Age Owner Branch data TLA 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) 1996-2025, 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/cluster.h"
41 : : #include "commands/defrem.h"
42 : : #include "commands/progress.h"
43 : : #include "commands/vacuum.h"
44 : : #include "miscadmin.h"
45 : : #include "nodes/makefuncs.h"
46 : : #include "pgstat.h"
47 : : #include "postmaster/autovacuum.h"
48 : : #include "postmaster/bgworker_internals.h"
49 : : #include "postmaster/interrupt.h"
50 : : #include "storage/bufmgr.h"
51 : : #include "storage/lmgr.h"
52 : : #include "storage/pmsignal.h"
53 : : #include "storage/proc.h"
54 : : #include "storage/procarray.h"
55 : : #include "utils/acl.h"
56 : : #include "utils/fmgroids.h"
57 : : #include "utils/guc.h"
58 : : #include "utils/guc_hooks.h"
59 : : #include "utils/injection_point.h"
60 : : #include "utils/memutils.h"
61 : : #include "utils/snapmgr.h"
62 : : #include "utils/syscache.h"
63 : :
64 : : /*
65 : : * Minimum interval for cost-based vacuum delay reports from a parallel worker.
66 : : * This aims to avoid sending too many messages and waking up the leader too
67 : : * frequently.
68 : : */
69 : : #define PARALLEL_VACUUM_DELAY_REPORT_INTERVAL_NS (NS_PER_S)
70 : :
71 : : /*
72 : : * GUC parameters
73 : : */
74 : : int vacuum_freeze_min_age;
75 : : int vacuum_freeze_table_age;
76 : : int vacuum_multixact_freeze_min_age;
77 : : int vacuum_multixact_freeze_table_age;
78 : : int vacuum_failsafe_age;
79 : : int vacuum_multixact_failsafe_age;
80 : : double vacuum_max_eager_freeze_failure_rate;
81 : : bool track_cost_delay_timing;
82 : : bool vacuum_truncate;
83 : :
84 : : /*
85 : : * Variables for cost-based vacuum delay. The defaults differ between
86 : : * autovacuum and vacuum. They should be set with the appropriate GUC value in
87 : : * vacuum code. They are initialized here to the defaults for client backends
88 : : * executing VACUUM or ANALYZE.
89 : : */
90 : : double vacuum_cost_delay = 0;
91 : : int vacuum_cost_limit = 200;
92 : :
93 : : /* Variable for reporting cost-based vacuum delay from parallel workers. */
94 : : int64 parallel_vacuum_worker_delay_ns = 0;
95 : :
96 : : /*
97 : : * VacuumFailsafeActive is a defined as a global so that we can determine
98 : : * whether or not to re-enable cost-based vacuum delay when vacuuming a table.
99 : : * If failsafe mode has been engaged, we will not re-enable cost-based delay
100 : : * for the table until after vacuuming has completed, regardless of other
101 : : * settings.
102 : : *
103 : : * Only VACUUM code should inspect this variable and only table access methods
104 : : * should set it to true. In Table AM-agnostic VACUUM code, this variable is
105 : : * inspected to determine whether or not to allow cost-based delays. Table AMs
106 : : * are free to set it if they desire this behavior, but it is false by default
107 : : * and reset to false in between vacuuming each relation.
108 : : */
109 : : bool VacuumFailsafeActive = false;
110 : :
111 : : /*
112 : : * Variables for cost-based parallel vacuum. See comments atop
113 : : * compute_parallel_delay to understand how it works.
114 : : */
115 : : pg_atomic_uint32 *VacuumSharedCostBalance = NULL;
116 : : pg_atomic_uint32 *VacuumActiveNWorkers = NULL;
117 : : int VacuumCostBalanceLocal = 0;
118 : :
119 : : /* non-export function prototypes */
120 : : static List *expand_vacuum_rel(VacuumRelation *vrel,
121 : : MemoryContext vac_context, int options);
122 : : static List *get_all_vacuum_rels(MemoryContext vac_context, int options);
123 : : static void vac_truncate_clog(TransactionId frozenXID,
124 : : MultiXactId minMulti,
125 : : TransactionId lastSaneFrozenXid,
126 : : MultiXactId lastSaneMinMulti);
127 : : static bool vacuum_rel(Oid relid, RangeVar *relation, VacuumParams params,
128 : : BufferAccessStrategy bstrategy);
129 : : static double compute_parallel_delay(void);
130 : : static VacOptValue get_vacoptval_from_boolean(DefElem *def);
131 : : static bool vac_tid_reaped(ItemPointer itemptr, void *state);
132 : :
133 : : /*
134 : : * GUC check function to ensure GUC value specified is within the allowable
135 : : * range.
136 : : */
137 : : bool
883 drowley@postgresql.o 138 :CBC 1067 : check_vacuum_buffer_usage_limit(int *newval, void **extra,
139 : : GucSource source)
140 : : {
141 : : /* Value upper and lower hard limits are inclusive */
142 [ + - + - ]: 1067 : if (*newval == 0 || (*newval >= MIN_BAS_VAC_RING_SIZE_KB &&
143 [ + - ]: 1067 : *newval <= MAX_BAS_VAC_RING_SIZE_KB))
144 : 1067 : return true;
145 : :
146 : : /* Value does not fall within any allowable range */
283 alvherre@alvh.no-ip. 147 :UBC 0 : GUC_check_errdetail("\"%s\" must be 0 or between %d kB and %d kB.",
148 : : "vacuum_buffer_usage_limit",
149 : : MIN_BAS_VAC_RING_SIZE_KB, MAX_BAS_VAC_RING_SIZE_KB);
150 : :
883 drowley@postgresql.o 151 : 0 : return false;
152 : : }
153 : :
154 : : /*
155 : : * Primary entry point for manual VACUUM and ANALYZE commands
156 : : *
157 : : * This is mainly a preparation wrapper for the real operations that will
158 : : * happen in vacuum().
159 : : */
160 : : void
2364 rhaas@postgresql.org 161 :CBC 6770 : ExecVacuum(ParseState *pstate, VacuumStmt *vacstmt, bool isTopLevel)
162 : : {
163 : : VacuumParams params;
884 drowley@postgresql.o 164 : 6770 : BufferAccessStrategy bstrategy = NULL;
2299 tgl@sss.pgh.pa.us 165 : 6770 : bool verbose = false;
166 : 6770 : bool skip_locked = false;
167 : 6770 : bool analyze = false;
168 : 6770 : bool freeze = false;
169 : 6770 : bool full = false;
170 : 6770 : bool disable_page_skipping = false;
915 michael@paquier.xyz 171 : 6770 : bool process_main = true;
1670 172 : 6770 : bool process_toast = true;
173 : : int ring_size;
974 tgl@sss.pgh.pa.us 174 : 6770 : bool skip_database_stats = false;
175 : 6770 : bool only_database_stats = false;
176 : : MemoryContext vac_context;
177 : : ListCell *lc;
178 : :
179 : : /* index_cleanup and truncate values unspecified for now */
1541 pg@bowt.ie 180 : 6770 : params.index_cleanup = VACOPTVALUE_UNSPECIFIED;
181 : 6770 : params.truncate = VACOPTVALUE_UNSPECIFIED;
182 : :
183 : : /* By default parallel vacuum is enabled */
2056 akapila@postgresql.o 184 : 6770 : params.nworkers = 0;
185 : :
186 : : /* Will be set later if we recurse to a TOAST table. */
542 nathan@postgresql.or 187 : 6770 : params.toast_parent = InvalidOid;
188 : :
189 : : /*
190 : : * Set this to an invalid value so it is clear whether or not a
191 : : * BUFFER_USAGE_LIMIT was specified when making the access strategy.
192 : : */
883 drowley@postgresql.o 193 : 6770 : ring_size = -1;
194 : :
195 : : /* Parse options list */
2364 rhaas@postgresql.org 196 [ + + + + : 13968 : foreach(lc, vacstmt->options)
+ + ]
197 : : {
2299 tgl@sss.pgh.pa.us 198 : 7216 : DefElem *opt = (DefElem *) lfirst(lc);
199 : :
200 : : /* Parse common options for VACUUM and ANALYZE */
2364 rhaas@postgresql.org 201 [ + + ]: 7216 : if (strcmp(opt->defname, "verbose") == 0)
2353 202 : 17 : verbose = defGetBoolean(opt);
2364 203 [ + + ]: 7199 : else if (strcmp(opt->defname, "skip_locked") == 0)
2353 204 : 167 : skip_locked = defGetBoolean(opt);
883 drowley@postgresql.o 205 [ + + ]: 7032 : else if (strcmp(opt->defname, "buffer_usage_limit") == 0)
206 : : {
207 : : const char *hintmsg;
208 : : int result;
209 : : char *vac_buffer_size;
210 : :
211 : 27 : vac_buffer_size = defGetString(opt);
212 : :
213 : : /*
214 : : * Check that the specified value is valid and the size falls
215 : : * within the hard upper and lower limits if it is not 0.
216 : : */
879 217 [ + + ]: 27 : if (!parse_int(vac_buffer_size, &result, GUC_UNIT_KB, &hintmsg) ||
218 [ + + ]: 24 : (result != 0 &&
219 [ + + + + ]: 18 : (result < MIN_BAS_VAC_RING_SIZE_KB || result > MAX_BAS_VAC_RING_SIZE_KB)))
220 : : {
883 221 [ + - + + ]: 9 : ereport(ERROR,
222 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
223 : : errmsg("BUFFER_USAGE_LIMIT option must be 0 or between %d kB and %d kB",
224 : : MIN_BAS_VAC_RING_SIZE_KB, MAX_BAS_VAC_RING_SIZE_KB),
225 : : hintmsg ? errhint("%s", _(hintmsg)) : 0));
226 : : }
227 : :
228 : 18 : ring_size = result;
229 : : }
2364 rhaas@postgresql.org 230 [ + + ]: 7005 : else if (!vacstmt->is_vacuumcmd)
231 [ + - ]: 3 : ereport(ERROR,
232 : : (errcode(ERRCODE_SYNTAX_ERROR),
233 : : errmsg("unrecognized ANALYZE option \"%s\"", opt->defname),
234 : : parser_errposition(pstate, opt->location)));
235 : :
236 : : /* Parse options available on VACUUM */
237 [ + + ]: 7002 : else if (strcmp(opt->defname, "analyze") == 0)
2353 238 : 1375 : analyze = defGetBoolean(opt);
2364 239 [ + + ]: 5627 : else if (strcmp(opt->defname, "freeze") == 0)
2353 240 : 1270 : freeze = defGetBoolean(opt);
2364 241 [ + + ]: 4357 : else if (strcmp(opt->defname, "full") == 0)
2353 242 : 185 : full = defGetBoolean(opt);
2364 243 [ + + ]: 4172 : else if (strcmp(opt->defname, "disable_page_skipping") == 0)
2353 244 : 101 : disable_page_skipping = defGetBoolean(opt);
2347 245 [ + + ]: 4071 : else if (strcmp(opt->defname, "index_cleanup") == 0)
246 : : {
247 : : /* Interpret no string as the default, which is 'auto' */
1541 pg@bowt.ie 248 [ - + ]: 87 : if (!opt->arg)
1541 pg@bowt.ie 249 :UBC 0 : params.index_cleanup = VACOPTVALUE_AUTO;
250 : : else
251 : : {
1541 pg@bowt.ie 252 :CBC 87 : char *sval = defGetString(opt);
253 : :
254 : : /* Try matching on 'auto' string, or fall back on boolean */
255 [ + + ]: 87 : if (pg_strcasecmp(sval, "auto") == 0)
256 : 3 : params.index_cleanup = VACOPTVALUE_AUTO;
257 : : else
258 : 84 : params.index_cleanup = get_vacoptval_from_boolean(opt);
259 : : }
260 : : }
915 michael@paquier.xyz 261 [ + + ]: 3984 : else if (strcmp(opt->defname, "process_main") == 0)
262 : 77 : process_main = defGetBoolean(opt);
1670 263 [ + + ]: 3907 : else if (strcmp(opt->defname, "process_toast") == 0)
264 : 80 : process_toast = defGetBoolean(opt);
2313 fujii@postgresql.org 265 [ + + ]: 3827 : else if (strcmp(opt->defname, "truncate") == 0)
1541 pg@bowt.ie 266 : 79 : params.truncate = get_vacoptval_from_boolean(opt);
2056 akapila@postgresql.o 267 [ + + ]: 3748 : else if (strcmp(opt->defname, "parallel") == 0)
268 : : {
269 [ + + ]: 176 : if (opt->arg == NULL)
270 : : {
271 [ + - ]: 3 : ereport(ERROR,
272 : : (errcode(ERRCODE_SYNTAX_ERROR),
273 : : errmsg("parallel option requires a value between 0 and %d",
274 : : MAX_PARALLEL_WORKER_LIMIT),
275 : : parser_errposition(pstate, opt->location)));
276 : : }
277 : : else
278 : : {
279 : : int nworkers;
280 : :
281 : 173 : nworkers = defGetInt32(opt);
282 [ + + - + ]: 173 : if (nworkers < 0 || nworkers > MAX_PARALLEL_WORKER_LIMIT)
283 [ + - ]: 3 : ereport(ERROR,
284 : : (errcode(ERRCODE_SYNTAX_ERROR),
285 : : errmsg("parallel workers for vacuum must be between 0 and %d",
286 : : MAX_PARALLEL_WORKER_LIMIT),
287 : : parser_errposition(pstate, opt->location)));
288 : :
289 : : /*
290 : : * Disable parallel vacuum, if user has specified parallel
291 : : * degree as zero.
292 : : */
293 [ + + ]: 170 : if (nworkers == 0)
294 : 78 : params.nworkers = -1;
295 : : else
296 : 92 : params.nworkers = nworkers;
297 : : }
298 : : }
974 tgl@sss.pgh.pa.us 299 [ + + ]: 3572 : else if (strcmp(opt->defname, "skip_database_stats") == 0)
300 : 3491 : skip_database_stats = defGetBoolean(opt);
301 [ + - ]: 81 : else if (strcmp(opt->defname, "only_database_stats") == 0)
302 : 81 : only_database_stats = defGetBoolean(opt);
303 : : else
2364 rhaas@postgresql.org 304 [ # # ]:UBC 0 : ereport(ERROR,
305 : : (errcode(ERRCODE_SYNTAX_ERROR),
306 : : errmsg("unrecognized VACUUM option \"%s\"", opt->defname),
307 : : parser_errposition(pstate, opt->location)));
308 : : }
309 : :
310 : : /* Set vacuum options */
2353 rhaas@postgresql.org 311 :CBC 6752 : params.options =
312 [ + + ]: 6752 : (vacstmt->is_vacuumcmd ? VACOPT_VACUUM : VACOPT_ANALYZE) |
313 [ + + ]: 6752 : (verbose ? VACOPT_VERBOSE : 0) |
314 [ + + ]: 6752 : (skip_locked ? VACOPT_SKIP_LOCKED : 0) |
315 [ + + ]: 6752 : (analyze ? VACOPT_ANALYZE : 0) |
316 [ + + ]: 6752 : (freeze ? VACOPT_FREEZE : 0) |
317 [ + + ]: 6752 : (full ? VACOPT_FULL : 0) |
1670 michael@paquier.xyz 318 [ + + ]: 6752 : (disable_page_skipping ? VACOPT_DISABLE_PAGE_SKIPPING : 0) |
915 319 [ + + ]: 6752 : (process_main ? VACOPT_PROCESS_MAIN : 0) |
974 tgl@sss.pgh.pa.us 320 [ + + ]: 6752 : (process_toast ? VACOPT_PROCESS_TOAST : 0) |
321 [ + + ]: 6752 : (skip_database_stats ? VACOPT_SKIP_DATABASE_STATS : 0) |
322 [ + + ]: 6752 : (only_database_stats ? VACOPT_ONLY_DATABASE_STATS : 0);
323 : :
324 : : /* sanity checks on options */
2364 rhaas@postgresql.org 325 [ - + ]: 6752 : Assert(params.options & (VACOPT_VACUUM | VACOPT_ANALYZE));
326 [ + + - + ]: 6752 : Assert((params.options & VACOPT_VACUUM) ||
327 : : !(params.options & (VACOPT_FULL | VACOPT_FREEZE)));
328 : :
1969 akapila@postgresql.o 329 [ + + + + ]: 6752 : if ((params.options & VACOPT_FULL) && params.nworkers > 0)
2056 330 [ + - ]: 3 : ereport(ERROR,
331 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
332 : : errmsg("VACUUM FULL cannot be performed in parallel")));
333 : :
334 : : /*
335 : : * BUFFER_USAGE_LIMIT does nothing for VACUUM (FULL) so just raise an
336 : : * ERROR for that case. VACUUM (FULL, ANALYZE) does make use of it, so
337 : : * we'll permit that.
338 : : */
883 drowley@postgresql.o 339 [ + + + + ]: 6749 : if (ring_size != -1 && (params.options & VACOPT_FULL) &&
340 [ + - ]: 3 : !(params.options & VACOPT_ANALYZE))
341 [ + - ]: 3 : ereport(ERROR,
342 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
343 : : errmsg("BUFFER_USAGE_LIMIT cannot be specified for VACUUM FULL")));
344 : :
345 : : /*
346 : : * Make sure VACOPT_ANALYZE is specified if any column lists are present.
347 : : */
2364 rhaas@postgresql.org 348 [ + + ]: 6746 : if (!(params.options & VACOPT_ANALYZE))
349 : : {
2895 tgl@sss.pgh.pa.us 350 [ + + + + : 5936 : foreach(lc, vacstmt->rels)
+ + ]
351 : : {
352 : 2912 : VacuumRelation *vrel = lfirst_node(VacuumRelation, lc);
353 : :
354 [ + + ]: 2912 : if (vrel->va_cols != NIL)
355 [ + - ]: 3 : ereport(ERROR,
356 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
357 : : errmsg("ANALYZE option must be specified when a column list is provided")));
358 : : }
359 : : }
360 : :
361 : :
362 : : /*
363 : : * Sanity check DISABLE_PAGE_SKIPPING option.
