber num_all_frozen_pages,
1430 : : bool hasindex, TransactionId frozenxid,
1431 : : MultiXactId minmulti,
1432 : : bool *frozenxid_updated, bool *minmulti_updated,
1433 : : bool in_outer_xact)
1434 : : {
1435 : 1921 : Oid relid = RelationGetRelid(relation);
1436 : 1921 : Relation rd;
1437 : 1921 : ScanKeyData key[1];
1438 : 1921 : HeapTuple ctup;
1439 : 1921 : void *inplace_state;
1440 : 1921 : Form_pg_class pgcform;
1441 : 1921 : bool dirty,
1442 : : futurexid,
1443 : : futuremxid;
1444 : 1921 : TransactionId oldfrozenxid;
1445 : 1921 : MultiXactId oldminmulti;
1446 : :
1447 : 1921 : rd = table_open(RelationRelationId, RowExclusiveLock);
1448 : :
1449 : : /* Fetch a copy of the tuple to scribble on */
1450 : 3842 : ScanKeyInit(&key[0],
1451 : : Anum_pg_class_oid,
1452 : : BTEqualStrategyNumber, F_OIDEQ,
1453 : 1921 : ObjectIdGetDatum(relid));
1454 : 3842 : systable_inplace_update_begin(rd, ClassOidIndexId, true,
1455 : 1921 : NULL, 1, key, &ctup, &inplace_state);
1456 [ + - ]: 1921 : if (!HeapTupleIsValid(ctup))
1457 [ # # # # ]: 0 : elog(ERROR, "pg_class entry for relid %u vanished during vacuuming",
1458 : : relid);
1459 : 1921 : pgcform = (Form_pg_class) GETSTRUCT(ctup);
1460 : :
1461 : : /* Apply statistical updates, if any, to copied tuple */
1462 : :
1463 : 1921 : dirty = false;
1464 [ + + ]: 1921 : if (pgcform->relpages != (int32) num_pages)
1465 : : {
1466 : 719 : pgcform->relpages = (int32) num_pages;
1467 : 719 : dirty = true;
1468 : 719 : }
1469 [ + + ]: 1921 : if (pgcform->reltuples != (float4) num_tuples)
1470 : : {
1471 : 1050 : pgcform->reltuples = (float4) num_tuples;
1472 : 1050 : dirty = true;
1473 : 1050 : }
1474 [ + + ]: 1921 : if (pgcform->relallvisible != (int32) num_all_visible_pages)
1475 : : {
1476 : 234 : pgcform->relallvisible = (int32) num_all_visible_pages;
1477 : 234 : dirty = true;
1478 : 234 : }
1479 [ + + ]: 1921 : if (pgcform->relallfrozen != (int32) num_all_frozen_pages)
1480 : : {
1481 : 101 : pgcform->relallfrozen = (int32) num_all_frozen_pages;
1482 : 101 : dirty = true;
1483 : 101 : }
1484 : :
1485 : : /* Apply DDL updates, but not inside an outer transaction (see above) */
1486 : :
1487 [ + + ]: 1921 : if (!in_outer_xact)
1488 : : {
1489 : : /*
1490 : : * If we didn't find any indexes, reset relhasindex.
1491 : : */
1492 [ + + + + ]: 1874 : if (pgcform->relhasindex && !hasindex)
1493 : : {
1494 : 3 : pgcform->relhasindex = false;
1495 : 3 : dirty = true;
1496 : 3 : }
1497 : :
1498 : : /* We also clear relhasrules and relhastriggers if needed */
1499 [ + + + - ]: 1874 : if (pgcform->relhasrules && relation->rd_rules == NULL)
1500 : : {
1501 : 0 : pgcform->relhasrules = false;
1502 : 0 : dirty = true;
1503 : 0 : }
1504 [ + + + + ]: 1874 : if (pgcform->relhastriggers && relation->trigdesc == NULL)
1505 : : {
1506 : 1 : pgcform->relhastriggers = false;
1507 : 1 : dirty = true;
1508 : 1 : }
1509 : 1874 : }
1510 : :
1511 : : /*
1512 : : * Update relfrozenxid, unless caller passed InvalidTransactionId
1513 : : * indicating it has no new data.
1514 : : *
1515 : : * Ordinarily, we don't let relfrozenxid go backwards. However, if the
1516 : : * stored relfrozenxid is "in the future" then it seems best to assume
1517 : : * it's corrupt, and overwrite with the oldest remaining XID in the table.
1518 : : * This should match vac_update_datfrozenxid() concerning what we consider
1519 : : * to be "in the future".
1520 : : */
1521 : 1921 : oldfrozenxid = pgcform->relfrozenxid;
1522 : 1921 : futurexid = false;
1523 [ + + ]: 1921 : if (frozenxid_updated)
1524 : 645 : *frozenxid_updated = false;
1525 [ + + + + ]: 1921 : if (TransactionIdIsNormal(frozenxid) && oldfrozenxid != frozenxid)
1526 : : {
1527 : 573 : bool update = false;
1528 : :
1529 [ + + ]: 573 : if (TransactionIdPrecedes(oldfrozenxid, frozenxid))
1530 : 561 : update = true;
1531 [ + - ]: 12 : else if (TransactionIdPrecedes(ReadNextTransactionId(), oldfrozenxid))
1532 : 0 : futurexid = update = true;
1533 : :
1534 [ + + ]: 573 : if (update)
1535 : : {
1536 : 561 : pgcform->relfrozenxid = frozenxid;
1537 : 561 : dirty = true;
1538 [ - + ]: 561 : if (frozenxid_updated)
1539 : 561 : *frozenxid_updated = true;
1540 : 561 : }
1541 : 573 : }
1542 : :
1543 : : /* Similarly for relminmxid */
1544 : 1921 : oldminmulti = pgcform->relminmxid;
1545 : 1921 : futuremxid = false;
1546 [ + + ]: 1921 : if (minmulti_updated)
1547 : 645 : *minmulti_updated = false;
1548 [ + + + + ]: 1921 : if (MultiXactIdIsValid(minmulti) && oldminmulti != minmulti)
1549 : : {
1550 : 4 : bool update = false;
1551 : :
1552 [ + - ]: 4 : if (MultiXactIdPrecedes(oldminmulti, minmulti))
1553 : 4 : update = true;
1554 [ # # ]: 0 : else if (MultiXactIdPrecedes(ReadNextMultiXactId(), oldminmulti))
1555 : 0 : futuremxid = update = true;
1556 : :
1557 [ - + ]: 4 : if (update)
1558 : : {
1559 : 4 : pgcform->relminmxid = minmulti;
1560 : 4 : dirty = true;
1561 [ - + ]: 4 : if (minmulti_updated)
1562 : 4 : *minmulti_updated = true;
1563 : 4 : }
1564 : 4 : }
1565 : :
1566 : : /* If anything changed, write out the tuple. */
1567 [ + + ]: 1921 : if (dirty)
1568 : 1310 : systable_inplace_update_finish(inplace_state, ctup);
1569 : : else
1570 : 611 : systable_inplace_update_cancel(inplace_state);
1571 : :
1572 : 1921 : table_close(rd, RowExclusiveLock);
1573 : :
1574 [ + - ]: 1921 : if (futurexid)
1575 [ # # # # ]: 0 : ereport(WARNING,
1576 : : (errcode(ERRCODE_DATA_CORRUPTED),
1577 : : errmsg_internal("overwrote invalid relfrozenxid value %u with new value %u for table \"%s\"",
1578 : : oldfrozenxid, frozenxid,
1579 : : RelationGetRelationName(relation))));
1580 [ + - ]: 1921 : if (futuremxid)
1581 [ # # # # ]: 0 : ereport(WARNING,
1582 : : (errcode(ERRCODE_DATA_CORRUPTED),
1583 : : errmsg_internal("overwrote invalid relminmxid value %u with new value %u for table \"%s\"",
1584 : : oldminmulti, minmulti,
1585 : : RelationGetRelationName(relation))));
1586 : 1921 : }
1587 : :
1588 : :
1589 : : /*
1590 : : * vac_update_datfrozenxid() -- update pg_database.datfrozenxid for our DB
1591 : : *
1592 : : * Update pg_database's datfrozenxid entry for our database to be the
1593 : : * minimum of the pg_class.relfrozenxid values.
