Line data Source code
1 : /*
2 : * contrib/pgstattuple/pgstattuple.c
3 : *
4 : * Copyright (c) 2001,2002 Tatsuo Ishii
5 : *
6 : * Permission to use, copy, modify, and distribute this software and
7 : * its documentation for any purpose, without fee, and without a
8 : * written agreement is hereby granted, provided that the above
9 : * copyright notice and this paragraph and the following two
10 : * paragraphs appear in all copies.
11 : *
12 : * IN NO EVENT SHALL THE AUTHOR BE LIABLE TO ANY PARTY FOR DIRECT,
13 : * INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES, INCLUDING
14 : * LOST PROFITS, ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS
15 : * DOCUMENTATION, EVEN IF THE UNIVERSITY OF CALIFORNIA HAS BEEN ADVISED
16 : * OF THE POSSIBILITY OF SUCH DAMAGE.
17 : *
18 : * THE AUTHOR SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING, BUT NOT
19 : * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
20 : * A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS ON AN "AS
21 : * IS" BASIS, AND THE AUTHOR HAS NO OBLIGATIONS TO PROVIDE MAINTENANCE,
22 : * SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
23 : */
24 :
25 : #include "postgres.h"
26 :
27 : #include "access/gist_private.h"
28 : #include "access/hash.h"
29 : #include "access/heapam.h"
30 : #include "access/nbtree.h"
31 : #include "access/relscan.h"
32 : #include "access/tableam.h"
33 : #include "catalog/namespace.h"
34 : #include "catalog/pg_am_d.h"
35 : #include "funcapi.h"
36 : #include "miscadmin.h"
37 : #include "storage/bufmgr.h"
38 : #include "storage/lmgr.h"
39 : #include "utils/varlena.h"
40 :
41 0 : PG_MODULE_MAGIC_EXT(
42 : .name = "pgstattuple",
43 : .version = PG_VERSION
44 : );
45 :
46 0 : PG_FUNCTION_INFO_V1(pgstattuple);
47 0 : PG_FUNCTION_INFO_V1(pgstattuple_v1_5);
48 0 : PG_FUNCTION_INFO_V1(pgstattuplebyid);
49 0 : PG_FUNCTION_INFO_V1(pgstattuplebyid_v1_5);
50 :
51 : /*
52 : * struct pgstattuple_type
53 : *
54 : * tuple_percent, dead_tuple_percent and free_percent are computable,
55 : * so not defined here.
56 : */
57 : typedef struct pgstattuple_type
58 : {
59 : uint64 table_len;
60 : uint64 tuple_count;
61 : uint64 tuple_len;
62 : uint64 dead_tuple_count;
63 : uint64 dead_tuple_len;
64 : uint64 free_space; /* free/reusable space in bytes */
65 : } pgstattuple_type;
66 :
67 : typedef void (*pgstat_page) (pgstattuple_type *, Relation, BlockNumber,
68 : BufferAccessStrategy);
69 :
70 : static Datum build_pgstattuple_type(pgstattuple_type *stat,
71 : FunctionCallInfo fcinfo);
72 : static Datum pgstat_relation(Relation rel, FunctionCallInfo fcinfo);
73 : static Datum pgstat_heap(Relation rel, FunctionCallInfo fcinfo);
74 : static void pgstat_btree_page(pgstattuple_type *stat,
75 : Relation rel, BlockNumber blkno,
76 : BufferAccessStrategy bstrategy);
77 : static void pgstat_hash_page(pgstattuple_type *stat,
78 : Relation rel, BlockNumber blkno,
79 : BufferAccessStrategy bstrategy);
80 : static void pgstat_gist_page(pgstattuple_type *stat,
81 : Relation rel, BlockNumber blkno,