364 : : */
884 drowley@postgresql.o 365 [ + + ]: 6743 : if ((params.options & VACOPT_FULL) != 0 &&
366 [ - + ]: 173 : (params.options & VACOPT_DISABLE_PAGE_SKIPPING) != 0)
884 drowley@postgresql.o 367 [ # # ]:UBC 0 : ereport(ERROR,
368 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
369 : : errmsg("VACUUM option DISABLE_PAGE_SKIPPING cannot be used with FULL")));
370 : :
371 : : /* sanity check for PROCESS_TOAST */
884 drowley@postgresql.o 372 [ + + ]:CBC 6743 : if ((params.options & VACOPT_FULL) != 0 &&
373 [ + + ]: 173 : (params.options & VACOPT_PROCESS_TOAST) == 0)
374 [ + - ]: 3 : ereport(ERROR,
375 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
376 : : errmsg("PROCESS_TOAST required with VACUUM FULL")));
377 : :
378 : : /* sanity check for ONLY_DATABASE_STATS */
379 [ + + ]: 6740 : if (params.options & VACOPT_ONLY_DATABASE_STATS)
380 : : {
381 [ - + ]: 81 : Assert(params.options & VACOPT_VACUUM);
382 [ + + ]: 81 : if (vacstmt->rels != NIL)
383 [ + - ]: 3 : ereport(ERROR,
384 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
385 : : errmsg("ONLY_DATABASE_STATS cannot be specified with a list of tables")));
386 : : /* don't require people to turn off PROCESS_TOAST/MAIN explicitly */
387 [ - + ]: 78 : if (params.options & ~(VACOPT_VACUUM |
388 : : VACOPT_VERBOSE |
389 : : VACOPT_PROCESS_MAIN |
390 : : VACOPT_PROCESS_TOAST |
391 : : VACOPT_ONLY_DATABASE_STATS))
884 drowley@postgresql.o 392 [ # # ]:UBC 0 : ereport(ERROR,
393 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
394 : : errmsg("ONLY_DATABASE_STATS cannot be specified with other VACUUM options")));
395 : : }
396 : :
397 : : /*
398 : : * All freeze ages are zero if the FREEZE option is given; otherwise pass
399 : : * them as -1 which means to use the default values.
400 : : */
2364 rhaas@postgresql.org 401 [ + + ]:CBC 6737 : if (params.options & VACOPT_FREEZE)
402 : : {
3825 alvherre@alvh.no-ip. 403 : 1270 : params.freeze_min_age = 0;
404 : 1270 : params.freeze_table_age = 0;
405 : 1270 : params.multixact_freeze_min_age = 0;
406 : 1270 : params.multixact_freeze_table_age = 0;
407 : : }
408 : : else
409 : : {
410 : 5467 : params.freeze_min_age = -1;
411 : 5467 : params.freeze_table_age = -1;
412 : 5467 : params.multixact_freeze_min_age = -1;
413 : 5467 : params.multixact_freeze_table_age = -1;
414 : : }
415 : :
416 : : /* user-invoked vacuum is never "for wraparound" */
417 : 6737 : params.is_wraparound = false;
418 : :
419 : : /* user-invoked vacuum uses VACOPT_VERBOSE instead of log_min_duration */
3809 420 : 6737 : params.log_min_duration = -1;
421 : :
422 : : /*
423 : : * Later, in vacuum_rel(), we check if a reloption override was specified.
424 : : */
207 melanieplageman@gmai 425 : 6737 : params.max_eager_freeze_failure_rate = vacuum_max_eager_freeze_failure_rate;
426 : :
427 : : /*
428 : : * Create special memory context for cross-transaction storage.
429 : : *
430 : : * Since it is a child of PortalContext, it will go away eventually even
431 : : * if we suffer an error; there's no need for special abort cleanup logic.
432 : : */
884 drowley@postgresql.o 433 : 6737 : vac_context = AllocSetContextCreate(PortalContext,
434 : : "Vacuum",
435 : : ALLOCSET_DEFAULT_SIZES);
436 : :
437 : : /*
438 : : * Make a buffer strategy object in the cross-transaction memory context.
439 : : * We needn't bother making this for VACUUM (FULL) or VACUUM
440 : : * (ONLY_DATABASE_STATS) as they'll not make use of it. VACUUM (FULL,
441 : : * ANALYZE) is possible, so we'd better ensure that we make a strategy
442 : : * when we see ANALYZE.
443 : : */
444 [ + + ]: 6737 : if ((params.options & (VACOPT_ONLY_DATABASE_STATS |
445 : 248 : VACOPT_FULL)) == 0 ||
446 [ + + ]: 248 : (params.options & VACOPT_ANALYZE) != 0)
447 : : {
448 : :
449 : 6492 : MemoryContext old_context = MemoryContextSwitchTo(vac_context);
450 : :
883 451 [ - + ]: 6492 : Assert(ring_size >= -1);
452 : :
453 : : /*
454 : : * If BUFFER_USAGE_LIMIT was specified by the VACUUM or ANALYZE
455 : : * command, it overrides the value of VacuumBufferUsageLimit. Either
456 : : * value may be 0, in which case GetAccessStrategyWithSize() will
457 : : * return NULL, effectively allowing full use of shared buffers.
458 : : */
459 [ + + ]: 6492 : if (ring_size == -1)
460 : 6477 : ring_size = VacuumBufferUsageLimit;
461 : :
462 : 6492 : bstrategy = GetAccessStrategyWithSize(BAS_VACUUM, ring_size);
463 : :
884 464 : 6492 : MemoryContextSwitchTo(old_context);
465 : : }
466 : :
467 : : /* Now go through the common routine */
68 michael@paquier.xyz 468 :GNC 6737 : vacuum(vacstmt->rels, params, bstrategy, vac_context, isTopLevel);
469 : :
470 : : /* Finally, clean up the vacuum memory context */
884 drowley@postgresql.o 471 :CBC 6671 : MemoryContextDelete(vac_context);
3825 alvherre@alvh.no-ip. 472 : 6671 : }
473 : :
474 : : /*
475 : : * Internal entry point for autovacuum and the VACUUM / ANALYZE commands.
476 : : *
477 : : * relations, if not NIL, is a list of VacuumRelation to process; otherwise,
478 : : * we process all relevant tables in the database. For each VacuumRelation,
479 : : * if a valid OID is supplied, the table with that OID is what to process;
480 : : * otherwise, the VacuumRelation's RangeVar indicates what to process.
481 : : *
482 : : * params contains a set of parameters that can be used to customize the
483 : : * behavior.
484 : : *
485 : : * bstrategy may be passed in as NULL when the caller does not want to
486 : : * restrict the number of shared_buffers that VACUUM / ANALYZE can use,
487 : : * otherwise, the caller must build a BufferAccessStrategy with the number of
488 : : * shared_buffers that VACUUM / ANALYZE should try to limit themselves to
489 : : * using.
490 : : *
491 : : * isTopLevel should be passed down from ProcessUtility.
492 : : *
493 : : * It is the caller's responsibility that all parameters are allocated in a
494 : : * memory context that will not disappear at transaction commit.
495 : : */
496 : : void
68 michael@paquier.xyz 497 :GNC 7055 : vacuum(List *relations, const VacuumParams params, BufferAccessStrategy bstrategy,
498 : : MemoryContext vac_context, bool isTopLevel)
499 : : {
500 : : static bool in_vacuum = false;
501 : :
502 : : const char *stmttype;
503 : : volatile bool in_outer_xact,
504 : : use_own_xacts;
505 : :
506 [ + + ]: 7055 : stmttype = (params.options & VACOPT_VACUUM) ? "VACUUM" : "ANALYZE";
507 : :
508 : : /*
509 : : * We cannot run VACUUM inside a user transaction block; if we were inside
510 : : * a transaction, then our commit- and start-transaction-command calls
511 : : * would not have the intended effect! There are numerous other subtle
512 : : * dependencies on this, too.
513 : : *
514 : : * ANALYZE (without VACUUM) can run either way.
515 : : */
516 [ + + ]: 7055 : if (params.options & VACOPT_VACUUM)
517 : : {
2759 peter_e@gmx.net 518 :CBC 4560 : PreventInTransactionBlock(isTopLevel, stmttype);
7777 tgl@sss.pgh.pa.us 519 : 4550 : in_outer_xact = false;
520 : : }
521 : : else
2759 peter_e@gmx.net 522 : 2495 : in_outer_xact = IsInTransactionBlock(isTopLevel);
523 : :
524 : : /*
525 : : * Check for and disallow recursive calls. This could happen when VACUUM
526 : : * FULL or ANALYZE calls a hostile index expression that itself calls
527 : : * ANALYZE.
528 : : */
3895 noah@leadboat.com 529 [ + + ]: 7045 : if (in_vacuum)
3688 tgl@sss.pgh.pa.us 530 [ + - ]: 6 : ereport(ERROR,
531 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
532 : : errmsg("%s cannot be executed from VACUUM or ANALYZE",
533 : : stmttype)));
534 : :
535 : : /*
536 : : * Build list of relation(s) to process, putting any new data in
537 : : * vac_context for safekeeping.
538 : : */
68 michael@paquier.xyz 539 [ + + ]:GNC 7039 : if (params.options & VACOPT_ONLY_DATABASE_STATS)
540 : : {
541 : : /* We don't process any tables in this case */
974 tgl@sss.pgh.pa.us 542 [ - + ]:CBC 78 : Assert(relations == NIL);
543 : : }
544 [ + + ]: 6961 : else if (relations != NIL)
545 : : {
2895 546 : 6853 : List *newrels = NIL;
547 : : ListCell *lc;
548 : :
549 [ + - + + : 13757 : foreach(lc, relations)
+ + ]
550 : : {
551 : 6922 : VacuumRelation *vrel = lfirst_node(VacuumRelation, lc);
552 : : List *sublist;
553 : : MemoryContext old_context;
554 : :
68 michael@paquier.xyz 555 :GNC 6922 : sublist = expand_vacuum_rel(vrel, vac_context, params.options);
2895 tgl@sss.pgh.pa.us 556 :CBC 6904 : old_context = MemoryContextSwitchTo(vac_context);
557 : 6904 : newrels = list_concat(newrels, sublist);
558 : 6904 : MemoryContextSwitchTo(old_context);
559 : : }
560 : 6835 : relations = newrels;
561 : : }
562 : : else
68 michael@paquier.xyz 563 :GNC 108 : relations = get_all_vacuum_rels(vac_context, params.options);
564 : :
565 : : /*
566 : : * Decide whether we need to start/commit our own transactions.
567 : : *
568 : : * For VACUUM (with or without ANALYZE): always do so, so that we can
569 : : * release locks as soon as possible. (We could possibly use the outer
570 : : * transaction for a one-table VACUUM, but handling TOAST tables would be
571 : : * problematic.)
572 : : *
573 : : * For ANALYZE (no VACUUM): if inside a transaction block, we cannot
574 : : * start/commit our own transactions. Also, there's no need to do so if
575 : : * only processing one relation. For multiple relations when not within a
576 : : * transaction block, and also in an autovacuum worker, use own
577 : : * transactions so we can release locks sooner.
578 : : */
579 [ + + ]: 7021 : if (params.options & VACOPT_VACUUM)
7777 tgl@sss.pgh.pa.us 580 :CBC 4544 : use_own_xacts = true;
581 : : else
582 : : {
68 michael@paquier.xyz 583 [ - + ]:GNC 2477 : Assert(params.options & VACOPT_ANALYZE);
551 heikki.linnakangas@i 584 [ + + ]:CBC 2477 : if (AmAutoVacuumWorkerProcess())
6659 alvherre@alvh.no-ip. 585 : 151 : use_own_xacts = true;
586 [ + + ]: 2326 : else if (in_outer_xact)
7777 tgl@sss.pgh.pa.us 587 : 119 : use_own_xacts = false;
7773 neilc@samurai.com 588 [ + + ]: 2207 : else if (list_length(relations) > 1)
7777 tgl@sss.pgh.pa.us 589 : 393 : use_own_xacts = true;
590 : : else
591 : 1814 : use_own_xacts = false;
592 : : }
593 : :
594 : : /*
595 : : * vacuum_rel expects to be entered with no transaction active; it will
596 : : * start and commit its own transaction. But we are called by an SQL
597 : : * command, and so we are executing inside a transaction already. We
598 : : * commit the transaction started in PostgresMain() here, and start
599 : : * another one before exiting to match the commit waiting for us back in
600 : : * PostgresMain().
601 : : */
602 [ + + ]: 7021 : if (use_own_xacts)
603 : : {
3964 604 [ - + ]: 5088 : Assert(!in_outer_xact);
605 : :
606 : : /* ActiveSnapshot is not set by autovacuum */
6326 alvherre@alvh.no-ip. 607 [ + + ]: 5088 : if (ActiveSnapshotSet())
608 : 4770 : PopActiveSnapshot();
609 : :
610 : : /* matches the StartTransaction in PostgresMain() */
8151 tgl@sss.pgh.pa.us 611 : 5088 : CommitTransactionCommand();
612 : : }
613 : :
614 : : /* Turn vacuum cost accounting on or off, and set/clear in_vacuum */
7707 615 [ + + ]: 7021 : PG_TRY();
616 : : {
617 : : ListCell *cur;
618 : :
3895 noah@leadboat.com 619 : 7021 : in_vacuum = true;
883 dgustafsson@postgres 620 : 7021 : VacuumFailsafeActive = false;
621 : 7021 : VacuumUpdateCosts();
7707 tgl@sss.pgh.pa.us 622 : 7021 : VacuumCostBalance = 0;
2056 akapila@postgresql.o 623 : 7021 : VacuumCostBalanceLocal = 0;
624 : 7021 : VacuumSharedCostBalance = NULL;
625 : 7021 : VacuumActiveNWorkers = NULL;
626 : :
627 : : /*
628 : : * Loop to process each selected relation.
629 : : */
7707 tgl@sss.pgh.pa.us 630 [ + + + + : 22447 : foreach(cur, relations)
+ + ]
631 : : {
2895 632 : 15458 : VacuumRelation *vrel = lfirst_node(VacuumRelation, cur);
633 : :
68 michael@paquier.xyz 634 [ + + ]:GNC 15458 : if (params.options & VACOPT_VACUUM)
635 : : {
636 [ + + ]: 8868 : if (!vacuum_rel(vrel->oid, vrel->relation, params, bstrategy))
5325 rhaas@postgresql.org 637 :CBC 50 : continue;
638 : : }
639 : :
68 michael@paquier.xyz 640 [ + + ]:GNC 15405 : if (params.options & VACOPT_ANALYZE)
641 : : {
642 : : /*
643 : : * If using separate xacts, start one for analyze. Otherwise,
644 : : * we can use the outer transaction.
645 : : */
7707 tgl@sss.pgh.pa.us 646 [ + + ]:CBC 8118 : if (use_own_xacts)
647 : : {
648 : 6210 : StartTransactionCommand();
649 : : /* functions in indexes may want a snapshot set */
6326 alvherre@alvh.no-ip. 650 : 6210 : PushActiveSnapshot(GetTransactionSnapshot());
651 : : }
652 : :
2364 rhaas@postgresql.org 653 : 8118 : analyze_rel(vrel->oid, vrel->relation, params,
654 : : vrel->va_cols, in_outer_xact, bstrategy);
655 : :
7707 tgl@sss.pgh.pa.us 656 [ + + ]: 8089 : if (use_own_xacts)
657 : : {
6326 alvherre@alvh.no-ip. 658 : 6191 : PopActiveSnapshot();
659 : : /* standard_ProcessUtility() does CCI if !use_own_xacts */
139 noah@leadboat.com 660 : 6191 : CommandCounterIncrement();
7707 tgl@sss.pgh.pa.us 661 : 6191 : CommitTransactionCommand();
662 : : }
663 : : else
664 : : {
665 : : /*
666 : : * If we're not using separate xacts, better separate the
667 : : * ANALYZE actions with CCIs. This avoids trouble if user
668 : : * says "ANALYZE t, t".
669 : : */
2219 670 : 1898 : CommandCounterIncrement();
671 : : }
672 : : }
673 : :
674 : : /*
675 : : * Ensure VacuumFailsafeActive has been reset before vacuuming the
676 : : * next relation.
677 : : */
883 dgustafsson@postgres 678 : 15376 : VacuumFailsafeActive = false;
679 : : }
680 : : }
2136 peter@eisentraut.org 681 : 32 : PG_FINALLY();
682 : : {
3895 noah@leadboat.com 683 : 7021 : in_vacuum = false;
7707 tgl@sss.pgh.pa.us 684 : 7021 : VacuumCostActive = false;
883 dgustafsson@postgres 685 : 7021 : VacuumFailsafeActive = false;
686 : 7021 : VacuumCostBalance = 0;
687 : : }
7707 tgl@sss.pgh.pa.us 688 [ + + ]: 7021 : PG_END_TRY();
689 : :
690 : : /*
691 : : * Finish up processing.
692 : : */
7777 693 [ + + ]: 6989 : if (use_own_xacts)
694 : : {
695 : : /* here, we are not in a transaction */
696 : :
697 : : /*
698 : : * This matches the CommitTransaction waiting for us in
699 : : * PostgresMain().
700 : : */
8151 701 : 5066 : StartTransactionCommand();
702 : : }
703 : :
68 michael@paquier.xyz 704 [ + + ]:GNC 6989 : if ((params.options & VACOPT_VACUUM) &&
705 [ + + ]: 4528 : !(params.options & VACOPT_SKIP_DATABASE_STATS))
706 : : {
707 : : /*
708 : : * Update pg_database.datfrozenxid, and truncate pg_xact if possible.
709 : : */
6880 tgl@sss.pgh.pa.us 710 :CBC 871 : vac_update_datfrozenxid();
711 : : }
712 : :
10651 scrappy@hub.org 713 : 6989 : }
714 : :
715 : : /*
716 : : * Check if the current user has privileges to vacuum or analyze the relation.
717 : : * If not, issue a WARNING log message and return false to let the caller
718 : : * decide what to do with this relation. This routine is used to decide if a
719 : : * relation can be processed for VACUUM or ANALYZE.