1594 : : *
1595 : : * Similarly, update our datminmxid to be the minimum of the
1596 : : * pg_class.relminmxid values.
1597 : : *
1598 : : * If we are able to advance either pg_database value, also try to
1599 : : * truncate pg_xact and pg_multixact.
1600 : : *
1601 : : * We violate transaction semantics here by overwriting the database's
1602 : : * existing pg_database tuple with the new values. This is reasonably
1603 : : * safe since the new values are correct whether or not this transaction
1604 : : * commits. As with vac_update_relstats, this avoids leaving dead tuples
1605 : : * behind after a VACUUM.
1606 : : */
1607 : : void
1608 : 186 : vac_update_datfrozenxid(void)
1609 : : {
1610 : 186 : HeapTuple tuple;
1611 : 186 : Form_pg_database dbform;
1612 : 186 : Relation relation;
1613 : 186 : SysScanDesc scan;
1614 : 186 : HeapTuple classTup;
1615 : 186 : TransactionId newFrozenXid;
1616 : 186 : MultiXactId newMinMulti;
1617 : 186 : TransactionId lastSaneFrozenXid;
1618 : 186 : MultiXactId lastSaneMinMulti;
1619 : 186 : bool bogus = false;
1620 : 186 : bool dirty = false;
1621 : 186 : ScanKeyData key[1];
1622 : 186 : void *inplace_state;
1623 : :
1624 : : /*
1625 : : * Restrict this task to one backend per database. This avoids race
1626 : : * conditions that would move datfrozenxid or datminmxid backward. It
1627 : : * avoids calling vac_truncate_clog() with a datfrozenxid preceding a
1628 : : * datfrozenxid passed to an earlier vac_truncate_clog() call.
1629 : : */
1630 : 186 : LockDatabaseFrozenIds(ExclusiveLock);
1631 : :
1632 : : /*
1633 : : * Initialize the "min" calculation with
1634 : : * GetOldestNonRemovableTransactionId(), which is a reasonable
1635 : : * approximation to the minimum relfrozenxid for not-yet-committed
1636 : : * pg_class entries for new tables; see AddNewRelationTuple(). So we
1637 : : * cannot produce a wrong minimum by starting with this.
1638 : : */
1639 : 186 : newFrozenXid = GetOldestNonRemovableTransactionId(NULL);
1640 : :
1641 : : /*
1642 : : * Similarly, initialize the MultiXact "min" with the value that would be
1643 : : * used on pg_class for new tables. See AddNewRelationTuple().
1644 : : */
1645 : 186 : newMinMulti = GetOldestMultiXactId();
1646 : :
1647 : : /*
1648 : : * Identify the latest relfrozenxid and relminmxid values that we could
1649 : : * validly see during the scan. These are conservative values, but it's
1650 : : * not really worth trying to be more exact.
1651 : : */
1652 : 186 : lastSaneFrozenXid = ReadNextTransactionId();
1653 : 186 : lastSaneMinMulti = ReadNextMultiXactId();
1654 : :
1655 : : /*
1656 : : * We must seqscan pg_class to find the minimum Xid, because there is no
1657 : : * index that can help us here.
1658 : : *
1659 : : * See vac_truncate_clog() for the race condition to prevent.
1660 : : */
1661 : 186 : relation = table_open(RelationRelationId, AccessShareLock);
1662 : :
1663 : 186 : scan = systable_beginscan(relation, InvalidOid, false,
1664 : : NULL, 0, NULL);
1665 : :
1666 [ + + ]: 152403 : while ((classTup = systable_getnext(scan)) != NULL)
1667 : : {
1668 : 152217 : volatile FormData_pg_class *classForm = (Form_pg_class) GETSTRUCT(classTup);
1669 : 152217 : TransactionId relfrozenxid = classForm->relfrozenxid;
1670 : 152217 : TransactionId relminmxid = classForm->relminmxid;
1671 : :
1672 : : /*
1673 : : * Only consider relations able to hold unfrozen XIDs (anything else
1674 : : * should have InvalidTransactionId in relfrozenxid anyway).
1675 : : */
1676 [ + + ]: 152217 : if (classForm->relkind != RELKIND_RELATION &&
1677 [ + + + + ]: 108635 : classForm->relkind != RELKIND_MATVIEW &&
1678 : 108248 : classForm->relkind != RELKIND_TOASTVALUE)
1679 : : {
1680 [ - + ]: 89209 : Assert(!TransactionIdIsValid(relfrozenxid));
1681 [ - + ]: 89209 : Assert(!MultiXactIdIsValid(relminmxid));
1682 : 89209 : continue;
1683 : : }
1684 : :
1685 : : /*
1686 : : * Some table AMs might not need per-relation xid / multixid horizons.
1687 : : * It therefore seems reasonable to allow relfrozenxid and relminmxid
1688 : : * to not be set (i.e. set to their respective Invalid*Id)
1689 : : * independently. Thus validate and compute horizon for each only if
1690 : : * set.
1691 : : *
1692 : : * If things are working properly, no relation should have a
1693 : : * relfrozenxid or relminmxid that is "in the future". However, such
1694 : : * cases have been known to arise due to bugs in pg_upgrade. If we
1695 : : * see any entries that are "in the future", chicken out and don't do
1696 : : * anything. This ensures we won't truncate clog & multixact SLRUs
1697 : : * before those relations have been scanned and cleaned up.
1698 : : */
1699 : :
1700 [ - + ]: 63008 : if (TransactionIdIsValid(relfrozenxid))
1701 : : {
1702 [ + - ]: 63008 : Assert(TransactionIdIsNormal(relfrozenxid));
1703 : :
1704 : : /* check for values in the future */
1705 [ - + ]: 63008 : if (TransactionIdPrecedes(lastSaneFrozenXid, relfrozenxid))
1706 : : {
1707 : 0 : bogus = true;
1708 : 0 : break;
1709 : : }
1710 : :
1711 : : /* determine new horizon */
1712 [ + + ]: 63008 : if (TransactionIdPrecedes(relfrozenxid, newFrozenXid))
1713 : 435 : newFrozenXid = relfrozenxid;
1714 : 63008 : }
1715 : :
1716 [ - + ]: 63008 : if (MultiXactIdIsValid(relminmxid))
1717 : : {
1718 : : /* check for values in the future */
1719 [ - + ]: 63008 : if (MultiXactIdPrecedes(lastSaneMinMulti, relminmxid))
1720 : : {
1721 : 0 : bogus = true;
1722 : 0 : break;
1723 : : }
1724 : :
1725 : : /* determine new horizon */
1726 [ + + ]: 63008 : if (MultiXactIdPrecedes(relminmxid, newMinMulti))
1727 : 10 : newMinMulti = relminmxid;
1728 : 63008 : }
1729 [ + + ]: 152217 : }
1730 : :
1731 : : /* we're done with pg_class */
1732 : 186 : systable_endscan(scan);
1733 : 186 : table_close(relation, AccessShareLock);
1734 : :
1735 : : /* chicken out if bogus data found */
1736 [ - + ]: 186 : if (bogus)
1737 : 0 : return;
1738 : :
1739 [ + - ]: 186 : Assert(TransactionIdIsNormal(newFrozenXid));
1740 [ + - ]: 186 : Assert(MultiXactIdIsValid(newMinMulti));
1741 : :
1742 : : /* Now fetch the pg_database tuple we need to update. */
1743 : 186 : relation = table_open(DatabaseRelationId, RowExclusiveLock);
1744 : :
1745 : : /*
1746 : : * Fetch a copy of the tuple to scribble on. We could check the syscache
1747 : : * tuple first. If that concluded !dirty, we'd avoid waiting on
1748 : : * concurrent heap_update() and would avoid exclusive-locking the buffer.