82 : BufferAccessStrategy bstrategy);
83 : static Datum pgstat_index(Relation rel, BlockNumber start,
84 : pgstat_page pagefn, FunctionCallInfo fcinfo);
85 : static void pgstat_index_page(pgstattuple_type *stat, Page page,
86 : OffsetNumber minoff, OffsetNumber maxoff);
87 :
88 : /*
89 : * build_pgstattuple_type -- build a pgstattuple_type tuple
90 : */
91 : static Datum
92 0 : build_pgstattuple_type(pgstattuple_type *stat, FunctionCallInfo fcinfo)
93 : {
94 : #define NCOLUMNS 9
95 : #define NCHARS 314
96 :
97 0 : HeapTuple tuple;
98 0 : char *values[NCOLUMNS];
99 0 : char values_buf[NCOLUMNS][NCHARS];
100 0 : int i;
101 0 : double tuple_percent;
102 0 : double dead_tuple_percent;
103 0 : double free_percent; /* free/reusable space in % */
104 0 : TupleDesc tupdesc;
105 0 : AttInMetadata *attinmeta;
106 :
107 : /* Build a tuple descriptor for our result type */
108 0 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
109 0 : elog(ERROR, "return type must be a row type");
110 :
111 : /*
112 : * Generate attribute metadata needed later to produce tuples from raw C
113 : * strings
114 : */
115 0 : attinmeta = TupleDescGetAttInMetadata(tupdesc);
116 :
117 0 : if (stat->table_len == 0)
118 : {
119 0 : tuple_percent = 0.0;
120 0 : dead_tuple_percent = 0.0;
121 0 : free_percent = 0.0;
122 0 : }
123 : else
124 : {
125 0 : tuple_percent = 100.0 * stat->tuple_len / stat->table_len;
126 0 : dead_tuple_percent = 100.0 * stat->dead_tuple_len / stat->table_len;
127 0 : free_percent = 100.0 * stat->free_space / stat->table_len;
128 : }
129 :
130 : /*
131 : * Prepare a values array for constructing the tuple. This should be an
132 : * array of C strings which will be processed later by the appropriate
133 : * "in" functions.
134 : */
135 0 : for (i = 0; i < NCOLUMNS; i++)
136 0 : values[i] = values_buf[i];
137 0 : i = 0;
138 0 : snprintf(values[i++], NCHARS, INT64_FORMAT, stat->table_len);
139 0 : snprintf(values[i++], NCHARS, INT64_FORMAT, stat->tuple_count);
140 0 : snprintf(values[i++], NCHARS, INT64_FORMAT, stat->tuple_len);
141 0 : snprintf(values[i++], NCHARS, "%.2f", tuple_percent);
142 0 : snprintf(values[i++], NCHARS, INT64_FORMAT, stat->dead_tuple_count);
143 0 : snprintf(values[i++], NCHARS, INT64_FORMAT, stat->dead_tuple_len);
144 0 : snprintf(values[i++], NCHARS, "%.2f", dead_tuple_percent);
145 0 : snprintf(values[i++], NCHARS, INT64_FORMAT, stat->free_space);
146 0 : snprintf(values[i++], NCHARS, "%.2f", free_percent);
147 :
148 : /* build a tuple */
149 0 : tuple = BuildTupleFromCStrings(attinmeta, values);
150 :
151 : /* make the tuple into a datum */
152 0 : return HeapTupleGetDatum(tuple);
153 0 : }
154 :
155 : /* ----------
156 : * pgstattuple:
157 : * returns live/dead tuples info
158 : *
159 : * C FUNCTION definition
160 : * pgstattuple(text) returns pgstattuple_type
161 : *
162 : * The superuser() check here must be kept as the library might be upgraded
163 : * without the extension being upgraded, meaning that in pre-1.5 installations
164 : * these functions could be called by any user.
