postgresql/src/include/executor/hashjoin.h

/*-------------------------------------------------------------------------
 *
 * hashjoin.h
 *	  internal structures for hash joins
 *
 *
 * Portions Copyright (c) 1996-2014, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 * src/include/executor/hashjoin.h
 *
 *-------------------------------------------------------------------------
 */
#ifndef HASHJOIN_H
#define HASHJOIN_H

#include "nodes/execnodes.h"
#include "storage/buffile.h"

/* ----------------------------------------------------------------
 *				hash-join hash table structures
 *
 * Each active hashjoin has a HashJoinTable control block, which is
 * palloc'd in the executor's per-query context.  All other storage needed
 * for the hashjoin is kept in private memory contexts, two for each hashjoin.
 * This makes it easy and fast to release the storage when we don't need it
 * anymore.  (Exception: data associated with the temp files lives in the
 * per-query context too, since we always call buffile.c in that context.)
 *
 * The hashtable contexts are made children of the per-query context, ensuring
 * that they will be discarded at end of statement even if the join is
 * aborted early by an error.  (Likewise, any temporary files we make will
 * be cleaned up by the virtual file manager in event of an error.)
 *
 * Storage that should live through the entire join is allocated from the
 * "hashCxt", while storage that is only wanted for the current batch is
 * allocated in the "batchCxt".  By resetting the batchCxt at the end of
 * each batch, we free all the per-batch storage reliably and without tedium.
 *
 * During first scan of inner relation, we get its tuples from executor.
 * If nbatch > 1 then tuples that don't belong in first batch get saved
 * into inner-batch temp files. The same statements apply for the
 * first scan of the outer relation, except we write tuples to outer-batch
 * temp files.  After finishing the first scan, we do the following for
 * each remaining batch:
 *	1. Read tuples from inner batch file, load into hash buckets.
 *	2. Read tuples from outer batch file, match to hash buckets and output.
 *
 * It is possible to increase nbatch on the fly if the in-memory hash table
 * gets too big.  The hash-value-to-batch computation is arranged so that this
 * can only cause a tuple to go into a later batch than previously thought,
 * never into an earlier batch.  When we increase nbatch, we rescan the hash
 * table and dump out any tuples that are now of a later batch to the correct
 * inner batch file.  Subsequently, while reading either inner or outer batch
 * files, we might find tuples that no longer belong to the current batch;
 * if so, we just dump them out to the correct batch file.
 * ----------------------------------------------------------------
 */

/* these are in nodes/execnodes.h: */
/* typedef struct HashJoinTupleData *HashJoinTuple; */
/* typedef struct HashJoinTableData *HashJoinTable; */

typedef struct HashJoinTupleData
{
	struct HashJoinTupleData *next;		/* link to next tuple in same bucket */
	uint32		hashvalue;		/* tuple's hash code */
	/* Tuple data, in MinimalTuple format, follows on a MAXALIGN boundary */
}	HashJoinTupleData;

#define HJTUPLE_OVERHEAD  MAXALIGN(sizeof(HashJoinTupleData))
#define HJTUPLE_MINTUPLE(hjtup)  \
	((MinimalTuple) ((char *) (hjtup) + HJTUPLE_OVERHEAD))

/*
 * If the outer relation's distribution is sufficiently nonuniform, we attempt
 * to optimize the join by treating the hash values corresponding to the outer
 * relation's MCVs specially.  Inner relation tuples matching these hash
 * values go into the "skew" hashtable instead of the main hashtable, and
 * outer relation tuples with these hash values are matched against that
 * table instead of the main one.  Thus, tuples with these hash values are
 * effectively handled as part of the first batch and will never go to disk.
 * The skew hashtable is limited to SKEW_WORK_MEM_PERCENT of the total memory
 * allowed for the join; while building the hashtables, we decrease the number
 * of MCVs being specially treated if needed to stay under this limit.
 *
 * Note: you might wonder why we look at the outer relation stats for this,
 * rather than the inner.  One reason is that the outer relation is typically
 * bigger, so we get more I/O savings by optimizing for its most common values.
 * Also, for similarly-sized relations, the planner prefers to put the more
 * uniformly distributed relation on the inside, so we're more likely to find
 * interesting skew in the outer relation.
 */
typedef struct HashSkewBucket
{
	uint32		hashvalue;		/* common hash value */
	HashJoinTuple tuples;		/* linked list of inner-relation tuples */
} HashSkewBucket;

#define SKEW_BUCKET_OVERHEAD  MAXALIGN(sizeof(HashSkewBucket))
#define INVALID_SKEW_BUCKET_NO	(-1)
#define SKEW_WORK_MEM_PERCENT  2
#define SKEW_MIN_OUTER_FRACTION  0.01


typedef struct HashJoinTableData
{
	int			nbuckets;		/* # buckets in the in-memory hash table */
	int			log2_nbuckets;	/* its log2 (nbuckets must be a power of 2) */

	/* buckets[i] is head of list of tuples in i'th in-memory bucket */
	struct HashJoinTupleData **buckets;
	/* buckets array is per-batch storage, as are all the tuples */

	bool		keepNulls;		/* true to store unmatchable NULL tuples */

	bool		skewEnabled;	/* are we using skew optimization? */
	HashSkewBucket **skewBucket;	/* hashtable of skew buckets */
	int			skewBucketLen;	/* size of skewBucket array (a power of 2!) */
	int			nSkewBuckets;	/* number of active skew buckets */
	int		   *skewBucketNums; /* array indexes of active skew buckets */

	int			nbatch;			/* number of batches */
	int			curbatch;		/* current batch #; 0 during 1st pass */

	int			nbatch_original;	/* nbatch when we started inner scan */
	int			nbatch_outstart;	/* nbatch when we started outer scan */

	bool		growEnabled;	/* flag to shut off nbatch increases */

	double		totalTuples;	/* # tuples obtained from inner plan */

	/*
	 * These arrays are allocated for the life of the hash join, but only if
	 * nbatch > 1.  A file is opened only when we first write a tuple into it
	 * (otherwise its pointer remains NULL).  Note that the zero'th array
	 * elements never get used, since we will process rather than dump out any
	 * tuples of batch zero.
	 */
	BufFile   **innerBatchFile; /* buffered virtual temp file per batch */
	BufFile   **outerBatchFile; /* buffered virtual temp file per batch */

	/*
	 * Info about the datatype-specific hash functions for the datatypes being
	 * hashed. These are arrays of the same length as the number of hash join
	 * clauses (hash keys).
	 */
	FmgrInfo   *outer_hashfunctions;	/* lookup data for hash functions */
	FmgrInfo   *inner_hashfunctions;	/* lookup data for hash functions */
	bool	   *hashStrict;		/* is each hash join operator strict? */

	Size		spaceUsed;		/* memory space currently used by tuples */
	Size		spaceAllowed;	/* upper limit for space used */
	Size		spacePeak;		/* peak space used */
	Size		spaceUsedSkew;	/* skew hash table's current space usage */
	Size		spaceAllowedSkew;		/* upper limit for skew hashtable */

	MemoryContext hashCxt;		/* context for whole-hash-join storage */
	MemoryContext batchCxt;		/* context for this-batch-only storage */
}	HashJoinTableData;

#endif   /* HASHJOIN_H */