nodehashjoin.c

PostgreSQL 8.1.4的源码 适用于Linux下的开源数据库系统

	hjstate->hj_CurTuple = NULL;

	/*
	 * Deconstruct the hash clauses into outer and inner argument values, so
	 * that we can evaluate those subexpressions separately.  Also make a list
	 * of the hash operator OIDs, in preparation for looking up the hash
	 * functions to use.
	 */
	lclauses = NIL;
	rclauses = NIL;
	hoperators = NIL;
	foreach(l, hjstate->hashclauses)
	{
		FuncExprState *fstate = (FuncExprState *) lfirst(l);
		OpExpr	   *hclause;

		Assert(IsA(fstate, FuncExprState));
		hclause = (OpExpr *) fstate->xprstate.expr;
		Assert(IsA(hclause, OpExpr));
		lclauses = lappend(lclauses, linitial(fstate->args));
		rclauses = lappend(rclauses, lsecond(fstate->args));
		hoperators = lappend_oid(hoperators, hclause->opno);
	}
	hjstate->hj_OuterHashKeys = lclauses;
	hjstate->hj_InnerHashKeys = rclauses;
	hjstate->hj_HashOperators = hoperators;
	/* child Hash node needs to evaluate inner hash keys, too */
	((HashState *) innerPlanState(hjstate))->hashkeys = rclauses;

	hjstate->js.ps.ps_OuterTupleSlot = NULL;
	hjstate->js.ps.ps_TupFromTlist = false;
	hjstate->hj_NeedNewOuter = true;
	hjstate->hj_MatchedOuter = false;
	hjstate->hj_OuterNotEmpty = false;

	return hjstate;
}

int
ExecCountSlotsHashJoin(HashJoin *node)
{
	return ExecCountSlotsNode(outerPlan(node)) +
		ExecCountSlotsNode(innerPlan(node)) +
		HASHJOIN_NSLOTS;
}

/* ----------------------------------------------------------------
 *		ExecEndHashJoin
 *
 *		clean up routine for HashJoin node
 * ----------------------------------------------------------------
 */
void
ExecEndHashJoin(HashJoinState *node)
{
	/*
	 * Free hash table
	 */
	if (node->hj_HashTable)
	{
		ExecHashTableDestroy(node->hj_HashTable);
		node->hj_HashTable = NULL;
	}

	/*
	 * Free the exprcontext
	 */
	ExecFreeExprContext(&node->js.ps);

	/*
	 * clean out the tuple table
	 */
	ExecClearTuple(node->js.ps.ps_ResultTupleSlot);
	ExecClearTuple(node->hj_OuterTupleSlot);
	ExecClearTuple(node->hj_HashTupleSlot);

	/*
	 * clean up subtrees
	 */
	ExecEndNode(outerPlanState(node));
	ExecEndNode(innerPlanState(node));
}

/*
 * ExecHashJoinOuterGetTuple
 *
 *		get the next outer tuple for hashjoin: either by
 *		executing a plan node in the first pass, or from
 *		the temp files for the hashjoin batches.
 *
 * Returns a null slot if no more outer tuples.  On success, the tuple's
 * hash value is stored at *hashvalue --- this is either originally computed,
 * or re-read from the temp file.
 */
static TupleTableSlot *
ExecHashJoinOuterGetTuple(PlanState *outerNode,
						  HashJoinState *hjstate,
						  uint32 *hashvalue)
{
	HashJoinTable hashtable = hjstate->hj_HashTable;
	int			curbatch = hashtable->curbatch;
	TupleTableSlot *slot;

	if (curbatch == 0)
	{							/* if it is the first pass */

		/*
		 * Check to see if first outer tuple was already fetched by
		 * ExecHashJoin() and not used yet.
		 */
		slot = hjstate->hj_FirstOuterTupleSlot;
		if (!TupIsNull(slot))
			hjstate->hj_FirstOuterTupleSlot = NULL;
		else
			slot = ExecProcNode(outerNode);
		if (!TupIsNull(slot))
		{
			/*
			 * We have to compute the tuple's hash value.
			 */
			ExprContext *econtext = hjstate->js.ps.ps_ExprContext;

			econtext->ecxt_outertuple = slot;
			*hashvalue = ExecHashGetHashValue(hashtable, econtext,
											  hjstate->hj_OuterHashKeys);

			/* remember outer relation is not empty for possible rescan */
			hjstate->hj_OuterNotEmpty = true;

			return slot;
		}

		/*
		 * We have just reached the end of the first pass. Try to switch to a
		 * saved batch.
		 */
		curbatch = ExecHashJoinNewBatch(hjstate);
	}

	/*
	 * Try to read from a temp file. Loop allows us to advance to new batches
	 * as needed.  NOTE: nbatch could increase inside ExecHashJoinNewBatch, so
	 * don't try to optimize this loop.
	 */
	while (curbatch < hashtable->nbatch)
	{
		slot = ExecHashJoinGetSavedTuple(hjstate,
										 hashtable->outerBatchFile[curbatch],
										 hashvalue,
										 hjstate->hj_OuterTupleSlot);
		if (!TupIsNull(slot))
			return slot;
		curbatch = ExecHashJoinNewBatch(hjstate);
	}