1749 : : * For now, don't optimize that.
1750 : : */
1751 : 372 : ScanKeyInit(&key[0],
1752 : : Anum_pg_database_oid,
1753 : : BTEqualStrategyNumber, F_OIDEQ,
1754 : 186 : ObjectIdGetDatum(MyDatabaseId));
1755 : :
1756 : 372 : systable_inplace_update_begin(relation, DatabaseOidIndexId, true,
1757 : 186 : NULL, 1, key, &tuple, &inplace_state);
1758 : :
1759 [ + - ]: 186 : if (!HeapTupleIsValid(tuple))
1760 [ # # # # ]: 0 : elog(ERROR, "could not find tuple for database %u", MyDatabaseId);
1761 : :
1762 : 186 : dbform = (Form_pg_database) GETSTRUCT(tuple);
1763 : :
1764 : : /*
1765 : : * As in vac_update_relstats(), we ordinarily don't want to let
1766 : : * datfrozenxid go backward; but if it's "in the future" then it must be
1767 : : * corrupt and it seems best to overwrite it.
1768 : : */
1769 [ + + # # ]: 186 : if (dbform->datfrozenxid != newFrozenXid &&
1770 [ - + ]: 2 : (TransactionIdPrecedes(dbform->datfrozenxid, newFrozenXid) ||
1771 : 0 : TransactionIdPrecedes(lastSaneFrozenXid, dbform->datfrozenxid)))
1772 : : {
1773 : 2 : dbform->datfrozenxid = newFrozenXid;
1774 : 2 : dirty = true;
1775 : 2 : }
1776 : : else
1777 : 184 : newFrozenXid = dbform->datfrozenxid;
1778 : :
1779 : : /* Ditto for datminmxid */
1780 [ - + # # ]: 186 : if (dbform->datminmxid != newMinMulti &&
1781 [ # # ]: 0 : (MultiXactIdPrecedes(dbform->datminmxid, newMinMulti) ||
1782 : 0 : MultiXactIdPrecedes(lastSaneMinMulti, dbform->datminmxid)))
1783 : : {
1784 : 0 : dbform->datminmxid = newMinMulti;
1785 : 0 : dirty = true;
1786 : 0 : }
1787 : : else
1788 : 186 : newMinMulti = dbform->datminmxid;
1789 : :
1790 [ + + ]: 186 : if (dirty)
1791 : 2 : systable_inplace_update_finish(inplace_state, tuple);
1792 : : else
1793 : 184 : systable_inplace_update_cancel(inplace_state);
1794 : :
1795 : 186 : heap_freetuple(tuple);
1796 : 186 : table_close(relation, RowExclusiveLock);
1797 : :
1798 : : /*
1799 : : * If we were able to advance datfrozenxid or datminmxid, see if we can
1800 : : * truncate pg_xact and/or pg_multixact. Also do it if the shared
1801 : : * XID-wrap-limit info is stale, since this action will update that too.
1802 : : */
1803 [ + + - + ]: 186 : if (dirty || ForceTransactionIdLimitUpdate())
1804 : 4 : vac_truncate_clog(newFrozenXid, newMinMulti,
1805 : 2 : lastSaneFrozenXid, lastSaneMinMulti);
1806 : 186 : }
1807 : :
1808 : :
1809 : : /*
1810 : : * vac_truncate_clog() -- attempt to truncate the commit log
1811 : : *
1812 : : * Scan pg_database to determine the system-wide oldest datfrozenxid,
1813 : : * and use it to truncate the transaction commit log (pg_xact).
1814 : : * Also update the XID wrap limit info maintained by varsup.c.
1815 : : * Likewise for datminmxid.
1816 : : *
1817 : : * The passed frozenXID and minMulti are the updated values for my own
1818 : : * pg_database entry. They're used to initialize the "min" calculations.
1819 : : * The caller also passes the "last sane" XID and MXID, since it has
1820 : : * those at hand already.
1821 : : *
1822 : : * This routine is only invoked when we've managed to change our
1823 : : * DB's datfrozenxid/datminmxid values, or we found that the shared
1824 : : * XID-wrap-limit info is stale.
1825 : : */
1826 : : static void
1827 : 2 : vac_truncate_clog(TransactionId frozenXID,
1828 : : MultiXactId minMulti,
1829 : : TransactionId lastSaneFrozenXid,
1830 : : MultiXactId lastSaneMinMulti)
1831 : : {
1832 : 2 : TransactionId nextXID = ReadNextTransactionId();
1833 : 2 : Relation relation;
1834 : 2 : TableScanDesc scan;
1835 : 2 : HeapTuple tuple;
1836 : 2 : Oid oldestxid_datoid;
1837 : 2 : Oid minmulti_datoid;
1838 : 2 : bool bogus = false;
1839 : 2 : bool frozenAlreadyWrapped = false;
1840 : :
1841 : : /* Restrict task to one backend per cluster; see SimpleLruTruncate(). */
1842 : 2 : LWLockAcquire(WrapLimitsVacuumLock, LW_EXCLUSIVE);
1843 : :
1844 : : /* init oldest datoids to sync with my frozenXID/minMulti values */
1845 : 2 : oldestxid_datoid = MyDatabaseId;
1846 : 2 : minmulti_datoid = MyDatabaseId;
1847 : :
1848 : : /*
1849 : : * Scan pg_database to compute the minimum datfrozenxid/datminmxid
1850 : : *
1851 : : * Since vac_update_datfrozenxid updates datfrozenxid/datminmxid in-place,
1852 : : * the values could change while we look at them. Fetch each one just
1853 : : * once to ensure sane behavior of the comparison logic. (Here, as in
1854 : : * many other places, we assume that fetching or updating an XID in shared
1855 : : * storage is atomic.)
1856 : : *
1857 : : * Note: we need not worry about a race condition with new entries being
1858 : : * inserted by CREATE DATABASE. Any such entry will have a copy of some
1859 : : * existing DB's datfrozenxid, and that source DB cannot be ours because
1860 : : * of the interlock against copying a DB containing an active backend.
1861 : : * Hence the new entry will not reduce the minimum. Also, if two VACUUMs
1862 : : * concurrently modify the datfrozenxid's of different databases, the
1863 : : * worst possible outcome is that pg_xact is not truncated as aggressively
1864 : : * as it could be.
1865 : : */
1866 : 2 : relation = table_open(DatabaseRelationId, AccessShareLock);
1867 : :
1868 : 2 : scan = table_beginscan_catalog(relation, 0, NULL);
1869 : :
1870 [ + + ]: 7 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1871 : : {
1872 : 5 : volatile FormData_pg_database *dbform = (Form_pg_database) GETSTRUCT(tuple);
1873 : 5 : TransactionId datfrozenxid = dbform->datfrozenxid;
1874 : 5 : TransactionId datminmxid = dbform->datminmxid;
1875 : :
1876 [ + - ]: 5 : Assert(TransactionIdIsNormal(datfrozenxid));
1877 [ + - ]: 5 : Assert(MultiXactIdIsValid(datminmxid));
1878 : :
1879 : : /*
1880 : : * If database is in the process of getting dropped, or has been
1881 : : * interrupted while doing so, no connections to it are possible
1882 : : * anymore. Therefore we don't need to take it into account here.
1883 : : * Which is good, because it can't be processed by autovacuum either.
