コード例 #1
0
ファイル: gist.c プロジェクト: berkeley-cs186/course-fa07
/*
 *	gistSplit -- split a page in the tree.
 */
static IndexTuple *
gistSplit(Relation r,
		  Buffer buffer,
		  IndexTuple *itup,		/* contains compressed entry */
		  int *len,
		  GISTSTATE *giststate,
		  InsertIndexResult *res)
{
	Page		p;
	Buffer		leftbuf,
				rightbuf;
	Page		left,
				right;
	IndexTuple *lvectup,
			   *rvectup,
			   *newtup;
	BlockNumber lbknum,
				rbknum;
	GISTPageOpaque opaque;
	GIST_SPLITVEC v;
	GistEntryVector *entryvec;
	bool	   *decompvec;
	int			i,
				j,
				nlen;
	int			MaxGrpId = 1;
	Datum		datum;
	bool		IsNull;

	p = (Page) BufferGetPage(buffer);
	opaque = (GISTPageOpaque) PageGetSpecialPointer(p);

	/*
	 * The root of the tree is the first block in the relation.  If we're
	 * about to split the root, we need to do some hocus-pocus to enforce
	 * this guarantee.
	 */

	if (BufferGetBlockNumber(buffer) == GISTP_ROOT)
	{
		leftbuf = ReadBuffer(r, P_NEW);
		GISTInitBuffer(leftbuf, opaque->flags);
		lbknum = BufferGetBlockNumber(leftbuf);
		left = (Page) BufferGetPage(leftbuf);
	}
	else
	{
		leftbuf = buffer;
		IncrBufferRefCount(buffer);
		lbknum = BufferGetBlockNumber(buffer);
		left = (Page) PageGetTempPage(p, sizeof(GISTPageOpaqueData));
	}

	rightbuf = ReadBuffer(r, P_NEW);
	GISTInitBuffer(rightbuf, opaque->flags);
	rbknum = BufferGetBlockNumber(rightbuf);
	right = (Page) BufferGetPage(rightbuf);

	/* generate the item array */
	entryvec = palloc(GEVHDRSZ + (*len + 1) * sizeof(GISTENTRY));
	entryvec->n = *len + 1;
	decompvec = (bool *) palloc((*len + 1) * sizeof(bool));
	for (i = 1; i <= *len; i++)
	{
		datum = index_getattr(itup[i - 1], 1, giststate->tupdesc, &IsNull);
		gistdentryinit(giststate, 0, &(entryvec->vector[i]),
					   datum, r, p, i,
		   ATTSIZE(datum, giststate->tupdesc, 1, IsNull), FALSE, IsNull);
		if ((!isAttByVal(giststate, 0)) && entryvec->vector[i].key != datum)
			decompvec[i] = TRUE;
		else
			decompvec[i] = FALSE;
	}

	/*
	 * now let the user-defined picksplit function set up the split
	 * vector; in entryvec have no null value!!
	 */
	FunctionCall2(&giststate->picksplitFn[0],
				  PointerGetDatum(entryvec),
				  PointerGetDatum(&v));

	/* compatibility with old code */
	if (v.spl_left[v.spl_nleft - 1] == InvalidOffsetNumber)
		v.spl_left[v.spl_nleft - 1] = (OffsetNumber) *len;
	if (v.spl_right[v.spl_nright - 1] == InvalidOffsetNumber)
		v.spl_right[v.spl_nright - 1] = (OffsetNumber) *len;

	v.spl_lattr[0] = v.spl_ldatum;
	v.spl_rattr[0] = v.spl_rdatum;
	v.spl_lisnull[0] = false;
	v.spl_risnull[0] = false;

	/*
	 * if index is multikey, then we must to try get smaller bounding box
	 * for subkey(s)
	 */
	if (r->rd_att->natts > 1)
	{
		v.spl_idgrp = (int *) palloc0(sizeof(int) * (*len + 1));
		v.spl_grpflag = (char *) palloc0(sizeof(char) * (*len + 1));
		v.spl_ngrp = (int *) palloc(sizeof(int) * (*len + 1));

		MaxGrpId = gistfindgroup(giststate, entryvec->vector, &v);

		/* form union of sub keys for each page (l,p) */
		gistunionsubkey(r, giststate, itup, &v);

		/*
		 * if possible, we insert equivalent tuples with control by
		 * penalty for a subkey(s)
		 */
		if (MaxGrpId > 1)
			gistadjsubkey(r, itup, len, &v, giststate);

		pfree(v.spl_idgrp);
		pfree(v.spl_grpflag);
		pfree(v.spl_ngrp);
	}

	/* clean up the entry vector: its keys need to be deleted, too */
	for (i = 1; i <= *len; i++)
		if (decompvec[i])
			pfree(DatumGetPointer(entryvec->vector[i].key));
	pfree(entryvec);
	pfree(decompvec);

