nodehash.c

PostgreSQL7.4.6 for Linux

/*-------------------------------------------------------------------------
 *
 * nodeHash.c
 *	  Routines to hash relations for hashjoin
 *
 * Portions Copyright (c) 1996-2003, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 *
 * IDENTIFICATION
 *	  $Header: /cvsroot/pgsql/src/backend/executor/nodeHash.c,v 1.79 2003/08/04 02:39:59 momjian Exp $
 *
 *-------------------------------------------------------------------------
 */
/*
 * INTERFACE ROUTINES
 *		ExecHash		- generate an in-memory hash table of the relation
 *		ExecInitHash	- initialize node and subnodes
 *		ExecEndHash		- shutdown node and subnodes
 */
#include "postgres.h"

#include "executor/execdebug.h"
#include "executor/nodeHash.h"
#include "executor/nodeHashjoin.h"
#include "miscadmin.h"
#include "parser/parse_expr.h"
#include "utils/memutils.h"
#include "utils/lsyscache.h"


/* ----------------------------------------------------------------
 *		ExecHash
 *
 *		build hash table for hashjoin, all do partitioning if more
 *		than one batches are required.
 * ----------------------------------------------------------------
 */
TupleTableSlot *
ExecHash(HashState *node)
{
	EState	   *estate;
	PlanState  *outerNode;
	List	   *hashkeys;
	HashJoinTable hashtable;
	TupleTableSlot *slot;
	ExprContext *econtext;
	int			nbatch;
	int			i;

	/*
	 * get state info from node
	 */
	estate = node->ps.state;
	outerNode = outerPlanState(node);

	hashtable = node->hashtable;
	nbatch = hashtable->nbatch;

	if (nbatch > 0)
	{
		/*
		 * Open temp files for inner batches, if needed. Note that file
		 * buffers are palloc'd in regular executor context.
		 */
		for (i = 0; i < nbatch; i++)
			hashtable->innerBatchFile[i] = BufFileCreateTemp(false);
	}

	/*
	 * set expression context
	 */
	hashkeys = node->hashkeys;
	econtext = node->ps.ps_ExprContext;

	/*
	 * get all inner tuples and insert into the hash table (or temp files)
	 */
	for (;;)
	{
		slot = ExecProcNode(outerNode);
		if (TupIsNull(slot))
			break;
		econtext->ecxt_innertuple = slot;
		ExecHashTableInsert(hashtable, econtext, hashkeys);
		ExecClearTuple(slot);
	}

	/*
	 * Return the slot so that we have the tuple descriptor when we need
	 * to save/restore them.  -Jeff 11 July 1991
	 */
	return slot;
}

/* ----------------------------------------------------------------
 *		ExecInitHash
 *
 *		Init routine for Hash node
 * ----------------------------------------------------------------
 */
HashState *
ExecInitHash(Hash *node, EState *estate)
{
	HashState  *hashstate;

	SO_printf("ExecInitHash: initializing hash node\n");

	/*
	 * create state structure
	 */
	hashstate = makeNode(HashState);
	hashstate->ps.plan = (Plan *) node;
	hashstate->ps.state = estate;
	hashstate->hashtable = NULL;

	/*
	 * Miscellaneous initialization
	 *
	 * create expression context for node
	 */
	ExecAssignExprContext(estate, &hashstate->ps);

#define HASH_NSLOTS 1

	/*
	 * initialize our result slot
	 */
	ExecInitResultTupleSlot(estate, &hashstate->ps);

	/*
	 * initialize child expressions
	 */
	hashstate->ps.targetlist = (List *)
		ExecInitExpr((Expr *) node->plan.targetlist,
					 (PlanState *) hashstate);
	hashstate->ps.qual = (List *)
		ExecInitExpr((Expr *) node->plan.qual,
					 (PlanState *) hashstate);

	/*
	 * initialize child nodes
	 */
	outerPlanState(hashstate) = ExecInitNode(outerPlan(node), estate);

	/*
	 * initialize tuple type. no need to initialize projection info
	 * because this node doesn't do projections
	 */
	ExecAssignResultTypeFromOuterPlan(&hashstate->ps);
	hashstate->ps.ps_ProjInfo = NULL;

	return hashstate;
}

int
ExecCountSlotsHash(Hash *node)
{
	return ExecCountSlotsNode(outerPlan(node)) +
		ExecCountSlotsNode(innerPlan(node)) +
		HASH_NSLOTS;
}

/* ---------------------------------------------------------------
 *		ExecEndHash
 *
 *		clean up routine for Hash node
 * ----------------------------------------------------------------
 */
void
ExecEndHash(HashState *node)
{
	PlanState  *outerPlan;

	/*
	 * free exprcontext
	 */
	ExecFreeExprContext(&node->ps);

	/*
	 * shut down the subplan
	 */
	outerPlan = outerPlanState(node);
	ExecEndNode(outerPlan);
}


/* ----------------------------------------------------------------
 *		ExecHashTableCreate
 *
 *		create a hashtable in shared memory for hashjoin.
 * ----------------------------------------------------------------
 */
HashJoinTable
ExecHashTableCreate(Hash *node, List *hashOperators)
{
	HashJoinTable hashtable;
	Plan	   *outerNode;
	int			totalbuckets;
	int			nbuckets;
	int			nbatch;
	int			nkeys;
	int			i;
	List	   *ho;
	MemoryContext oldcxt;