720 : : */
721 : : bool
542 nathan@postgresql.or 722 : 37109 : vacuum_is_permitted_for_relation(Oid relid, Form_pg_class reltuple,
723 : : bits32 options)
724 : : {
725 : : char *relname;
726 : :
2567 michael@paquier.xyz 727 [ - + ]: 37109 : Assert((options & (VACOPT_VACUUM | VACOPT_ANALYZE)) != 0);
728 : :
729 : : /*----------
730 : : * A role has privileges to vacuum or analyze the relation if any of the
731 : : * following are true:
732 : : * - the role owns the current database and the relation is not shared
733 : : * - the role has the MAINTAIN privilege on the relation
734 : : *----------
735 : : */
542 nathan@postgresql.or 736 [ + + ]: 37109 : if ((object_ownercheck(DatabaseRelationId, MyDatabaseId, GetUserId()) &&
737 [ + + + + ]: 42089 : !reltuple->relisshared) ||
738 : 6759 : pg_class_aclcheck(relid, GetUserId(), ACL_MAINTAIN) == ACLCHECK_OK)
2567 michael@paquier.xyz 739 : 35568 : return true;
740 : :
741 : 1541 : relname = NameStr(reltuple->relname);
742 : :
743 [ + + ]: 1541 : if ((options & VACOPT_VACUUM) != 0)
744 : : {
1018 andrew@dunslane.net 745 [ + - ]: 112 : ereport(WARNING,
746 : : (errmsg("permission denied to vacuum \"%s\", skipping it",
747 : : relname)));
748 : :
749 : : /*
750 : : * For VACUUM ANALYZE, both logs could show up, but just generate
751 : : * information for VACUUM as that would be the first one to be
752 : : * processed.
753 : : */
2567 michael@paquier.xyz 754 : 112 : return false;
755 : : }
756 : :
757 [ + - ]: 1429 : if ((options & VACOPT_ANALYZE) != 0)
1018 andrew@dunslane.net 758 [ + - ]: 1429 : ereport(WARNING,
759 : : (errmsg("permission denied to analyze \"%s\", skipping it",
760 : : relname)));
761 : :
2567 michael@paquier.xyz 762 : 1429 : return false;
763 : : }
764 : :
765 : :
766 : : /*
767 : : * vacuum_open_relation
768 : : *
769 : : * This routine is used for attempting to open and lock a relation which
770 : : * is going to be vacuumed or analyzed. If the relation cannot be opened
771 : : * or locked, a log is emitted if possible.
772 : : */
773 : : Relation
1692 774 : 21529 : vacuum_open_relation(Oid relid, RangeVar *relation, bits32 options,
775 : : bool verbose, LOCKMODE lmode)
776 : : {
777 : : Relation rel;
2531 778 : 21529 : bool rel_lock = true;
779 : : int elevel;
780 : :
781 [ - + ]: 21529 : Assert((options & (VACOPT_VACUUM | VACOPT_ANALYZE)) != 0);
782 : :
783 : : /*
784 : : * Open the relation and get the appropriate lock on it.
785 : : *
786 : : * There's a race condition here: the relation may have gone away since
787 : : * the last time we saw it. If so, we don't need to vacuum or analyze it.
788 : : *
789 : : * If we've been asked not to wait for the relation lock, acquire it first
790 : : * in non-blocking mode, before calling try_relation_open().
791 : : */
792 [ + + ]: 21529 : if (!(options & VACOPT_SKIP_LOCKED))
1615 pg@bowt.ie 793 : 20850 : rel = try_relation_open(relid, lmode);
2531 michael@paquier.xyz 794 [ + + ]: 679 : else if (ConditionalLockRelationOid(relid, lmode))
1615 pg@bowt.ie 795 : 667 : rel = try_relation_open(relid, NoLock);
796 : : else
797 : : {
798 : 12 : rel = NULL;
2531 michael@paquier.xyz 799 : 12 : rel_lock = false;
800 : : }
801 : :
802 : : /* if relation is opened, leave */
1615 pg@bowt.ie 803 [ + + ]: 21529 : if (rel)
804 : 21511 : return rel;
805 : :
806 : : /*
807 : : * Relation could not be opened, hence generate if possible a log
808 : : * informing on the situation.
809 : : *
810 : : * If the RangeVar is not defined, we do not have enough information to
811 : : * provide a meaningful log statement. Chances are that the caller has
812 : : * intentionally not provided this information so that this logging is
813 : : * skipped, anyway.
814 : : */
2531 michael@paquier.xyz 815 [ + + ]: 18 : if (relation == NULL)
816 : 9 : return NULL;
817 : :
818 : : /*
819 : : * Determine the log level.
820 : : *
821 : : * For manual VACUUM or ANALYZE, we emit a WARNING to match the log
822 : : * statements in the permission checks; otherwise, only log if the caller
823 : : * so requested.
824 : : */
551 heikki.linnakangas@i 825 [ + + ]: 9 : if (!AmAutoVacuumWorkerProcess())
2531 michael@paquier.xyz 826 : 7 : elevel = WARNING;
2364 rhaas@postgresql.org 827 [ + - ]: 2 : else if (verbose)
2531 michael@paquier.xyz 828 : 2 : elevel = LOG;
829 : : else
2531 michael@paquier.xyz 830 :UBC 0 : return NULL;
831 : :
2531 michael@paquier.xyz 832 [ + + ]:CBC 9 : if ((options & VACOPT_VACUUM) != 0)
833 : : {
834 [ + + ]: 5 : if (!rel_lock)
835 [ + - ]: 3 : ereport(elevel,
836 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
837 : : errmsg("skipping vacuum of \"%s\" --- lock not available",
838 : : relation->relname)));
839 : : else
840 [ + - ]: 2 : ereport(elevel,
841 : : (errcode(ERRCODE_UNDEFINED_TABLE),
842 : : errmsg("skipping vacuum of \"%s\" --- relation no longer exists",
843 : : relation->relname)));
844 : :
845 : : /*
846 : : * For VACUUM ANALYZE, both logs could show up, but just generate
847 : : * information for VACUUM as that would be the first one to be
848 : : * processed.
849 : : */
850 : 5 : return NULL;
851 : : }
852 : :
853 [ + - ]: 4 : if ((options & VACOPT_ANALYZE) != 0)
854 : : {
855 [ + + ]: 4 : if (!rel_lock)
856 [ + - ]: 3 : ereport(elevel,
857 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
858 : : errmsg("skipping analyze of \"%s\" --- lock not available",
859 : : relation->relname)));
860 : : else
861 [ + - ]: 1 : ereport(elevel,
862 : : (errcode(ERRCODE_UNDEFINED_TABLE),
863 : : errmsg("skipping analyze of \"%s\" --- relation no longer exists",
864 : : relation->relname)));
865 : : }
866 : :
867 : 4 : return NULL;
868 : : }
869 : :
870 : :
871 : : /*
872 : : * Given a VacuumRelation, fill in the table OID if it wasn't specified,
873 : : * and optionally add VacuumRelations for partitions or inheritance children.
874 : : *
875 : : * If a VacuumRelation does not have an OID supplied and is a partitioned
876 : : * table, an extra entry will be added to the output for each partition.
877 : : * Presently, only autovacuum supplies OIDs when calling vacuum(), and
878 : : * it does not want us to expand partitioned tables.
879 : : *
880 : : * We take care not to modify the input data structure, but instead build
881 : : * new VacuumRelation(s) to return. (But note that they will reference
882 : : * unmodified parts of the input, eg column lists.) New data structures
883 : : * are made in vac_context.
884 : : */
885 : : static List *
887 drowley@postgresql.o 886 : 6922 : expand_vacuum_rel(VacuumRelation *vrel, MemoryContext vac_context,
887 : : int options)
888 : : {
2895 tgl@sss.pgh.pa.us 889 : 6922 : List *vacrels = NIL;
890 : : MemoryContext oldcontext;
891 : :
892 : : /* If caller supplied OID, there's nothing we need do here. */
893 [ + + ]: 6922 : if (OidIsValid(vrel->oid))
894 : : {
6302 alvherre@alvh.no-ip. 895 : 318 : oldcontext = MemoryContextSwitchTo(vac_context);
2895 tgl@sss.pgh.pa.us 896 : 318 : vacrels = lappend(vacrels, vrel);
6302 alvherre@alvh.no-ip. 897 : 318 : MemoryContextSwitchTo(oldcontext);
898 : : }
899 : : else
900 : : {
901 : : /*
902 : : * Process a specific relation, and possibly partitions or child
903 : : * tables thereof.
904 : : */
905 : : Oid relid;
906 : : HeapTuple tuple;
907 : : Form_pg_class classForm;
908 : : bool include_children;
909 : : bool is_partitioned_table;
910 : : int rvr_opts;
911 : :
912 : : /*
913 : : * Since autovacuum workers supply OIDs when calling vacuum(), no
914 : : * autovacuum worker should reach this code.
915 : : */
551 heikki.linnakangas@i 916 [ - + ]: 6604 : Assert(!AmAutoVacuumWorkerProcess());
917 : :
918 : : /*
919 : : * We transiently take AccessShareLock to protect the syscache lookup
920 : : * below, as well as find_all_inheritors's expectation that the caller
921 : : * holds some lock on the starting relation.
922 : : */
2529 michael@paquier.xyz 923 : 6604 : rvr_opts = (options & VACOPT_SKIP_LOCKED) ? RVR_SKIP_LOCKED : 0;
924 : 6604 : relid = RangeVarGetRelidExtended(vrel->relation,
925 : : AccessShareLock,
926 : : rvr_opts,
927 : : NULL, NULL);
928 : :
929 : : /*
930 : : * If the lock is unavailable, emit the same log statement that
931 : : * vacuum_rel() and analyze_rel() would.
932 : : */
933 [ + + ]: 6586 : if (!OidIsValid(relid))
934 : : {
935 [ + + ]: 4 : if (options & VACOPT_VACUUM)
936 [ + - ]: 3 : ereport(WARNING,
937 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
938 : : errmsg("skipping vacuum of \"%s\" --- lock not available",
939 : : vrel->relation->relname)));
940 : : else
941 [ + - ]: 1 : ereport(WARNING,
942 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
943 : : errmsg("skipping analyze of \"%s\" --- lock not available",
944 : : vrel->relation->relname)));
945 : 4 : return vacrels;
946 : : }
947 : :
948 : : /*
949 : : * To check whether the relation is a partitioned table and its
950 : : * ownership, fetch its syscache entry.
951 : : */
3110 rhaas@postgresql.org 952 : 6582 : tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(relid));
953 [ - + ]: 6582 : if (!HeapTupleIsValid(tuple))
3110 rhaas@postgresql.org 954 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for relation %u", relid);
3110 rhaas@postgresql.org 955 :CBC 6582 : classForm = (Form_pg_class) GETSTRUCT(tuple);
956 : :
957 : : /*
958 : : * Make a returnable VacuumRelation for this rel if the user has the
959 : : * required privileges.
960 : : */
542 nathan@postgresql.or 961 [ + + ]: 6582 : if (vacuum_is_permitted_for_relation(relid, classForm, options))
962 : : {
2567 michael@paquier.xyz 963 : 6466 : oldcontext = MemoryContextSwitchTo(vac_context);
964 : 6466 : vacrels = lappend(vacrels, makeVacuumRelation(vrel->relation,
965 : : relid,
966 : : vrel->va_cols));
967 : 6466 : MemoryContextSwitchTo(oldcontext);
968 : : }
969 : :
970 : : /*
971 : : * Vacuuming a partitioned table with ONLY will not do anything since
972 : : * the partitioned table itself is empty. Issue a warning if the user
973 : : * requests this.
974 : : */
347 drowley@postgresql.o 975 : 6582 : include_children = vrel->relation->inh;
976 : 6582 : is_partitioned_table = (classForm->relkind == RELKIND_PARTITIONED_TABLE);
977 [ + + + + : 6582 : if ((options & VACOPT_VACUUM) && is_partitioned_table && !include_children)
+ + ]
978 [ + - ]: 3 : ereport(WARNING,
979 : : (errmsg("VACUUM ONLY of partitioned table \"%s\" has no effect",
980 : : vrel->relation->relname)));
981 : :
3110 rhaas@postgresql.org 982 : 6582 : ReleaseSysCache(tuple);
983 : :
984 : : /*
985 : : * Unless the user has specified ONLY, make relation list entries for
986 : : * its partitions or inheritance child tables. Note that the list
987 : : * returned by find_all_inheritors() includes the passed-in OID, so we
988 : : * have to skip that. There's no point in taking locks on the
989 : : * individual partitions or child tables yet, and doing so would just
990 : : * add unnecessary deadlock risk. For this last reason, we do not yet
991 : : * check the ownership of the partitions/tables, which get added to
992 : : * the list to process. Ownership will be checked later on anyway.
993 : : */
347 drowley@postgresql.o 994 [ + + ]: 6582 : if (include_children)
995 : : {
2895 tgl@sss.pgh.pa.us 996 : 6567 : List *part_oids = find_all_inheritors(relid, NoLock, NULL);
997 : : ListCell *part_lc;
998 : :
999 [ + - + + : 14187 : foreach(part_lc, part_oids)
+ + ]
1000 : : {
1001 : 7620 : Oid part_oid = lfirst_oid(part_lc);
1002 : :
1003 [ + + ]: 7620 : if (part_oid == relid)
1004 : 6567 : continue; /* ignore original table */
1005 : :
1006 : : /*
1007 : : * We omit a RangeVar since it wouldn't be appropriate to
1008 : : * complain about failure to open one of these relations
1009 : : * later.
1010 : : */
1011 : 1053 : oldcontext = MemoryContextSwitchTo(vac_context);
1012 : 1053 : vacrels = lappend(vacrels, makeVacuumRelation(NULL,
1013 : : part_oid,
1014 : : vrel->va_cols));
1015 : 1053 : MemoryContextSwitchTo(oldcontext);
1016 : : }
1017 : : }
1018 : :
1019 : : /*
1020 : : * Release lock again. This means that by the time we actually try to
1021 : : * process the table, it might be gone or renamed. In the former case
1022 : : * we'll silently ignore it; in the latter case we'll process it
1023 : : * anyway, but we must beware that the RangeVar doesn't necessarily
1024 : : * identify it anymore. This isn't ideal, perhaps, but there's little
1025 : : * practical alternative, since we're typically going to commit this
1026 : : * transaction and begin a new one between now and then. Moreover,
1027 : : * holding locks on multiple relations would create significant risk
1028 : : * of deadlock.
1029 : : */
2899 1030 : 6582 : UnlockRelationOid(relid, AccessShareLock);
1031 : : }
1032 : :
2895 1033 : 6900 : return vacrels;
1034 : : }
1035 : :
1036 : : /*
1037 : : * Construct a list of VacuumRelations for all vacuumable rels in
1038 : : * the current database. The list is built in vac_context.
1039 : : */
1040 : : static List *
887 drowley@postgresql.o 1041 : 108 : get_all_vacuum_rels(MemoryContext vac_context, int options)
1042 : : {
2895 tgl@sss.pgh.pa.us 1043 : 108 : List *vacrels = NIL;
1044 : : Relation pgclass;
1045 : : TableScanDesc scan;
1046 : : HeapTuple tuple;
1047 : :
2420 andres@anarazel.de 1048 : 108 : pgclass = table_open(RelationRelationId, AccessShareLock);
1049 : :
2371 1050 : 108 : scan = table_beginscan_catalog(pgclass, 0, NULL);
1051 : :
2895 tgl@sss.pgh.pa.us 1052 [ + + ]: 48855 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1053 : : {
1054 : 48747 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
1055 : : MemoryContext oldcontext;
2482 andres@anarazel.de 1056 : 48747 : Oid relid = classForm->oid;
1057 : :
1058 : : /*
1059 : : * We include partitioned tables here; depending on which operation is
1060 : : * to be performed, caller will decide whether to process or ignore
1061 : : * them.
1062 : : */
2895 tgl@sss.pgh.pa.us 1063 [ + + ]: 48747 : if (classForm->relkind != RELKIND_RELATION &&
1064 [ + + ]: 39837 : classForm->relkind != RELKIND_MATVIEW &&
1065 [ + + ]: 39813 : classForm->relkind != RELKIND_PARTITIONED_TABLE)
1066 : 39731 : continue;
1067 : :
1068 : : /* check permissions of relation */
542 nathan@postgresql.or 1069 [ + + ]: 9016 : if (!vacuum_is_permitted_for_relation(relid, classForm, options))
967 jdavis@postgresql.or 1070 : 1371 : continue;
1071 : :
1072 : : /*
1073 : : * Build VacuumRelation(s) specifying the table OIDs to be processed.
1074 : : * We omit a RangeVar since it wouldn't be appropriate to complain
1075 : : * about failure to open one of these relations later.
1076 : : */
2895 tgl@sss.pgh.pa.us 1077 : 7645 : oldcontext = MemoryContextSwitchTo(vac_context);
1078 : 7645 : vacrels = lappend(vacrels, makeVacuumRelation(NULL,
1079 : : relid,
1080 : : NIL));
1081 : 7645 : MemoryContextSwitchTo(oldcontext);
1082 : : }
1083 : :
2371 andres@anarazel.de 1084 : 108 : table_endscan(scan);
2420 1085 : 108 : table_close(pgclass, AccessShareLock);
1086 : :
2895 tgl@sss.pgh.pa.us 1087 : 108 : return vacrels;
1088 : : }
1089 : :
1090 : : /*
1091 : : * vacuum_get_cutoffs() -- compute OldestXmin and freeze cutoff points
1092 : : *
1093 : : * The target relation and VACUUM parameters are our inputs.
1094 : : *
1095 : : * Output parameters are the cutoffs that VACUUM caller should use.
1096 : : *
1097 : : * Return value indicates if vacuumlazy.c caller should make its VACUUM
1098 : : * operation aggressive. An aggressive VACUUM must advance relfrozenxid up to
1099 : : * FreezeLimit (at a minimum), and relminmxid up to MultiXactCutoff (at a
1100 : : * minimum).
