vacuum.c 59.5 KB
Newer Older
1 2 3 4 5 6 7 8 9
/*-------------------------------------------------------------------------
 *
 * vacuum.c--
 *    the postgres vacuum cleaner
 *
 * Copyright (c) 1994, Regents of the University of California
 *
 *
 * IDENTIFICATION
10
 *    $Header: /cvsroot/pgsql/src/backend/commands/vacuum.c,v 1.27 1997/04/02 03:48:01 vadim Exp $
11 12 13 14
 *
 *-------------------------------------------------------------------------
 */
#include <sys/file.h>
15
#include <string.h>
16 17 18 19
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
20

21 22
#include <postgres.h>

23
#include <fmgr.h>
24 25 26 27 28 29 30
#include <utils/portal.h>
#include <access/genam.h>
#include <access/heapam.h>
#include <access/xact.h>
#include <storage/bufmgr.h>
#include <access/transam.h>
#include <catalog/pg_index.h>
31
#include <catalog/index.h>
32
#include <catalog/catname.h>
33
#include <catalog/catalog.h>
34 35
#include <catalog/pg_class.h>
#include <catalog/pg_proc.h>
36 37 38
#include <catalog/pg_statistic.h>
#include <catalog/pg_type.h>
#include <catalog/pg_operator.h>
39 40
#include <storage/smgr.h>
#include <storage/lmgr.h>
41
#include <utils/inval.h>
42
#include <utils/mcxt.h>
43
#include <utils/inval.h>
44
#include <utils/syscache.h>
45
#include <commands/vacuum.h>
46
#include <parser/catalog_utils.h>
47
#include <storage/bufpage.h>
48
#include "storage/shmem.h"
49
#ifndef HAVE_GETRUSAGE
50
# include <rusagestub.h>
51
#else 
52 53
# include <sys/time.h>
# include <sys/resource.h>
54
#endif 
55

56
bool VacuumRunning =	false;
57
static int MESSAGE_LEVEL;	/* message level */
58

59 60 61 62 63 64 65 66
#define swapLong(a,b)	{long tmp; tmp=a; a=b; b=tmp;}
#define swapInt(a,b)	{int tmp; tmp=a; a=b; b=tmp;}
#define swapDatum(a,b)	{Datum tmp; tmp=a; a=b; b=tmp;}
#define VacAttrStatsEqValid(stats) ( RegProcedureIsValid(stats->cmpeq))
#define VacAttrStatsLtGtValid(stats) ( RegProcedureIsValid(stats->cmplt) && \
				   RegProcedureIsValid(stats->cmpgt) && \
				   RegProcedureIsValid(stats->outfunc) )
  
67

68
/* non-export function prototypes */
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98
static void vc_init(void);
static void vc_shutdown(void);
static void vc_vacuum(NameData *VacRelP);
static VRelList vc_getrels(Portal p, NameData *VacRelP);
static void vc_vacone (Oid relid);
static void vc_scanheap (VRelStats *vacrelstats, Relation onerel, VPageList Vvpl, VPageList Fvpl);
static void vc_rpfheap (VRelStats *vacrelstats, Relation onerel, VPageList Vvpl, VPageList Fvpl, int nindices, Relation *Irel);
static void vc_vacheap (VRelStats *vacrelstats, Relation onerel, VPageList vpl);
static void vc_vacpage (Page page, VPageDescr vpd, Relation archrel);
static void vc_vaconeind (VPageList vpl, Relation indrel, int nhtups);
static void vc_scanoneind (Relation indrel, int nhtups);
static void vc_attrstats(Relation onerel, VacAttrStats *vacattrstats, HeapTuple htup);
static void vc_bucketcpy(AttributeTupleForm attr, Datum value, Datum *bucket, int16 *bucket_len);
static void vc_updstats(Oid relid, int npages, int ntups, bool hasindex, VacAttrStats *vacattrstats);
static void vc_delhilowstats(Oid relid);
static void vc_setpagelock(Relation rel, BlockNumber blkno);
static VPageDescr vc_tidreapped (ItemPointer itemptr, VPageList vpl);
static void vc_reappage (VPageList vpl, VPageDescr vpc);
static void vc_vpinsert (VPageList vpl, VPageDescr vpnew);
static void vc_free(Portal p, VRelList vrl);
static void vc_getindices (Oid relid, int *nindices, Relation **Irel);
static void vc_clsindices (int nindices, Relation *Irel);
static Relation vc_getarchrel(Relation heaprel);
static void vc_archive(Relation archrel, HeapTuple htup);
static bool vc_isarchrel(char *rname);
static void vc_mkindesc (Relation onerel, int nindices, Relation *Irel, IndDesc **Idesc);
static char * vc_find_eq (char *bot, int nelem, int size, char *elm, int (*compar)(char *, char *));
static int vc_cmp_blk (char *left, char *right);
static int vc_cmp_offno (char *left, char *right);
static bool vc_enough_space (VPageDescr vpd, Size len);
99 100

void
101
vacuum(char *vacrel, bool verbose)
102
{
103 104
    NameData VacRel;

105
    if (verbose)
106
	MESSAGE_LEVEL = NOTICE;
107
    else
108
	MESSAGE_LEVEL = DEBUG;
109

110 111
    /* vacrel gets de-allocated on transaction commit */
    	
112
    /* initialize vacuum cleaner */
113
    vc_init();
114 115

    /* vacuum the database */
116 117 118
    if (vacrel)
    {
	strcpy(VacRel.data,vacrel);
119
	vc_vacuum(&VacRel);
120 121
    }
    else
122
	vc_vacuum(NULL);
123 124

    /* clean up */
125
    vc_shutdown();
126 127 128
}

/*
129
 *  vc_init(), vc_shutdown() -- start up and shut down the vacuum cleaner.
130 131 132 133 134 135 136 137 138 139 140
 *
 *	We run exactly one vacuum cleaner at a time.  We use the file system
 *	to guarantee an exclusive lock on vacuuming, since a single vacuum
 *	cleaner instantiation crosses transaction boundaries, and we'd lose
 *	postgres-style locks at the end of every transaction.
 *
 *	The strangeness with committing and starting transactions in the
 *	init and shutdown routines is due to the fact that the vacuum cleaner
 *	is invoked via a sql command, and so is already executing inside
 *	a transaction.  We need to leave ourselves in a predictable state
 *	on entry and exit to the vacuum cleaner.  We commit the transaction
141 142
 *	started in PostgresMain() inside vc_init(), and start one in
 *	vc_shutdown() to match the commit waiting for us back in
143 144 145
 *	PostgresMain().
 */
static void
146
vc_init()
147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168
{
    int fd;

    if ((fd = open("pg_vlock", O_CREAT|O_EXCL, 0600)) < 0)
	elog(WARN, "can't create lock file -- another vacuum cleaner running?");

    close(fd);

    /*
     *  By here, exclusive open on the lock file succeeded.  If we abort
     *  for any reason during vacuuming, we need to remove the lock file.
     *  This global variable is checked in the transaction manager on xact
     *  abort, and the routine vc_abort() is called if necessary.
     */

    VacuumRunning = true;

    /* matches the StartTransaction in PostgresMain() */
    CommitTransactionCommand();
}

static void
169
vc_shutdown()
170 171 172 173 174 175 176 177 178 179
{
    /* on entry, not in a transaction */
    if (unlink("pg_vlock") < 0)
	elog(WARN, "vacuum: can't destroy lock file!");

    /* okay, we're done */
    VacuumRunning = false;

    /* matches the CommitTransaction in PostgresMain() */
    StartTransactionCommand();
180

181 182 183 184 185 186 187 188 189 190 191 192
}

void
vc_abort()
{
    /* on abort, remove the vacuum cleaner lock file */
    (void) unlink("pg_vlock");

    VacuumRunning = false;
}

/*
193
 *  vc_vacuum() -- vacuum the database.
194 195 196 197 198 199 200
 *
 *	This routine builds a list of relations to vacuum, and then calls
 *	code that vacuums them one at a time.  We are careful to vacuum each
 *	relation in a separate transaction in order to avoid holding too many
 *	locks at one time.
 */
static void
201
vc_vacuum(NameData *VacRelP)
202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219
{
    VRelList vrl, cur;
    char *pname;
    Portal p;

    /*
     *  Create a portal for safe memory across transctions.  We need to
     *  palloc the name space for it because our hash function expects
     *	the name to be on a longword boundary.  CreatePortal copies the
     *  name to safe storage for us.
     */

    pname = (char *) palloc(strlen(VACPNAME) + 1);
    strcpy(pname, VACPNAME);
    p = CreatePortal(pname);
    pfree(pname);

    /* get list of relations */
220
    vrl = vc_getrels(p, VacRelP);
221

222 223 224 225 226 227 228
    if ( vrl != NULL )
    {
    	if (VacRelP != NULL)
    	    vc_delhilowstats(vrl->vrl_relid);
    	else
    	    vc_delhilowstats(InvalidOid);
    }
229
    	
230 231
    /* vacuum each heap relation */
    for (cur = vrl; cur != (VRelList) NULL; cur = cur->vrl_next)
232
	vc_vacone (cur->vrl_relid);
233

234
    vc_free(p, vrl);
235 236 237 238 239

    PortalDestroy(&p);
}

static VRelList
240
vc_getrels(Portal p, NameData *VacRelP)
241 242 243 244 245 246 247 248
{
    Relation pgclass;
    TupleDesc pgcdesc;
    HeapScanDesc pgcscan;
    HeapTuple pgctup;
    Buffer buf;
    PortalVariableMemory portalmem;
    MemoryContext old;
249
    VRelList vrl, cur;
250 251
    Datum d;
    char *rname;
252
    char rkind;
253 254 255
    int16 smgrno;
    bool n;
    ScanKeyData  pgckey;
256
    bool found = false;
257 258 259

    StartTransactionCommand();

260
    if (VacRelP->data) {
261 262
	ScanKeyEntryInitialize(&pgckey, 0x0, Anum_pg_class_relname,
			       NameEqualRegProcedure, 
263
			       PointerGetDatum(VacRelP->data));
264 265 266 267
    } else {
	ScanKeyEntryInitialize(&pgckey, 0x0, Anum_pg_class_relkind,
			       CharacterEqualRegProcedure, CharGetDatum('r'));
    }
268
 
269
    portalmem = PortalGetVariableMemory(p);
270
    vrl = cur = (VRelList) NULL;
271 272 273 274 275 276 277 278

    pgclass = heap_openr(RelationRelationName);
    pgcdesc = RelationGetTupleDescriptor(pgclass);

    pgcscan = heap_beginscan(pgclass, false, NowTimeQual, 1, &pgckey);

    while (HeapTupleIsValid(pgctup = heap_getnext(pgcscan, 0, &buf))) {

279 280
	found = true;
	