1884 : : */
1885 [ - + ]: 5 : if (database_is_invalid_form((Form_pg_database) dbform))
1886 : : {
1887 [ # # # # ]: 0 : elog(DEBUG2,
1888 : : "skipping invalid database \"%s\" while computing relfrozenxid",
1889 : : NameStr(dbform->datname));
1890 : 0 : continue;
1891 : : }
1892 : :
1893 : : /*
1894 : : * If things are working properly, no database should have a
1895 : : * datfrozenxid or datminmxid that is "in the future". However, such
1896 : : * cases have been known to arise due to bugs in pg_upgrade. If we
1897 : : * see any entries that are "in the future", chicken out and don't do
1898 : : * anything. This ensures we won't truncate clog before those
1899 : : * databases have been scanned and cleaned up. (We will issue the
1900 : : * "already wrapped" warning if appropriate, though.)
1901 : : */
1902 [ + - - + ]: 5 : if (TransactionIdPrecedes(lastSaneFrozenXid, datfrozenxid) ||
1903 : 5 : MultiXactIdPrecedes(lastSaneMinMulti, datminmxid))
1904 : 0 : bogus = true;
1905 : :
1906 [ - + ]: 5 : if (TransactionIdPrecedes(nextXID, datfrozenxid))
1907 : 0 : frozenAlreadyWrapped = true;
1908 [ + + ]: 5 : else if (TransactionIdPrecedes(datfrozenxid, frozenXID))
1909 : : {
1910 : 1 : frozenXID = datfrozenxid;
1911 : 1 : oldestxid_datoid = dbform->oid;
1912 : 1 : }
1913 : :
1914 [ + - ]: 5 : if (MultiXactIdPrecedes(datminmxid, minMulti))
1915 : : {
1916 : 0 : minMulti = datminmxid;
1917 : 0 : minmulti_datoid = dbform->oid;
1918 : 0 : }
1919 [ - + ]: 5 : }
1920 : :
1921 : 2 : table_endscan(scan);
1922 : :
1923 : 2 : table_close(relation, AccessShareLock);
1924 : :
1925 : : /*
1926 : : * Do not truncate CLOG if we seem to have suffered wraparound already;
1927 : : * the computed minimum XID might be bogus. This case should now be
1928 : : * impossible due to the defenses in GetNewTransactionId, but we keep the
1929 : : * test anyway.
1930 : : */
1931 [ - + ]: 2 : if (frozenAlreadyWrapped)
1932 : : {
1933 [ # # # # ]: 0 : ereport(WARNING,
1934 : : (errmsg("some databases have not been vacuumed in over 2 billion transactions"),
1935 : : errdetail("You might have already suffered transaction-wraparound data loss.")));
1936 : 0 : LWLockRelease(WrapLimitsVacuumLock);
1937 : 0 : return;
1938 : : }
1939 : :
1940 : : /* chicken out if data is bogus in any other way */
1941 [ - + ]: 2 : if (bogus)
1942 : : {
1943 : 0 : LWLockRelease(WrapLimitsVacuumLock);
1944 : 0 : return;
1945 : : }
1946 : :
1947 : : /*
1948 : : * Freeze any old transaction IDs in the async notification queue before
1949 : : * CLOG truncation.
1950 : : */
1951 : 2 : AsyncNotifyFreezeXids(frozenXID);
1952 : :
1953 : : /*
1954 : : * Advance the oldest value for commit timestamps before truncating, so
1955 : : * that if a user requests a timestamp for a transaction we're truncating
1956 : : * away right after this point, they get NULL instead of an ugly "file not
1957 : : * found" error from slru.c. This doesn't matter for xact/multixact
1958 : : * because they are not subject to arbitrary lookups from users.
1959 : : */
1960 : 2 : AdvanceOldestCommitTsXid(frozenXID);
1961 : :
1962 : : /*
1963 : : * Truncate CLOG, multixact and CommitTs to the oldest computed value.
1964 : : */
1965 : 2 : TruncateCLOG(frozenXID, oldestxid_datoid);
1966 : 2 : TruncateCommitTs(frozenXID);
1967 : 2 : TruncateMultiXact(minMulti, minmulti_datoid);
1968 : :
1969 : : /*
1970 : : * Update the wrap limit for GetNewTransactionId and creation of new
1971 : : * MultiXactIds. Note: these functions will also signal the postmaster
1972 : : * for an(other) autovac cycle if needed. XXX should we avoid possibly
1973 : : * signaling twice?
1974 : : */
1975 : 2 : SetTransactionIdLimit(frozenXID, oldestxid_datoid);
1976 : 2 : SetMultiXactIdLimit(minMulti, minmulti_datoid);
1977 : :
1978 : 2 : LWLockRelease(WrapLimitsVacuumLock);
1979 : 2 : }
1980 : :
1981 : :
1982 : : /*
1983 : : * vacuum_rel() -- vacuum one heap relation
1984 : : *
1985 : : * relid identifies the relation to vacuum. If relation is supplied,
1986 : : * use the name therein for reporting any failure to open/lock the rel;
1987 : : * do not use it once we've successfully opened the rel, since it might
1988 : : * be stale.
1989 : : *
1990 : : * Returns true if it's okay to proceed with a requested ANALYZE
1991 : : * operation on this table.
1992 : : *
1993 : : * Doing one heap at a time incurs extra overhead, since we need to
1994 : : * check that the heap exists again just before we vacuum it. The
1995 : : * reason that we do this is so that vacuuming can be spread across
1996 : : * many small transactions. Otherwise, two-phase locking would require
1997 : : * us to lock the entire database during one pass of the vacuum cleaner.
1998 : : *
1999 : : * At entry and exit, we are not inside a transaction.
2000 : : */
2001 : : static bool
2002 : 718 : vacuum_rel(Oid relid, RangeVar *relation, VacuumParams params,
2003 : : BufferAccessStrategy bstrategy)
2004 : : {
2005 : 718 : LOCKMODE lmode;
2006 : 718 : Relation rel;
2007 : 718 : LockRelId lockrelid;
2008 : 718 : Oid priv_relid;
2009 : 718 : Oid toast_relid;
2010 : 718 : Oid save_userid;
2011 : 718 : int save_sec_context;
2012 : 718 : int save_nestlevel;
2013 : 718 : VacuumParams toast_vacuum_params;
2014 : :
2015 : : /*
2016 : : * This function scribbles on the parameters, so make a copy early to
2017 : : * avoid affecting the TOAST table (if we do end up recursing to it).
2018 : : */
2019 : 718 : memcpy(&toast_vacuum_params, ¶ms, sizeof(VacuumParams));
2020 : :
2021 : : /* Begin a transaction for vacuuming this relation */
2022 : 718 : StartTransactionCommand();
2023 : :
2024 [ + + ]: 718 : if (!(params.options & VACOPT_FULL))
2025 : : {
2026 : : /*
2027 : : * In lazy vacuum, we can set the PROC_IN_VACUUM flag, which lets
2028 : : * other concurrent VACUUMs know that they can ignore this one while
2029 : : * determining their OldestXmin. (The reason we don't set it during a
2030 : : * full VACUUM is exactly that we may have to run user-defined
2031 : : * functions for functional indexes, and we want to make sure that if
2032 : : * they use the snapshot set above, any tuples it requires can't get
2033 : : * removed from other tables. An index function that depends on the
2034 : : * contents of other tables is arguably broken, but we won't break it
2035 : : * here by violating transaction semantics.)
2036 : : *
2037 : : * We also set the VACUUM_FOR_WRAPAROUND flag, which is passed down by
2038 : : * autovacuum; it's used to avoid canceling a vacuum that was invoked
2039 : : * in an emergency.
2040 : : *
2041 : : * Note: these flags remain set until CommitTransaction or
2042 : : * AbortTransaction. We don't want to clear them until we reset
2043 : : * MyProc->xid/xmin, otherwise GetOldestNonRemovableTransactionId()
2044 : : * might appear to go backwards, which is probably Not Good. (We also
2045 : : * set PROC_IN_VACUUM *before* taking our own snapshot, so that our
2046 : : * xmin doesn't become visible ahead of setting the flag.)