165 : * ----------
166 : */
167 :
168 : Datum
169 0 : pgstattuple(PG_FUNCTION_ARGS)
170 : {
171 0 : text *relname = PG_GETARG_TEXT_PP(0);
172 0 : RangeVar *relrv;
173 0 : Relation rel;
174 :
175 0 : if (!superuser())
176 0 : ereport(ERROR,
177 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
178 : errmsg("must be superuser to use pgstattuple functions")));
179 :
180 : /* open relation */
181 0 : relrv = makeRangeVarFromNameList(textToQualifiedNameList(relname));
182 0 : rel = relation_openrv(relrv, AccessShareLock);
183 :
184 0 : PG_RETURN_DATUM(pgstat_relation(rel, fcinfo));
185 0 : }
186 :
187 : /*
188 : * As of pgstattuple version 1.5, we no longer need to check if the user
189 : * is a superuser because we REVOKE EXECUTE on the function from PUBLIC.
190 : * Users can then grant access to it based on their policies.
191 : *
192 : * Otherwise identical to pgstattuple (above).
193 : */
194 : Datum
195 0 : pgstattuple_v1_5(PG_FUNCTION_ARGS)
196 : {
197 0 : text *relname = PG_GETARG_TEXT_PP(0);
198 0 : RangeVar *relrv;
199 0 : Relation rel;
200 :
201 : /* open relation */
202 0 : relrv = makeRangeVarFromNameList(textToQualifiedNameList(relname));
203 0 : rel = relation_openrv(relrv, AccessShareLock);
204 :
205 0 : PG_RETURN_DATUM(pgstat_relation(rel, fcinfo));
206 0 : }
207 :
208 : /* Must keep superuser() check, see above. */
209 : Datum
210 0 : pgstattuplebyid(PG_FUNCTION_ARGS)
211 : {
212 0 : Oid relid = PG_GETARG_OID(0);
213 0 : Relation rel;
214 :
215 0 : if (!superuser())
216 0 : ereport(ERROR,
217 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
218 : errmsg("must be superuser to use pgstattuple functions")));
219 :
220 : /* open relation */
221 0 : rel = relation_open(relid, AccessShareLock);
222 :
223 0 : PG_RETURN_DATUM(pgstat_relation(rel, fcinfo));
224 0 : }
225 :
226 : /* Remove superuser() check for 1.5 version, see above */
227 : Datum
228 0 : pgstattuplebyid_v1_5(PG_FUNCTION_ARGS)
229 : {
230 0 : Oid relid = PG_GETARG_OID(0);
231 0 : Relation rel;
232 :
233 : /* open relation */
234 0 : rel = relation_open(relid, AccessShareLock);
235 :
236 0 : PG_RETURN_DATUM(pgstat_relation(rel, fcinfo));
237 0 : }
238 :
239 : /*
240 : * pgstat_relation
241 : */
242 : static Datum
243 0 : pgstat_relation(Relation rel, FunctionCallInfo fcinfo)
244 : {
245 0 : const char *err;
246 :
247 : /*
248 : * Reject attempts to read non-local temporary relations; we would be
249 : * likely to get wrong data since we have no visibility into the owning
250 : * session's local buffers.