281 282 283 284 285 286 287 288 289 290 291 292
	/*
	 *  We have to be careful not to vacuum the archive (since it
	 *  already contains vacuumed tuples), and not to vacuum
	 *  relations on write-once storage managers like the Sony
	 *  jukebox at Berkeley.
	 */

	d = (Datum) heap_getattr(pgctup, buf, Anum_pg_class_relname,
				 pgcdesc, &n);
	rname = (char*)d;

	/* skip archive relations */
293
	if (vc_isarchrel(rname)) {
294 295 296 297
	    ReleaseBuffer(buf);
	    continue;
	}

298 299 300 301 302 303 304 305 306 307 308
	/* don't vacuum large objects for now - something breaks when we do */
	if ( (strlen(rname) > 4) && rname[0] == 'X' &&
		rname[1] == 'i' && rname[2] == 'n' &&
		(rname[3] == 'v' || rname[3] == 'x'))
	{
	    elog (NOTICE, "Rel %.*s: can't vacuum LargeObjects now", 
			NAMEDATALEN, rname);
	    ReleaseBuffer(buf);
	    continue;
	}

309 310 311 312 313 314 315 316 317 318
	d = (Datum) heap_getattr(pgctup, buf, Anum_pg_class_relsmgr,
				 pgcdesc, &n);
	smgrno = DatumGetInt16(d);

	/* skip write-once storage managers */
	if (smgriswo(smgrno)) {
	    ReleaseBuffer(buf);
	    continue;
	}

319 320 321 322 323 324 325 326 327 328 329 330
	d = (Datum) heap_getattr(pgctup, buf, Anum_pg_class_relkind,
				 pgcdesc, &n);

	rkind = DatumGetChar(d);

	/* skip system relations */
	if (rkind != 'r') {
	    ReleaseBuffer(buf);
	    elog(NOTICE, "Vacuum: can not process index and certain system tables" );
	    continue;
	}
				 
331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346
	/* get a relation list entry for this guy */
	old = MemoryContextSwitchTo((MemoryContext)portalmem);
	if (vrl == (VRelList) NULL) {
	    vrl = cur = (VRelList) palloc(sizeof(VRelListData));
	} else {
	    cur->vrl_next = (VRelList) palloc(sizeof(VRelListData));
	    cur = cur->vrl_next;
	}
	(void) MemoryContextSwitchTo(old);

	cur->vrl_relid = pgctup->t_oid;
	cur->vrl_next = (VRelList) NULL;

	/* wei hates it if you forget to do this */
	ReleaseBuffer(buf);
    }
347 348
    if (found == false)
    	elog(NOTICE, "Vacuum: table not found" );
349

350
    
351
    heap_endscan(pgcscan);
352
    heap_close(pgclass);
353 354 355 356 357 358 359

    CommitTransactionCommand();

    return (vrl);
}

/*
360
 *  vc_vacone() -- vacuum one heap relation
361 362
 *
 *	This routine vacuums a single heap, cleans out its indices, and
363
 *	updates its statistics npages and ntups statistics.
364 365 366 367 368 369 370 371
 *
 *	Doing one heap at a time incurs extra overhead, since we need to
 *	check that the heap exists again just before we vacuum it.  The
 *	reason that we do this is so that vacuuming can be spread across
 *	many small transactions.  Otherwise, two-phase locking would require
 *	us to lock the entire database during one pass of the vacuum cleaner.
 */
static void
372
vc_vacone (Oid relid)
373 374 375
{
    Relation pgclass;
    TupleDesc pgcdesc;
376
    HeapTuple pgctup, pgttup;
377 378 379 380
    Buffer pgcbuf;
    HeapScanDesc pgcscan;
    Relation onerel;
    ScanKeyData pgckey;
381 382 383 384
    VPageListData Vvpl;	/* List of pages to vacuum and/or clean indices */
    VPageListData Fvpl;	/* List of pages with space enough for re-using */
    VPageDescr *vpp;
    Relation *Irel;
385 386 387 388
    int32 nindices, i, attr_cnt;
    AttributeTupleForm *attr;
    VRelStats *vacrelstats;
    
389 390 391 392
    StartTransactionCommand();

    ScanKeyEntryInitialize(&pgckey, 0x0, ObjectIdAttributeNumber,
			   ObjectIdEqualRegProcedure,
393
			   ObjectIdGetDatum(relid));
394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411

    pgclass = heap_openr(RelationRelationName);
    pgcdesc = RelationGetTupleDescriptor(pgclass);
    pgcscan = heap_beginscan(pgclass, false, NowTimeQual, 1, &pgckey);

    /*
     *  Race condition -- if the pg_class tuple has gone away since the
     *  last time we saw it, we don't need to vacuum it.
     */

    if (!HeapTupleIsValid(pgctup = heap_getnext(pgcscan, 0, &pgcbuf))) {
	heap_endscan(pgcscan);
	heap_close(pgclass);
	CommitTransactionCommand();
	return;
    }

    /* now open the class and vacuum it */
412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432
    onerel = heap_open(relid);

    attr_cnt = onerel->rd_att->natts;
    attr = onerel->rd_att->attrs;

    vacrelstats = (VRelStats *) palloc(sizeof(VRelStats));
    vacrelstats->relid = relid;
    vacrelstats->npages = vacrelstats->ntups = 0;
    vacrelstats->hasindex = false;
    vacrelstats->vacattrstats =
		(VacAttrStats *) palloc(attr_cnt * sizeof(VacAttrStats));
    
    for (i = 0; i < attr_cnt; i++) {
	Operator func_operator;
	OperatorTupleForm pgopform;
	VacAttrStats *stats = &vacrelstats->vacattrstats[i];

	stats->attr = palloc(ATTRIBUTE_TUPLE_SIZE);
	memmove(stats->attr,attr[i],ATTRIBUTE_TUPLE_SIZE);
	stats->best = stats->guess1 = stats->guess2 = 0;
	stats->max = stats->min = 0;
433 434
	stats->best_len = stats->guess1_len = stats->guess2_len = 0;
	stats->max_len = stats->min_len = 0;
435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
	stats->initialized = false;
	stats->best_cnt = stats->guess1_cnt = stats->guess1_hits = stats->guess2_hits = 0;
	stats->max_cnt = stats->min_cnt = stats->null_cnt = stats->nonnull_cnt = 0;
	
        func_operator = oper("=",stats->attr->atttypid,stats->attr->atttypid,true);
	if (func_operator != NULL) {
	    pgopform = (OperatorTupleForm) GETSTRUCT(func_operator);
	    stats->cmpeq = pgopform->oprcode;
	}
	else	stats->cmpeq = InvalidOid;
        func_operator = oper("<",stats->attr->atttypid,stats->attr->atttypid,true);
	if (func_operator != NULL) {
	    pgopform = (OperatorTupleForm) GETSTRUCT(func_operator);
	    stats->cmplt = pgopform->oprcode;
	}
	else	stats->cmplt = InvalidOid;
        func_operator = oper(">",stats->attr->atttypid,stats->attr->atttypid,true);
	if (func_operator != NULL) {
	    pgopform = (OperatorTupleForm) GETSTRUCT(func_operator);
	    stats->cmpgt = pgopform->oprcode;
	}
	else	stats->cmpgt = InvalidOid;
	pgttup = SearchSysCacheTuple(TYPOID,
                                    ObjectIdGetDatum(stats->attr->atttypid),
                                    0,0,0);
        if (HeapTupleIsValid(pgttup))
            stats->outfunc = ((TypeTupleForm) GETSTRUCT(pgttup))->typoutput;
	else	stats->outfunc = InvalidOid;
    }
464 465 466 467

    /* we require the relation to be locked until the indices are cleaned */
    RelationSetLockForWrite(onerel);

468 469
    /* scan it */
    Vvpl.vpl_npages = Fvpl.vpl_npages = 0;
470
    vc_scanheap(vacrelstats, onerel, &Vvpl, &Fvpl);
471

472
    /* Now open indices */
473
    Irel = (Relation *) NULL;
474
    vc_getindices(vacrelstats->relid, &nindices, &Irel);
475 476
    
    if ( nindices > 0 )
477
	vacrelstats->hasindex = true;
478
    else
479
	vacrelstats->hasindex = false;
480

481
    /* Clean/scan index relation(s) */
482 483
    if ( Irel != (Relation*) NULL )
    {
484 485 486
    	if ( Vvpl.vpl_npages > 0 )
    	{
	    for (i = 0; i < nindices; i++)
487
	    	vc_vaconeind (&Vvpl, Irel[i], vacrelstats->ntups);
488 489 490 491
    	}
    	else	/* just scan indices to update statistic */
    	{
	    for (i = 0; i < nindices; i++)
492
	    	vc_scanoneind (Irel[i], vacrelstats->ntups);
493
    	}
494 495 496
    }

    if ( Fvpl.vpl_npages > 0 )		/* Try to shrink heap */
497
    	vc_rpfheap (vacrelstats, onerel, &Vvpl, &Fvpl, nindices, Irel);
498
    else 
Vadim B. Mikheev's avatar
Vadim B. Mikheev committed
499 500
    {
    	if ( Irel != (Relation*) NULL )
501
	    vc_clsindices (nindices, Irel);
502
    	if ( Vvpl.vpl_npages > 0 )	/* Clean pages from Vvpl list */
503
	    vc_vacheap (vacrelstats, onerel, &Vvpl);
Vadim B. Mikheev's avatar
Vadim B. Mikheev committed
504
    }
505 506 507 508 509 510 511 512 513 514 515

    /* ok - free Vvpl list of reapped pages */
    if ( Vvpl.vpl_npages > 0 )
    {
    	vpp = Vvpl.vpl_pgdesc;
    	for (i = 0; i < Vvpl.vpl_npages; i++, vpp++)
	    pfree(*vpp);
    	pfree (Vvpl.vpl_pgdesc);
    	if ( Fvpl.vpl_npages > 0 )
	    pfree (Fvpl.vpl_pgdesc);
    }
516 517 518 519 520 521 522

    /* all done with this class */
    heap_close(onerel);
    heap_endscan(pgcscan);
    heap_close(pgclass);

    /* update statistics in pg_class */
523 524 525
    vc_updstats(vacrelstats->relid, vacrelstats->npages, vacrelstats->ntups,
			 vacrelstats->hasindex, vacrelstats->vacattrstats);