1101 : : */
1102 : : bool
68 michael@paquier.xyz 1103 :GNC 13287 : vacuum_get_cutoffs(Relation rel, const VacuumParams params,
1104 : : struct VacuumCutoffs *cutoffs)
1105 : : {
1106 : : int freeze_min_age,
1107 : : multixact_freeze_min_age,
1108 : : freeze_table_age,
1109 : : multixact_freeze_table_age,
1110 : : effective_multixact_freeze_max_age;
1111 : : TransactionId nextXID,
1112 : : safeOldestXmin,
1113 : : aggressiveXIDCutoff;
1114 : : MultiXactId nextMXID,
1115 : : safeOldestMxact,
1116 : : aggressiveMXIDCutoff;
1117 : :
1118 : : /* Use mutable copies of freeze age parameters */
1119 : 13287 : freeze_min_age = params.freeze_min_age;
1120 : 13287 : multixact_freeze_min_age = params.multixact_freeze_min_age;
1121 : 13287 : freeze_table_age = params.freeze_table_age;
1122 : 13287 : multixact_freeze_table_age = params.multixact_freeze_table_age;
1123 : :
1124 : : /* Set pg_class fields in cutoffs */
989 pg@bowt.ie 1125 :CBC 13287 : cutoffs->relfrozenxid = rel->rd_rel->relfrozenxid;
1126 : 13287 : cutoffs->relminmxid = rel->rd_rel->relminmxid;
1127 : :
1128 : : /*
1129 : : * Acquire OldestXmin.
1130 : : *
1131 : : * We can always ignore processes running lazy vacuum. This is because we
1132 : : * use these values only for deciding which tuples we must keep in the
1133 : : * tables. Since lazy vacuum doesn't write its XID anywhere (usually no
1134 : : * XID assigned), it's safe to ignore it. In theory it could be
1135 : : * problematic to ignore lazy vacuums in a full vacuum, but keep in mind
1136 : : * that only one vacuum process can be working on a particular table at
1137 : : * any time, and that each vacuum is always an independent transaction.
1138 : : */
1139 : 13287 : cutoffs->OldestXmin = GetOldestNonRemovableTransactionId(rel);
1140 : :
1141 [ - + ]: 13287 : Assert(TransactionIdIsNormal(cutoffs->OldestXmin));
1142 : :
1143 : : /* Acquire OldestMxact */
1144 : 13287 : cutoffs->OldestMxact = GetOldestMultiXactId();
1145 [ - + ]: 13287 : Assert(MultiXactIdIsValid(cutoffs->OldestMxact));
1146 : :
1147 : : /* Acquire next XID/next MXID values used to apply age-based settings */
1102 1148 : 13287 : nextXID = ReadNextTransactionId();
1149 : 13287 : nextMXID = ReadNextMultiXactId();
1150 : :
1151 : : /*
1152 : : * Also compute the multixact age for which freezing is urgent. This is
1153 : : * normally autovacuum_multixact_freeze_max_age, but may be less if we are
1154 : : * short of multixact member space.
1155 : : */
989 1156 : 13287 : effective_multixact_freeze_max_age = MultiXactMemberFreezeThreshold();
1157 : :
1158 : : /*
1159 : : * Almost ready to set freeze output parameters; check if OldestXmin or
1160 : : * OldestMxact are held back to an unsafe degree before we start on that
1161 : : */
1162 : 13287 : safeOldestXmin = nextXID - autovacuum_freeze_max_age;
1163 [ - + ]: 13287 : if (!TransactionIdIsNormal(safeOldestXmin))
989 pg@bowt.ie 1164 :UBC 0 : safeOldestXmin = FirstNormalTransactionId;
989 pg@bowt.ie 1165 :CBC 13287 : safeOldestMxact = nextMXID - effective_multixact_freeze_max_age;
1166 [ - + ]: 13287 : if (safeOldestMxact < FirstMultiXactId)
989 pg@bowt.ie 1167 :UBC 0 : safeOldestMxact = FirstMultiXactId;
989 pg@bowt.ie 1168 [ - + ]:CBC 13287 : if (TransactionIdPrecedes(cutoffs->OldestXmin, safeOldestXmin))
989 pg@bowt.ie 1169 [ # # ]:UBC 0 : ereport(WARNING,
1170 : : (errmsg("cutoff for removing and freezing tuples is far in the past"),
1171 : : errhint("Close open transactions soon to avoid wraparound problems.\n"
1172 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
989 pg@bowt.ie 1173 [ - + ]:CBC 13287 : if (MultiXactIdPrecedes(cutoffs->OldestMxact, safeOldestMxact))
989 pg@bowt.ie 1174 [ # # ]:UBC 0 : ereport(WARNING,
1175 : : (errmsg("cutoff for freezing multixacts is far in the past"),
1176 : : errhint("Close open transactions soon to avoid wraparound problems.\n"
1177 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1178 : :
1179 : : /*
1180 : : * Determine the minimum freeze age to use: as specified by the caller, or
1181 : : * vacuum_freeze_min_age, but in any case not more than half
1182 : : * autovacuum_freeze_max_age, so that autovacuums to prevent XID
1183 : : * wraparound won't occur too frequently.
1184 : : */
1102 pg@bowt.ie 1185 [ + + ]:CBC 13287 : if (freeze_min_age < 0)
1186 : 5368 : freeze_min_age = vacuum_freeze_min_age;
1187 : 13287 : freeze_min_age = Min(freeze_min_age, autovacuum_freeze_max_age / 2);
1188 [ - + ]: 13287 : Assert(freeze_min_age >= 0);
1189 : :
1190 : : /* Compute FreezeLimit, being careful to generate a normal XID */
989 1191 : 13287 : cutoffs->FreezeLimit = nextXID - freeze_min_age;
1192 [ - + ]: 13287 : if (!TransactionIdIsNormal(cutoffs->FreezeLimit))
989 pg@bowt.ie 1193 :UBC 0 : cutoffs->FreezeLimit = FirstNormalTransactionId;
1194 : : /* FreezeLimit must always be <= OldestXmin */
989 pg@bowt.ie 1195 [ + + ]:CBC 13287 : if (TransactionIdPrecedes(cutoffs->OldestXmin, cutoffs->FreezeLimit))
1196 : 438 : cutoffs->FreezeLimit = cutoffs->OldestXmin;
1197 : :
1198 : : /*
1199 : : * Determine the minimum multixact freeze age to use: as specified by
1200 : : * caller, or vacuum_multixact_freeze_min_age, but in any case not more
1201 : : * than half effective_multixact_freeze_max_age, so that autovacuums to
1202 : : * prevent MultiXact wraparound won't occur too frequently.
1203 : : */
1102 1204 [ + + ]: 13287 : if (multixact_freeze_min_age < 0)
1205 : 5368 : multixact_freeze_min_age = vacuum_multixact_freeze_min_age;
1206 : 13287 : multixact_freeze_min_age = Min(multixact_freeze_min_age,
1207 : : effective_multixact_freeze_max_age / 2);
1208 [ - + ]: 13287 : Assert(multixact_freeze_min_age >= 0);
1209 : :
1210 : : /* Compute MultiXactCutoff, being careful to generate a valid value */
989 1211 : 13287 : cutoffs->MultiXactCutoff = nextMXID - multixact_freeze_min_age;
1212 [ - + ]: 13287 : if (cutoffs->MultiXactCutoff < FirstMultiXactId)
989 pg@bowt.ie 1213 :UBC 0 : cutoffs->MultiXactCutoff = FirstMultiXactId;
1214 : : /* MultiXactCutoff must always be <= OldestMxact */
989 pg@bowt.ie 1215 [ + + ]:CBC 13287 : if (MultiXactIdPrecedes(cutoffs->OldestMxact, cutoffs->MultiXactCutoff))
1216 : 2 : cutoffs->MultiXactCutoff = cutoffs->OldestMxact;
1217 : :
1218 : : /*
1219 : : * Finally, figure out if caller needs to do an aggressive VACUUM or not.
1220 : : *
1221 : : * Determine the table freeze age to use: as specified by the caller, or
1222 : : * the value of the vacuum_freeze_table_age GUC, but in any case not more
1223 : : * than autovacuum_freeze_max_age * 0.95, so that if you have e.g nightly
1224 : : * VACUUM schedule, the nightly VACUUM gets a chance to freeze XIDs before
1225 : : * anti-wraparound autovacuum is launched.
1226 : : */
1102 1227 [ + + ]: 13287 : if (freeze_table_age < 0)
1228 : 5368 : freeze_table_age = vacuum_freeze_table_age;
1229 [ + - ]: 13287 : freeze_table_age = Min(freeze_table_age, autovacuum_freeze_max_age * 0.95);
1230 [ - + ]: 13287 : Assert(freeze_table_age >= 0);
1231 : 13287 : aggressiveXIDCutoff = nextXID - freeze_table_age;
1232 [ - + ]: 13287 : if (!TransactionIdIsNormal(aggressiveXIDCutoff))
1102 pg@bowt.ie 1233 :UBC 0 : aggressiveXIDCutoff = FirstNormalTransactionId;
495 noah@leadboat.com 1234 [ + + ]:CBC 13287 : if (TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid,
1235 : : aggressiveXIDCutoff))
1303 pg@bowt.ie 1236 : 7766 : return true;
1237 : :
1238 : : /*
1239 : : * Similar to the above, determine the table freeze age to use for
1240 : : * multixacts: as specified by the caller, or the value of the
1241 : : * vacuum_multixact_freeze_table_age GUC, but in any case not more than
1242 : : * effective_multixact_freeze_max_age * 0.95, so that if you have e.g.
1243 : : * nightly VACUUM schedule, the nightly VACUUM gets a chance to freeze
1244 : : * multixacts before anti-wraparound autovacuum is launched.
1245 : : */
1102 1246 [ + + ]: 5521 : if (multixact_freeze_table_age < 0)
1247 : 5368 : multixact_freeze_table_age = vacuum_multixact_freeze_table_age;
1248 : 5521 : multixact_freeze_table_age =
1249 [ + - ]: 5521 : Min(multixact_freeze_table_age,
1250 : : effective_multixact_freeze_max_age * 0.95);
1251 [ - + ]: 5521 : Assert(multixact_freeze_table_age >= 0);
1252 : 5521 : aggressiveMXIDCutoff = nextMXID - multixact_freeze_table_age;
1253 [ - + ]: 5521 : if (aggressiveMXIDCutoff < FirstMultiXactId)
1102 pg@bowt.ie 1254 :UBC 0 : aggressiveMXIDCutoff = FirstMultiXactId;
495 noah@leadboat.com 1255 [ - + ]:CBC 5521 : if (MultiXactIdPrecedesOrEquals(cutoffs->relminmxid,
1256 : : aggressiveMXIDCutoff))
1303 pg@bowt.ie 1257 :UBC 0 : return true;
1258 : :
1259 : : /* Non-aggressive VACUUM */
1303 pg@bowt.ie 1260 :CBC 5521 : return false;
1261 : : }
1262 : :
1263 : : /*
1264 : : * vacuum_xid_failsafe_check() -- Used by VACUUM's wraparound failsafe
1265 : : * mechanism to determine if its table's relfrozenxid and relminmxid are now
1266 : : * dangerously far in the past.
1267 : : *
1268 : : * When we return true, VACUUM caller triggers the failsafe.
1269 : : */
1270 : : bool
989 1271 : 14967 : vacuum_xid_failsafe_check(const struct VacuumCutoffs *cutoffs)
1272 : : {
1273 : 14967 : TransactionId relfrozenxid = cutoffs->relfrozenxid;
1274 : 14967 : MultiXactId relminmxid = cutoffs->relminmxid;
1275 : : TransactionId xid_skip_limit;
1276 : : MultiXactId multi_skip_limit;
1277 : : int skip_index_vacuum;
1278 : :
1613 1279 [ - + ]: 14967 : Assert(TransactionIdIsNormal(relfrozenxid));
1280 [ - + ]: 14967 : Assert(MultiXactIdIsValid(relminmxid));
1281 : :
1282 : : /*
1283 : : * Determine the index skipping age to use. In any case no less than
1284 : : * autovacuum_freeze_max_age * 1.05.
1285 : : */
1286 [ + - ]: 14967 : skip_index_vacuum = Max(vacuum_failsafe_age, autovacuum_freeze_max_age * 1.05);
1287 : :
1288 : 14967 : xid_skip_limit = ReadNextTransactionId() - skip_index_vacuum;
1289 [ - + ]: 14967 : if (!TransactionIdIsNormal(xid_skip_limit))
1613 pg@bowt.ie 1290 :UBC 0 : xid_skip_limit = FirstNormalTransactionId;
1291 : :
1613 pg@bowt.ie 1292 [ - + ]:CBC 14967 : if (TransactionIdPrecedes(relfrozenxid, xid_skip_limit))
1293 : : {
1294 : : /* The table's relfrozenxid is too old */
1613 pg@bowt.ie 1295 :UBC 0 : return true;
1296 : : }
1297 : :
1298 : : /*
1299 : : * Similar to above, determine the index skipping age to use for
1300 : : * multixact. In any case no less than autovacuum_multixact_freeze_max_age *
1301 : : * 1.05.
1302 : : */
1613 pg@bowt.ie 1303 [ + - ]:CBC 14967 : skip_index_vacuum = Max(vacuum_multixact_failsafe_age,
1304 : : autovacuum_multixact_freeze_max_age * 1.05);
1305 : :
1306 : 14967 : multi_skip_limit = ReadNextMultiXactId() - skip_index_vacuum;
1307 [ - + ]: 14967 : if (multi_skip_limit < FirstMultiXactId)
1613 pg@bowt.ie 1308 :UBC 0 : multi_skip_limit = FirstMultiXactId;
1309 : :
1613 pg@bowt.ie 1310 [ - + ]:CBC 14967 : if (MultiXactIdPrecedes(relminmxid, multi_skip_limit))
1311 : : {
1312 : : /* The table's relminmxid is too old */
1613 pg@bowt.ie 1313 :UBC 0 : return true;
1314 : : }
1315 : :
1613 pg@bowt.ie 1316 :CBC 14967 : return false;
1317 : : }
1318 : :
1319 : : /*
1320 : : * vac_estimate_reltuples() -- estimate the new value for pg_class.reltuples
1321 : : *
1322 : : * If we scanned the whole relation then we should just use the count of
1323 : : * live tuples seen; but if we did not, we should not blindly extrapolate
1324 : : * from that number, since VACUUM may have scanned a quite nonrandom
1325 : : * subset of the table. When we have only partial information, we take
1326 : : * the old value of pg_class.reltuples/pg_class.relpages as a measurement
1327 : : * of the tuple density in the unscanned pages.
1328 : : *
1329 : : * Note: scanned_tuples should count only *live* tuples, since
1330 : : * pg_class.reltuples is defined that way.
1331 : : */
1332 : : double
2734 tgl@sss.pgh.pa.us 1333 : 13014 : vac_estimate_reltuples(Relation relation,
1334 : : BlockNumber total_pages,
1335 : : BlockNumber scanned_pages,
1336 : : double scanned_tuples)
1337 : : {
5203 bruce@momjian.us 1338 : 13014 : BlockNumber old_rel_pages = relation->rd_rel->relpages;
5213 tgl@sss.pgh.pa.us 1339 : 13014 : double old_rel_tuples = relation->rd_rel->reltuples;
1340 : : double old_density;
1341 : : double unscanned_pages;
1342 : : double total_tuples;
1343 : :
1344 : : /* If we did scan the whole table, just use the count as-is */
1345 [ + + ]: 13014 : if (scanned_pages >= total_pages)
1346 : 12759 : return scanned_tuples;
1347 : :
1348 : : /*
1349 : : * When successive VACUUM commands scan the same few pages again and
1350 : : * again, without anything from the table really changing, there is a risk
1351 : : * that our beliefs about tuple density will gradually become distorted.
1352 : : * This might be caused by vacuumlazy.c implementation details, such as
1353 : : * its tendency to always scan the last heap page. Handle that here.
1354 : : *
1355 : : * If the relation is _exactly_ the same size according to the existing
1356 : : * pg_class entry, and only a few of its pages (less than 2%) were
1357 : : * scanned, keep the existing value of reltuples. Also keep the existing
1358 : : * value when only a subset of rel's pages <= a single page were scanned.
1359 : : *
1360 : : * (Note: we might be returning -1 here.)
1361 : : */
1298 pg@bowt.ie 1362 [ + + ]: 255 : if (old_rel_pages == total_pages &&
1363 [ + + ]: 240 : scanned_pages < (double) total_pages * 0.02)
1364 : 160 : return old_rel_tuples;
1114 1365 [ + + ]: 95 : if (scanned_pages <= 1)
1366 : 71 : return old_rel_tuples;
1367 : :
1368 : : /*
1369 : : * If old density is unknown, we can't do much except scale up
1370 : : * scanned_tuples to match total_pages.
1371 : : */
1833 tgl@sss.pgh.pa.us 1372 [ + + - + ]: 24 : if (old_rel_tuples < 0 || old_rel_pages == 0)
5213 1373 : 1 : return floor((scanned_tuples / scanned_pages) * total_pages + 0.5);
1374 : :
1375 : : /*
1376 : : * Okay, we've covered the corner cases. The normal calculation is to
1377 : : * convert the old measurement to a density (tuples per page), then
1378 : : * estimate the number of tuples in the unscanned pages using that figure,
1379 : : * and finally add on the number of tuples in the scanned pages.
1380 : : */
1381 : 23 : old_density = old_rel_tuples / old_rel_pages;
2734 1382 : 23 : unscanned_pages = (double) total_pages - (double) scanned_pages;
1383 : 23 : total_tuples = old_density * unscanned_pages + scanned_tuples;
1384 : 23 : return floor(total_tuples + 0.5);
1385 : : }
1386 : :
1387 : :
1388 : : /*
1389 : : * vac_update_relstats() -- update statistics for one relation
1390 : : *
1391 : : * Update the whole-relation statistics that are kept in its pg_class
1392 : : * row. There are additional stats that will be updated if we are
1393 : : * doing ANALYZE, but we always update these stats. This routine works
1394 : : * for both index and heap relation entries in pg_class.