	/*
	 * Get information about the size of the relation to be hashed (it's
	 * the "outer" subtree of this node, but the inner relation of the
	 * hashjoin).  Compute the appropriate size of the hash table.
	 */
	outerNode = outerPlan(node);

	ExecChooseHashTableSize(outerNode->plan_rows, outerNode->plan_width,
							&totalbuckets, &nbuckets, &nbatch);

#ifdef HJDEBUG
	printf("nbatch = %d, totalbuckets = %d, nbuckets = %d\n",
		   nbatch, totalbuckets, nbuckets);
#endif

	/*
	 * Initialize the hash table control block.
	 *
	 * The hashtable control block is just palloc'd from the executor's
	 * per-query memory context.
	 */
	hashtable = (HashJoinTable) palloc(sizeof(HashTableData));
	hashtable->nbuckets = nbuckets;
	hashtable->totalbuckets = totalbuckets;
	hashtable->buckets = NULL;
	hashtable->nbatch = nbatch;
	hashtable->curbatch = 0;
	hashtable->innerBatchFile = NULL;
	hashtable->outerBatchFile = NULL;
	hashtable->innerBatchSize = NULL;
	hashtable->outerBatchSize = NULL;

	/*
	 * Get info about the hash functions to be used for each hash key.
	 */
	nkeys = length(hashOperators);
	hashtable->hashfunctions = (FmgrInfo *) palloc(nkeys * sizeof(FmgrInfo));
	i = 0;
	foreach(ho, hashOperators)
	{
		Oid			hashfn;

		hashfn = get_op_hash_function(lfirsto(ho));
		if (!OidIsValid(hashfn))
			elog(ERROR, "could not find hash function for hash operator %u",
				 lfirsto(ho));
		fmgr_info(hashfn, &hashtable->hashfunctions[i]);
		i++;
	}

	/*
	 * Create temporary memory contexts in which to keep the hashtable
	 * working storage.  See notes in executor/hashjoin.h.
	 */
	hashtable->hashCxt = AllocSetContextCreate(CurrentMemoryContext,
											   "HashTableContext",
											   ALLOCSET_DEFAULT_MINSIZE,
											   ALLOCSET_DEFAULT_INITSIZE,
											   ALLOCSET_DEFAULT_MAXSIZE);

	hashtable->batchCxt = AllocSetContextCreate(hashtable->hashCxt,
												"HashBatchContext",
												ALLOCSET_DEFAULT_MINSIZE,
												ALLOCSET_DEFAULT_INITSIZE,
												ALLOCSET_DEFAULT_MAXSIZE);

	/* Allocate data that will live for the life of the hashjoin */

	oldcxt = MemoryContextSwitchTo(hashtable->hashCxt);

	if (nbatch > 0)
	{
		/*
		 * allocate and initialize the file arrays in hashCxt
		 */
		hashtable->innerBatchFile = (BufFile **)
			palloc0(nbatch * sizeof(BufFile *));
		hashtable->outerBatchFile = (BufFile **)
			palloc0(nbatch * sizeof(BufFile *));
		hashtable->innerBatchSize = (long *)
			palloc0(nbatch * sizeof(long));
		hashtable->outerBatchSize = (long *)
			palloc0(nbatch * sizeof(long));
		/* The files will not be opened until later... */
	}

	/*
	 * Prepare context for the first-scan space allocations; allocate the
	 * hashbucket array therein, and set each bucket "empty".
	 */
	MemoryContextSwitchTo(hashtable->batchCxt);

	hashtable->buckets = (HashJoinTuple *)
		palloc0(nbuckets * sizeof(HashJoinTuple));

	MemoryContextSwitchTo(oldcxt);

	return hashtable;
}


/*
 * Compute appropriate size for hashtable given the estimated size of the
 * relation to be hashed (number of rows and average row width).
 *
 * Caution: the input is only the planner's estimates, and so can't be
 * trusted too far.  Apply a healthy fudge factor.
 *
 * This is exported so that the planner's costsize.c can use it.
 */

/* Target bucket loading (tuples per bucket) */
#define NTUP_PER_BUCKET			10
/* Fudge factor to allow for inaccuracy of input estimates */
#define FUDGE_FAC				2.0

void
ExecChooseHashTableSize(double ntuples, int tupwidth,
						int *virtualbuckets,
						int *physicalbuckets,
						int *numbatches)
{
	int			tupsize;
	double		inner_rel_bytes;
	long		hash_table_bytes;
	double		dtmp;
	int			nbatch;
	int			nbuckets;
	int			totalbuckets;
	int			bucketsize;

	/* Force a plausible relation size if no info */
	if (ntuples <= 0.0)
		ntuples = 1000.0;

	/*
	 * Estimate tupsize based on footprint of tuple in hashtable... but
	 * what about palloc overhead?
	 */
	tupsize = MAXALIGN(tupwidth) + MAXALIGN(sizeof(HashJoinTupleData));
	inner_rel_bytes = ntuples * tupsize * FUDGE_FAC;

	/*