526
    /* next command frees attribute stats */
527 528 529 530 531

    CommitTransactionCommand();
}

/*
532
 *  vc_scanheap() -- scan an open heap relation
533
 *
534 535 536 537 538
 *	This routine sets commit times, constructs Vvpl list of 
 *	empty/uninitialized pages and pages with dead tuples and
 *	~LP_USED line pointers, constructs Fvpl list of pages
 *	appropriate for purposes of shrinking and maintains statistics 
 *	on the number of live tuples in a heap.
539 540
 */
static void
541
vc_scanheap (VRelStats *vacrelstats, Relation onerel, 
542
			VPageList Vvpl, VPageList Fvpl)
543 544 545
{
    int nblocks, blkno;
    ItemId itemid;
546
    ItemPointer itemptr;
547 548
    HeapTuple htup;
    Buffer buf;
549
    Page page, tempPage = NULL;
550
    OffsetNumber offnum, maxoff;
551
    bool pgchanged, tupgone, dobufrel, notup;
552 553
    AbsoluteTime purgetime, expiretime;
    RelativeTime preservetime;
554
    char *relname;
555 556 557 558 559
    VPageDescr vpc, vp;
    uint32 nvac, ntups, nunused, ncrash, nempg, nnepg, nchpg, nemend;
    Size frsize, frsusf;
    Size min_tlen = MAXTUPLEN;
    Size max_tlen = 0;
560
    int32 i/*, attr_cnt*/;
561 562 563 564
    struct rusage ru0, ru1;

    getrusage(RUSAGE_SELF, &ru0);

565 566
    nvac = ntups = nunused = ncrash = nempg = nnepg = nchpg = nemend = 0;
    frsize = frsusf = 0;
567

568
    relname = (RelationGetRelationName(onerel))->data;
569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586

    nblocks = RelationGetNumberOfBlocks(onerel);

    /* calculate the purge time: tuples that expired before this time
       will be archived or deleted */
    purgetime = GetCurrentTransactionStartTime();
    expiretime = (AbsoluteTime)onerel->rd_rel->relexpires;
    preservetime = (RelativeTime)onerel->rd_rel->relpreserved;

    if (RelativeTimeIsValid(preservetime) && (preservetime)) {
	purgetime -= preservetime;
	if (AbsoluteTimeIsBackwardCompatiblyValid(expiretime) &&
	    expiretime > purgetime)
	    purgetime = expiretime;
    }

    else if (AbsoluteTimeIsBackwardCompatiblyValid(expiretime))
	purgetime = expiretime;
587 588

    vpc = (VPageDescr) palloc (sizeof(VPageDescrData) + MaxOffsetNumber*sizeof(OffsetNumber));
589
    vpc->vpd_nusd = 0;
590 591 592 593
	    
    for (blkno = 0; blkno < nblocks; blkno++) {
	buf = ReadBuffer(onerel, blkno);
	page = BufferGetPage(buf);
594 595 596 597 598 599 600
	vpc->vpd_blkno = blkno;
	vpc->vpd_noff = 0;

	if (PageIsNew(page)) {
	    elog (NOTICE, "Rel %.*s: Uninitialized page %u - fixing",
		NAMEDATALEN, relname, blkno);
	    PageInit (page, BufferGetPageSize (buf), 0);
601 602
	    vpc->vpd_free = ((PageHeader)page)->pd_upper - ((PageHeader)page)->pd_lower;
	    frsize += (vpc->vpd_free - sizeof (ItemIdData));
603
	    nnepg++;
604
	    nemend++;
605
	    vc_reappage (Vvpl, vpc);
606 607 608
	    WriteBuffer(buf);
	    continue;
	}
609 610

	if (PageIsEmpty(page)) {
611 612
	    vpc->vpd_free = ((PageHeader)page)->pd_upper - ((PageHeader)page)->pd_lower;
	    frsize += (vpc->vpd_free - sizeof (ItemIdData));
613
	    nempg++;
614
	    nemend++;
615
	    vc_reappage (Vvpl, vpc);
616 617 618 619 620
	    ReleaseBuffer(buf);
	    continue;
	}

	pgchanged = false;
621
	notup = true;
622 623 624 625 626 627
	maxoff = PageGetMaxOffsetNumber(page);
	for (offnum = FirstOffsetNumber;
	     offnum <= maxoff;
	     offnum = OffsetNumberNext(offnum)) {
	    itemid = PageGetItemId(page, offnum);

628 629 630 631
	    /*
	     * Collect un-used items too - it's possible to have
	     * indices pointing here after crash.
	     */
632 633
	    if (!ItemIdIsUsed(itemid)) {
	    	vpc->vpd_voff[vpc->vpd_noff++] = offnum;
634
	    	nunused++;
635
		continue;
636
	    }
637 638 639 640 641 642 643 644 645 646 647 648

	    htup = (HeapTuple) PageGetItem(page, itemid);
	    tupgone = false;

	    if (!AbsoluteTimeIsBackwardCompatiblyValid(htup->t_tmin) && 
		TransactionIdIsValid((TransactionId)htup->t_xmin)) {

		if (TransactionIdDidAbort(htup->t_xmin)) {
		    tupgone = true;
		} else if (TransactionIdDidCommit(htup->t_xmin)) {
		    htup->t_tmin = TransactionIdGetCommitTime(htup->t_xmin);
		    pgchanged = true;
649 650 651 652 653 654 655 656 657
		} else if ( !TransactionIdIsInProgress (htup->t_xmin) ) {
		    /* 
		     * Not Aborted, Not Committed, Not in Progress -
		     * so it from crashed process. - vadim 11/26/96
		     */
		    ncrash++;
		    tupgone = true;
		}
		else {
658
		    elog (MESSAGE_LEVEL, "Rel %.*s: InsertTransactionInProgress %u for TID %u/%u",
659
			NAMEDATALEN, relname, htup->t_xmin, blkno, offnum);
660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678
		}
	    }

	    if (TransactionIdIsValid((TransactionId)htup->t_xmax)) {
		if (TransactionIdDidAbort(htup->t_xmax)) {
		    StoreInvalidTransactionId(&(htup->t_xmax));
		    pgchanged = true;
		} else if (TransactionIdDidCommit(htup->t_xmax)) {
		    if (!AbsoluteTimeIsBackwardCompatiblyReal(htup->t_tmax)) {

			htup->t_tmax = TransactionIdGetCommitTime(htup->t_xmax);  
			pgchanged = true;
		    }

		    /*
		     *  Reap the dead tuple if its expiration time is
		     *  before purgetime.
		     */

679
		    if (htup->t_tmax < purgetime) {
680 681 682 683 684
			tupgone = true;
		    }
		}
	    }

685 686 687 688 689 690 691 692 693 694 695
	    /*
	     * Is it possible at all ? - vadim 11/26/96
	     */
	    if ( !TransactionIdIsValid((TransactionId)htup->t_xmin) )
	    {
		elog (NOTICE, "TID %u/%u: INSERT_TRANSACTION_ID IS INVALID. \
DELETE_TRANSACTION_ID_VALID %d, TUPGONE %d.", 
			TransactionIdIsValid((TransactionId)htup->t_xmax),
			tupgone);
	    }
	    
696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722
	    /*
	     * It's possibly! But from where it comes ?
	     * And should we fix it ?  - vadim 11/28/96
	     */
	    itemptr = &(htup->t_ctid);
	    if ( !ItemPointerIsValid (itemptr) || 
	    		BlockIdGetBlockNumber(&(itemptr->ip_blkid)) != blkno )
	    {
	    	elog (NOTICE, "ITEM POINTER IS INVALID: %u/%u FOR %u/%u. TUPGONE %d.", 
	    		BlockIdGetBlockNumber(&(itemptr->ip_blkid)), 
	    		itemptr->ip_posid, blkno, offnum, tupgone);
	    }

	    /*
	     * Other checks...
	     */
	    if ( htup->t_len != itemid->lp_len )
	    {
	    	elog (NOTICE, "PAGEHEADER' LEN %u IS NOT THE SAME AS HTUP' %u FOR %u/%u.TUPGONE %d.", 
	    		itemid->lp_len, htup->t_len, blkno, offnum, tupgone);
	    }
	    if ( !OidIsValid(htup->t_oid) )
	    {
	    	elog (NOTICE, "OID IS INVALID FOR %u/%u.TUPGONE %d.", 
	    		blkno, offnum, tupgone);
	    }
	    
723
	    if (tupgone) {
724 725 726 727 728 729 730 731 732 733 734 735
		ItemId lpp;
                                                    
		if ( tempPage == (Page) NULL )
		{
		    Size pageSize;
		    
		    pageSize = PageGetPageSize(page);
		    tempPage = (Page) palloc(pageSize);
		    memmove (tempPage, page, pageSize);
		}
		
		lpp = &(((PageHeader) tempPage)->pd_linp[offnum - 1]);
736 737 738 739

		/* mark it unused */
		lpp->lp_flags &= ~LP_USED;

740
	    	vpc->vpd_voff[vpc->vpd_noff++] = offnum;
741
	    	nvac++;
742

743 744
	    } else {
		ntups++;
745 746 747 748 749
		notup = false;
		if ( htup->t_len < min_tlen )
		    min_tlen = htup->t_len;
		if ( htup->t_len > max_tlen )
		    max_tlen = htup->t_len;
750
		vc_attrstats(onerel, vacrelstats->vacattrstats, htup);
751 752 753 754 755
	    }
	}

	if (pgchanged) {
	    WriteBuffer(buf);
756
	    dobufrel = false;
757
	    nchpg++;
758
	}
759 760 761 762 763 764 765
	else
	    dobufrel = true;
	if ( tempPage != (Page) NULL )
	{ /* Some tuples are gone */
	    PageRepairFragmentation(tempPage);
	    vpc->vpd_free = ((PageHeader)tempPage)->pd_upper - ((PageHeader)tempPage)->pd_lower;
	    frsize += vpc->vpd_free;
766
	    vc_reappage (Vvpl, vpc);
767 768 769 770 771 772 773
	    pfree (tempPage);
	    tempPage = (Page) NULL;
	}
	else if ( vpc->vpd_noff > 0 )
	{ /* there are only ~LP_USED line pointers */
	    vpc->vpd_free = ((PageHeader)page)->pd_upper - ((PageHeader)page)->pd_lower;
	    frsize += vpc->vpd_free;
774
	    vc_reappage (Vvpl, vpc);
775 776 777 778 779 780 781
	}
	if ( dobufrel )
	    ReleaseBuffer(buf);
	if ( notup )
	    nemend++;
	else
	    nemend = 0;
782 783
    }

784
    pfree (vpc);
785 786

    /* save stats in the rel list for use later */
787 788 789
    vacrelstats->ntups = ntups;
    vacrelstats->npages = nblocks;
/*    vacrelstats->natts = attr_cnt;*/
790 791
    if ( ntups == 0 )
    	min_tlen = max_tlen = 0;
792 793
    vacrelstats->min_tlen = min_tlen;
    vacrelstats->max_tlen = max_tlen;
794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812
    