1395 : : *
1396 : : * We violate transaction semantics here by overwriting the rel's
1397 : : * existing pg_class tuple with the new values. This is reasonably
1398 : : * safe as long as we're sure that the new values are correct whether or
1399 : : * not this transaction commits. The reason for doing this is that if
1400 : : * we updated these tuples in the usual way, vacuuming pg_class itself
1401 : : * wouldn't work very well --- by the time we got done with a vacuum
1402 : : * cycle, most of the tuples in pg_class would've been obsoleted. Of
1403 : : * course, this only works for fixed-size not-null columns, but these are.
1404 : : *
1405 : : * Another reason for doing it this way is that when we are in a lazy
1406 : : * VACUUM and have PROC_IN_VACUUM set, we mustn't do any regular updates.
1407 : : * Somebody vacuuming pg_class might think they could delete a tuple
1408 : : * marked with xmin = our xid.
1409 : : *
1410 : : * In addition to fundamentally nontransactional statistics such as
1411 : : * relpages and relallvisible, we try to maintain certain lazily-updated
1412 : : * DDL flags such as relhasindex, by clearing them if no longer correct.
1413 : : * It's safe to do this in VACUUM, which can't run in parallel with
1414 : : * CREATE INDEX/RULE/TRIGGER and can't be part of a transaction block.
1415 : : * However, it's *not* safe to do it in an ANALYZE that's within an
1416 : : * outer transaction, because for example the current transaction might
1417 : : * have dropped the last index; then we'd think relhasindex should be
1418 : : * cleared, but if the transaction later rolls back this would be wrong.
1419 : : * So we refrain from updating the DDL flags if we're inside an outer
1420 : : * transaction. This is OK since postponing the flag maintenance is
1421 : : * always allowable.
1422 : : *
1423 : : * Note: num_tuples should count only *live* tuples, since
1424 : : * pg_class.reltuples is defined that way.
1425 : : *
1426 : : * This routine is shared by VACUUM and ANALYZE.
1427 : : */
1428 : : void
6144 1429 : 33734 : vac_update_relstats(Relation relation,
1430 : : BlockNumber num_pages, double num_tuples,
1431 : : BlockNumber num_all_visible_pages,
1432 : : BlockNumber num_all_frozen_pages,
1433 : : bool hasindex, TransactionId frozenxid,
1434 : : MultiXactId minmulti,
1435 : : bool *frozenxid_updated, bool *minmulti_updated,
1436 : : bool in_outer_xact)
1437 : : {
1438 : 33734 : Oid relid = RelationGetRelid(relation);
1439 : : Relation rd;
1440 : : ScanKeyData key[1];
1441 : : HeapTuple ctup;
1442 : : void *inplace_state;
1443 : : Form_pg_class pgcform;
1444 : : bool dirty,
1445 : : futurexid,
1446 : : futuremxid;
1447 : : TransactionId oldfrozenxid;
1448 : : MultiXactId oldminmulti;
1449 : :
2420 andres@anarazel.de 1450 : 33734 : rd = table_open(RelationRelationId, RowExclusiveLock);
1451 : :
1452 : : /* Fetch a copy of the tuple to scribble on */
347 noah@leadboat.com 1453 : 33734 : ScanKeyInit(&key[0],
1454 : : Anum_pg_class_oid,
1455 : : BTEqualStrategyNumber, F_OIDEQ,
1456 : : ObjectIdGetDatum(relid));
1457 : 33734 : systable_inplace_update_begin(rd, ClassOidIndexId, true,
1458 : : NULL, 1, key, &ctup, &inplace_state);
8822 tgl@sss.pgh.pa.us 1459 [ - + ]: 33734 : if (!HeapTupleIsValid(ctup))
8822 tgl@sss.pgh.pa.us 1460 [ # # ]:UBC 0 : elog(ERROR, "pg_class entry for relid %u vanished during vacuuming",
1461 : : relid);
7059 tgl@sss.pgh.pa.us 1462 :CBC 33734 : pgcform = (Form_pg_class) GETSTRUCT(ctup);
1463 : :
1464 : : /* Apply statistical updates, if any, to copied tuple */
1465 : :
1466 : 33734 : dirty = false;
1467 [ + + ]: 33734 : if (pgcform->relpages != (int32) num_pages)
1468 : : {
1469 : 4742 : pgcform->relpages = (int32) num_pages;
1470 : 4742 : dirty = true;
1471 : : }
1472 [ + + ]: 33734 : if (pgcform->reltuples != (float4) num_tuples)
1473 : : {
1474 : 10039 : pgcform->reltuples = (float4) num_tuples;
1475 : 10039 : dirty = true;
1476 : : }
5076 1477 [ + + ]: 33734 : if (pgcform->relallvisible != (int32) num_all_visible_pages)
1478 : : {
1479 : 2914 : pgcform->relallvisible = (int32) num_all_visible_pages;
1480 : 2914 : dirty = true;
1481 : : }
187 melanieplageman@gmai 1482 [ + + ]: 33734 : if (pgcform->relallfrozen != (int32) num_all_frozen_pages)
1483 : : {
1484 : 2633 : pgcform->relallfrozen = (int32) num_all_frozen_pages;
1485 : 2633 : dirty = true;
1486 : : }
1487 : :
1488 : : /* Apply DDL updates, but not inside an outer transaction (see above) */
1489 : :
3964 tgl@sss.pgh.pa.us 1490 [ + + ]: 33734 : if (!in_outer_xact)
1491 : : {
1492 : : /*
1493 : : * If we didn't find any indexes, reset relhasindex.
1494 : : */
3965 1495 [ + + + + ]: 33577 : if (pgcform->relhasindex && !hasindex)
1496 : : {
1497 : 10 : pgcform->relhasindex = false;
1498 : 10 : dirty = true;
1499 : : }
1500 : :
1501 : : /* We also clear relhasrules and relhastriggers if needed */
1502 [ + + - + ]: 33577 : if (pgcform->relhasrules && relation->rd_rules == NULL)
1503 : : {
3965 tgl@sss.pgh.pa.us 1504 :UBC 0 : pgcform->relhasrules = false;
1505 : 0 : dirty = true;
1506 : : }
3965 tgl@sss.pgh.pa.us 1507 [ + + + + ]:CBC 33577 : if (pgcform->relhastriggers && relation->trigdesc == NULL)
1508 : : {
1509 : 3 : pgcform->relhastriggers = false;
1510 : 3 : dirty = true;
1511 : : }
1512 : : }
1513 : :
1514 : : /*
1515 : : * Update relfrozenxid, unless caller passed InvalidTransactionId
1516 : : * indicating it has no new data.
1517 : : *
1518 : : * Ordinarily, we don't let relfrozenxid go backwards. However, if the
1519 : : * stored relfrozenxid is "in the future" then it seems best to assume
1520 : : * it's corrupt, and overwrite with the oldest remaining XID in the table.
1521 : : * This should match vac_update_datfrozenxid() concerning what we consider
1522 : : * to be "in the future".
1523 : : */
1250 pg@bowt.ie 1524 : 33734 : oldfrozenxid = pgcform->relfrozenxid;
1525 : 33734 : futurexid = false;
1303 1526 [ + + ]: 33734 : if (frozenxid_updated)
1527 : 13012 : *frozenxid_updated = false;
1250 1528 [ + + + + ]: 33734 : if (TransactionIdIsNormal(frozenxid) && oldfrozenxid != frozenxid)
1529 : : {
1213 tgl@sss.pgh.pa.us 1530 : 11546 : bool update = false;
1531 : :
1250 pg@bowt.ie 1532 [ + + ]: 11546 : if (TransactionIdPrecedes(oldfrozenxid, frozenxid))
1533 : 11500 : update = true;
1534 [ - + ]: 46 : else if (TransactionIdPrecedes(ReadNextTransactionId(), oldfrozenxid))
1250 pg@bowt.ie 1535 :UBC 0 : futurexid = update = true;
1536 : :
1250 pg@bowt.ie 1537 [ + + ]:CBC 11546 : if (update)
1538 : : {
1539 : 11500 : pgcform->relfrozenxid = frozenxid;
1540 : 11500 : dirty = true;
1541 [ + - ]: 11500 : if (frozenxid_updated)
1542 : 11500 : *frozenxid_updated = true;
1543 : : }
1544 : : }
1545 : :
1546 : : /* Similarly for relminmxid */
1547 : 33734 : oldminmulti = pgcform->relminmxid;
1548 : 33734 : futuremxid = false;
1303 1549 [ + + ]: 33734 : if (minmulti_updated)
1550 : 13012 : *minmulti_updated = false;
1250 1551 [ + + + + ]: 33734 : if (MultiXactIdIsValid(minmulti) && oldminmulti != minmulti)
1552 : : {
1213 tgl@sss.pgh.pa.us 1553 : 207 : bool update = false;
1554 : :
1250 pg@bowt.ie 1555 [ + - ]: 207 : if (MultiXactIdPrecedes(oldminmulti, minmulti))
1556 : 207 : update = true;
1250 pg@bowt.ie 1557 [ # # ]:UBC 0 : else if (MultiXactIdPrecedes(ReadNextMultiXactId(), oldminmulti))
1558 : 0 : futuremxid = update = true;
1559 : :
1250 pg@bowt.ie 1560 [ + - ]:CBC 207 : if (update)
1561 : : {
1562 : 207 : pgcform->relminmxid = minmulti;
1563 : 207 : dirty = true;
1564 [ + - ]: 207 : if (minmulti_updated)
1565 : 207 : *minmulti_updated = true;
1566 : : }
1567 : : }
1568 : :
1569 : : /* If anything changed, write out the tuple. */
7059 tgl@sss.pgh.pa.us 1570 [ + + ]: 33734 : if (dirty)
347 noah@leadboat.com 1571 : 18720 : systable_inplace_update_finish(inplace_state, ctup);
1572 : : else
1573 : 15014 : systable_inplace_update_cancel(inplace_state);
1574 : :
2420 andres@anarazel.de 1575 : 33734 : table_close(rd, RowExclusiveLock);
1576 : :
1250 pg@bowt.ie 1577 [ - + ]: 33734 : if (futurexid)
1250 pg@bowt.ie 1578 [ # # ]:UBC 0 : ereport(WARNING,
1579 : : (errcode(ERRCODE_DATA_CORRUPTED),
1580 : : errmsg_internal("overwrote invalid relfrozenxid value %u with new value %u for table \"%s\"",
1581 : : oldfrozenxid, frozenxid,
1582 : : RelationGetRelationName(relation))));
1250 pg@bowt.ie 1583 [ - + ]:CBC 33734 : if (futuremxid)
1250 pg@bowt.ie 1584 [ # # ]:UBC 0 : ereport(WARNING,
1585 : : (errcode(ERRCODE_DATA_CORRUPTED),
1586 : : errmsg_internal("overwrote invalid relminmxid value %u with new value %u for table \"%s\"",
1587 : : oldminmulti, minmulti,
1588 : : RelationGetRelationName(relation))));
8822 tgl@sss.pgh.pa.us 1589 :CBC 33734 : }
1590 : :
1591 : :
1592 : : /*
1593 : : * vac_update_datfrozenxid() -- update pg_database.datfrozenxid for our DB
1594 : : *
1595 : : * Update pg_database's datfrozenxid entry for our database to be the
1596 : : * minimum of the pg_class.relfrozenxid values.
1597 : : *
1598 : : * Similarly, update our datminmxid to be the minimum of the
1599 : : * pg_class.relminmxid values.
1600 : : *
1601 : : * If we are able to advance either pg_database value, also try to
1602 : : * truncate pg_xact and pg_multixact.
1603 : : *
1604 : : * We violate transaction semantics here by overwriting the database's
1605 : : * existing pg_database tuple with the new values. This is reasonably
1606 : : * safe since the new values are correct whether or not this transaction
1607 : : * commits. As with vac_update_relstats, this avoids leaving dead tuples
1608 : : * behind after a VACUUM.
1609 : : */
1610 : : void
6880 1611 : 904 : vac_update_datfrozenxid(void)
1612 : : {
1613 : : HeapTuple tuple;
1614 : : Form_pg_database dbform;
1615 : : Relation relation;
1616 : : SysScanDesc scan;
1617 : : HeapTuple classTup;
1618 : : TransactionId newFrozenXid;
1619 : : MultiXactId newMinMulti;
1620 : : TransactionId lastSaneFrozenXid;
1621 : : MultiXactId lastSaneMinMulti;
4065 1622 : 904 : bool bogus = false;
6998 alvherre@alvh.no-ip. 1623 : 904 : bool dirty = false;
1624 : : ScanKeyData key[1];
1625 : : void *inplace_state;
1626 : :
1627 : : /*
1628 : : * Restrict this task to one backend per database. This avoids race
1629 : : * conditions that would move datfrozenxid or datminmxid backward. It
1630 : : * avoids calling vac_truncate_clog() with a datfrozenxid preceding a
1631 : : * datfrozenxid passed to an earlier vac_truncate_clog() call.
1632 : : */
1848 noah@leadboat.com 1633 : 904 : LockDatabaseFrozenIds(ExclusiveLock);
1634 : :
1635 : : /*
1636 : : * Initialize the "min" calculation with
1637 : : * GetOldestNonRemovableTransactionId(), which is a reasonable
1638 : : * approximation to the minimum relfrozenxid for not-yet-committed
1639 : : * pg_class entries for new tables; see AddNewRelationTuple(). So we
1640 : : * cannot produce a wrong minimum by starting with this.
1641 : : */
1851 andres@anarazel.de 1642 : 904 : newFrozenXid = GetOldestNonRemovableTransactionId(NULL);
1643 : :
1644 : : /*
1645 : : * Similarly, initialize the MultiXact "min" with the value that would be
1646 : : * used on pg_class for new tables. See AddNewRelationTuple().
1647 : : */
4065 tgl@sss.pgh.pa.us 1648 : 904 : newMinMulti = GetOldestMultiXactId();
1649 : :
1650 : : /*
1651 : : * Identify the latest relfrozenxid and relminmxid values that we could
1652 : : * validly see during the scan. These are conservative values, but it's
1653 : : * not really worth trying to be more exact.
1654 : : */
1664 tmunro@postgresql.or 1655 : 904 : lastSaneFrozenXid = ReadNextTransactionId();
4065 tgl@sss.pgh.pa.us 1656 : 904 : lastSaneMinMulti = ReadNextMultiXactId();
1657 : :
1658 : : /*
1659 : : * We must seqscan pg_class to find the minimum Xid, because there is no
1660 : : * index that can help us here.
1661 : : *
1662 : : * See vac_truncate_clog() for the race condition to prevent.
1663 : : */
2420 andres@anarazel.de 1664 : 904 : relation = table_open(RelationRelationId, AccessShareLock);
1665 : :
6998 alvherre@alvh.no-ip. 1666 : 904 : scan = systable_beginscan(relation, InvalidOid, false,
1667 : : NULL, 0, NULL);
1668 : :
1669 [ + + ]: 593378 : while ((classTup = systable_getnext(scan)) != NULL)
1670 : : {
495 noah@leadboat.com 1671 : 592474 : volatile FormData_pg_class *classForm = (Form_pg_class) GETSTRUCT(classTup);
1672 : 592474 : TransactionId relfrozenxid = classForm->relfrozenxid;
1673 : 592474 : TransactionId relminmxid = classForm->relminmxid;
1674 : :
1675 : : /*
1676 : : * Only consider relations able to hold unfrozen XIDs (anything else
1677 : : * should have InvalidTransactionId in relfrozenxid anyway).
1678 : : */
6998 alvherre@alvh.no-ip. 1679 [ + + ]: 592474 : if (classForm->relkind != RELKIND_RELATION &&
4570 kgrittn@postgresql.o 1680 [ + + ]: 442888 : classForm->relkind != RELKIND_MATVIEW &&
6998 alvherre@alvh.no-ip. 1681 [ + + ]: 441739 : classForm->relkind != RELKIND_TOASTVALUE)
1682 : : {
495 noah@leadboat.com 1683 [ - + ]: 372010 : Assert(!TransactionIdIsValid(relfrozenxid));
1684 [ - + ]: 372010 : Assert(!MultiXactIdIsValid(relminmxid));
6998 alvherre@alvh.no-ip. 1685 : 372010 : continue;
1686 : : }
1687 : :
1688 : : /*
1689 : : * Some table AMs might not need per-relation xid / multixid horizons.
1690 : : * It therefore seems reasonable to allow relfrozenxid and relminmxid
1691 : : * to not be set (i.e. set to their respective Invalid*Id)
1692 : : * independently. Thus validate and compute horizon for each only if
1693 : : * set.
1694 : : *
1695 : : * If things are working properly, no relation should have a
1696 : : * relfrozenxid or relminmxid that is "in the future". However, such
1697 : : * cases have been known to arise due to bugs in pg_upgrade. If we
1698 : : * see any entries that are "in the future", chicken out and don't do
1699 : : * anything. This ensures we won't truncate clog & multixact SLRUs
1700 : : * before those relations have been scanned and cleaned up.