	/* form left and right vector */
	lvectup = (IndexTuple *) palloc(sizeof(IndexTuple) * v.spl_nleft);
	rvectup = (IndexTuple *) palloc(sizeof(IndexTuple) * v.spl_nright);

	for (i = 0; i < v.spl_nleft; i++)
		lvectup[i] = itup[v.spl_left[i] - 1];

	for (i = 0; i < v.spl_nright; i++)
		rvectup[i] = itup[v.spl_right[i] - 1];


	/* write on disk (may be need another split) */
	if (gistnospace(right, rvectup, v.spl_nright))
	{
		nlen = v.spl_nright;
		newtup = gistSplit(r, rightbuf, rvectup, &nlen, giststate,
			  (res && rvectup[nlen - 1] == itup[*len - 1]) ? res : NULL);
		ReleaseBuffer(rightbuf);
		for (j = 1; j < r->rd_att->natts; j++)
			if ((!isAttByVal(giststate, j)) && !v.spl_risnull[j])
				pfree(DatumGetPointer(v.spl_rattr[j]));
	}
	else
	{
		OffsetNumber l;

		l = gistwritebuffer(r, right, rvectup, v.spl_nright, FirstOffsetNumber);
		WriteBuffer(rightbuf);

		if (res)
			ItemPointerSet(&((*res)->pointerData), rbknum, l);

		nlen = 1;
		newtup = (IndexTuple *) palloc(sizeof(IndexTuple) * 1);
		newtup[0] = gistFormTuple(giststate, r, v.spl_rattr, v.spl_rattrsize, v.spl_risnull);
		ItemPointerSet(&(newtup[0]->t_tid), rbknum, 1);
	}


	if (gistnospace(left, lvectup, v.spl_nleft))
	{
		int			llen = v.spl_nleft;
		IndexTuple *lntup;

		lntup = gistSplit(r, leftbuf, lvectup, &llen, giststate,
			  (res && lvectup[llen - 1] == itup[*len - 1]) ? res : NULL);
		ReleaseBuffer(leftbuf);

		for (j = 1; j < r->rd_att->natts; j++)
			if ((!isAttByVal(giststate, j)) && !v.spl_lisnull[j])
				pfree(DatumGetPointer(v.spl_lattr[j]));

		newtup = gistjoinvector(newtup, &nlen, lntup, llen);
		pfree(lntup);
	}
	else
	{
		OffsetNumber l;

		l = gistwritebuffer(r, left, lvectup, v.spl_nleft, FirstOffsetNumber);
		if (BufferGetBlockNumber(buffer) != GISTP_ROOT)
			PageRestoreTempPage(left, p);

		WriteBuffer(leftbuf);

		if (res)
			ItemPointerSet(&((*res)->pointerData), lbknum, l);

		nlen += 1;
		newtup = (IndexTuple *) repalloc((void *) newtup, sizeof(IndexTuple) * nlen);
		newtup[nlen - 1] = gistFormTuple(giststate, r, v.spl_lattr, v.spl_lattrsize, v.spl_lisnull);
		ItemPointerSet(&(newtup[nlen - 1]->t_tid), lbknum, 1);
	}

	/* !!! pfree */
	pfree(rvectup);
	pfree(lvectup);
	pfree(v.spl_left);
	pfree(v.spl_right);

	*len = nlen;
	return newtup;
}
コード例 #2
0
/*
 * Calls user picksplit method for attno columns to split vector to
 * two vectors. May use attno+n columns data to
 * get better split.
 * Returns TRUE and v->spl_equiv = NULL if left and right unions of attno columns are the same,
 * so caller may find better split
 * Returns TRUE and v->spl_equiv != NULL if there is tuples which may be freely moved
 */
static bool
gistUserPicksplit(Relation r, GistEntryVector *entryvec, int attno, GistSplitVector *v,
				  IndexTuple *itup, int len, GISTSTATE *giststate)
{
	GIST_SPLITVEC *sv = &v->splitVector;

	/*
	 * now let the user-defined picksplit function set up the split vector; in
	 * entryvec there is no null value!!
	 */

	sv->spl_ldatum_exists = (v->spl_lisnull[attno]) ? false : true;
	sv->spl_rdatum_exists = (v->spl_risnull[attno]) ? false : true;
	sv->spl_ldatum = v->spl_lattr[attno];
	sv->spl_rdatum = v->spl_rattr[attno];