    Vvpl->vpl_nemend = nemend;
    Fvpl->vpl_nemend = nemend;

    /* 
     * Try to make Fvpl keeping in mind that we can't use free space 
     * of "empty" end-pages and last page if it reapped.
     */
    if ( Vvpl->vpl_npages - nemend > 0 )
    {
	int nusf;		/* blocks usefull for re-using */
	
	nusf = Vvpl->vpl_npages - nemend;
	if ( (Vvpl->vpl_pgdesc[nusf-1])->vpd_blkno == nblocks - nemend - 1 )
	    nusf--;
    
	for (i = 0; i < nusf; i++)
    	{
	    vp = Vvpl->vpl_pgdesc[i];
813
	    if ( vc_enough_space (vp, min_tlen) )
814
	    {
815
		vc_vpinsert (Fvpl, vp);
816 817 818 819
		frsusf += vp->vpd_free;
	    }
	}
    }
820 821 822

    getrusage(RUSAGE_SELF, &ru1);
    
823
    elog (MESSAGE_LEVEL, "Rel %.*s: Pages %u: Changed %u, Reapped %u, Empty %u, New %u; \
824
Tup %u: Vac %u, Crash %u, UnUsed %u, MinLen %u, MaxLen %u; Re-using: Free/Avail. Space %u/%u; EndEmpty/Avail. Pages %u/%u. Elapsed %u/%u sec.",
825
	NAMEDATALEN, relname, 
826 827 828
	nblocks, nchpg, Vvpl->vpl_npages, nempg, nnepg,
	ntups, nvac, ncrash, nunused, min_tlen, max_tlen, 
	frsize, frsusf, nemend, Fvpl->vpl_npages,
829 830 831
	ru1.ru_stime.tv_sec - ru0.ru_stime.tv_sec, 
	ru1.ru_utime.tv_sec - ru0.ru_utime.tv_sec);

832
} /* vc_scanheap */
833

834 835

/*
836
 *  vc_rpfheap() -- try to repaire relation' fragmentation
837
 *
838 839 840 841 842 843
 *	This routine marks dead tuples as unused and tries re-use dead space
 *	by moving tuples (and inserting indices if needed). It constructs 
 *	Nvpl list of free-ed pages (moved tuples) and clean indices
 *	for them after committing (in hack-manner - without losing locks
 *	and freeing memory!) current transaction. It truncates relation
 *	if some end-blocks are gone away.
844 845
 */
static void
846
vc_rpfheap (VRelStats *vacrelstats, Relation onerel, 
847
		VPageList Vvpl, VPageList Fvpl, int nindices, Relation *Irel)
848
{
849 850
    TransactionId myXID;
    CommandId myCID;
851
    AbsoluteTime myCTM = 0;
852 853
    Buffer buf, ToBuf;
    int nblocks, blkno;
854 855
    Page page, ToPage = NULL;
    OffsetNumber offnum = 0, maxoff = 0, newoff, moff;
856 857
    ItemId itemid, newitemid;
    HeapTuple htup, newtup;
858 859 860
    TupleDesc tupdesc = NULL;
    Datum *idatum = NULL;
    char *inulls = NULL;
861 862
    InsertIndexResult iresult;
    VPageListData Nvpl;
863
    VPageDescr ToVpd = NULL, Fvplast, Vvplast, vpc, *vpp;
864
    int ToVpI = 0;
865 866 867 868 869 870 871 872
    IndDesc *Idesc, *idcur;
    int Fblklast, Vblklast, i;
    Size tlen;
    int nmoved, Fnpages, Vnpages;
    int nchkmvd, ntups;
    bool isempty, dowrite;
    Relation archrel;
    struct rusage ru0, ru1;
873

874
    getrusage(RUSAGE_SELF, &ru0);
875

876 877 878 879 880
    myXID = GetCurrentTransactionId();
    myCID = GetCurrentCommandId();
	
    if ( Irel != (Relation*) NULL )	/* preparation for index' inserts */
    {
881
	vc_mkindesc (onerel, nindices, Irel, &Idesc);
882 883 884 885
	tupdesc = RelationGetTupleDescriptor(onerel);
	idatum = (Datum *) palloc(INDEX_MAX_KEYS * sizeof (*idatum));
	inulls = (char *) palloc(INDEX_MAX_KEYS * sizeof (*inulls));
    }
886

887 888 889
    /* if the relation has an archive, open it */
    if (onerel->rd_rel->relarch != 'n')
    {
890
	archrel = vc_getarchrel(onerel);
891 892 893 894 895 896 897 898 899
	/* Archive tuples from "empty" end-pages */
	for ( vpp = Vvpl->vpl_pgdesc + Vvpl->vpl_npages - 1, 
				i = Vvpl->vpl_nemend; i > 0; i--, vpp-- )
	{
	    if ( (*vpp)->vpd_noff > 0 )
	    {
	    	buf = ReadBuffer(onerel, (*vpp)->vpd_blkno);
	    	page = BufferGetPage(buf);
	    	Assert ( !PageIsEmpty(page) );
900
		vc_vacpage (page, *vpp, archrel);
901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918
		WriteBuffer (buf);
	    }
	}
    }
    else
	archrel = (Relation) NULL;

    Nvpl.vpl_npages = 0;
    Fnpages = Fvpl->vpl_npages;
    Fvplast = Fvpl->vpl_pgdesc[Fnpages - 1];
    Fblklast = Fvplast->vpd_blkno;
    Assert ( Vvpl->vpl_npages > Vvpl->vpl_nemend );
    Vnpages = Vvpl->vpl_npages - Vvpl->vpl_nemend;
    Vvplast = Vvpl->vpl_pgdesc[Vnpages - 1];
    Vblklast = Vvplast->vpd_blkno;
    Assert ( Vblklast >= Fblklast );
    ToBuf = InvalidBuffer;
    nmoved = 0;
919

920
    vpc = (VPageDescr) palloc (sizeof(VPageDescrData) + MaxOffsetNumber*sizeof(OffsetNumber));
921
    vpc->vpd_nusd = vpc->vpd_noff = 0;
922
	
923
    nblocks = vacrelstats->npages;
924 925 926 927 928
    for (blkno = nblocks - Vvpl->vpl_nemend - 1; ; blkno--)
    {
	/* if it's reapped page and it was used by me - quit */
	if ( blkno == Fblklast && Fvplast->vpd_nusd > 0 )
	    break;
929

930 931
	buf = ReadBuffer(onerel, blkno);
	page = BufferGetPage(buf);
932

933 934 935 936
	vpc->vpd_noff = 0;

	isempty = PageIsEmpty(page);

937
	dowrite = false;
938 939 940 941 942
	if ( blkno == Vblklast )		/* it's reapped page */
	{
	    if ( Vvplast->vpd_noff > 0 )	/* there are dead tuples */
	    {					/* on this page - clean */
		Assert ( ! isempty );
943
		vc_vacpage (page, Vvplast, archrel);
944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987
		dowrite = true;
	    }
	    else
		Assert ( isempty );
	    Assert ( --Vnpages > 0 );
	    /* get prev reapped page from Vvpl */
	    Vvplast = Vvpl->vpl_pgdesc[Vnpages - 1];
	    Vblklast = Vvplast->vpd_blkno;
	    if ( blkno == Fblklast )	/* this page in Fvpl too */
	    {
		Assert ( --Fnpages > 0 );
		Assert ( Fvplast->vpd_nusd == 0 );
		/* get prev reapped page from Fvpl */
		Fvplast = Fvpl->vpl_pgdesc[Fnpages - 1];
		Fblklast = Fvplast->vpd_blkno;
	    }
	    Assert ( Fblklast <= Vblklast );
	    if ( isempty )
	    {
		ReleaseBuffer(buf);
		continue;
	    }
	}
	else
	{
	    Assert ( ! isempty );
	}

	vpc->vpd_blkno = blkno;
	maxoff = PageGetMaxOffsetNumber(page);
	for (offnum = FirstOffsetNumber;
		offnum <= maxoff;
		offnum = OffsetNumberNext(offnum))
	{
	    itemid = PageGetItemId(page, offnum);

	    if (!ItemIdIsUsed(itemid))
		continue;

	    htup = (HeapTuple) PageGetItem(page, itemid);
	    tlen = htup->t_len;
		
	    /* try to find new page for this tuple */
	    if ( ToBuf == InvalidBuffer ||
988
		! vc_enough_space (ToVpd, tlen) )
989 990
	    {
		if ( ToBuf != InvalidBuffer )
991
		{
992
		    WriteBuffer(ToBuf);
993 994 995 996 997
		    ToBuf = InvalidBuffer;
 		    /*
 		     * If no one tuple can't be added to this page -
 		     * remove page from Fvpl. - vadim 11/27/96
 		     */ 
998
 		    if ( !vc_enough_space (ToVpd, vacrelstats->min_tlen) )
999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015
 		    {
 		    	if ( ToVpd != Fvplast )
 		    	{
 		    	    Assert ( Fnpages > ToVpI + 1 );
 		    	    memmove (Fvpl->vpl_pgdesc + ToVpI, 
 		    	    	Fvpl->vpl_pgdesc + ToVpI + 1, 
 		    	    	sizeof (VPageDescr*) * (Fnpages - ToVpI - 1));
 		    	}
 		    	Assert ( Fnpages >= 1 );
 		    	Fnpages--;
 		    	if ( Fnpages == 0 )
 		    	    break;
			/* get prev reapped page from Fvpl */
			Fvplast = Fvpl->vpl_pgdesc[Fnpages - 1];
			Fblklast = Fvplast->vpd_blkno;
 		    }
 		}
1016 1017
		for (i=0; i < Fnpages; i++)
		{
1018
		    if ( vc_enough_space (Fvpl->vpl_pgdesc[i], tlen) )
1019 1020 1021 1022
			break;
		}
		if ( i == Fnpages )
		    break;			/* can't move item anywhere */
1023 1024
 		ToVpI = i;
 		ToVpd = Fvpl->vpl_pgdesc[ToVpI];
1025 1026 1027 1028
		ToBuf = ReadBuffer(onerel, ToVpd->vpd_blkno);
		ToPage = BufferGetPage(ToBuf);
		/* if this page was not used before - clean it */
		if ( ! PageIsEmpty(ToPage) && ToVpd->vpd_nusd == 0 )
1029
		    vc_vacpage (ToPage, ToVpd, archrel);
1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087
	    }
		
	    /* copy tuple */
	    newtup = (HeapTuple) palloc (tlen);
	    memmove((char *) newtup, (char *) htup, tlen);