2047 : : */
2048 : 687 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
2049 : 687 : MyProc->statusFlags |= PROC_IN_VACUUM;
2050 [ + - ]: 687 : if (params.is_wraparound)
2051 : 0 : MyProc->statusFlags |= PROC_VACUUM_FOR_WRAPAROUND;
2052 : 687 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
2053 : 687 : LWLockRelease(ProcArrayLock);
2054 : 687 : }
2055 : :
2056 : : /*
2057 : : * Need to acquire a snapshot to prevent pg_subtrans from being truncated,
2058 : : * cutoff xids in local memory wrapping around, and to have updated xmin
2059 : : * horizons.
2060 : : */
2061 : 718 : PushActiveSnapshot(GetTransactionSnapshot());
2062 : :
2063 : : /*
2064 : : * Check for user-requested abort. Note we want this to be inside a
2065 : : * transaction, so xact.c doesn't issue useless WARNING.
2066 : : */
2067 [ + - ]: 718 : CHECK_FOR_INTERRUPTS();
2068 : :
2069 : : /*
2070 : : * Determine the type of lock we want --- hard exclusive lock for a FULL
2071 : : * vacuum, but just ShareUpdateExclusiveLock for concurrent vacuum. Either
2072 : : * way, we can be sure that no other backend is vacuuming the same table.
2073 : : */
2074 : 718 : lmode = (params.options & VACOPT_FULL) ?
2075 : : AccessExclusiveLock : ShareUpdateExclusiveLock;
2076 : :
2077 : : /* open the relation and get the appropriate lock on it */
2078 : 1436 : rel = vacuum_open_relation(relid, relation, params.options,
2079 : 718 : params.log_vacuum_min_duration >= 0, lmode);
2080 : :
2081 : : /* leave if relation could not be opened or locked */
2082 [ + - ]: 718 : if (!rel)
2083 : : {
2084 : 0 : PopActiveSnapshot();
2085 : 0 : CommitTransactionCommand();
2086 : 0 : return false;
2087 : : }
2088 : :
2089 : : /*
2090 : : * When recursing to a TOAST table, check privileges on the parent. NB:
2091 : : * This is only safe to do because we hold a session lock on the main
2092 : : * relation that prevents concurrent deletion.
2093 : : */
2094 [ + + ]: 718 : if (OidIsValid(params.toast_parent))
2095 : 200 : priv_relid = params.toast_parent;
2096 : : else
2097 : 518 : priv_relid = RelationGetRelid(rel);
2098 : :
2099 : : /*
2100 : : * Check if relation needs to be skipped based on privileges. This check
2101 : : * happens also when building the relation list to vacuum for a manual
2102 : : * operation, and needs to be done additionally here as VACUUM could
2103 : : * happen across multiple transactions where privileges could have changed
2104 : : * in-between. Make sure to only generate logs for VACUUM in this case.
2105 : : */
2106 [ + + + + ]: 1436 : if (!vacuum_is_permitted_for_relation(priv_relid,
2107 : 718 : rel->rd_rel,
2108 : 718 : params.options & ~VACOPT_ANALYZE))
2109 : : {
2110 : 12 : relation_close(rel, lmode);
2111 : 12 : PopActiveSnapshot();
2112 : 12 : CommitTransactionCommand();
2113 : 12 : return false;
2114 : : }
2115 : :
2116 : : /*
2117 : : * Check that it's of a vacuumable relkind.
2118 : : */
2119 [ + + ]: 706 : if (rel->rd_rel->relkind != RELKIND_RELATION &&
2120 [ + + ]: 232 : rel->rd_rel->relkind != RELKIND_MATVIEW &&
2121 [ + + + - ]: 231 : rel->rd_rel->relkind != RELKIND_TOASTVALUE &&
2122 : 31 : rel->rd_rel->relkind != RELKIND_PARTITIONED_TABLE)
2123 : : {
2124 [ # # # # ]: 0 : ereport(WARNING,
2125 : : (errmsg("skipping \"%s\" --- cannot vacuum non-tables or special system tables",
2126 : : RelationGetRelationName(rel))));
2127 : 0 : relation_close(rel, lmode);
2128 : 0 : PopActiveSnapshot();
2129 : 0 : CommitTransactionCommand();
2130 : 0 : return false;
2131 : : }
2132 : :
2133 : : /*
2134 : : * Silently ignore tables that are temp tables of other backends ---
2135 : : * trying to vacuum these will lead to great unhappiness, since their
2136 : : * contents are probably not up-to-date on disk. (We don't throw a
2137 : : * warning here; it would just lead to chatter during a database-wide
2138 : : * VACUUM.)
2139 : : */
2140 [ + + + - ]: 706 : if (RELATION_IS_OTHER_TEMP(rel))
2141 : : {
2142 : 0 : relation_close(rel, lmode);
2143 : 0 : PopActiveSnapshot();
2144 : 0 : CommitTransactionCommand();
2145 : 0 : return false;
2146 : : }
2147 : :
2148 : : /*
2149 : : * Silently ignore partitioned tables as there is no work to be done. The
2150 : : * useful work is on their child partitions, which have been queued up for
2151 : : * us separately.
2152 : : */
2153 [ + + ]: 706 : if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
2154 : : {
2155 : 31 : relation_close(rel, lmode);
2156 : 31 : PopActiveSnapshot();
2157 : 31 : CommitTransactionCommand();
2158 : : /* It's OK to proceed with ANALYZE on this table */
2159 : 31 : return true;
2160 : : }
2161 : :
2162 : : /*
2163 : : * Get a session-level lock too. This will protect our access to the
2164 : : * relation across multiple transactions, so that we can vacuum the
2165 : : * relation's TOAST table (if any) secure in the knowledge that no one is
2166 : : * deleting the parent relation.
2167 : : *
2168 : : * NOTE: this cannot block, even if someone else is waiting for access,
2169 : : * because the lock manager knows that both lock requests are from the
2170 : : * same process.
2171 : : */
2172 : 675 : lockrelid = rel->rd_lockInfo.lockRelId;
2173 : 675 : LockRelationIdForSession(&lockrelid, lmode);
2174 : :
2175 : : /*
2176 : : * Set index_cleanup option based on index_cleanup reloption if it wasn't
2177 : : * specified in VACUUM command, or when running in an autovacuum worker
2178 : : */
2179 [ + + ]: 675 : if (params.index_cleanup == VACOPTVALUE_UNSPECIFIED)
2180 : : {
2181 : 668 : StdRdOptIndexCleanup vacuum_index_cleanup;
2182 : :
2183 [ + + ]: 668 : if (rel->rd_options == NULL)
2184 : 612 : vacuum_index_cleanup = STDRD_OPTION_VACUUM_INDEX_CLEANUP_AUTO;
2185 : : else
2186 : 56 : vacuum_index_cleanup =
2187 : 56 : ((StdRdOptions *) rel->rd_options)->vacuum_index_cleanup;
2188 : :
2189 [ + + ]: 668 : if (vacuum_index_cleanup == STDRD_OPTION_VACUUM_INDEX_CLEANUP_AUTO)
2190 : 662 : params.index_cleanup = VACOPTVALUE_AUTO;
2191 [ + + ]: 6 : else if (vacuum_index_cleanup == STDRD_OPTION_VACUUM_INDEX_CLEANUP_ON)
2192 : 3 : params.index_cleanup = VACOPTVALUE_ENABLED;
2193 : : else
2194 : : {
2195 [ + - ]: 3 : Assert(vacuum_index_cleanup ==
2196 : : STDRD_OPTION_VACUUM_INDEX_CLEANUP_OFF);
2197 : 3 : params.index_cleanup = VACOPTVALUE_DISABLED;
2198 : : }
2199 : 668 : }
2200 : :
2201 : : #ifdef USE_INJECTION_POINTS
2202 : : if (params.index_cleanup == VACOPTVALUE_AUTO)
2203 : : INJECTION_POINT("vacuum-index-cleanup-auto", NULL);
2204 : : else if (params.index_cleanup == VACOPTVALUE_DISABLED)
2205 : : INJECTION_POINT("vacuum-index-cleanup-disabled", NULL);
2206 : : else if (params.index_cleanup == VACOPTVALUE_ENABLED)
2207 : : INJECTION_POINT("vacuum-index-cleanup-enabled", NULL);
2208 : : #endif
2209 : :
2210 : : /*
2211 : : * Check if the vacuum_max_eager_freeze_failure_rate table storage
2212 : : * parameter was specified. This overrides the GUC value.