1701 : : */
1702 : :
495 noah@leadboat.com 1703 [ + - ]: 220464 : if (TransactionIdIsValid(relfrozenxid))
1704 : : {
1705 [ - + ]: 220464 : Assert(TransactionIdIsNormal(relfrozenxid));
1706 : :
1707 : : /* check for values in the future */
1708 [ - + ]: 220464 : if (TransactionIdPrecedes(lastSaneFrozenXid, relfrozenxid))
1709 : : {
2328 andres@anarazel.de 1710 :UBC 0 : bogus = true;
1711 : 0 : break;
1712 : : }
1713 : :
1714 : : /* determine new horizon */
495 noah@leadboat.com 1715 [ + + ]:CBC 220464 : if (TransactionIdPrecedes(relfrozenxid, newFrozenXid))
1716 : 1590 : newFrozenXid = relfrozenxid;
1717 : : }
1718 : :
1719 [ + - ]: 220464 : if (MultiXactIdIsValid(relminmxid))
1720 : : {
1721 : : /* check for values in the future */
1722 [ - + ]: 220464 : if (MultiXactIdPrecedes(lastSaneMinMulti, relminmxid))
1723 : : {
2328 andres@anarazel.de 1724 :UBC 0 : bogus = true;
1725 : 0 : break;
1726 : : }
1727 : :
1728 : : /* determine new horizon */
495 noah@leadboat.com 1729 [ + + ]:CBC 220464 : if (MultiXactIdPrecedes(relminmxid, newMinMulti))
1730 : 116 : newMinMulti = relminmxid;
1731 : : }
1732 : : }
1733 : :
1734 : : /* we're done with pg_class */
6998 alvherre@alvh.no-ip. 1735 : 904 : systable_endscan(scan);
2420 andres@anarazel.de 1736 : 904 : table_close(relation, AccessShareLock);
1737 : :
1738 : : /* chicken out if bogus data found */
4065 tgl@sss.pgh.pa.us 1739 [ - + ]: 904 : if (bogus)
4065 tgl@sss.pgh.pa.us 1740 :UBC 0 : return;
1741 : :
6880 tgl@sss.pgh.pa.us 1742 [ - + ]:CBC 904 : Assert(TransactionIdIsNormal(newFrozenXid));
4373 alvherre@alvh.no-ip. 1743 [ - + ]: 904 : Assert(MultiXactIdIsValid(newMinMulti));
1744 : :
1745 : : /* Now fetch the pg_database tuple we need to update. */
2420 andres@anarazel.de 1746 : 904 : relation = table_open(DatabaseRelationId, RowExclusiveLock);
1747 : :
1748 : : /*
1749 : : * Fetch a copy of the tuple to scribble on. We could check the syscache
1750 : : * tuple first. If that concluded !dirty, we'd avoid waiting on
1751 : : * concurrent heap_update() and would avoid exclusive-locking the buffer.
1752 : : * For now, don't optimize that.
1753 : : */
1733 michael@paquier.xyz 1754 : 904 : ScanKeyInit(&key[0],
1755 : : Anum_pg_database_oid,
1756 : : BTEqualStrategyNumber, F_OIDEQ,
1757 : : ObjectIdGetDatum(MyDatabaseId));
1758 : :
347 noah@leadboat.com 1759 : 904 : systable_inplace_update_begin(relation, DatabaseOidIndexId, true,
1760 : : NULL, 1, key, &tuple, &inplace_state);
1761 : :
8777 tgl@sss.pgh.pa.us 1762 [ - + ]: 904 : if (!HeapTupleIsValid(tuple))
6880 tgl@sss.pgh.pa.us 1763 [ # # ]:UBC 0 : elog(ERROR, "could not find tuple for database %u", MyDatabaseId);
1764 : :
8777 tgl@sss.pgh.pa.us 1765 :CBC 904 : dbform = (Form_pg_database) GETSTRUCT(tuple);
1766 : :
1767 : : /*
1768 : : * As in vac_update_relstats(), we ordinarily don't want to let
1769 : : * datfrozenxid go backward; but if it's "in the future" then it must be
1770 : : * corrupt and it seems best to overwrite it.
1771 : : */
4065 1772 [ + + - + ]: 1002 : if (dbform->datfrozenxid != newFrozenXid &&
1773 [ - - ]: 98 : (TransactionIdPrecedes(dbform->datfrozenxid, newFrozenXid) ||
4065 tgl@sss.pgh.pa.us 1774 :UBC 0 : TransactionIdPrecedes(lastSaneFrozenXid, dbform->datfrozenxid)))
1775 : : {
6880 tgl@sss.pgh.pa.us 1776 :CBC 98 : dbform->datfrozenxid = newFrozenXid;
6998 alvherre@alvh.no-ip. 1777 : 98 : dirty = true;
1778 : : }
1779 : : else
4065 tgl@sss.pgh.pa.us 1780 : 806 : newFrozenXid = dbform->datfrozenxid;
1781 : :
1782 : : /* Ditto for datminmxid */
1783 [ + + - + ]: 905 : if (dbform->datminmxid != newMinMulti &&
1784 [ - - ]: 1 : (MultiXactIdPrecedes(dbform->datminmxid, newMinMulti) ||
4065 tgl@sss.pgh.pa.us 1785 :UBC 0 : MultiXactIdPrecedes(lastSaneMinMulti, dbform->datminmxid)))
1786 : : {
4373 alvherre@alvh.no-ip. 1787 :CBC 1 : dbform->datminmxid = newMinMulti;
4609 1788 : 1 : dirty = true;
1789 : : }
1790 : : else
4065 tgl@sss.pgh.pa.us 1791 : 903 : newMinMulti = dbform->datminmxid;
1792 : :
6998 alvherre@alvh.no-ip. 1793 [ + + ]: 904 : if (dirty)
347 noah@leadboat.com 1794 : 98 : systable_inplace_update_finish(inplace_state, tuple);
1795 : : else
1796 : 806 : systable_inplace_update_cancel(inplace_state);
1797 : :
6998 alvherre@alvh.no-ip. 1798 : 904 : heap_freetuple(tuple);
2420 andres@anarazel.de 1799 : 904 : table_close(relation, RowExclusiveLock);
1800 : :
1801 : : /*
1802 : : * If we were able to advance datfrozenxid or datminmxid, see if we can
1803 : : * truncate pg_xact and/or pg_multixact. Also do it if the shared
1804 : : * XID-wrap-limit info is stale, since this action will update that too.
1805 : : */
5849 tgl@sss.pgh.pa.us 1806 [ + + - + ]: 904 : if (dirty || ForceTransactionIdLimitUpdate())
4065 1807 : 98 : vac_truncate_clog(newFrozenXid, newMinMulti,
1808 : : lastSaneFrozenXid, lastSaneMinMulti);
1809 : : }
1810 : :
1811 : :
1812 : : /*
1813 : : * vac_truncate_clog() -- attempt to truncate the commit log
1814 : : *
1815 : : * Scan pg_database to determine the system-wide oldest datfrozenxid,
1816 : : * and use it to truncate the transaction commit log (pg_xact).
1817 : : * Also update the XID wrap limit info maintained by varsup.c.
1818 : : * Likewise for datminmxid.
1819 : : *
1820 : : * The passed frozenXID and minMulti are the updated values for my own
1821 : : * pg_database entry. They're used to initialize the "min" calculations.
1822 : : * The caller also passes the "last sane" XID and MXID, since it has
1823 : : * those at hand already.
1824 : : *
1825 : : * This routine is only invoked when we've managed to change our
1826 : : * DB's datfrozenxid/datminmxid values, or we found that the shared
1827 : : * XID-wrap-limit info is stale.
1828 : : */
1829 : : static void
1830 : 98 : vac_truncate_clog(TransactionId frozenXID,
1831 : : MultiXactId minMulti,
1832 : : TransactionId lastSaneFrozenXid,
1833 : : MultiXactId lastSaneMinMulti)
1834 : : {
1664 tmunro@postgresql.or 1835 : 98 : TransactionId nextXID = ReadNextTransactionId();
1836 : : Relation relation;
1837 : : TableScanDesc scan;
1838 : : HeapTuple tuple;
1839 : : Oid oldestxid_datoid;
1840 : : Oid minmulti_datoid;
4065 tgl@sss.pgh.pa.us 1841 : 98 : bool bogus = false;
6880 1842 : 98 : bool frozenAlreadyWrapped = false;
1843 : :
1844 : : /* Restrict task to one backend per cluster; see SimpleLruTruncate(). */
1848 noah@leadboat.com 1845 : 98 : LWLockAcquire(WrapLimitsVacuumLock, LW_EXCLUSIVE);
1846 : :
1847 : : /* init oldest datoids to sync with my frozenXID/minMulti values */
4609 alvherre@alvh.no-ip. 1848 : 98 : oldestxid_datoid = MyDatabaseId;
4373 1849 : 98 : minmulti_datoid = MyDatabaseId;
1850 : :
1851 : : /*
1852 : : * Scan pg_database to compute the minimum datfrozenxid/datminmxid
1853 : : *
1854 : : * Since vac_update_datfrozenxid updates datfrozenxid/datminmxid in-place,
1855 : : * the values could change while we look at them. Fetch each one just
1856 : : * once to ensure sane behavior of the comparison logic. (Here, as in
1857 : : * many other places, we assume that fetching or updating an XID in shared
1858 : : * storage is atomic.)
1859 : : *
1860 : : * Note: we need not worry about a race condition with new entries being
1861 : : * inserted by CREATE DATABASE. Any such entry will have a copy of some
1862 : : * existing DB's datfrozenxid, and that source DB cannot be ours because
1863 : : * of the interlock against copying a DB containing an active backend.
1864 : : * Hence the new entry will not reduce the minimum. Also, if two VACUUMs
1865 : : * concurrently modify the datfrozenxid's of different databases, the
1866 : : * worst possible outcome is that pg_xact is not truncated as aggressively
1867 : : * as it could be.
1868 : : */
2420 andres@anarazel.de 1869 : 98 : relation = table_open(DatabaseRelationId, AccessShareLock);
1870 : :
2371 1871 : 98 : scan = table_beginscan_catalog(relation, 0, NULL);
1872 : :
8510 tgl@sss.pgh.pa.us 1873 [ + + ]: 305 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1874 : : {
3392 1875 : 207 : volatile FormData_pg_database *dbform = (Form_pg_database) GETSTRUCT(tuple);
1876 : 207 : TransactionId datfrozenxid = dbform->datfrozenxid;
1877 : 207 : TransactionId datminmxid = dbform->datminmxid;
1878 : :
1879 [ - + ]: 207 : Assert(TransactionIdIsNormal(datfrozenxid));
1880 [ - + ]: 207 : Assert(MultiXactIdIsValid(datminmxid));
1881 : :
1882 : : /*
1883 : : * If database is in the process of getting dropped, or has been
1884 : : * interrupted while doing so, no connections to it are possible
1885 : : * anymore. Therefore we don't need to take it into account here.
1886 : : * Which is good, because it can't be processed by autovacuum either.
1887 : : */
786 andres@anarazel.de 1888 [ + + ]: 207 : if (database_is_invalid_form((Form_pg_database) dbform))
1889 : : {
1890 [ - + ]: 1 : elog(DEBUG2,
1891 : : "skipping invalid database \"%s\" while computing relfrozenxid",
1892 : : NameStr(dbform->datname));
1893 : 1 : continue;
1894 : : }
1895 : :
1896 : : /*
1897 : : * If things are working properly, no database should have a
1898 : : * datfrozenxid or datminmxid that is "in the future". However, such
1899 : : * cases have been known to arise due to bugs in pg_upgrade. If we
1900 : : * see any entries that are "in the future", chicken out and don't do
1901 : : * anything. This ensures we won't truncate clog before those
1902 : : * databases have been scanned and cleaned up. (We will issue the
1903 : : * "already wrapped" warning if appropriate, though.)
1904 : : */
3392 tgl@sss.pgh.pa.us 1905 [ + - - + ]: 412 : if (TransactionIdPrecedes(lastSaneFrozenXid, datfrozenxid) ||
1906 : 206 : MultiXactIdPrecedes(lastSaneMinMulti, datminmxid))
4065 tgl@sss.pgh.pa.us 1907 :UBC 0 : bogus = true;
1908 : :
3392 tgl@sss.pgh.pa.us 1909 [ - + ]:CBC 206 : if (TransactionIdPrecedes(nextXID, datfrozenxid))
6880 tgl@sss.pgh.pa.us 1910 :UBC 0 : frozenAlreadyWrapped = true;
3392 tgl@sss.pgh.pa.us 1911 [ + + ]:CBC 206 : else if (TransactionIdPrecedes(datfrozenxid, frozenXID))
1912 : : {
1913 : 51 : frozenXID = datfrozenxid;
2482 andres@anarazel.de 1914 : 51 : oldestxid_datoid = dbform->oid;
1915 : : }
1916 : :
3392 tgl@sss.pgh.pa.us 1917 [ + + ]: 206 : if (MultiXactIdPrecedes(datminmxid, minMulti))
1918 : : {
1919 : 2 : minMulti = datminmxid;
2482 andres@anarazel.de 1920 : 2 : minmulti_datoid = dbform->oid;
1921 : : }
1922 : : }
1923 : :
2371 1924 : 98 : table_endscan(scan);
1925 : :
2420 1926 : 98 : table_close(relation, AccessShareLock);
1927 : :
1928 : : /*
1929 : : * Do not truncate CLOG if we seem to have suffered wraparound already;
1930 : : * the computed minimum XID might be bogus. This case should now be
1931 : : * impossible due to the defenses in GetNewTransactionId, but we keep the
1932 : : * test anyway.
1933 : : */
6880 tgl@sss.pgh.pa.us 1934 [ - + ]: 98 : if (frozenAlreadyWrapped)
1935 : : {
8084 tgl@sss.pgh.pa.us 1936 [ # # ]:UBC 0 : ereport(WARNING,
1937 : : (errmsg("some databases have not been vacuumed in over 2 billion transactions"),
1938 : : errdetail("You might have already suffered transaction-wraparound data loss.")));
786 andres@anarazel.de 1939 : 0 : LWLockRelease(WrapLimitsVacuumLock);
8558 tgl@sss.pgh.pa.us 1940 : 0 : return;
1941 : : }
1942 : :
1943 : : /* chicken out if data is bogus in any other way */
4065 tgl@sss.pgh.pa.us 1944 [ - + ]:CBC 98 : if (bogus)
1945 : : {
786 andres@anarazel.de 1946 :UBC 0 : LWLockRelease(WrapLimitsVacuumLock);
4065 tgl@sss.pgh.pa.us 1947 : 0 : return;
1948 : : }
1949 : :
1950 : : /*
1951 : : * Advance the oldest value for commit timestamps before truncating, so
1952 : : * that if a user requests a timestamp for a transaction we're truncating
1953 : : * away right after this point, they get NULL instead of an ugly "file not
1954 : : * found" error from slru.c. This doesn't matter for xact/multixact
1955 : : * because they are not subject to arbitrary lookups from users.
1956 : : */
3152 alvherre@alvh.no-ip. 1957 :CBC 98 : AdvanceOldestCommitTsXid(frozenXID);
1958 : :
1959 : : /*
1960 : : * Truncate CLOG, multixact and CommitTs to the oldest computed value.
1961 : : */
3089 rhaas@postgresql.org 1962 : 98 : TruncateCLOG(frozenXID, oldestxid_datoid);
3602 alvherre@alvh.no-ip. 1963 : 98 : TruncateCommitTs(frozenXID);
3633 andres@anarazel.de 1964 : 98 : TruncateMultiXact(minMulti, minmulti_datoid);
1965 : :
1966 : : /*
1967 : : * Update the wrap limit for GetNewTransactionId and creation of new
1968 : : * MultiXactIds. Note: these functions will also signal the postmaster
1969 : : * for an(other) autovac cycle if needed. XXX should we avoid possibly
1970 : : * signaling twice?
1971 : : */
4609 alvherre@alvh.no-ip. 1972 : 98 : SetTransactionIdLimit(frozenXID, oldestxid_datoid);
3098 tgl@sss.pgh.pa.us 1973 : 98 : SetMultiXactIdLimit(minMulti, minmulti_datoid, false);
1974 : :
1848 noah@leadboat.com 1975 : 98 : LWLockRelease(WrapLimitsVacuumLock);
1976 : : }
1977 : :
1978 : :
1979 : : /*
1980 : : * vacuum_rel() -- vacuum one heap relation
1981 : : *
1982 : : * relid identifies the relation to vacuum. If relation is supplied,
1983 : : * use the name therein for reporting any failure to open/lock the rel;
1984 : : * do not use it once we've successfully opened the rel, since it might
1985 : : * be stale.
1986 : : *
1987 : : * Returns true if it's okay to proceed with a requested ANALYZE
1988 : : * operation on this table.
1989 : : *
1990 : : * Doing one heap at a time incurs extra overhead, since we need to
1991 : : * check that the heap exists again just before we vacuum it. The
1992 : : * reason that we do this is so that vacuuming can be spread across
1993 : : * many small transactions. Otherwise, two-phase locking would require
1994 : : * us to lock the entire database during one pass of the vacuum cleaner.
1995 : : *
1996 : : * At entry and exit, we are not inside a transaction.
1997 : : */
1998 : : static bool
68 michael@paquier.xyz 1999 :GNC 13411 : vacuum_rel(Oid relid, RangeVar *relation, VacuumParams params,
2000 : : BufferAccessStrategy bstrategy)
2001 : : {
2002 : : LOCKMODE lmode;
2003 : : Relation rel;
2004 : : LockRelId lockrelid;
2005 : : Oid priv_relid;
2006 : : Oid toast_relid;
2007 : : Oid save_userid;
2008 : : int save_sec_context;
2009 : : int save_nestlevel;
2010 : : VacuumParams toast_vacuum_params;
2011 : :
2012 : : /*
2013 : : * This function scribbles on the parameters, so make a copy early to
2014 : : * avoid affecting the TOAST table (if we do end up recursing to it).
2015 : : */
2016 : 13411 : memcpy(&toast_vacuum_params, ¶ms, sizeof(VacuumParams));
2017 : :
2018 : : /* Begin a transaction for vacuuming this relation */
8151 tgl@sss.pgh.pa.us 2019 :CBC 13411 : StartTransactionCommand();
2020 : :
68 michael@paquier.xyz 2021 [ + + ]:GNC 13411 : if (!(params.options & VACOPT_FULL))
2022 : : {
2023 : : /*
2024 : : * In lazy vacuum, we can set the PROC_IN_VACUUM flag, which lets
2025 : : * other concurrent VACUUMs know that they can ignore this one while
2026 : : * determining their OldestXmin. (The reason we don't set it during a
2027 : : * full VACUUM is exactly that we may have to run user-defined
2028 : : * functions for functional indexes, and we want to make sure that if
2029 : : * they use the snapshot set above, any tuples it requires can't get
2030 : : * removed from other tables. An index function that depends on the
2031 : : * contents of other tables is arguably broken, but we won't break it
2032 : : * here by violating transaction semantics.)