	    /* store transaction information */
	    TransactionIdStore(myXID, &(newtup->t_xmin));
	    newtup->t_cmin = myCID;
	    StoreInvalidTransactionId(&(newtup->t_xmax));
	    newtup->t_tmin = INVALID_ABSTIME;
	    newtup->t_tmax = CURRENT_ABSTIME;
	    ItemPointerSetInvalid(&newtup->t_chain);

	    /* add tuple to the page */
	    newoff = PageAddItem (ToPage, (Item)newtup, tlen, 
				InvalidOffsetNumber, LP_USED);
	    if ( newoff == InvalidOffsetNumber )
	    {
		elog (WARN, "\
failed to add item with len = %u to page %u (free space %u, nusd %u, noff %u)",
		tlen, ToVpd->vpd_blkno, ToVpd->vpd_free, 
		ToVpd->vpd_nusd, ToVpd->vpd_noff);
	    }
	    newitemid = PageGetItemId(ToPage, newoff);
	    pfree (newtup);
	    newtup = (HeapTuple) PageGetItem(ToPage, newitemid);
	    ItemPointerSet(&(newtup->t_ctid), ToVpd->vpd_blkno, newoff);

	    /* now logically delete end-tuple */
	    TransactionIdStore(myXID, &(htup->t_xmax));
	    htup->t_cmax = myCID;
	    memmove ((char*)&(htup->t_chain), (char*)&(newtup->t_ctid), sizeof (newtup->t_ctid));

	    ToVpd->vpd_nusd++;
	    nmoved++;
	    ToVpd->vpd_free = ((PageHeader)ToPage)->pd_upper - ((PageHeader)ToPage)->pd_lower;
	    vpc->vpd_voff[vpc->vpd_noff++] = offnum;
		
	    /* insert index' tuples if needed */
	    if ( Irel != (Relation*) NULL )
	    {
		for (i = 0, idcur = Idesc; i < nindices; i++, idcur++)
		{
		    FormIndexDatum (
		    		idcur->natts,
		    		(AttrNumber *)&(idcur->tform->indkey[0]),
				newtup, 
				tupdesc,
				InvalidBuffer,
				idatum,
				inulls,
				idcur->finfoP);
		    iresult = index_insert (
				Irel[i],
				idatum,
				inulls,
				&(newtup->t_ctid),
1088
				onerel);
1089 1090 1091 1092 1093 1094 1095 1096
		    if (iresult) pfree(iresult);
		}
	    }
		
	} /* walk along page */

	if ( vpc->vpd_noff > 0 )		/* some tuples were moved */
	{
1097
	    vc_reappage (&Nvpl, vpc);
1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115
	    WriteBuffer(buf);
	}
	else if ( dowrite )
	    WriteBuffer(buf);
	else
	    ReleaseBuffer(buf);
	    
	if ( offnum <= maxoff )
	    break;				/* some item(s) left */
	    
    } /* walk along relation */
	
    blkno++;				/* new number of blocks */

    if ( ToBuf != InvalidBuffer )
    {
	Assert (nmoved > 0);
	WriteBuffer(ToBuf);
1116 1117
    }

1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145
    if ( nmoved > 0 )
    {
	/* 
	 * We have to commit our tuple' movings before we'll truncate 
	 * relation, but we shouldn't lose our locks. And so - quick hack: 
	 * flush buffers and record status of current transaction
	 * as committed, and continue. - vadim 11/13/96
	 */
	FlushBufferPool(!TransactionFlushEnabled());
	TransactionIdCommit(myXID);
	FlushBufferPool(!TransactionFlushEnabled());
	myCTM = TransactionIdGetCommitTime(myXID);
    }
	
    /* 
     * Clean uncleaned reapped pages from Vvpl list 
     * and set commit' times  for inserted tuples
     */
    nchkmvd = 0;
    for (i = 0, vpp = Vvpl->vpl_pgdesc; i < Vnpages; i++, vpp++)
    {
	Assert ( (*vpp)->vpd_blkno < blkno );
	buf = ReadBuffer(onerel, (*vpp)->vpd_blkno);
	page = BufferGetPage(buf);
	if ( (*vpp)->vpd_nusd == 0 )	/* this page was not used */
	{
	    /* noff == 0 in empty pages only - such pages should be re-used */
	    Assert ( (*vpp)->vpd_noff > 0 );
1146
	    vc_vacpage (page, *vpp, archrel);
1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171
	}
	else				/* this page was used */
	{
	    ntups = 0;
	    moff = PageGetMaxOffsetNumber(page);
	    for (newoff = FirstOffsetNumber;
			newoff <= moff;
			newoff = OffsetNumberNext(newoff))
	    {
	    	itemid = PageGetItemId(page, newoff);
	    	if (!ItemIdIsUsed(itemid))
		    continue;
	    	htup = (HeapTuple) PageGetItem(page, itemid);
	    	if ( TransactionIdEquals((TransactionId)htup->t_xmin, myXID) )
	    	{
	    	    htup->t_tmin = myCTM;
	    	    ntups++;
	    	}
	    }
	    Assert ( (*vpp)->vpd_nusd == ntups );
	    nchkmvd += ntups;
	}
    	WriteBuffer (buf);
    }
    Assert ( nmoved == nchkmvd );
1172

1173 1174
    getrusage(RUSAGE_SELF, &ru1);
    
1175
    elog (MESSAGE_LEVEL, "Rel %.*s: Pages: %u --> %u; Tuple(s) moved: %u. \
1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196
Elapsed %u/%u sec.",
		NAMEDATALEN, (RelationGetRelationName(onerel))->data, 
		nblocks, blkno, nmoved,
		ru1.ru_stime.tv_sec - ru0.ru_stime.tv_sec, 
		ru1.ru_utime.tv_sec - ru0.ru_utime.tv_sec);

    if ( Nvpl.vpl_npages > 0 )
    {
	/* vacuum indices again if needed */
	if ( Irel != (Relation*) NULL )
	{
	    VPageDescr *vpleft, *vpright, vpsave;
		
	    /* re-sort Nvpl.vpl_pgdesc */
	    for (vpleft = Nvpl.vpl_pgdesc, 
		vpright = Nvpl.vpl_pgdesc + Nvpl.vpl_npages - 1;
		vpleft < vpright; vpleft++, vpright--)
	    {
		vpsave = *vpleft; *vpleft = *vpright; *vpright = vpsave;
	    }
	    for (i = 0; i < nindices; i++)
1197
		vc_vaconeind (&Nvpl, Irel[i], vacrelstats->ntups);
1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239
	}

	/* 
	 * clean moved tuples from last page in Nvpl list
	 * if some tuples left there
	 */
	if ( vpc->vpd_noff > 0 && offnum <= maxoff )
	{
	    Assert (vpc->vpd_blkno == blkno - 1);
	    buf = ReadBuffer(onerel, vpc->vpd_blkno);
	    page = BufferGetPage (buf);
	    ntups = 0;
	    maxoff = offnum;
	    for (offnum = FirstOffsetNumber;
			offnum < maxoff;
			offnum = OffsetNumberNext(offnum))
	    {
	    	itemid = PageGetItemId(page, offnum);
	    	if (!ItemIdIsUsed(itemid))
		    continue;
	    	htup = (HeapTuple) PageGetItem(page, itemid);
	    	Assert ( TransactionIdEquals((TransactionId)htup->t_xmax, myXID) );
		itemid->lp_flags &= ~LP_USED;
		ntups++;
	    }
	    Assert ( vpc->vpd_noff == ntups );
	    PageRepairFragmentation(page);
	    WriteBuffer (buf);
	}

	/* now - free new list of reapped pages */
	vpp = Nvpl.vpl_pgdesc;
	for (i = 0; i < Nvpl.vpl_npages; i++, vpp++)
	    pfree(*vpp);
	pfree (Nvpl.vpl_pgdesc);
    }
	
    /* truncate relation */
    if ( blkno < nblocks )
    {
    	blkno = smgrtruncate (onerel->rd_rel->relsmgr, onerel, blkno);
	Assert ( blkno >= 0 );
1240
	vacrelstats->npages = blkno;	/* set new number of blocks */
1241 1242 1243 1244 1245 1246 1247 1248 1249 1250
    }

    if ( archrel != (Relation) NULL )
	heap_close(archrel);

    if ( Irel != (Relation*) NULL )	/* pfree index' allocations */
    {
	pfree (Idesc);
	pfree (idatum);
	pfree (inulls);
1251
	vc_clsindices (nindices, Irel);
1252 1253 1254 1255
    }

    pfree (vpc);

1256
} /* vc_rpfheap */
1257 1258

/*
1259
 *  vc_vacheap() -- free dead tuples
1260 1261 1262 1263 1264
 *
 *	This routine marks dead tuples as unused and truncates relation
 *	if there are "empty" end-blocks.
 */
static void
1265
vc_vacheap (VRelStats *vacrelstats, Relation onerel, VPageList Vvpl)
1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276
{
    Buffer buf;
    Page page;
    VPageDescr *vpp;
    Relation archrel;
    int nblocks;
    int i;

    nblocks = Vvpl->vpl_npages;
    /* if the relation has an archive, open it */
    if (onerel->rd_rel->relarch != 'n')
1277
	archrel = vc_getarchrel(onerel);
1278
    else
1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289
    {
	archrel = (Relation) NULL;
	nblocks -= Vvpl->vpl_nemend;	/* nothing to do with them */
    }
	
    for (i = 0, vpp = Vvpl->vpl_pgdesc; i < nblocks; i++, vpp++)
    {
	if ( (*vpp)->vpd_noff > 0 )
	{
	    buf = ReadBuffer(onerel, (*vpp)->vpd_blkno);
	    page = BufferGetPage (buf);
1290
	    vc_vacpage (page, *vpp, archrel);
1291 1292 1293 1294 1295 1296 1297
	    WriteBuffer (buf);
	}
    }

    /* truncate relation if there are some empty end-pages */
    if ( Vvpl->vpl_nemend > 0 )
    {
1298 1299 1300
	Assert ( vacrelstats->npages >= Vvpl->vpl_nemend );
	nblocks = vacrelstats->npages - Vvpl->vpl_nemend;
	elog (MESSAGE_LEVEL, "Rel %.*s: Pages: %u --> %u.",
1301
		NAMEDATALEN, (RelationGetRelationName(onerel))->data, 
1302
		vacrelstats->npages, nblocks);
1303 1304 1305 1306 1307 1308 1309 1310 1311

	/* 
	 * we have to flush "empty" end-pages (if changed, but who knows it)
	 * before truncation 
	 */
	FlushBufferPool(!TransactionFlushEnabled());

    	nblocks = smgrtruncate (onerel->rd_rel->relsmgr, onerel, nblocks);
	Assert ( nblocks >= 0 );
1312
	vacrelstats->npages = nblocks;	/* set new number of blocks */
1313 1314 1315 1316 1317
    }

    if ( archrel != (Relation) NULL )
	heap_close(archrel);