	/* Out of batches... */
	return NULL;
}

/*
 * ExecHashJoinNewBatch
 *		switch to a new hashjoin batch
 *
 * Returns the number of the new batch (1..nbatch-1), or nbatch if no more.
 * We will never return a batch number that has an empty outer batch file.
 */
static int
ExecHashJoinNewBatch(HashJoinState *hjstate)
{
	HashJoinTable hashtable = hjstate->hj_HashTable;
	int			nbatch;
	int			curbatch;
	BufFile    *innerFile;
	TupleTableSlot *slot;
	uint32		hashvalue;

start_over:
	nbatch = hashtable->nbatch;
	curbatch = hashtable->curbatch;

	if (curbatch > 0)
	{
		/*
		 * We no longer need the previous outer batch file; close it right
		 * away to free disk space.
		 */
		if (hashtable->outerBatchFile[curbatch])
			BufFileClose(hashtable->outerBatchFile[curbatch]);
		hashtable->outerBatchFile[curbatch] = NULL;
	}

	/*
	 * We can always skip over any batches that are completely empty on both
	 * sides.  We can sometimes skip over batches that are empty on only one
	 * side, but there are exceptions:
	 *
	 * 1. In a LEFT JOIN, we have to process outer batches even if the inner
	 * batch is empty.
	 *
	 * 2. If we have increased nbatch since the initial estimate, we have to
	 * scan inner batches since they might contain tuples that need to be
	 * reassigned to later inner batches.
	 *
	 * 3. Similarly, if we have increased nbatch since starting the outer
	 * scan, we have to rescan outer batches in case they contain tuples that
	 * need to be reassigned.
	 */
	curbatch++;
	while (curbatch < nbatch &&
		   (hashtable->outerBatchFile[curbatch] == NULL ||
			hashtable->innerBatchFile[curbatch] == NULL))
	{
		if (hashtable->outerBatchFile[curbatch] &&
			hjstate->js.jointype == JOIN_LEFT)
			break;				/* must process due to rule 1 */
		if (hashtable->innerBatchFile[curbatch] &&
			nbatch != hashtable->nbatch_original)
			break;				/* must process due to rule 2 */
		if (hashtable->outerBatchFile[curbatch] &&
			nbatch != hashtable->nbatch_outstart)
			break;				/* must process due to rule 3 */
		/* We can ignore this batch. */
		/* Release associated temp files right away. */
		if (hashtable->innerBatchFile[curbatch])
			BufFileClose(hashtable->innerBatchFile[curbatch]);
		hashtable->innerBatchFile[curbatch] = NULL;
		if (hashtable->outerBatchFile[curbatch])
			BufFileClose(hashtable->outerBatchFile[curbatch]);
		hashtable->outerBatchFile[curbatch] = NULL;
		curbatch++;
	}

	if (curbatch >= nbatch)
		return curbatch;		/* no more batches */

	hashtable->curbatch = curbatch;

	/*
	 * Reload the hash table with the new inner batch (which could be empty)
	 */
	ExecHashTableReset(hashtable);

	innerFile = hashtable->innerBatchFile[curbatch];

	if (innerFile != NULL)
	{
		if (BufFileSeek(innerFile, 0, 0L, SEEK_SET))
			ereport(ERROR,
					(errcode_for_file_access(),
				   errmsg("could not rewind hash-join temporary file: %m")));

		while ((slot = ExecHashJoinGetSavedTuple(hjstate,
												 innerFile,
												 &hashvalue,
												 hjstate->hj_HashTupleSlot)))
		{
			/*
			 * NOTE: some tuples may be sent to future batches.  Also, it is
			 * possible for hashtable->nbatch to be increased here!
			 */
			ExecHashTableInsert(hashtable,
								ExecFetchSlotTuple(slot),
								hashvalue);
		}

		/*
		 * after we build the hash table, the inner batch file is no longer
		 * needed
		 */
		BufFileClose(innerFile);
		hashtable->innerBatchFile[curbatch] = NULL;
	}

	/*
	 * If there's no outer batch file, advance to next batch.
	 */
	if (hashtable->outerBatchFile[curbatch] == NULL)
		goto start_over;