251 : */
252 0 : if (RELATION_IS_OTHER_TEMP(rel))
253 0 : ereport(ERROR,
254 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
255 : errmsg("cannot access temporary tables of other sessions")));
256 :
257 0 : if (RELKIND_HAS_TABLE_AM(rel->rd_rel->relkind) ||
258 0 : rel->rd_rel->relkind == RELKIND_SEQUENCE)
259 : {
260 0 : return pgstat_heap(rel, fcinfo);
261 : }
262 0 : else if (rel->rd_rel->relkind == RELKIND_INDEX)
263 : {
264 : /* see pgstatindex_impl */
265 0 : if (!rel->rd_index->indisvalid)
266 0 : ereport(ERROR,
267 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
268 : errmsg("index \"%s\" is not valid",
269 : RelationGetRelationName(rel))));
270 :
271 0 : switch (rel->rd_rel->relam)
272 : {
273 : case BTREE_AM_OID:
274 0 : return pgstat_index(rel, BTREE_METAPAGE + 1,
275 0 : pgstat_btree_page, fcinfo);
276 : case HASH_AM_OID:
277 0 : return pgstat_index(rel, HASH_METAPAGE + 1,
278 0 : pgstat_hash_page, fcinfo);
279 : case GIST_AM_OID:
280 0 : return pgstat_index(rel, GIST_ROOT_BLKNO + 1,
281 0 : pgstat_gist_page, fcinfo);
282 : case GIN_AM_OID:
283 0 : err = "gin index";
284 0 : break;
285 : case SPGIST_AM_OID:
286 0 : err = "spgist index";
287 0 : break;
288 : case BRIN_AM_OID:
289 0 : err = "brin index";
290 0 : break;
291 : default:
292 0 : err = "unknown index";
293 0 : break;
294 : }
295 0 : ereport(ERROR,
296 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
297 : errmsg("index \"%s\" (%s) is not supported",
298 : RelationGetRelationName(rel), err)));
299 0 : }
300 : else
301 : {
302 0 : ereport(ERROR,
303 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
304 : errmsg("cannot get tuple-level statistics for relation \"%s\"",
305 : RelationGetRelationName(rel)),
306 : errdetail_relkind_not_supported(rel->rd_rel->relkind)));
307 : }
308 :
309 0 : return 0; /* should not happen */
310 0 : }
311 :
312 : /*
313 : * pgstat_heap -- returns live/dead tuples info in a heap
314 : */
315 : static Datum
316 0 : pgstat_heap(Relation rel, FunctionCallInfo fcinfo)
317 : {
318 0 : TableScanDesc scan;
319 0 : HeapScanDesc hscan;
320 0 : HeapTuple tuple;
321 0 : BlockNumber nblocks;
322 0 : BlockNumber block = 0; /* next block to count free space in */
323 0 : BlockNumber tupblock;
324 0 : Buffer buffer;
325 0 : pgstattuple_type stat = {0};
326 0 : SnapshotData SnapshotDirty;
327 :
328 : /*
329 : * Sequences always use heap AM, but they don't show that in the catalogs.
330 : */
331 0 : if (rel->rd_rel->relkind != RELKIND_SEQUENCE &&
332 0 : rel->rd_rel->relam != HEAP_TABLE_AM_OID)
333 0 : ereport(ERROR,
334 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
335 : errmsg("only heap AM is supported")));
336 :
337 : /* Disable syncscan because we assume we scan from block zero upwards */
338 0 : scan = table_beginscan_strat(rel, SnapshotAny, 0, NULL, true, false);
339 0 : hscan = (HeapScanDesc) scan;
340 :
341 0 : InitDirtySnapshot(SnapshotDirty);
342 :
343 0 : nblocks = hscan->rs_nblocks; /* # blocks to be scanned */
344 :
345 : /* scan the relation */
346 0 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
347 : {
348 0 : CHECK_FOR_INTERRUPTS();
349 :
350 : /* must hold a buffer lock to call HeapTupleSatisfiesVisibility */
351 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
352 :
353 0 : if (HeapTupleSatisfiesVisibility(tuple, &SnapshotDirty, hscan->rs_cbuf))
354 : {
355 0 : stat.tuple_len += tuple->t_len;
356 0 : stat.tuple_count++;
357 0 : }
358 : else
359 : {
360 0 : stat.dead_tuple_len += tuple->t_len;
361 0 : stat.dead_tuple_count++;
362 : }
363 :
364 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
365 :
366 : /*
367 : * To avoid physically reading the table twice, try to do the
368 : * free-space scan in parallel with the heap scan. However,
369 : * heap_getnext may find no tuples on a given page, so we cannot
370 : * simply examine the pages returned by the heap scan.