1318
} /* vc_vacheap */
1319 1320

/*
1321
 *  vc_vacpage() -- free (and archive if needed) dead tuples on a page
1322 1323 1324
 *		     and repaire its fragmentation.
 */
static void
1325
vc_vacpage (Page page, VPageDescr vpd, Relation archrel)
1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337
{
    ItemId itemid;
    HeapTuple htup;
    int i;
    
    Assert ( vpd->vpd_nusd == 0 );
    for (i=0; i < vpd->vpd_noff; i++)
    {
	itemid = &(((PageHeader) page)->pd_linp[vpd->vpd_voff[i] - 1]);
	if ( archrel != (Relation) NULL && ItemIdIsUsed(itemid) )
	{
	    htup = (HeapTuple) PageGetItem (page, itemid);
1338
	    vc_archive (archrel, htup);
1339 1340 1341 1342 1343
	}
	itemid->lp_flags &= ~LP_USED;
    }
    PageRepairFragmentation(page);

1344
} /* vc_vacpage */
1345

1346 1347 1348 1349 1350
/*
 *  _vc_scanoneind() -- scan one index relation to update statistic.
 *
 */
static void
1351
vc_scanoneind (Relation indrel, int nhtups)
1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375
{
    RetrieveIndexResult res;
    IndexScanDesc iscan;
    int nitups;
    int nipages;
    struct rusage ru0, ru1;

    getrusage(RUSAGE_SELF, &ru0);

    /* walk through the entire index */
    iscan = index_beginscan(indrel, false, 0, (ScanKey) NULL);
    nitups = 0;

    while ((res = index_getnext(iscan, ForwardScanDirection))
					!= (RetrieveIndexResult) NULL)
    {
	nitups++;
	pfree(res);
    }

    index_endscan(iscan);

    /* now update statistics in pg_class */
    nipages = RelationGetNumberOfBlocks(indrel);
1376
    vc_updstats(indrel->rd_id, nipages, nitups, false, NULL);
1377 1378 1379

    getrusage(RUSAGE_SELF, &ru1);

1380
    elog (MESSAGE_LEVEL, "Ind %.*s: Pages %u; Tuples %u. Elapsed %u/%u sec.",
1381 1382 1383 1384 1385 1386 1387 1388
	NAMEDATALEN, indrel->rd_rel->relname.data, nipages, nitups, 
	ru1.ru_stime.tv_sec - ru0.ru_stime.tv_sec, 
	ru1.ru_utime.tv_sec - ru0.ru_utime.tv_sec);

    if ( nitups != nhtups )
    	elog (NOTICE, "NUMBER OF INDEX' TUPLES (%u) IS NOT THE SAME AS HEAP' (%u)",
    		nitups, nhtups);

1389
} /* vc_scanoneind */
1390

1391
/*
1392
 *  vc_vaconeind() -- vacuum one index relation.
1393
 *
1394 1395
 *	Vpl is the VPageList of the heap we're currently vacuuming.
 *	It's locked. Indrel is an index relation on the vacuumed heap. 
1396 1397 1398
 *	We don't set locks on the index	relation here, since the indexed 
 *	access methods support locking at different granularities. 
 *	We let them handle it.
1399 1400 1401 1402 1403
 *
 *	Finally, we arrange to update the index relation's statistics in
 *	pg_class.
 */
static void
1404
vc_vaconeind(VPageList vpl, Relation indrel, int nhtups)
1405 1406 1407 1408 1409 1410 1411
{
    RetrieveIndexResult res;
    IndexScanDesc iscan;
    ItemPointer heapptr;
    int nvac;
    int nitups;
    int nipages;
1412
    VPageDescr vp;
1413 1414 1415
    struct rusage ru0, ru1;

    getrusage(RUSAGE_SELF, &ru0);
1416 1417 1418 1419 1420 1421 1422 1423 1424 1425

    /* walk through the entire index */
    iscan = index_beginscan(indrel, false, 0, (ScanKey) NULL);
    nvac = 0;
    nitups = 0;

    while ((res = index_getnext(iscan, ForwardScanDirection))
	   != (RetrieveIndexResult) NULL) {
	heapptr = &res->heap_iptr;

1426
	if ( (vp = vc_tidreapped (heapptr, vpl)) != (VPageDescr) NULL)
1427
	{
1428 1429 1430 1431 1432 1433 1434
#if 0
	    elog(DEBUG, "<%x,%x> -> <%x,%x>",
		 ItemPointerGetBlockNumber(&(res->index_iptr)),
		 ItemPointerGetOffsetNumber(&(res->index_iptr)),
		 ItemPointerGetBlockNumber(&(res->heap_iptr)),
		 ItemPointerGetOffsetNumber(&(res->heap_iptr)));
#endif
1435 1436 1437 1438 1439 1440
	    if ( vp->vpd_noff == 0 ) 
	    {				/* this is EmptyPage !!! */
	    	elog (NOTICE, "Ind %.*s: pointer to EmptyPage (blk %u off %u) - fixing",
			NAMEDATALEN, indrel->rd_rel->relname.data,
	    		vp->vpd_blkno, ItemPointerGetOffsetNumber(heapptr));
	    }
1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454
	    ++nvac;
	    index_delete(indrel, &res->index_iptr);
	} else {
	    nitups++;
	}

	/* be tidy */
	pfree(res);
    }

    index_endscan(iscan);

    /* now update statistics in pg_class */
    nipages = RelationGetNumberOfBlocks(indrel);
1455
    vc_updstats(indrel->rd_id, nipages, nitups, false, NULL);
1456 1457 1458

    getrusage(RUSAGE_SELF, &ru1);

1459
    elog (MESSAGE_LEVEL, "Ind %.*s: Pages %u; Tuples %u: Deleted %u. Elapsed %u/%u sec.",
1460 1461 1462 1463
	NAMEDATALEN, indrel->rd_rel->relname.data, nipages, nitups, nvac,
	ru1.ru_stime.tv_sec - ru0.ru_stime.tv_sec, 
	ru1.ru_utime.tv_sec - ru0.ru_utime.tv_sec);

1464 1465 1466 1467
    if ( nitups != nhtups )
    	elog (NOTICE, "NUMBER OF INDEX' TUPLES (%u) IS NOT THE SAME AS HEAP' (%u)",
    		nitups, nhtups);

1468
} /* vc_vaconeind */
1469 1470

/*
1471
 *  vc_tidreapped() -- is a particular tid reapped?
1472 1473 1474 1475
 *
 *	vpl->VPageDescr_array is sorted in right order.
 */
static VPageDescr
1476
vc_tidreapped(ItemPointer itemptr, VPageList vpl)
1477 1478 1479 1480 1481 1482 1483 1484 1485 1486
{
    OffsetNumber ioffno;
    OffsetNumber *voff;
    VPageDescr vp, *vpp;
    VPageDescrData vpd;

    vpd.vpd_blkno = ItemPointerGetBlockNumber(itemptr);
    ioffno = ItemPointerGetOffsetNumber(itemptr);
	
    vp = &vpd;
1487
    vpp = (VPageDescr*) vc_find_eq ((char*)(vpl->vpl_pgdesc), 
1488
		vpl->vpl_npages, sizeof (VPageDescr), (char*)&vp, 
1489
		vc_cmp_blk);
1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500

    if ( vpp == (VPageDescr*) NULL )
	return ((VPageDescr)NULL);
    vp = *vpp;

    /* ok - we are on true page */

    if ( vp->vpd_noff == 0 ) {		/* this is EmptyPage !!! */
	return (vp);
    }
    
1501
    voff = (OffsetNumber*) vc_find_eq ((char*)(vp->vpd_voff), 
1502
		vp->vpd_noff, sizeof (OffsetNumber), (char*)&ioffno, 
1503
		vc_cmp_offno);
1504 1505 1506 1507 1508 1509

    if ( voff == (OffsetNumber*) NULL )
	return ((VPageDescr)NULL);

    return (vp);

1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571
} /* vc_tidreapped */

/*
 *  vc_attrstats() -- compute column statistics used by the optimzer
 *
 *  We compute the column min, max, null and non-null counts.
 *  Plus we attempt to find the count of the value that occurs most
 *  frequently in each column
 *  These figures are used to compute the selectivity of the column
 *
 *  We use a three-bucked cache to get the most frequent item
 *  The 'guess' buckets count hits.  A cache miss causes guess1
 *  to get the most hit 'guess' item in the most recent cycle, and
 *  the new item goes into guess2.  Whenever the total count of hits
 *  of a 'guess' entry is larger than 'best', 'guess' becomes 'best'.
 *
 *  This method works perfectly for columns with unique values, and columns
 *  with only two unique values, plus nulls.
 *
 *  It becomes less perfect as the number of unique values increases and
 *  their distribution in the table becomes more random.
 *
 */
static void
vc_attrstats(Relation onerel, VacAttrStats *vacattrstats, HeapTuple htup)
{
    int i, attr_cnt;
    AttributeTupleForm *attr;
    TupleDesc tupDesc;
    Datum value;
    bool isnull;

    attr_cnt = onerel->rd_att->natts;
    attr = onerel->rd_att->attrs;
    tupDesc = onerel->rd_att;

    for (i = 0; i < attr_cnt; i++) {
	VacAttrStats *stats = &vacattrstats[i];
	bool value_hit = true;

	value = (Datum) heap_getattr(htup, InvalidBuffer, i+1, tupDesc, &isnull);

	if (!VacAttrStatsEqValid(stats))
		continue;

	if (isnull)
	    stats->null_cnt++;
	else {
	    stats->nonnull_cnt++;
	    if (stats->initialized == false) {
		vc_bucketcpy(stats->attr, value, &stats->best, &stats->best_len);
		/* best_cnt gets incremented later */
		vc_bucketcpy(stats->attr, value, &stats->guess1, &stats->guess1_len);
		stats->guess1_cnt = stats->guess1_hits = 1;
		vc_bucketcpy(stats->attr, value, &stats->guess2, &stats->guess2_len);
		stats->guess2_hits = 1;
		if (VacAttrStatsLtGtValid(stats)) {
		    vc_bucketcpy(stats->attr, value, &stats->max , &stats->max_len);
		    vc_bucketcpy(stats->attr, value, &stats->min, &stats->min_len);
		}
		stats->initialized = true;
	    }
Bruce Momjian's avatar
Bruce Momjian committed
1572
	    if (VacAttrStatsLtGtValid(stats)) {
1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584
		if (fmgr(stats->cmplt,value,stats->min)) {
		    vc_bucketcpy(stats->attr, value, &stats->min, &stats->min_len);
		    stats->min_cnt = 0;
	    	}
		if (fmgr(stats->cmpgt,value,stats->max)) {
		    vc_bucketcpy(stats->attr, value, &stats->max, &stats->max_len);
		    stats->max_cnt = 0;
	    	}
	    	if (fmgr(stats->cmpeq,value,stats->min))
		    stats->min_cnt++;
		else if (fmgr(stats->cmpeq,value,stats->max))
		    stats->max_cnt++;
1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603
	    }
	    if (fmgr(stats->cmpeq,value,stats->best))
		stats->	best_cnt++;
	    else if (fmgr(stats->cmpeq,value,stats->guess1)) {
		stats->guess1_cnt++;
		stats->	guess1_hits++;
	    }
	    else if (fmgr(stats->cmpeq,value,stats->guess2))
		stats->guess2_hits++;
	    else value_hit = false;