2213 : : */
2214 [ + + + - ]: 675 : if (rel->rd_options != NULL &&
2215 : 58 : ((StdRdOptions *) rel->rd_options)->vacuum_max_eager_freeze_failure_rate >= 0)
2216 : 0 : params.max_eager_freeze_failure_rate =
2217 : 0 : ((StdRdOptions *) rel->rd_options)->vacuum_max_eager_freeze_failure_rate;
2218 : :
2219 : : /*
2220 : : * Set truncate option based on truncate reloption or GUC if it wasn't
2221 : : * specified in VACUUM command, or when running in an autovacuum worker
2222 : : */
2223 [ + + ]: 675 : if (params.truncate == VACOPTVALUE_UNSPECIFIED)
2224 : : {
2225 : 670 : StdRdOptions *opts = (StdRdOptions *) rel->rd_options;
2226 : :
2227 [ + + + + ]: 670 : if (opts && opts->vacuum_truncate != PG_TERNARY_UNSET)
2228 : : {
2229 [ + + ]: 4 : if (opts->vacuum_truncate == PG_TERNARY_TRUE)
2230 : 1 : params.truncate = VACOPTVALUE_ENABLED;
2231 : : else
2232 : 3 : params.truncate = VACOPTVALUE_DISABLED;
2233 : 4 : }
2234 [ + + ]: 666 : else if (vacuum_truncate)
2235 : 663 : params.truncate = VACOPTVALUE_ENABLED;
2236 : : else
2237 : 3 : params.truncate = VACOPTVALUE_DISABLED;
2238 : 670 : }
2239 : :
2240 : : #ifdef USE_INJECTION_POINTS
2241 : : if (params.truncate == VACOPTVALUE_AUTO)
2242 : : INJECTION_POINT("vacuum-truncate-auto", NULL);
2243 : : else if (params.truncate == VACOPTVALUE_DISABLED)
2244 : : INJECTION_POINT("vacuum-truncate-disabled", NULL);
2245 : : else if (params.truncate == VACOPTVALUE_ENABLED)
2246 : : INJECTION_POINT("vacuum-truncate-enabled", NULL);
2247 : : #endif
2248 : :
2249 : : /*
2250 : : * Remember the relation's TOAST relation for later, if the caller asked
2251 : : * us to process it. In VACUUM FULL, though, the toast table is
2252 : : * automatically rebuilt by cluster_rel so we shouldn't recurse to it,
2253 : : * unless PROCESS_MAIN is disabled.
2254 : : */
2255 [ + + + + ]: 703 : if ((params.options & VACOPT_PROCESS_TOAST) != 0 &&
2256 [ + + ]: 673 : ((params.options & VACOPT_FULL) == 0 ||
2257 : 28 : (params.options & VACOPT_PROCESS_MAIN) == 0))
2258 : 646 : toast_relid = rel->rd_rel->reltoastrelid;
2259 : : else
2260 : 29 : toast_relid = InvalidOid;
2261 : :
2262 : : /*
2263 : : * Switch to the table owner's userid, so that any index functions are run
2264 : : * as that user. Also lock down security-restricted operations and
2265 : : * arrange to make GUC variable changes local to this command. (This is
2266 : : * unnecessary, but harmless, for lazy VACUUM.)
2267 : : */
2268 : 675 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
2269 : 1350 : SetUserIdAndSecContext(rel->rd_rel->relowner,
2270 : 675 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
2271 : 675 : save_nestlevel = NewGUCNestLevel();
2272 : 675 : RestrictSearchPath();
2273 : :
2274 : : /*
2275 : : * If PROCESS_MAIN is set (the default), it's time to vacuum the main
2276 : : * relation. Otherwise, we can skip this part. If processing the TOAST
2277 : : * table is required (e.g., PROCESS_TOAST is set), we force PROCESS_MAIN
2278 : : * to be set when we recurse to the TOAST table.
2279 : : */
2280 [ + + ]: 675 : if (params.options & VACOPT_PROCESS_MAIN)
2281 : : {
2282 : : /*
2283 : : * Do the actual work --- either FULL or "lazy" vacuum
2284 : : */
2285 [ + + ]: 672 : if (params.options & VACOPT_FULL)
2286 : : {
2287 : 27 : ClusterParams cluster_params = {0};
2288 : :
2289 [ + - ]: 27 : if ((params.options & VACOPT_VERBOSE) != 0)
2290 : 0 : cluster_params.options |= CLUOPT_VERBOSE;
2291 : :
2292 : : /* VACUUM FULL is now a variant of CLUSTER; see cluster.c */
2293 : 27 : cluster_rel(rel, InvalidOid, &cluster_params);
2294 : : /* cluster_rel closes the relation, but keeps lock */
2295 : :
2296 : 27 : rel = NULL;
2297 : 27 : }
2298 : : else
2299 : 645 : table_relation_vacuum(rel, params, bstrategy);
2300 : 672 : }
2301 : :
2302 : : /* Roll back any GUC changes executed by index functions */
2303 : 675 : AtEOXact_GUC(false, save_nestlevel);
2304 : :
2305 : : /* Restore userid and security context */
2306 : 675 : SetUserIdAndSecContext(save_userid, save_sec_context);
2307 : :
2308 : : /* all done with this class, but hold lock until commit */
2309 [ + + ]: 675 : if (rel)
2310 : 648 : relation_close(rel, NoLock);
2311 : :
2312 : : /*
2313 : : * Complete the transaction and free all temporary memory used.
2314 : : */
2315 : 675 : PopActiveSnapshot();
2316 : 675 : CommitTransactionCommand();
2317 : :
2318 : : /*
2319 : : * If the relation has a secondary toast rel, vacuum that too while we
2320 : : * still hold the session lock on the main table. Note however that
2321 : : * "analyze" will not get done on the toast table. This is good, because
2322 : : * the toaster always uses hardcoded index access and statistics are
2323 : : * totally unimportant for toast relations.
2324 : : */
2325 [ + + ]: 675 : if (toast_relid != InvalidOid)
2326 : : {
2327 : : /*
2328 : : * Force VACOPT_PROCESS_MAIN so vacuum_rel() processes it. Likewise,
2329 : : * set toast_parent so that the privilege checks are done on the main
2330 : : * relation. NB: This is only safe to do because we hold a session
2331 : : * lock on the main relation that prevents concurrent deletion.
2332 : : */
2333 : 200 : toast_vacuum_params.options |= VACOPT_PROCESS_MAIN;
2334 : 200 : toast_vacuum_params.toast_parent = relid;
2335 : :
2336 : 200 : vacuum_rel(toast_relid, NULL, toast_vacuum_params, bstrategy);
2337 : 200 : }
2338 : :
2339 : : /*
2340 : : * Now release the session-level lock on the main table.
2341 : : */
2342 : 675 : UnlockRelationIdForSession(&lockrelid, lmode);
2343 : :
2344 : : /* Report that we really did it. */
2345 : 675 : return true;
2346 : 718 : }
2347 : :
2348 : :
2349 : : /*
2350 : : * Open all the vacuumable indexes of the given relation, obtaining the
2351 : : * specified kind of lock on each. Return an array of Relation pointers for
2352 : : * the indexes into *Irel, and the number of indexes into *nindexes.
2353 : : *
2354 : : * We consider an index vacuumable if it is marked insertable (indisready).