2033 : : *
2034 : : * We also set the VACUUM_FOR_WRAPAROUND flag, which is passed down by
2035 : : * autovacuum; it's used to avoid canceling a vacuum that was invoked
2036 : : * in an emergency.
2037 : : *
2038 : : * Note: these flags remain set until CommitTransaction or
2039 : : * AbortTransaction. We don't want to clear them until we reset
2040 : : * MyProc->xid/xmin, otherwise GetOldestNonRemovableTransactionId()
2041 : : * might appear to go backwards, which is probably Not Good. (We also
2042 : : * set PROC_IN_VACUUM *before* taking our own snapshot, so that our
2043 : : * xmin doesn't become visible ahead of setting the flag.)
2044 : : */
1745 alvherre@alvh.no-ip. 2045 :CBC 13215 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
1755 2046 : 13215 : MyProc->statusFlags |= PROC_IN_VACUUM;
68 michael@paquier.xyz 2047 [ - + ]:GNC 13215 : if (params.is_wraparound)
1755 alvherre@alvh.no-ip. 2048 :UBC 0 : MyProc->statusFlags |= PROC_VACUUM_FOR_WRAPAROUND;
1755 alvherre@alvh.no-ip. 2049 :CBC 13215 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
6527 2050 : 13215 : LWLockRelease(ProcArrayLock);
2051 : : }
2052 : :
2053 : : /*
2054 : : * Need to acquire a snapshot to prevent pg_subtrans from being truncated,
2055 : : * cutoff xids in local memory wrapping around, and to have updated xmin
2056 : : * horizons.
2057 : : */
1745 2058 : 13411 : PushActiveSnapshot(GetTransactionSnapshot());
2059 : :
2060 : : /*
2061 : : * Check for user-requested abort. Note we want this to be inside a
2062 : : * transaction, so xact.c doesn't issue useless WARNING.
2063 : : */
9001 tgl@sss.pgh.pa.us 2064 [ - + ]: 13411 : CHECK_FOR_INTERRUPTS();
2065 : :
2066 : : /*
2067 : : * Determine the type of lock we want --- hard exclusive lock for a FULL
2068 : : * vacuum, but just ShareUpdateExclusiveLock for concurrent vacuum. Either
2069 : : * way, we can be sure that no other backend is vacuuming the same table.
2070 : : */
68 michael@paquier.xyz 2071 :GNC 26822 : lmode = (params.options & VACOPT_FULL) ?
2364 rhaas@postgresql.org 2072 [ + + ]:CBC 13411 : AccessExclusiveLock : ShareUpdateExclusiveLock;
2073 : :
2074 : : /* open the relation and get the appropriate lock on it */
68 michael@paquier.xyz 2075 :GNC 13411 : rel = vacuum_open_relation(relid, relation, params.options,
2076 : 13411 : params.log_min_duration >= 0, lmode);
2077 : :
2078 : : /* leave if relation could not be opened or locked */
1615 pg@bowt.ie 2079 [ + + ]:CBC 13411 : if (!rel)
2080 : : {
6204 alvherre@alvh.no-ip. 2081 : 12 : PopActiveSnapshot();
6959 tgl@sss.pgh.pa.us 2082 : 12 : CommitTransactionCommand();
5325 rhaas@postgresql.org 2083 : 12 : return false;
2084 : : }
2085 : :
2086 : : /*
2087 : : * When recursing to a TOAST table, check privileges on the parent. NB:
2088 : : * This is only safe to do because we hold a session lock on the main
2089 : : * relation that prevents concurrent deletion.
2090 : : */
68 michael@paquier.xyz 2091 [ + + ]:GNC 13399 : if (OidIsValid(params.toast_parent))
2092 : 4543 : priv_relid = params.toast_parent;
2093 : : else
542 nathan@postgresql.or 2094 :CBC 8856 : priv_relid = RelationGetRelid(rel);
2095 : :
2096 : : /*
2097 : : * Check if relation needs to be skipped based on privileges. This check
2098 : : * happens also when building the relation list to vacuum for a manual
2099 : : * operation, and needs to be done additionally here as VACUUM could
2100 : : * happen across multiple transactions where privileges could have changed
2101 : : * in-between. Make sure to only generate logs for VACUUM in this case.
2102 : : */
2103 [ + + ]: 13399 : if (!vacuum_is_permitted_for_relation(priv_relid,
2104 : : rel->rd_rel,
68 michael@paquier.xyz 2105 :GNC 13399 : params.options & ~VACOPT_ANALYZE))
2106 : : {
1615 pg@bowt.ie 2107 :CBC 36 : relation_close(rel, lmode);
6204 alvherre@alvh.no-ip. 2108 : 36 : PopActiveSnapshot();
8151 tgl@sss.pgh.pa.us 2109 : 36 : CommitTransactionCommand();
998 jdavis@postgresql.or 2110 : 36 : return false;
2111 : : }
2112 : :
2113 : : /*
2114 : : * Check that it's of a vacuumable relkind.
2115 : : */
1615 pg@bowt.ie 2116 [ + + ]: 13363 : if (rel->rd_rel->relkind != RELKIND_RELATION &&
2117 [ + + ]: 4644 : rel->rd_rel->relkind != RELKIND_MATVIEW &&
2118 [ + + ]: 4640 : rel->rd_rel->relkind != RELKIND_TOASTVALUE &&
2119 [ + + ]: 94 : rel->rd_rel->relkind != RELKIND_PARTITIONED_TABLE)
2120 : : {
8084 tgl@sss.pgh.pa.us 2121 [ + - ]: 1 : ereport(WARNING,
2122 : : (errmsg("skipping \"%s\" --- cannot vacuum non-tables or special system tables",
2123 : : RelationGetRelationName(rel))));
1615 pg@bowt.ie 2124 : 1 : relation_close(rel, lmode);
6204 alvherre@alvh.no-ip. 2125 : 1 : PopActiveSnapshot();
8151 tgl@sss.pgh.pa.us 2126 : 1 : CommitTransactionCommand();
5325 rhaas@postgresql.org 2127 : 1 : return false;
2128 : : }
2129 : :
2130 : : /*
2131 : : * Silently ignore tables that are temp tables of other backends ---
2132 : : * trying to vacuum these will lead to great unhappiness, since their
2133 : : * contents are probably not up-to-date on disk. (We don't throw a
2134 : : * warning here; it would just lead to chatter during a database-wide
2135 : : * VACUUM.)
2136 : : */
1615 pg@bowt.ie 2137 [ + + + + ]: 13362 : if (RELATION_IS_OTHER_TEMP(rel))
2138 : : {
2139 : 1 : relation_close(rel, lmode);
6204 alvherre@alvh.no-ip. 2140 : 1 : PopActiveSnapshot();
8151 tgl@sss.pgh.pa.us 2141 : 1 : CommitTransactionCommand();
5325 rhaas@postgresql.org 2142 : 1 : return false;
2143 : : }
2144 : :
2145 : : /*
2146 : : * Silently ignore partitioned tables as there is no work to be done. The
2147 : : * useful work is on their child partitions, which have been queued up for
2148 : : * us separately.
2149 : : */
1615 pg@bowt.ie 2150 [ + + ]: 13361 : if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
2151 : : {
2152 : 93 : relation_close(rel, lmode);
3110 rhaas@postgresql.org 2153 : 93 : PopActiveSnapshot();
2154 : 93 : CommitTransactionCommand();
2155 : : /* It's OK to proceed with ANALYZE on this table */
2156 : 93 : return true;
2157 : : }
2158 : :
2159 : : /*
2160 : : * Get a session-level lock too. This will protect our access to the
2161 : : * relation across multiple transactions, so that we can vacuum the
2162 : : * relation's TOAST table (if any) secure in the knowledge that no one is
2163 : : * deleting the parent relation.
2164 : : *
2165 : : * NOTE: this cannot block, even if someone else is waiting for access,
2166 : : * because the lock manager knows that both lock requests are from the
2167 : : * same process.
2168 : : */
1615 pg@bowt.ie 2169 : 13268 : lockrelid = rel->rd_lockInfo.lockRelId;
2170 : 13268 : LockRelationIdForSession(&lockrelid, lmode);
2171 : :
2172 : : /*
2173 : : * Set index_cleanup option based on index_cleanup reloption if it wasn't
2174 : : * specified in VACUUM command, or when running in an autovacuum worker
2175 : : */
68 michael@paquier.xyz 2176 [ + + ]:GNC 13268 : if (params.index_cleanup == VACOPTVALUE_UNSPECIFIED)
2177 : : {
2178 : : StdRdOptIndexCleanup vacuum_index_cleanup;
2179 : :
513 akorotkov@postgresql 2180 [ + + ]:CBC 13139 : if (rel->rd_options == NULL)
1541 pg@bowt.ie 2181 : 12963 : vacuum_index_cleanup = STDRD_OPTION_VACUUM_INDEX_CLEANUP_AUTO;
2182 : : else
2183 : 176 : vacuum_index_cleanup =
513 akorotkov@postgresql 2184 : 176 : ((StdRdOptions *) rel->rd_options)->vacuum_index_cleanup;
2185 : :
1541 pg@bowt.ie 2186 [ + + ]: 13139 : if (vacuum_index_cleanup == STDRD_OPTION_VACUUM_INDEX_CLEANUP_AUTO)
68 michael@paquier.xyz 2187 :GNC 13121 : params.index_cleanup = VACOPTVALUE_AUTO;
1541 pg@bowt.ie 2188 [ + + ]:CBC 18 : else if (vacuum_index_cleanup == STDRD_OPTION_VACUUM_INDEX_CLEANUP_ON)
68 michael@paquier.xyz 2189 :GNC 9 : params.index_cleanup = VACOPTVALUE_ENABLED;
2190 : : else
2191 : : {
1541 pg@bowt.ie 2192 [ - + ]:CBC 9 : Assert(vacuum_index_cleanup ==
2193 : : STDRD_OPTION_VACUUM_INDEX_CLEANUP_OFF);
68 michael@paquier.xyz 2194 :GNC 9 : params.index_cleanup = VACOPTVALUE_DISABLED;
2195 : : }
2196 : : }
2197 : :
2198 : : #ifdef USE_INJECTION_POINTS
2199 : : if (params.index_cleanup == VACOPTVALUE_AUTO)
2200 : : INJECTION_POINT("vacuum-index-cleanup-auto", NULL);
2201 : : else if (params.index_cleanup == VACOPTVALUE_DISABLED)
2202 : : INJECTION_POINT("vacuum-index-cleanup-disabled", NULL);
2203 : : else if (params.index_cleanup == VACOPTVALUE_ENABLED)
2204 : : INJECTION_POINT("vacuum-index-cleanup-enabled", NULL);
2205 : : #endif
2206 : :
2207 : : /*
2208 : : * Check if the vacuum_max_eager_freeze_failure_rate table storage
2209 : : * parameter was specified. This overrides the GUC value.
2210 : : */
207 melanieplageman@gmai 2211 [ + + ]:CBC 13268 : if (rel->rd_options != NULL &&
2212 [ - + ]: 182 : ((StdRdOptions *) rel->rd_options)->vacuum_max_eager_freeze_failure_rate >= 0)
68 michael@paquier.xyz 2213 :UNC 0 : params.max_eager_freeze_failure_rate =
207 melanieplageman@gmai 2214 :UBC 0 : ((StdRdOptions *) rel->rd_options)->vacuum_max_eager_freeze_failure_rate;
2215 : :
2216 : : /*
2217 : : * Set truncate option based on truncate reloption or GUC if it wasn't
2218 : : * specified in VACUUM command, or when running in an autovacuum worker
2219 : : */
68 michael@paquier.xyz 2220 [ + + ]:GNC 13268 : if (params.truncate == VACOPTVALUE_UNSPECIFIED)
2221 : : {
170 nathan@postgresql.or 2222 :CBC 13142 : StdRdOptions *opts = (StdRdOptions *) rel->rd_options;
2223 : :
2224 [ + + + + ]: 13142 : if (opts && opts->vacuum_truncate_set)
2225 : : {
2226 [ + + ]: 12 : if (opts->vacuum_truncate)
68 michael@paquier.xyz 2227 :GNC 3 : params.truncate = VACOPTVALUE_ENABLED;
2228 : : else
2229 : 9 : params.truncate = VACOPTVALUE_DISABLED;
2230 : : }
170 nathan@postgresql.or 2231 [ + + ]:CBC 13130 : else if (vacuum_truncate)
68 michael@paquier.xyz 2232 :GNC 13121 : params.truncate = VACOPTVALUE_ENABLED;
2233 : : else
2234 : 9 : params.truncate = VACOPTVALUE_DISABLED;
2235 : : }
2236 : :
2237 : : #ifdef USE_INJECTION_POINTS
2238 : : if (params.truncate == VACOPTVALUE_AUTO)
2239 : : INJECTION_POINT("vacuum-truncate-auto", NULL);
2240 : : else if (params.truncate == VACOPTVALUE_DISABLED)
2241 : : INJECTION_POINT("vacuum-truncate-disabled", NULL);
2242 : : else if (params.truncate == VACOPTVALUE_ENABLED)
2243 : : INJECTION_POINT("vacuum-truncate-enabled", NULL);
2244 : : #endif
2245 : :
2246 : : /*
2247 : : * Remember the relation's TOAST relation for later, if the caller asked
2248 : : * us to process it. In VACUUM FULL, though, the toast table is
2249 : : * automatically rebuilt by cluster_rel so we shouldn't recurse to it,
2250 : : * unless PROCESS_MAIN is disabled.
2251 : : */
2252 [ + + ]: 13268 : if ((params.options & VACOPT_PROCESS_TOAST) != 0 &&
2253 [ + + ]: 13027 : ((params.options & VACOPT_FULL) == 0 ||
2254 [ + + ]: 182 : (params.options & VACOPT_PROCESS_MAIN) == 0))
1615 pg@bowt.ie 2255 :CBC 12848 : toast_relid = rel->rd_rel->reltoastrelid;
2256 : : else
6233 alvherre@alvh.no-ip. 2257 : 420 : toast_relid = InvalidOid;
2258 : :
2259 : : /*
2260 : : * Switch to the table owner's userid, so that any index functions are run
2261 : : * as that user. Also lock down security-restricted operations and
2262 : : * arrange to make GUC variable changes local to this command. (This is
2263 : : * unnecessary, but harmless, for lazy VACUUM.)
2264 : : */
5750 tgl@sss.pgh.pa.us 2265 : 13268 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1615 pg@bowt.ie 2266 : 13268 : SetUserIdAndSecContext(rel->rd_rel->relowner,
2267 : : save_sec_context | SECURITY_RESTRICTED_OPERATION);
5750 tgl@sss.pgh.pa.us 2268 : 13268 : save_nestlevel = NewGUCNestLevel();
551 jdavis@postgresql.or 2269 : 13268 : RestrictSearchPath();
2270 : :
2271 : : /*
2272 : : * If PROCESS_MAIN is set (the default), it's time to vacuum the main
2273 : : * relation. Otherwise, we can skip this part. If processing the TOAST
2274 : : * table is required (e.g., PROCESS_TOAST is set), we force PROCESS_MAIN
2275 : : * to be set when we recurse to the TOAST table.
2276 : : */
68 michael@paquier.xyz 2277 [ + + ]:GNC 13268 : if (params.options & VACOPT_PROCESS_MAIN)
2278 : : {
2279 : : /*
2280 : : * Do the actual work --- either FULL or "lazy" vacuum
2281 : : */
2282 [ + + ]: 13191 : if (params.options & VACOPT_FULL)
2283 : : {
913 michael@paquier.xyz 2284 :CBC 179 : ClusterParams cluster_params = {0};
2285 : :
68 michael@paquier.xyz 2286 [ - + ]:GNC 179 : if ((params.options & VACOPT_VERBOSE) != 0)
913 michael@paquier.xyz 2287 :UBC 0 : cluster_params.options |= CLUOPT_VERBOSE;
2288 : :
2289 : : /* VACUUM FULL is now a variant of CLUSTER; see cluster.c */
239 alvherre@alvh.no-ip. 2290 :CBC 179 : cluster_rel(rel, InvalidOid, &cluster_params);
2291 : : /* cluster_rel closes the relation, but keeps lock */
2292 : :
2293 : 176 : rel = NULL;
2294 : : }
2295 : : else
887 drowley@postgresql.o 2296 : 13012 : table_relation_vacuum(rel, params, bstrategy);
2297 : : }
2298 : :
2299 : : /* Roll back any GUC changes executed by index functions */
5750 tgl@sss.pgh.pa.us 2300 : 13265 : AtEOXact_GUC(false, save_nestlevel);
2301 : :
2302 : : /* Restore userid and security context */
2303 : 13265 : SetUserIdAndSecContext(save_userid, save_sec_context);
2304 : :
2305 : : /* all done with this class, but hold lock until commit */
1615 pg@bowt.ie 2306 [ + + ]: 13265 : if (rel)
2307 : 13089 : relation_close(rel, NoLock);
2308 : :
2309 : : /*
2310 : : * Complete the transaction and free all temporary memory used.
2311 : : */
6204 alvherre@alvh.no-ip. 2312 : 13265 : PopActiveSnapshot();
8151 tgl@sss.pgh.pa.us 2313 : 13265 : CommitTransactionCommand();
2314 : :
2315 : : /*
2316 : : * If the relation has a secondary toast rel, vacuum that too while we
2317 : : * still hold the session lock on the main table. Note however that
2318 : : * "analyze" will not get done on the toast table. This is good, because
2319 : : * the toaster always uses hardcoded index access and statistics are
2320 : : * totally unimportant for toast relations.
2321 : : */
8822 2322 [ + + ]: 13265 : if (toast_relid != InvalidOid)
2323 : : {
2324 : : /*
2325 : : * Force VACOPT_PROCESS_MAIN so vacuum_rel() processes it. Likewise,
2326 : : * set toast_parent so that the privilege checks are done on the main
2327 : : * relation. NB: This is only safe to do because we hold a session
2328 : : * lock on the main relation that prevents concurrent deletion.