	    if (stats->guess2_hits > stats->guess1_hits) {
		swapDatum(stats->guess1,stats->guess2);
		swapInt(stats->guess1_len,stats->guess2_len);
		stats->guess1_cnt = stats->guess2_hits;
		swapLong(stats->guess1_hits, stats->guess2_hits);
	    }
	    if (stats->guess1_cnt > stats->best_cnt) {
		swapDatum(stats->best,stats->guess1);
1604
		swapInt(stats->best_len,stats->guess1_len);
1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630
		swapLong(stats->best_cnt,stats->guess1_cnt);
		stats->guess1_hits = 1;
		stats->	guess2_hits = 1;
	    }
	    if (!value_hit) {
		vc_bucketcpy(stats->attr, value, &stats->guess2, &stats->guess2_len);
		stats->guess1_hits = 1;
		stats->	guess2_hits = 1;
	    }
	}
    }
    return;
}

/*
 *  vc_bucketcpy() -- update pg_class statistics for one relation
 *
 */
static void
vc_bucketcpy(AttributeTupleForm attr, Datum value, Datum *bucket, int16 *bucket_len)
{
    if (attr->attbyval && attr->attlen != -1)
    	*bucket = value;
    else {
    	int len = (attr->attlen != -1 ? attr->attlen : VARSIZE(value));

1631
 	if (len > *bucket_len)
1632
    	{
1633
	    if (*bucket_len != 0)
1634
	    	pfree(DatumGetPointer(*bucket));
1635
	    *bucket = PointerGetDatum(palloc(len));
1636
	    *bucket_len = len;
1637
	}
1638
    	memmove(DatumGetPointer(*bucket), DatumGetPointer(value), len);
1639 1640
    }
}
1641 1642

/*
1643
 *  vc_updstats() -- update pg_class statistics for one relation
1644 1645 1646
 *
 *	This routine works for both index and heap relation entries in
 *	pg_class.  We violate no-overwrite semantics here by storing new
1647
 *	values for ntups, npages, and hasindex directly in the pg_class
1648 1649 1650 1651 1652 1653
 *	tuple that's already on the page.  The reason for this is that if
 *	we updated these tuples in the usual way, then every tuple in pg_class
 *	would be replaced every day.  This would make planning and executing
 *	historical queries very expensive.
 */
static void
1654
vc_updstats(Oid relid, int npages, int ntups, bool hasindex, VacAttrStats *vacattrstats)
1655
{
1656 1657 1658 1659 1660
    Relation rd, ad, sd;
    HeapScanDesc rsdesc, asdesc;
    TupleDesc sdesc;
    HeapTuple rtup, atup, stup;
    Buffer rbuf, abuf;
1661
    Form_pg_class pgcform;
1662 1663
    ScanKeyData rskey, askey;
    AttributeTupleForm attp;
1664 1665 1666 1667

    /*
     * update number of tuples and number of pages in pg_class
     */
1668
    ScanKeyEntryInitialize(&rskey, 0x0, ObjectIdAttributeNumber,
1669 1670 1671 1672
			   ObjectIdEqualRegProcedure,
			   ObjectIdGetDatum(relid));

    rd = heap_openr(RelationRelationName);
1673
    rsdesc = heap_beginscan(rd, false, NowTimeQual, 1, &rskey);
1674

1675
    if (!HeapTupleIsValid(rtup = heap_getnext(rsdesc, 0, &rbuf)))
1676 1677 1678 1679
	elog(WARN, "pg_class entry for relid %d vanished during vacuuming",
		   relid);

    /* overwrite the existing statistics in the tuple */
1680 1681 1682
    vc_setpagelock(rd, BufferGetBlockNumber(rbuf));
    pgcform = (Form_pg_class) GETSTRUCT(rtup);
    pgcform->reltuples = ntups;
1683 1684
    pgcform->relpages = npages;
    pgcform->relhasindex = hasindex;
1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783

    if (vacattrstats != NULL)
    {
    	ad = heap_openr(AttributeRelationName);
	sd = heap_openr(StatisticRelationName);
        ScanKeyEntryInitialize(&askey, 0, Anum_pg_attribute_attrelid,
		   F_INT4EQ, relid);

	asdesc = heap_beginscan(ad, false, NowTimeQual, 1, &askey);

	while (HeapTupleIsValid(atup = heap_getnext(asdesc, 0, &abuf))) {
	    int slot, i;
	    double selratio;  /* average ratio of rows selected for a random constant */
	    VacAttrStats *stats;
	    Datum values[ Natts_pg_statistic ];
    	    char nulls[ Natts_pg_statistic ];

	    attp = (AttributeTupleForm) GETSTRUCT(atup);
	    slot = attp->attnum - 1;
	    if (slot < 0) /* skip system attributes for now,
			   they are unique anyway */
		continue;            
	    stats = &vacattrstats[slot];

	    /* overwrite the existing statistics in the tuple */
	    if (VacAttrStatsEqValid(stats)) {

		vc_setpagelock(ad, BufferGetBlockNumber(abuf));

		if (stats->nonnull_cnt + stats->null_cnt == 0 ||
		    (stats->null_cnt <= 1 && stats->best_cnt == 1))
		    selratio = 0;
		else if (VacAttrStatsLtGtValid(stats) && stats->min_cnt + stats->max_cnt == stats->nonnull_cnt)
		{
		    double min_cnt_d = stats->min_cnt, 
			   max_cnt_d = stats->max_cnt, 
			   null_cnt_d = stats->null_cnt,
			   nonnullcnt_d = stats->nonnull_cnt; /* prevent overflow */
		    selratio = (min_cnt_d*min_cnt_d+max_cnt_d*max_cnt_d+null_cnt_d*null_cnt_d)/
				(nonnullcnt_d+null_cnt_d)/(nonnullcnt_d+null_cnt_d);
		}
		else {
		    double most = (double)(stats->best_cnt > stats->null_cnt ? stats->best_cnt : stats->null_cnt);
		    double total = ((double)stats->nonnull_cnt)+((double)stats->null_cnt);
		    /* we assume count of other values are 20% 
		       of best count in table */
		    selratio = (most*most + 0.20*most*(total-most))/total/total;
		}
		if (selratio > 1.0)
		    selratio = 1.0;
	    	attp->attnvals = (selratio ? (selratio * ATTNVALS_SCALE) : 0);
		WriteNoReleaseBuffer(abuf);

		/* DO PG_STATISTIC INSERTS */

		/* doing system relations, especially pg_statistic is a problem */
		if (VacAttrStatsLtGtValid(stats) && stats->initialized /* &&
			!IsSystemRelationName(pgcform->relname.data)*/) {
		    func_ptr out_function;
		    char *out_string;
		    int dummy;
		
		    for (i = 0; i < Natts_pg_statistic; ++i) nulls[i] = ' ';

		    /* ----------------
	    	     *	initialize values[]
		     * ----------------
		     */
		    i = 0;
		    values[i++] = (Datum) relid;	/* 1 */
		    values[i++] = (Datum) attp->attnum;	/* 2 */
		    values[i++] = (Datum) InvalidOid; 	/* 3 */
		    fmgr_info(stats->outfunc, &out_function, &dummy);
		    out_string = (*out_function)(stats->min, stats->attr->atttypid);
		    values[i++] = (Datum) fmgr(TextInRegProcedure,out_string);
		    pfree(out_string);
		    out_string = (char *)(*out_function)(stats->max, stats->attr->atttypid);
		    values[i++] = (Datum) fmgr(TextInRegProcedure,out_string);
		    pfree(out_string);

		    sdesc = sd->rd_att;
	
		    stup = heap_formtuple(sdesc, values, nulls);

		    /* ----------------
		     *	insert the tuple in the relation and get the tuple's oid.
		     * ----------------
		     */
		    heap_insert(sd, stup);
		    pfree(DatumGetPointer(values[3]));
		    pfree(DatumGetPointer(values[4]));
		    pfree(stup);
		}
	    }
	}
	heap_endscan(asdesc);
	heap_close(ad);
	heap_close(sd);
    }
1784 1785
 
    /* XXX -- after write, should invalidate relcache in other backends */
1786
    WriteNoReleaseBuffer(rbuf);	/* heap_endscan release scan' buffers ? */
1787

1788 1789 1790
    /* invalidating system relations confuses the function cache
       of pg_operator and pg_opclass */
    if ( !IsSystemRelationName(pgcform->relname.data))
1791
    	RelationInvalidateHeapTuple(rd, rtup);
1792

1793
    /* that's all, folks */
1794
    heap_endscan(rsdesc);
1795 1796 1797 1798
    heap_close(rd);

}

1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830
/*
 *  vc_delhilowstats() -- delete pg_statistics rows
 *
 */
static void
vc_delhilowstats(Oid relid)
{
    Relation pgstatistic;
    HeapScanDesc pgsscan;
    HeapTuple pgstup;
    ScanKeyData  pgskey;

    pgstatistic = heap_openr(StatisticRelationName);

    if (relid != InvalidOid ) {
	ScanKeyEntryInitialize(&pgskey, 0x0, Anum_pg_statistic_starelid,
				   ObjectIdEqualRegProcedure,
				   ObjectIdGetDatum(relid));
    	pgsscan = heap_beginscan(pgstatistic, false, NowTimeQual, 1, &pgskey);
    }
    else
	pgsscan = heap_beginscan(pgstatistic, false, NowTimeQual, 0, NULL);

    while (HeapTupleIsValid(pgstup = heap_getnext(pgsscan, 0, NULL))) {
	heap_delete(pgstatistic, &pgstup->t_ctid);
    }

    heap_endscan(pgsscan);
    heap_close(pgstatistic);
}

static void vc_setpagelock(Relation rel, BlockNumber blkno)
1831 1832 1833 1834 1835 1836 1837 1838 1839 1840
{
    ItemPointerData itm;

    ItemPointerSet(&itm, blkno, 1);