	FunctionCall2(&giststate->picksplitFn[attno],
				  PointerGetDatum(entryvec),
				  PointerGetDatum(sv));

	if ( sv->spl_nleft == 0 || sv->spl_nright == 0 )
	{
		ereport(DEBUG1,
				(errcode(ERRCODE_INTERNAL_ERROR),
				 errmsg("picksplit method for column %d of index \"%s\" failed",
											attno+1, RelationGetRelationName(r)),
				 errhint("The index is not optimal. To optimize it, contact a developer, or try to use the column as the second one in the CREATE INDEX command.")));

		/*
		 * Reinit GIST_SPLITVEC. Although that fields are not used
		 * by genericPickSplit(), let us set up it for further processing 
		 */
		sv->spl_ldatum_exists = (v->spl_lisnull[attno]) ? false : true;
		sv->spl_rdatum_exists = (v->spl_risnull[attno]) ? false : true;
		sv->spl_ldatum = v->spl_lattr[attno];
		sv->spl_rdatum = v->spl_rattr[attno];

		genericPickSplit(giststate, entryvec, sv, attno);

		if (sv->spl_ldatum_exists || sv->spl_rdatum_exists)
			supportSecondarySplit(r, giststate, attno, sv, v->spl_lattr[attno], v->spl_rattr[attno]);
	}
	else
	{
		/* compatibility with old code */
		if (sv->spl_left[sv->spl_nleft - 1] == InvalidOffsetNumber)
			sv->spl_left[sv->spl_nleft - 1] = (OffsetNumber) (entryvec->n - 1);
		if (sv->spl_right[sv->spl_nright - 1] == InvalidOffsetNumber)
			sv->spl_right[sv->spl_nright - 1] = (OffsetNumber) (entryvec->n - 1);

		if (sv->spl_ldatum_exists || sv->spl_rdatum_exists)
		{
			elog(LOG, "PickSplit method of %d columns of index '%s' doesn't support secondary split",
				 attno + 1, RelationGetRelationName(r));

			supportSecondarySplit(r, giststate, attno, sv, v->spl_lattr[attno], v->spl_rattr[attno]);
		}
	}

	v->spl_lattr[attno] = sv->spl_ldatum;
	v->spl_rattr[attno] = sv->spl_rdatum;
	v->spl_lisnull[attno] = false;
	v->spl_risnull[attno] = false;

	/*
	 * if index is multikey, then we must to try get smaller bounding box for
	 * subkey(s)
	 */
	v->spl_equiv = NULL;

	if (giststate->tupdesc->natts > 1 && attno + 1 != giststate->tupdesc->natts)
	{
		if (gistKeyIsEQ(giststate, attno, sv->spl_ldatum, sv->spl_rdatum))
		{
			/*
			 * Left and right key's unions are equial, so we can get better
			 * split by following columns. Note, unions for attno columns are
			 * already done.
			 */

			return true;
		}
		else
		{
			int			LenEquiv;

			v->spl_equiv = (bool *) palloc0(sizeof(bool) * (entryvec->n + 1));

			LenEquiv = gistfindgroup(r, giststate, entryvec->vector, v, attno);

			/*
			 * if possible, we should distribute equivalent tuples
			 */
			if (LenEquiv == 0)
			{
				gistunionsubkey(giststate, itup, v, attno + 1);
			}
			else
			{
				cleanupOffsets(sv->spl_left, &sv->spl_nleft, v->spl_equiv, &LenEquiv);
				cleanupOffsets(sv->spl_right, &sv->spl_nright, v->spl_equiv, &LenEquiv);

				gistunionsubkey(giststate, itup, v, attno + 1);
				if (LenEquiv == 1)
				{
					/*
					 * In case with one tuple we just choose left-right by
					 * penalty. It's simplify user-defined pickSplit
					 */
					OffsetNumber toMove = InvalidOffsetNumber;

					for (toMove = FirstOffsetNumber; toMove < entryvec->n; toMove++)
						if (v->spl_equiv[toMove])
							break;
					Assert(toMove < entryvec->n);

					placeOne(r, giststate, v, itup[toMove - 1], toMove, attno + 1);

					/*
					 * redo gistunionsubkey(): it will not degradate
					 * performance, because it's very rarely
					 */
					v->spl_equiv = NULL;
					gistunionsubkey(giststate, itup, v, attno + 1);

					return false;
				}
				else if (LenEquiv > 1)
					return true;
			}
		}
	}

	return false;
}