2355 : : * If it isn't, probably a CREATE INDEX CONCURRENTLY command failed early in
2356 : : * execution, and what we have is too corrupt to be processable. We will
2357 : : * vacuum even if the index isn't indisvalid; this is important because in a
2358 : : * unique index, uniqueness checks will be performed anyway and had better not
2359 : : * hit dangling index pointers.
2360 : : */
2361 : : void
2362 : 1311 : vac_open_indexes(Relation relation, LOCKMODE lockmode,
2363 : : int *nindexes, Relation **Irel)
2364 : : {
2365 : 1311 : List *indexoidlist;
2366 : 1311 : ListCell *indexoidscan;
2367 : 1311 : int i;
2368 : :
2369 [ + - ]: 1311 : Assert(lockmode != NoLock);
2370 : :
2371 : 1311 : indexoidlist = RelationGetIndexList(relation);
2372 : :
2373 : : /* allocate enough memory for all indexes */
2374 : 1311 : i = list_length(indexoidlist);
2375 : :
2376 [ + + ]: 1311 : if (i > 0)
2377 : 724 : *Irel = (Relation *) palloc(i * sizeof(Relation));
2378 : : else
2379 : 587 : *Irel = NULL;
2380 : :
2381 : : /* collect just the ready indexes */
2382 : 1311 : i = 0;
2383 [ + + + + : 2349 : foreach(indexoidscan, indexoidlist)
+ + ]
2384 : : {
2385 : 1038 : Oid indexoid = lfirst_oid(indexoidscan);
2386 : 1038 : Relation indrel;
2387 : :
2388 : 1038 : indrel = index_open(indexoid, lockmode);
2389 [ + - ]: 1038 : if (indrel->rd_index->indisready)
2390 : 1038 : (*Irel)[i++] = indrel;
2391 : : else
2392 : 0 : index_close(indrel, lockmode);
2393 : 1038 : }
2394 : :
2395 : 1311 : *nindexes = i;
2396 : :
2397 : 1311 : list_free(indexoidlist);
2398 : 1311 : }
2399 : :
2400 : : /*
2401 : : * Release the resources acquired by vac_open_indexes. Optionally release
2402 : : * the locks (say NoLock to keep 'em).
2403 : : */
2404 : : void
2405 : 1452 : vac_close_indexes(int nindexes, Relation *Irel, LOCKMODE lockmode)
2406 : : {
2407 [ + + ]: 1452 : if (Irel == NULL)
2408 : 729 : return;
2409 : :
2410 [ + + ]: 1759 : while (nindexes--)
2411 : : {
2412 : 1036 : Relation ind = Irel[nindexes];
2413 : :
2414 : 1036 : index_close(ind, lockmode);
2415 : 1036 : }
2416 : 723 : pfree(Irel);
2417 : 1452 : }
2418 : :
2419 : : /*
2420 : : * vacuum_delay_point --- check for interrupts and cost-based delay.
2421 : : *
2422 : : * This should be called in each major loop of VACUUM processing,
2423 : : * typically once per page processed.
2424 : : */
2425 : : void
2426 : 3980462 : vacuum_delay_point(bool is_analyze)
2427 : : {
2428 : 3980462 : double msec = 0;
2429 : :
2430 : : /* Always check for interrupts */
2431 [ + - ]: 3980462 : CHECK_FOR_INTERRUPTS();
2432 : :
2433 [ + - - + ]: 7960924 : if (InterruptPending ||
2434 [ + - ]: 3980462 : (!VacuumCostActive && !ConfigReloadPending))
2435 : 3980462 : return;
2436 : :
2437 : : /*
2438 : : * Autovacuum workers should reload the configuration file if requested.
2439 : : * This allows changes to [autovacuum_]vacuum_cost_limit and
2440 : : * [autovacuum_]vacuum_cost_delay to take effect while a table is being
2441 : : * vacuumed or analyzed.
2442 : : */
2443 [ # # # # ]: 0 : if (ConfigReloadPending && AmAutoVacuumWorkerProcess())
2444 : : {
2445 : 0 : ConfigReloadPending = false;
2446 : 0 : ProcessConfigFile(PGC_SIGHUP);
2447 : 0 : VacuumUpdateCosts();
2448 : 0 : }
2449 : :
2450 : : /*
2451 : : * If we disabled cost-based delays after reloading the config file,
2452 : : * return.
2453 : : */
2454 [ # # ]: 0 : if (!VacuumCostActive)
2455 : 0 : return;
2456 : :
2457 : : /*
2458 : : * For parallel vacuum, the delay is computed based on the shared cost
2459 : : * balance. See compute_parallel_delay.
2460 : : */
2461 [ # # ]: 0 : if (VacuumSharedCostBalance != NULL)
2462 : 0 : msec = compute_parallel_delay();
2463 [ # # ]: 0 : else if (VacuumCostBalance >= vacuum_cost_limit)
2464 : 0 : msec = vacuum_cost_delay * VacuumCostBalance / vacuum_cost_limit;
2465 : :
2466 : : /* Nap if appropriate */
2467 [ # # ]: 0 : if (msec > 0)
2468 : : {
2469 : 0 : instr_time delay_start;
2470 : :
2471 [ # # ]: 0 : if (msec > vacuum_cost_delay * 4)
2472 : 0 : msec = vacuum_cost_delay * 4;
2473 : :
2474 [ # # ]: 0 : if (track_cost_delay_timing)
2475 : 0 : INSTR_TIME_SET_CURRENT(delay_start);
2476 : :
2477 : 0 : pgstat_report_wait_start(WAIT_EVENT_VACUUM_DELAY);
2478 : 0 : pg_usleep(msec * 1000);
2479 : 0 : pgstat_report_wait_end();
2480 : :
2481 [ # # ]: 0 : if (track_cost_delay_timing)
2482 : : {
2483 : 0 : instr_time delay_end;
2484 : 0 : instr_time delay;
2485 : :
2486 : 0 : INSTR_TIME_SET_CURRENT(delay_end);
2487 : 0 : INSTR_TIME_SET_ZERO(delay);
2488 : 0 : INSTR_TIME_ACCUM_DIFF(delay, delay_end, delay_start);
2489 : :
2490 : : /*
2491 : : * For parallel workers, we only report the delay time every once
2492 : : * in a while to avoid overloading the leader with messages and
2493 : : * interrupts.
2494 : : */
2495 [ # # ]: 0 : if (IsParallelWorker())
2496 : : {
2497 : : static instr_time last_report_time;
2498 : 0 : instr_time time_since_last_report;
2499 : :
2500 [ # # ]: 0 : Assert(!is_analyze);
2501 : :
2502 : : /* Accumulate the delay time */
2503 : 0 : parallel_vacuum_worker_delay_ns += INSTR_TIME_GET_NANOSEC(delay);
2504 : :
2505 : : /* Calculate interval since last report */
2506 : 0 : INSTR_TIME_SET_ZERO(time_since_last_report);
2507 : 0 : INSTR_TIME_ACCUM_DIFF(time_since_last_report, delay_end, last_report_time);
2508 : :
2509 : : /* If we haven't reported in a while, do so now */
2510 [ # # ]: 0 : if (INSTR_TIME_GET_NANOSEC(time_since_last_report) >=
2511 : : PARALLEL_VACUUM_DELAY_REPORT_INTERVAL_NS)
2512 : : {
2513 : 0 : pgstat_progress_parallel_incr_param(PROGRESS_VACUUM_DELAY_TIME,
2514 : 0 : parallel_vacuum_worker_delay_ns);
2515 : :
2516 : : /* Reset variables */
2517 : 0 : last_report_time = delay_end;
2518 : 0 : parallel_vacuum_worker_delay_ns = 0;
2519 : 0 : }
2520 : 0 : }
2521 [ # # ]: 0 : else if (is_analyze)
2522 : 0 : pgstat_progress_incr_param(PROGRESS_ANALYZE_DELAY_TIME,
2523 : 0 : INSTR_TIME_GET_NANOSEC(delay));
2524 : : else
2525 : 0 : pgstat_progress_incr_param(PROGRESS_VACUUM_DELAY_TIME,
2526 : 0 : INSTR_TIME_GET_NANOSEC(delay));
2527 : 0 : }
2528 : :
2529 : : /*
2530 : : * We don't want to ignore postmaster death during very long vacuums
2531 : : * with vacuum_cost_delay configured. We can't use the usual
2532 : : * WaitLatch() approach here because we want microsecond-based sleep
2533 : : * durations above.
