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