    RelationSetLockForWritePage(rel, &itm);
}


/*
1841
 *  vc_reappage() -- save a page on the array of reapped pages.
1842 1843
 *
 *	As a side effect of the way that the vacuuming loop for a given
1844 1845
 *	relation works, higher pages come after lower pages in the array
 *	(and highest tid on a page is last).
1846
 */
1847
static void 
1848
vc_reappage(VPageList vpl, VPageDescr vpc)
1849
{
1850
    VPageDescr newvpd;
1851

1852
    /* allocate a VPageDescrData entry */
1853
    newvpd = (VPageDescr) palloc(sizeof(VPageDescrData) + vpc->vpd_noff*sizeof(OffsetNumber));
1854 1855

    /* fill it in */
1856 1857 1858 1859
    if ( vpc->vpd_noff > 0 )
    	memmove (newvpd->vpd_voff, vpc->vpd_voff, vpc->vpd_noff*sizeof(OffsetNumber));
    newvpd->vpd_blkno = vpc->vpd_blkno;
    newvpd->vpd_free = vpc->vpd_free;
1860
    newvpd->vpd_nusd = vpc->vpd_nusd;
1861 1862
    newvpd->vpd_noff = vpc->vpd_noff;

1863
    /* insert this page into vpl list */
1864
    vc_vpinsert (vpl, newvpd);
1865
    
1866
} /* vc_reappage */
1867 1868

static void
1869
vc_vpinsert (VPageList vpl, VPageDescr vpnew)
1870 1871 1872 1873 1874 1875 1876 1877 1878 1879
{

    /* allocate a VPageDescr entry if needed */
    if ( vpl->vpl_npages == 0 )
    	vpl->vpl_pgdesc = (VPageDescr*) palloc(100*sizeof(VPageDescr));
    else if ( vpl->vpl_npages % 100 == 0 )
    	vpl->vpl_pgdesc = (VPageDescr*) repalloc(vpl->vpl_pgdesc, (vpl->vpl_npages+100)*sizeof(VPageDescr));
    vpl->vpl_pgdesc[vpl->vpl_npages] = vpnew;
    (vpl->vpl_npages)++;
    
1880 1881 1882
}

static void
1883
vc_free(Portal p, VRelList vrl)
1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903
{
    VRelList p_vrl;
    MemoryContext old;
    PortalVariableMemory pmem;

    pmem = PortalGetVariableMemory(p);
    old = MemoryContextSwitchTo((MemoryContext)pmem);

    while (vrl != (VRelList) NULL) {

	/* free rel list entry */
	p_vrl = vrl;
	vrl = vrl->vrl_next;
	pfree(p_vrl);
    }

    (void) MemoryContextSwitchTo(old);
}

/*
1904
 *  vc_getarchrel() -- open the archive relation for a heap relation
1905 1906 1907 1908 1909 1910 1911 1912
 *
 *	The archive relation is named 'a,XXXXX' for the heap relation
 *	whose relid is XXXXX.
 */

#define ARCHIVE_PREFIX	"a,"

static Relation
1913
vc_getarchrel(Relation heaprel)
1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927
{
    Relation archrel;
    char *archrelname;

    archrelname = palloc(sizeof(ARCHIVE_PREFIX) + NAMEDATALEN); /* bogus */
    sprintf(archrelname, "%s%d", ARCHIVE_PREFIX, heaprel->rd_id);

    archrel = heap_openr(archrelname);

    pfree(archrelname);
    return (archrel);
}

/*
1928
 *  vc_archive() -- write a tuple to an archive relation
1929 1930 1931 1932 1933
 *
 *	In the future, this will invoke the archived accessd method.  For
 *	now, archive relations are on mag disk.
 */
static void
1934
vc_archive(Relation archrel, HeapTuple htup)
1935 1936 1937 1938 1939
{
    doinsert(archrel, htup);
}

static bool
1940
vc_isarchrel(char *rname)
1941 1942 1943 1944 1945 1946
{
    if (strncmp(ARCHIVE_PREFIX, rname,strlen(ARCHIVE_PREFIX)) == 0)
	return (true);

    return (false);
}
1947 1948

static char *
1949
vc_find_eq (char *bot, int nelem, int size, char *elm, int (*compar)(char *, char *))
1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998
{
    int res;
    int last = nelem - 1;
    int celm = nelem / 2;
    bool last_move, first_move;
    
    last_move = first_move = true;
    for ( ; ; )
    {
	if ( first_move == true )
	{
	    res = compar (bot, elm);
	    if ( res > 0 )
		return (NULL);
	    if ( res == 0 )
		return (bot);
	    first_move = false;
	}
	if ( last_move == true )
	{
	    res = compar (elm, bot + last*size);
	    if ( res > 0 )
		return (NULL);
	    if ( res == 0 )
		return (bot + last*size);
	    last_move = false;
	}
    	res = compar (elm, bot + celm*size);
    	if ( res == 0 )
	    return (bot + celm*size);
	if ( res < 0 )
	{
	    if ( celm == 0 )
		return (NULL);
	    last = celm - 1;
	    celm = celm / 2;
	    last_move = true;
	    continue;
	}
	
	if ( celm == last )
	    return (NULL);
	    
	last = last - celm - 1;
	bot = bot + (celm+1)*size;
	celm = (last + 1) / 2;
	first_move = true;
    }

1999
} /* vc_find_eq */
2000 2001

static int 
2002
vc_cmp_blk (char *left, char *right)
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
{
    BlockNumber lblk, rblk;

    lblk = (*((VPageDescr*)left))->vpd_blkno;
    rblk = (*((VPageDescr*)right))->vpd_blkno;

    if ( lblk < rblk )
    	return (-1);
    if ( lblk == rblk )
    	return (0);
    return (1);

2015
} /* vc_cmp_blk */
2016 2017

static int 
2018
vc_cmp_offno (char *left, char *right)
2019 2020 2021 2022 2023 2024 2025 2026
{

    if ( *(OffsetNumber*)left < *(OffsetNumber*)right )
    	return (-1);
    if ( *(OffsetNumber*)left == *(OffsetNumber*)right )
    	return (0);
    return (1);

2027
} /* vc_cmp_offno */
2028 2029 2030


static void
2031
vc_getindices (Oid relid, int *nindices, Relation **Irel)
2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100
{
    Relation pgindex;
    Relation irel;
    TupleDesc pgidesc;
    HeapTuple pgitup;
    HeapScanDesc pgiscan;
    Datum d;
    int i, k;
    bool n;
    ScanKeyData pgikey;
    Oid *ioid;

    *nindices = i = 0;
    
    ioid = (Oid *) palloc(10*sizeof(Oid));

    /* prepare a heap scan on the pg_index relation */
    pgindex = heap_openr(IndexRelationName);
    pgidesc = RelationGetTupleDescriptor(pgindex);

    ScanKeyEntryInitialize(&pgikey, 0x0, Anum_pg_index_indrelid,
			   ObjectIdEqualRegProcedure,
			   ObjectIdGetDatum(relid));

    pgiscan = heap_beginscan(pgindex, false, NowTimeQual, 1, &pgikey);

    while (HeapTupleIsValid(pgitup = heap_getnext(pgiscan, 0, NULL))) {
	d = (Datum) heap_getattr(pgitup, InvalidBuffer, Anum_pg_index_indexrelid,
				 pgidesc, &n);
	i++;
	if ( i % 10 == 0 )
	    ioid = (Oid *) repalloc(ioid, (i+10)*sizeof(Oid));
	ioid[i-1] = DatumGetObjectId(d);
    }

    heap_endscan(pgiscan);
    heap_close(pgindex);

    if ( i == 0 ) {	/* No one index found */
	pfree(ioid);
	return;
    }

    if ( Irel != (Relation **) NULL )
	*Irel = (Relation *) palloc(i * sizeof(Relation));
    
    for (k = 0; i > 0; )
    {
	irel = index_open(ioid[--i]);
	if ( irel != (Relation) NULL )
	{
	    if ( Irel != (Relation **) NULL )
		(*Irel)[k] = irel;
	    else
		index_close (irel);
	    k++;
	}
	else
	    elog (NOTICE, "CAN't OPEN INDEX %u - SKIP IT", ioid[i]);
    }
    *nindices = k;
    pfree(ioid);

    if ( Irel != (Relation **) NULL && *nindices == 0 )
    {
	pfree (*Irel);
	*Irel = (Relation *) NULL;
    }

2101
} /* vc_getindices */
2102 2103 2104


static void
2105
vc_clsindices (int nindices, Relation *Irel)
2106 2107 2108 2109 2110 2111 2112 2113 2114 2115
{

    if ( Irel == (Relation*) NULL )
    	return;

    while (nindices--) {
	index_close (Irel[nindices]);
    }
    pfree (Irel);

2116
} /* vc_clsindices */
2117 2118 2119


static void
2120
vc_mkindesc (Relation onerel, int nindices, Relation *Irel, IndDesc **Idesc)
2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151
{
    IndDesc *idcur;
    HeapTuple pgIndexTup;
    AttrNumber *attnumP;
    int natts;
    int i;

    *Idesc = (IndDesc *) palloc (nindices * sizeof (IndDesc));
    
    for (i = 0, idcur = *Idesc; i < nindices; i++, idcur++) {
	pgIndexTup =
		SearchSysCacheTuple(INDEXRELID,
				ObjectIdGetDatum(Irel[i]->rd_id),
				0,0,0);
	Assert(pgIndexTup);
	idcur->tform = (IndexTupleForm)GETSTRUCT(pgIndexTup);
	for (attnumP = &(idcur->tform->indkey[0]), natts = 0;
		*attnumP != InvalidAttrNumber && natts != INDEX_MAX_KEYS;
		attnumP++, natts++);
	if (idcur->tform->indproc != InvalidOid) {
	    idcur->finfoP = &(idcur->finfo);
	    FIgetnArgs(idcur->finfoP) = natts;
	    natts = 1;
	    FIgetProcOid(idcur->finfoP) = idcur->tform->indproc;
	    *(FIgetname(idcur->finfoP)) = '\0';
	} else
	    idcur->finfoP = (FuncIndexInfo *) NULL;
	
	idcur->natts = natts;
    }
    
2152
} /* vc_mkindesc */
2153 2154 2155


static bool
2156
vc_enough_space (VPageDescr vpd, Size len)
2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172
{

    len = DOUBLEALIGN(len);

    if ( len > vpd->vpd_free )
    	return (false);
    
    if ( vpd->vpd_nusd < vpd->vpd_noff )	/* there are free itemid(s) */
    	return (true);				/* and len <= free_space */
    
    /* ok. noff_usd >= noff_free and so we'll have to allocate new itemid */
    if ( len <= vpd->vpd_free - sizeof (ItemIdData) )
    	return (true);
    
    return (false);
    
2173
} /* vc_enough_space */