	/*
	 * Rewind outer batch file, so that we can start reading it.
	 */
	if (BufFileSeek(hashtable->outerBatchFile[curbatch], 0, 0L, SEEK_SET))
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not rewind hash-join temporary file: %m")));

	return curbatch;
}

/*
 * ExecHashJoinSaveTuple
 *		save a tuple to a batch file.
 *
 * The data recorded in the file for each tuple is its hash value,
 * then an image of its HeapTupleData (with meaningless t_data pointer)
 * followed by the HeapTupleHeader and tuple data.
 *
 * Note: it is important always to call this in the regular executor
 * context, not in a shorter-lived context; else the temp file buffers
 * will get messed up.
 */
void
ExecHashJoinSaveTuple(HeapTuple heapTuple, uint32 hashvalue,
					  BufFile **fileptr)
{
	BufFile    *file = *fileptr;
	size_t		written;

	if (file == NULL)
	{
		/* First write to this batch file, so open it. */
		file = BufFileCreateTemp(false);
		*fileptr = file;
	}

	written = BufFileWrite(file, (void *) &hashvalue, sizeof(uint32));
	if (written != sizeof(uint32))
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not write to hash-join temporary file: %m")));

	written = BufFileWrite(file, (void *) heapTuple, sizeof(HeapTupleData));
	if (written != sizeof(HeapTupleData))
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not write to hash-join temporary file: %m")));

	written = BufFileWrite(file, (void *) heapTuple->t_data, heapTuple->t_len);
	if (written != (size_t) heapTuple->t_len)
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not write to hash-join temporary file: %m")));
}

/*
 * ExecHashJoinGetSavedTuple
 *		read the next tuple from a batch file.	Return NULL if no more.
 *
 * On success, *hashvalue is set to the tuple's hash value, and the tuple
 * itself is stored in the given slot.
 */
static TupleTableSlot *
ExecHashJoinGetSavedTuple(HashJoinState *hjstate,
						  BufFile *file,
						  uint32 *hashvalue,
						  TupleTableSlot *tupleSlot)
{
	HeapTupleData htup;
	size_t		nread;
	HeapTuple	heapTuple;

	nread = BufFileRead(file, (void *) hashvalue, sizeof(uint32));
	if (nread == 0)
		return NULL;			/* end of file */
	if (nread != sizeof(uint32))
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not read from hash-join temporary file: %m")));
	nread = BufFileRead(file, (void *) &htup, sizeof(HeapTupleData));
	if (nread != sizeof(HeapTupleData))
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not read from hash-join temporary file: %m")));
	heapTuple = palloc(HEAPTUPLESIZE + htup.t_len);
	memcpy((char *) heapTuple, (char *) &htup, sizeof(HeapTupleData));
	heapTuple->t_datamcxt = CurrentMemoryContext;
	heapTuple->t_data = (HeapTupleHeader)
		((char *) heapTuple + HEAPTUPLESIZE);
	nread = BufFileRead(file, (void *) heapTuple->t_data, htup.t_len);
	if (nread != (size_t) htup.t_len)
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not read from hash-join temporary file: %m")));
	return ExecStoreTuple(heapTuple, tupleSlot, InvalidBuffer, true);
}


void
ExecReScanHashJoin(HashJoinState *node, ExprContext *exprCtxt)
{
	/*
	 * In a multi-batch join, we currently have to do rescans the hard way,
	 * primarily because batch temp files may have already been released. But
	 * if it's a single-batch join, and there is no parameter change for the
	 * inner subnode, then we can just re-use the existing hash table without
	 * rebuilding it.
	 */
	if (node->hj_HashTable != NULL)
	{
		if (node->hj_HashTable->nbatch == 1 &&
			((PlanState *) node)->righttree->chgParam == NULL)
		{
			/*
			 * okay to reuse the hash table; needn't rescan inner, either.
			 *
			 * What we do need to do is reset our state about the emptiness
			 * of the outer relation, so that the new scan of the outer will
			 * update it correctly if it turns out to be empty this time.
			 * (There's no harm in clearing it now because ExecHashJoin won't
			 * need the info.  In the other cases, where the hash table
			 * doesn't exist or we are destroying it, we leave this state
			 * alone because ExecHashJoin will need it the first time
			 * through.)
			 */
			node->hj_OuterNotEmpty = false;
		}
		else
		{
			/* must destroy and rebuild hash table */
			ExecHashTableDestroy(node->hj_HashTable);
			node->hj_HashTable = NULL;

			/*
			 * if chgParam of subnode is not null then plan will be re-scanned
			 * by first ExecProcNode.
			 */
			if (((PlanState *) node)->righttree->chgParam == NULL)
				ExecReScan(((PlanState *) node)->righttree, exprCtxt);
		}
	}

	/* Always reset intra-tuple state */
	node->hj_CurHashValue = 0;
	node->hj_CurBucketNo = 0;
	node->hj_CurTuple = NULL;

	node->js.ps.ps_OuterTupleSlot = NULL;
	node->js.ps.ps_TupFromTlist = false;
	node->hj_NeedNewOuter = true;
	node->hj_MatchedOuter = false;
	node->hj_FirstOuterTupleSlot = NULL;

	/*
	 * if chgParam of subnode is not null then plan will be re-scanned by
	 * first ExecProcNode.
	 */
	if (((PlanState *) node)->lefttree->chgParam == NULL)
		ExecReScan(((PlanState *) node)->lefttree, exprCtxt);
}