371 : */
372 0 : tupblock = ItemPointerGetBlockNumber(&tuple->t_self);
373 :
374 0 : while (block <= tupblock)
375 : {
376 0 : CHECK_FOR_INTERRUPTS();
377 :
378 0 : buffer = ReadBufferExtended(rel, MAIN_FORKNUM, block,
379 0 : RBM_NORMAL, hscan->rs_strategy);
380 0 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
381 0 : stat.free_space += PageGetExactFreeSpace(BufferGetPage(buffer));
382 0 : UnlockReleaseBuffer(buffer);
383 0 : block++;
384 : }
385 : }
386 :
387 0 : while (block < nblocks)
388 : {
389 0 : CHECK_FOR_INTERRUPTS();
390 :
391 0 : buffer = ReadBufferExtended(rel, MAIN_FORKNUM, block,
392 0 : RBM_NORMAL, hscan->rs_strategy);
393 0 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
394 0 : stat.free_space += PageGetExactFreeSpace(BufferGetPage(buffer));
395 0 : UnlockReleaseBuffer(buffer);
396 0 : block++;
397 : }
398 :
399 0 : table_endscan(scan);
400 0 : relation_close(rel, AccessShareLock);
401 :
402 0 : stat.table_len = (uint64) nblocks * BLCKSZ;
403 :
404 0 : return build_pgstattuple_type(&stat, fcinfo);
405 0 : }
406 :
407 : /*
408 : * pgstat_btree_page -- check tuples in a btree page
409 : */
410 : static void
411 0 : pgstat_btree_page(pgstattuple_type *stat, Relation rel, BlockNumber blkno,
412 : BufferAccessStrategy bstrategy)
413 : {
414 0 : Buffer buf;
415 0 : Page page;
416 :
417 0 : buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, bstrategy);
418 0 : LockBuffer(buf, BT_READ);
419 0 : page = BufferGetPage(buf);
420 :
421 : /* Page is valid, see what to do with it */
422 0 : if (PageIsNew(page))
423 : {
424 : /* fully empty page */
425 0 : stat->free_space += BLCKSZ;
426 0 : }
427 0 : else if (PageGetSpecialSize(page) == MAXALIGN(sizeof(BTPageOpaqueData)))
428 : {
429 0 : BTPageOpaque opaque;
430 :
431 0 : opaque = BTPageGetOpaque(page);
432 0 : if (P_IGNORE(opaque))
433 : {
434 : /* deleted or half-dead page */
435 0 : stat->free_space += BLCKSZ;
436 0 : }
437 0 : else if (P_ISLEAF(opaque))
438 : {
439 0 : pgstat_index_page(stat, page, P_FIRSTDATAKEY(opaque),
440 0 : PageGetMaxOffsetNumber(page));
441 0 : }
442 : else
443 : {
444 : /* internal page */
445 : }
446 0 : }
447 :
448 0 : _bt_relbuf(rel, buf);
449 0 : }
450 :
451 : /*
452 : * pgstat_hash_page -- check tuples in a hash page
453 : */
454 : static void
455 0 : pgstat_hash_page(pgstattuple_type *stat, Relation rel, BlockNumber blkno,
456 : BufferAccessStrategy bstrategy)
457 : {
458 0 : Buffer buf;
459 0 : Page page;
460 :
461 0 : buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, bstrategy);
462 0 : LockBuffer(buf, HASH_READ);
463 0 : page = BufferGetPage(buf);
464 :
465 0 : if (PageIsNew(page))
466 : {
467 : /* fully empty page */
468 0 : stat->free_space += BLCKSZ;
469 0 : }
470 0 : else if (PageGetSpecialSize(page) == MAXALIGN(sizeof(HashPageOpaqueData)))
471 : {
472 0 : HashPageOpaque opaque;
473 :
474 0 : opaque = HashPageGetOpaque(page);
475 0 : switch (opaque->hasho_flag & LH_PAGE_TYPE)
476 : {
477 : case LH_UNUSED_PAGE:
478 0 : stat->free_space += BLCKSZ;
479 0 : break;
480 : case LH_BUCKET_PAGE:
481 : case LH_OVERFLOW_PAGE:
482 0 : pgstat_index_page(stat, page, FirstOffsetNumber,
483 0 : PageGetMaxOffsetNumber(page));
484 0 : break;
485 : case LH_BITMAP_PAGE:
486 0 : case LH_META_PAGE:
487 : default:
488 0 : break;
489 : }
490 0 : }
491 : else
492 : {
493 : /* maybe corrupted */
494 : }
495 :
496 0 : _hash_relbuf(rel, buf);
497 0 : }
498 :
499 : /*
500 : * pgstat_gist_page -- check tuples in a gist page
501 : */
502 : static void
503 0 : pgstat_gist_page(pgstattuple_type *stat, Relation rel, BlockNumber blkno,
504 : BufferAccessStrategy bstrategy)
505 : {
506 0 : Buffer buf;
507 0 : Page page;
508 :
509 0 : buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, bstrategy);
510 0 : LockBuffer(buf, GIST_SHARE);
511 0 : page = BufferGetPage(buf);
512 0 : if (PageIsNew(page))
513 : {
514 : /* fully empty page */
515 0 : stat->free_space += BLCKSZ;
516 0 : }
517 0 : else if (PageGetSpecialSize(page) == MAXALIGN(sizeof(GISTPageOpaqueData)))
518 : {
519 0 : if (GistPageIsLeaf(page))
520 : {
521 0 : pgstat_index_page(stat, page, FirstOffsetNumber,
522 0 : PageGetMaxOffsetNumber(page));
523 0 : }
524 : else
525 : {
526 : /* root or node */
527 : }
528 0 : }
529 :
530 0 : UnlockReleaseBuffer(buf);
531 0 : }
532 :
533 : /*
534 : * pgstat_index -- returns live/dead tuples info in a generic index
535 : */
536 : static Datum
537 0 : pgstat_index(Relation rel, BlockNumber start, pgstat_page pagefn,
538 : FunctionCallInfo fcinfo)
539 : {
540 0 : BlockNumber nblocks;
541 0 : BlockNumber blkno;
542 0 : BufferAccessStrategy bstrategy;
543 0 : pgstattuple_type stat = {0};
544 :
545 : /* prepare access strategy for this index */
546 0 : bstrategy = GetAccessStrategy(BAS_BULKREAD);
547 :
548 0 : blkno = start;
549 0 : for (;;)
550 : {
551 : /* Get the current relation length */
552 0 : LockRelationForExtension(rel, ExclusiveLock);
553 0 : nblocks = RelationGetNumberOfBlocks(rel);
554 0 : UnlockRelationForExtension(rel, ExclusiveLock);
555 :
556 : /* Quit if we've scanned the whole relation */
557 0 : if (blkno >= nblocks)
558 : {
559 0 : stat.table_len = (uint64) nblocks * BLCKSZ;
560 :
561 0 : break;
562 : }
563 :
564 0 : for (; blkno < nblocks; blkno++)
565 : {
566 0 : CHECK_FOR_INTERRUPTS();
567 :
568 0 : pagefn(&stat, rel, blkno, bstrategy);
569 0 : }
570 : }
571 :
572 0 : relation_close(rel, AccessShareLock);
573 :
574 0 : return build_pgstattuple_type(&stat, fcinfo);
575 0 : }
576 :
577 : /*
578 : * pgstat_index_page -- for generic index page
579 : */
580 : static void
581 0 : pgstat_index_page(pgstattuple_type *stat, Page page,
582 : OffsetNumber minoff, OffsetNumber maxoff)
583 : {
584 0 : OffsetNumber i;
585 :
586 0 : stat->free_space += PageGetExactFreeSpace(page);
587 :
588 0 : for (i = minoff; i <= maxoff; i = OffsetNumberNext(i))
589 : {
590 0 : ItemId itemid = PageGetItemId(page, i);
591 :
592 0 : if (ItemIdIsDead(itemid))
593 : {
594 0 : stat->dead_tuple_count++;
595 0 : stat->dead_tuple_len += ItemIdGetLength(itemid);
596 0 : }
597 : else
598 : {
599 0 : stat->tuple_count++;
600 0 : stat->tuple_len += ItemIdGetLength(itemid);
601 : }
602 0 : }
603 0 : }
|