2329 : : */
915 michael@paquier.xyz 2330 : 4543 : toast_vacuum_params.options |= VACOPT_PROCESS_MAIN;
542 nathan@postgresql.or 2331 : 4543 : toast_vacuum_params.toast_parent = relid;
2332 : :
68 michael@paquier.xyz 2333 :GNC 4543 : vacuum_rel(toast_relid, NULL, toast_vacuum_params, bstrategy);
2334 : : }
2335 : :
2336 : : /*
2337 : : * Now release the session-level lock on the main table.
2338 : : */
1615 pg@bowt.ie 2339 :CBC 13265 : UnlockRelationIdForSession(&lockrelid, lmode);
2340 : :
2341 : : /* Report that we really did it. */
5325 rhaas@postgresql.org 2342 : 13265 : return true;
2343 : : }
2344 : :
2345 : :
2346 : : /*
2347 : : * Open all the vacuumable indexes of the given relation, obtaining the
2348 : : * specified kind of lock on each. Return an array of Relation pointers for
2349 : : * the indexes into *Irel, and the number of indexes into *nindexes.
2350 : : *
2351 : : * We consider an index vacuumable if it is marked insertable (indisready).
2352 : : * If it isn't, probably a CREATE INDEX CONCURRENTLY command failed early in
2353 : : * execution, and what we have is too corrupt to be processable. We will
2354 : : * vacuum even if the index isn't indisvalid; this is important because in a
2355 : : * unique index, uniqueness checks will be performed anyway and had better not
2356 : : * hit dangling index pointers.
2357 : : */
2358 : : void
5689 tgl@sss.pgh.pa.us 2359 : 20671 : vac_open_indexes(Relation relation, LOCKMODE lockmode,
2360 : : int *nindexes, Relation **Irel)
2361 : : {
2362 : : List *indexoidlist;
2363 : : ListCell *indexoidscan;
2364 : : int i;
2365 : :
2366 [ - + ]: 20671 : Assert(lockmode != NoLock);
2367 : :
2368 : 20671 : indexoidlist = RelationGetIndexList(relation);
2369 : :
2370 : : /* allocate enough memory for all indexes */
4665 2371 : 20671 : i = list_length(indexoidlist);
2372 : :
2373 [ + + ]: 20671 : if (i > 0)
2374 : 18022 : *Irel = (Relation *) palloc(i * sizeof(Relation));
2375 : : else
5689 2376 : 2649 : *Irel = NULL;
2377 : :
2378 : : /* collect just the ready indexes */
2379 : 20671 : i = 0;
2380 [ + + + + : 50816 : foreach(indexoidscan, indexoidlist)
+ + ]
2381 : : {
2382 : 30145 : Oid indexoid = lfirst_oid(indexoidscan);
2383 : : Relation indrel;
2384 : :
4665 2385 : 30145 : indrel = index_open(indexoid, lockmode);
2445 peter_e@gmx.net 2386 [ + - ]: 30145 : if (indrel->rd_index->indisready)
4665 tgl@sss.pgh.pa.us 2387 : 30145 : (*Irel)[i++] = indrel;
2388 : : else
4665 tgl@sss.pgh.pa.us 2389 :UBC 0 : index_close(indrel, lockmode);
2390 : : }
2391 : :
4665 tgl@sss.pgh.pa.us 2392 :CBC 20671 : *nindexes = i;
2393 : :
5689 2394 : 20671 : list_free(indexoidlist);
9231 bruce@momjian.us 2395 : 20671 : }
2396 : :
2397 : : /*
2398 : : * Release the resources acquired by vac_open_indexes. Optionally release
2399 : : * the locks (say NoLock to keep 'em).
2400 : : */
2401 : : void
5689 tgl@sss.pgh.pa.us 2402 : 21090 : vac_close_indexes(int nindexes, Relation *Irel, LOCKMODE lockmode)
2403 : : {
2404 [ + + ]: 21090 : if (Irel == NULL)
2405 : 3071 : return;
2406 : :
2407 [ + + ]: 48158 : while (nindexes--)
2408 : : {
2409 : 30139 : Relation ind = Irel[nindexes];
2410 : :
6977 2411 : 30139 : index_close(ind, lockmode);
2412 : : }
10226 bruce@momjian.us 2413 : 18019 : pfree(Irel);
2414 : : }
2415 : :
2416 : : /*
2417 : : * vacuum_delay_point --- check for interrupts and cost-based delay.
2418 : : *
2419 : : * This should be called in each major loop of VACUUM processing,
2420 : : * typically once per page processed.
2421 : : */
2422 : : void
207 nathan@postgresql.or 2423 : 42654875 : vacuum_delay_point(bool is_analyze)
2424 : : {
2056 akapila@postgresql.o 2425 : 42654875 : double msec = 0;
2426 : :
2427 : : /* Always check for interrupts */
7879 tgl@sss.pgh.pa.us 2428 [ - + ]: 42654875 : CHECK_FOR_INTERRUPTS();
2429 : :
883 dgustafsson@postgres 2430 [ + - ]: 42654875 : if (InterruptPending ||
2431 [ + + + - ]: 42654875 : (!VacuumCostActive && !ConfigReloadPending))
2432 : 37912628 : return;
2433 : :
2434 : : /*
2435 : : * Autovacuum workers should reload the configuration file if requested.
2436 : : * This allows changes to [autovacuum_]vacuum_cost_limit and
2437 : : * [autovacuum_]vacuum_cost_delay to take effect while a table is being
2438 : : * vacuumed or analyzed.
2439 : : */
551 heikki.linnakangas@i 2440 [ - + - - ]: 4742247 : if (ConfigReloadPending && AmAutoVacuumWorkerProcess())
2441 : : {
883 dgustafsson@postgres 2442 :UBC 0 : ConfigReloadPending = false;
2443 : 0 : ProcessConfigFile(PGC_SIGHUP);
2444 : 0 : VacuumUpdateCosts();
2445 : : }
2446 : :
2447 : : /*
2448 : : * If we disabled cost-based delays after reloading the config file,
2449 : : * return.
2450 : : */
883 dgustafsson@postgres 2451 [ - + ]:CBC 4742247 : if (!VacuumCostActive)
2056 akapila@postgresql.o 2452 :UBC 0 : return;
2453 : :
2454 : : /*
2455 : : * For parallel vacuum, the delay is computed based on the shared cost
2456 : : * balance. See compute_parallel_delay.
2457 : : */
2056 akapila@postgresql.o 2458 [ - + ]:CBC 4742247 : if (VacuumSharedCostBalance != NULL)
2056 akapila@postgresql.o 2459 :UBC 0 : msec = compute_parallel_delay();
883 dgustafsson@postgres 2460 [ + + ]:CBC 4742247 : else if (VacuumCostBalance >= vacuum_cost_limit)
2461 : 1265 : msec = vacuum_cost_delay * VacuumCostBalance / vacuum_cost_limit;
2462 : :
2463 : : /* Nap if appropriate */
2056 akapila@postgresql.o 2464 [ + + ]: 4742247 : if (msec > 0)
2465 : : {
2466 : : instr_time delay_start;
2467 : :
883 dgustafsson@postgres 2468 [ + + ]: 1265 : if (msec > vacuum_cost_delay * 4)
2469 : 8 : msec = vacuum_cost_delay * 4;
2470 : :
207 nathan@postgresql.or 2471 [ - + ]: 1265 : if (track_cost_delay_timing)
207 nathan@postgresql.or 2472 :UBC 0 : INSTR_TIME_SET_CURRENT(delay_start);
2473 : :
906 tmunro@postgresql.or 2474 :CBC 1265 : pgstat_report_wait_start(WAIT_EVENT_VACUUM_DELAY);
2475 : 1265 : pg_usleep(msec * 1000);
2476 : 1265 : pgstat_report_wait_end();
2477 : :
207 nathan@postgresql.or 2478 [ - + ]: 1265 : if (track_cost_delay_timing)
2479 : : {
2480 : : instr_time delay_end;
2481 : : instr_time delay;
2482 : :
207 nathan@postgresql.or 2483 :UBC 0 : INSTR_TIME_SET_CURRENT(delay_end);
2484 : 0 : INSTR_TIME_SET_ZERO(delay);
2485 : 0 : INSTR_TIME_ACCUM_DIFF(delay, delay_end, delay_start);
2486 : :
2487 : : /*
2488 : : * For parallel workers, we only report the delay time every once
2489 : : * in a while to avoid overloading the leader with messages and
2490 : : * interrupts.
2491 : : */
2492 [ # # ]: 0 : if (IsParallelWorker())
2493 : : {
2494 : : static instr_time last_report_time;
2495 : : instr_time time_since_last_report;
2496 : :
2497 [ # # ]: 0 : Assert(!is_analyze);
2498 : :
2499 : : /* Accumulate the delay time */
2500 : 0 : parallel_vacuum_worker_delay_ns += INSTR_TIME_GET_NANOSEC(delay);
2501 : :
2502 : : /* Calculate interval since last report */
2503 : 0 : INSTR_TIME_SET_ZERO(time_since_last_report);
2504 : 0 : INSTR_TIME_ACCUM_DIFF(time_since_last_report, delay_end, last_report_time);
2505 : :
2506 : : /* If we haven't reported in a while, do so now */
2507 [ # # ]: 0 : if (INSTR_TIME_GET_NANOSEC(time_since_last_report) >=
2508 : : PARALLEL_VACUUM_DELAY_REPORT_INTERVAL_NS)
2509 : : {
2510 : 0 : pgstat_progress_parallel_incr_param(PROGRESS_VACUUM_DELAY_TIME,
2511 : : parallel_vacuum_worker_delay_ns);
2512 : :
2513 : : /* Reset variables */
2514 : 0 : last_report_time = delay_end;
2515 : 0 : parallel_vacuum_worker_delay_ns = 0;
2516 : : }
2517 : : }
2518 [ # # ]: 0 : else if (is_analyze)
2519 : 0 : pgstat_progress_incr_param(PROGRESS_ANALYZE_DELAY_TIME,
2520 : 0 : INSTR_TIME_GET_NANOSEC(delay));
2521 : : else
2522 : 0 : pgstat_progress_incr_param(PROGRESS_VACUUM_DELAY_TIME,
2523 : 0 : INSTR_TIME_GET_NANOSEC(delay));
2524 : : }
2525 : :
2526 : : /*
2527 : : * We don't want to ignore postmaster death during very long vacuums
2528 : : * with vacuum_cost_delay configured. We can't use the usual
2529 : : * WaitLatch() approach here because we want microsecond-based sleep
2530 : : * durations above.
2531 : : */
906 tmunro@postgresql.or 2532 [ + - - + ]:CBC 1265 : if (IsUnderPostmaster && !PostmasterIsAlive())
906 tmunro@postgresql.or 2533 :UBC 0 : exit(1);
2534 : :
7879 tgl@sss.pgh.pa.us 2535 :CBC 1265 : VacuumCostBalance = 0;
2536 : :
2537 : : /*
2538 : : * Balance and update limit values for autovacuum workers. We must do
2539 : : * this periodically, as the number of workers across which we are
2540 : : * balancing the limit may have changed.
2541 : : *
2542 : : * TODO: There may be better criteria for determining when to do this
2543 : : * besides "check after napping".
2544 : : */
883 dgustafsson@postgres 2545 : 1265 : AutoVacuumUpdateCostLimit();
2546 : :
2547 : : /* Might have gotten an interrupt while sleeping */
7879 tgl@sss.pgh.pa.us 2548 [ - + ]: 1265 : CHECK_FOR_INTERRUPTS();
2549 : : }
2550 : : }
2551 : :
2552 : : /*
2553 : : * Computes the vacuum delay for parallel workers.
2554 : : *
2555 : : * The basic idea of a cost-based delay for parallel vacuum is to allow each
2556 : : * worker to sleep in proportion to the share of work it's done. We achieve this
2557 : : * by allowing all parallel vacuum workers including the leader process to
2558 : : * have a shared view of cost related parameters (mainly VacuumCostBalance).
2559 : : * We allow each worker to update it as and when it has incurred any cost and
2560 : : * then based on that decide whether it needs to sleep. We compute the time
2561 : : * to sleep for a worker based on the cost it has incurred
2562 : : * (VacuumCostBalanceLocal) and then reduce the VacuumSharedCostBalance by
2563 : : * that amount. This avoids putting to sleep those workers which have done less
2564 : : * I/O than other workers and therefore ensure that workers
2565 : : * which are doing more I/O got throttled more.
2566 : : *
2567 : : * We allow a worker to sleep only if it has performed I/O above a certain
2568 : : * threshold, which is calculated based on the number of active workers
2569 : : * (VacuumActiveNWorkers), and the overall cost balance is more than
2570 : : * VacuumCostLimit set by the system. Testing reveals that we achieve
2571 : : * the required throttling if we force a worker that has done more than 50%
2572 : : * of its share of work to sleep.
2573 : : */
2574 : : static double
2056 akapila@postgresql.o 2575 :UBC 0 : compute_parallel_delay(void)
2576 : : {
2577 : 0 : double msec = 0;
2578 : : uint32 shared_balance;
2579 : : int nworkers;
2580 : :
2581 : : /* Parallel vacuum must be active */
2582 [ # # ]: 0 : Assert(VacuumSharedCostBalance);
2583 : :
2584 : 0 : nworkers = pg_atomic_read_u32(VacuumActiveNWorkers);
2585 : :
2586 : : /* At least count itself */
2587 [ # # ]: 0 : Assert(nworkers >= 1);
2588 : :
2589 : : /* Update the shared cost balance value atomically */
2590 : 0 : shared_balance = pg_atomic_add_fetch_u32(VacuumSharedCostBalance, VacuumCostBalance);
2591 : :
2592 : : /* Compute the total local balance for the current worker */
2593 : 0 : VacuumCostBalanceLocal += VacuumCostBalance;
2594 : :
883 dgustafsson@postgres 2595 [ # # ]: 0 : if ((shared_balance >= vacuum_cost_limit) &&
2596 [ # # ]: 0 : (VacuumCostBalanceLocal > 0.5 * ((double) vacuum_cost_limit / nworkers)))
2597 : : {
2598 : : /* Compute sleep time based on the local cost balance */
2599 : 0 : msec = vacuum_cost_delay * VacuumCostBalanceLocal / vacuum_cost_limit;
2056 akapila@postgresql.o 2600 : 0 : pg_atomic_sub_fetch_u32(VacuumSharedCostBalance, VacuumCostBalanceLocal);
2601 : 0 : VacuumCostBalanceLocal = 0;
2602 : : }
2603 : :
2604 : : /*
2605 : : * Reset the local balance as we accumulated it into the shared value.
2606 : : */
2607 : 0 : VacuumCostBalance = 0;
2608 : :
2609 : 0 : return msec;
2610 : : }
2611 : :
2612 : : /*
2613 : : * A wrapper function of defGetBoolean().
2614 : : *
2615 : : * This function returns VACOPTVALUE_ENABLED and VACOPTVALUE_DISABLED instead
2616 : : * of true and false.
2617 : : */
2618 : : static VacOptValue
1541 pg@bowt.ie 2619 :CBC 163 : get_vacoptval_from_boolean(DefElem *def)
2620 : : {
2621 [ + + ]: 163 : return defGetBoolean(def) ? VACOPTVALUE_ENABLED : VACOPTVALUE_DISABLED;
2622 : : }
2623 : :
2624 : : /*
2625 : : * vac_bulkdel_one_index() -- bulk-deletion for index relation.
2626 : : *
2627 : : * Returns bulk delete stats derived from input stats
2628 : : */
2629 : : IndexBulkDeleteResult *
1354 akapila@postgresql.o 2630 : 1302 : vac_bulkdel_one_index(IndexVacuumInfo *ivinfo, IndexBulkDeleteResult *istat,
2631 : : TidStore *dead_items, VacDeadItemsInfo *dead_items_info)
2632 : : {
2633 : : /* Do bulk deletion */
2634 : 1302 : istat = index_bulk_delete(ivinfo, istat, vac_tid_reaped,
2635 : : dead_items);
2636 : :
2637 [ + + ]: 1302 : ereport(ivinfo->message_level,
2638 : : (errmsg("scanned index \"%s\" to remove %" PRId64 " row versions",
2639 : : RelationGetRelationName(ivinfo->index),
2640 : : dead_items_info->num_items)));
2641 : :
2642 : 1302 : return istat;
2643 : : }
2644 : :
2645 : : /*
2646 : : * vac_cleanup_one_index() -- do post-vacuum cleanup for index relation.
2647 : : *
2648 : : * Returns bulk delete stats derived from input stats
2649 : : */
2650 : : IndexBulkDeleteResult *
2651 : 18501 : vac_cleanup_one_index(IndexVacuumInfo *ivinfo, IndexBulkDeleteResult *istat)
2652 : : {
2653 : 18501 : istat = index_vacuum_cleanup(ivinfo, istat);
2654 : :
2655 [ + + ]: 18501 : if (istat)
2656 [ + + ]: 1434 : ereport(ivinfo->message_level,
2657 : : (errmsg("index \"%s\" now contains %.0f row versions in %u pages",
2658 : : RelationGetRelationName(ivinfo->index),
2659 : : istat->num_index_tuples,
2660 : : istat->num_pages),
2661 : : errdetail("%.0f index row versions were removed.\n"
2662 : : "%u index pages were newly deleted.\n"
2663 : : "%u index pages are currently deleted, of which %u are currently reusable.",
2664 : : istat->tuples_removed,
2665 : : istat->pages_newly_deleted,
2666 : : istat->pages_deleted, istat->pages_free)));
2667 : :
2668 : 18501 : return istat;
2669 : : }
2670 : :
2671 : : /*
2672 : : * vac_tid_reaped() -- is a particular tid deletable?
2673 : : *
2674 : : * This has the right signature to be an IndexBulkDeleteCallback.
2675 : : */
2676 : : static bool
2677 : 3575508 : vac_tid_reaped(ItemPointer itemptr, void *state)
2678 : : {
522 msawada@postgresql.o 2679 : 3575508 : TidStore *dead_items = (TidStore *) state;
2680 : :
2681 : 3575508 : return TidStoreIsMember(dead_items, itemptr);
2682 : : }
|