2534 : : */
2535 [ # # # # ]: 0 : if (IsUnderPostmaster && !PostmasterIsAlive())
2536 : 0 : exit(1);
2537 : :
2538 : 0 : VacuumCostBalance = 0;
2539 : :
2540 : : /*
2541 : : * Balance and update limit values for autovacuum workers. We must do
2542 : : * this periodically, as the number of workers across which we are
2543 : : * balancing the limit may have changed.
2544 : : *
2545 : : * TODO: There may be better criteria for determining when to do this
2546 : : * besides "check after napping".
2547 : : */
2548 : 0 : AutoVacuumUpdateCostLimit();
2549 : :
2550 : : /* Might have gotten an interrupt while sleeping */
2551 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
2552 : 0 : }
2553 [ - + ]: 3980462 : }
2554 : :
2555 : : /*
2556 : : * Computes the vacuum delay for parallel workers.
2557 : : *
2558 : : * The basic idea of a cost-based delay for parallel vacuum is to allow each
2559 : : * worker to sleep in proportion to the share of work it's done. We achieve this
2560 : : * by allowing all parallel vacuum workers including the leader process to
2561 : : * have a shared view of cost related parameters (mainly VacuumCostBalance).
2562 : : * We allow each worker to update it as and when it has incurred any cost and
2563 : : * then based on that decide whether it needs to sleep. We compute the time
2564 : : * to sleep for a worker based on the cost it has incurred
2565 : : * (VacuumCostBalanceLocal) and then reduce the VacuumSharedCostBalance by
2566 : : * that amount. This avoids putting to sleep those workers which have done less
2567 : : * I/O than other workers and therefore ensure that workers
2568 : : * which are doing more I/O got throttled more.
2569 : : *
2570 : : * We allow a worker to sleep only if it has performed I/O above a certain
2571 : : * threshold, which is calculated based on the number of active workers
2572 : : * (VacuumActiveNWorkers), and the overall cost balance is more than
2573 : : * VacuumCostLimit set by the system. Testing reveals that we achieve
2574 : : * the required throttling if we force a worker that has done more than 50%
2575 : : * of its share of work to sleep.
2576 : : */
2577 : : static double
2578 : 0 : compute_parallel_delay(void)
2579 : : {
2580 : 0 : double msec = 0;
2581 : 0 : uint32 shared_balance;
2582 : 0 : int nworkers;
2583 : :
2584 : : /* Parallel vacuum must be active */
2585 [ # # ]: 0 : Assert(VacuumSharedCostBalance);
2586 : :
2587 : 0 : nworkers = pg_atomic_read_u32(VacuumActiveNWorkers);
2588 : :
2589 : : /* At least count itself */
2590 [ # # ]: 0 : Assert(nworkers >= 1);
2591 : :
2592 : : /* Update the shared cost balance value atomically */
2593 : 0 : shared_balance = pg_atomic_add_fetch_u32(VacuumSharedCostBalance, VacuumCostBalance);
2594 : :
2595 : : /* Compute the total local balance for the current worker */
2596 : 0 : VacuumCostBalanceLocal += VacuumCostBalance;
2597 : :
2598 [ # # # # ]: 0 : if ((shared_balance >= vacuum_cost_limit) &&
2599 : 0 : (VacuumCostBalanceLocal > 0.5 * ((double) vacuum_cost_limit / nworkers)))
2600 : : {
2601 : : /* Compute sleep time based on the local cost balance */
2602 : 0 : msec = vacuum_cost_delay * VacuumCostBalanceLocal / vacuum_cost_limit;
2603 : 0 : pg_atomic_sub_fetch_u32(VacuumSharedCostBalance, VacuumCostBalanceLocal);
2604 : 0 : VacuumCostBalanceLocal = 0;
2605 : 0 : }
2606 : :
2607 : : /*
2608 : : * Reset the local balance as we accumulated it into the shared value.
2609 : : */
2610 : 0 : VacuumCostBalance = 0;
2611 : :
2612 : 0 : return msec;
2613 : 0 : }
2614 : :
2615 : : /*
2616 : : * A wrapper function of defGetBoolean().
2617 : : *
2618 : : * This function returns VACOPTVALUE_ENABLED and VACOPTVALUE_DISABLED instead
2619 : : * of true and false.
2620 : : */
2621 : : static VacOptValue
2622 : 8 : get_vacoptval_from_boolean(DefElem *def)
2623 : : {
2624 : 8 : return defGetBoolean(def) ? VACOPTVALUE_ENABLED : VACOPTVALUE_DISABLED;
2625 : : }
2626 : :
2627 : : /*
2628 : : * vac_bulkdel_one_index() -- bulk-deletion for index relation.
2629 : : *
2630 : : * Returns bulk delete stats derived from input stats
2631 : : */
2632 : : IndexBulkDeleteResult *
2633 : 126 : vac_bulkdel_one_index(IndexVacuumInfo *ivinfo, IndexBulkDeleteResult *istat,
2634 : : TidStore *dead_items, VacDeadItemsInfo *dead_items_info)
2635 : : {
2636 : : /* Do bulk deletion */
2637 : 252 : istat = index_bulk_delete(ivinfo, istat, vac_tid_reaped,
2638 : 126 : dead_items);
2639 : :
2640 [ - + # # : 126 : ereport(ivinfo->message_level,
- + - + #
# ]
2641 : : (errmsg("scanned index \"%s\" to remove %" PRId64 " row versions",
2642 : : RelationGetRelationName(ivinfo->index),
2643 : : dead_items_info->num_items)));
2644 : :
2645 : 126 : return istat;
2646 : : }
2647 : :
2648 : : /*
2649 : : * vac_cleanup_one_index() -- do post-vacuum cleanup for index relation.
2650 : : *
2651 : : * Returns bulk delete stats derived from input stats
2652 : : */
2653 : : IndexBulkDeleteResult *
2654 : 626 : vac_cleanup_one_index(IndexVacuumInfo *ivinfo, IndexBulkDeleteResult *istat)
2655 : : {
2656 : 626 : istat = index_vacuum_cleanup(ivinfo, istat);
2657 : :
2658 [ + + ]: 626 : if (istat)
2659 [ - + # # : 155 : ereport(ivinfo->message_level,
- + - + #
# ]
2660 : : (errmsg("index \"%s\" now contains %.0f row versions in %u pages",
2661 : : RelationGetRelationName(ivinfo->index),
2662 : : istat->num_index_tuples,
2663 : : istat->num_pages),
2664 : : errdetail("%.0f index row versions were removed.\n"
2665 : : "%u index pages were newly deleted.\n"
2666 : : "%u index pages are currently deleted, of which %u are currently reusable.",
2667 : : istat->tuples_removed,
2668 : : istat->pages_newly_deleted,
2669 : : istat->pages_deleted, istat->pages_free)));
2670 : :
2671 : 626 : return istat;
2672 : : }
2673 : :
2674 : : /*
2675 : : * vac_tid_reaped() -- is a particular tid deletable?
2676 : : *
2677 : : * This has the right signature to be an IndexBulkDeleteCallback.
2678 : : */
2679 : : static bool
2680 : 299243 : vac_tid_reaped(ItemPointer itemptr, void *state)
2681 : : {
2682 : 299243 : TidStore *dead_items = (TidStore *) state;
2683 : :
2684 : 598486 : return TidStoreIsMember(dead_items, itemptr);
2685 : 299243 : }
|
