From: | "Xiao Meng" <mx(dot)cogito(at)gmail(dot)com> |
---|---|
To: | pgsql-hackers(at)postgresql(dot)org |
Cc: | "Heikki Linnakangas" <heikki(at)enterprisedb(dot)com>, "Jonah H(dot) Harris" <jonah(dot)harris(at)gmail(dot)com>, "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us>, "Kenneth Marshall" <ktm(at)rice(dot)edu> |
Subject: | gsoc, store hash index tuple with hash code only |
Date: | 2008-07-16 00:23:31 |
Message-ID: | ded849dd0807151723y55cfb0ddrf6d5c300107668b9@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Hi, hackers.
I'm working on my gsoc project of improving hash index.
I posted a thread to ask some questions about implementation a few hours ago.
http://archives.postgresql.org/pgsql-hackers/2008-07/msg00721.php
Someone's advice reminds me of the problem of design decision again.
I store hash index tuple as a regular index tuple currently. An
alternative design is to totally change the layout of hash index tuple
(using a different structure).
Here is my design.
We convert hash code into a datum using UINT32GetDatum() and store it
in a regular index tuple.
When we check the tuple, we get the hashkey back using DatumGetUINT32.
We also need a TupleDesc with uint32 attribute. I don't know how to
create it. Alternatively, I create a TupleDesc with int32 attribute.
The hash item should be in order to support binary search later. I
haven't implement it because I'd like to finish the first available
one ASAP.
I post a patch below. The patch didn't passed regression test yet. I
just posted it here to show my design.
Any comments are welcome.
diff --git a/src/backend/access/hash/hash.c b/src/backend/access/hash/hash.c
index 6a5c000..fbbd7dc 100644
--- a/src/backend/access/hash/hash.c
+++ b/src/backend/access/hash/hash.c
@@ -129,7 +129,11 @@ hashbuildCallback(Relation index,
IndexTuple itup;
/* form an index tuple and point it at the heap tuple */
+#ifdef HASHVALUE_ONLY
+ itup = _hash_form_tuple(index, values,isnull);
+#else
itup = index_form_tuple(RelationGetDescr(index), values, isnull);
+#endif
itup->t_tid = htup->t_self;
/* Hash indexes don't index nulls, see notes in hashinsert */
@@ -171,7 +175,12 @@ hashinsert(PG_FUNCTION_ARGS)
IndexTuple itup;
/* generate an index tuple */
+#ifdef HASHVALUE_ONLY
+ itup = _hash_form_tuple(rel,values,isnull);
+#else
itup = index_form_tuple(RelationGetDescr(rel), values, isnull);
+#endif
+
itup->t_tid = *ht_ctid;
/*
@@ -212,7 +221,11 @@ hashgettuple(PG_FUNCTION_ARGS)
bool res;
/* Hash indexes are never lossy (at the moment anyway) */
- scan->xs_recheck = false;
+#ifdef HASHVALUE_ONLY
+ scan->xs_recheck = true;
+#else
+ scan->xs_recheck = false;
+#endif
/*
* We hold pin but not lock on current buffer while outside the hash AM.
diff --git a/src/backend/access/hash/hashpage.c
b/src/backend/access/hash/hashpage.c
index b0b5874..068337e 100644
--- a/src/backend/access/hash/hashpage.c
+++ b/src/backend/access/hash/hashpage.c
@@ -785,7 +785,7 @@ _hash_splitbucket(Relation rel,
OffsetNumber omaxoffnum;
Page opage;
Page npage;
- TupleDesc itupdesc = RelationGetDescr(rel);
+ TupleDesc itupdesc = _create_hash_desc();
/*
* It should be okay to simultaneously write-lock pages from each bucket,
@@ -854,9 +854,13 @@ _hash_splitbucket(Relation rel,
itup = (IndexTuple) PageGetItem(opage, PageGetItemId(opage, ooffnum));
datum = index_getattr(itup, 1, itupdesc, &null);
Assert(!null);
-
+#ifdef HASHVALUE_ONLY
+ bucket = _hash_hashkey2bucket(DatumGetUInt32(datum),
+ maxbucket, highmask, lowmask);
+#else
bucket = _hash_hashkey2bucket(_hash_datum2hashkey(rel, datum),
maxbucket, highmask, lowmask);
+#endif
if (bucket == nbucket)
{
diff --git a/src/backend/access/hash/hashsearch.c
b/src/backend/access/hash/hashsearch.c
index 258526b..5e7668d 100644
--- a/src/backend/access/hash/hashsearch.c
+++ b/src/backend/access/hash/hashsearch.c
@@ -177,7 +177,7 @@ _hash_first(IndexScanDesc scan, ScanDirection dir)
else
hashkey = _hash_datum2hashkey_type(rel, cur->sk_argument,
cur->sk_subtype);
-
+ so->hashso_sk_hash = hashkey;
/*
* Acquire shared split lock so we can compute the target bucket safely
* (see README).
diff --git a/src/backend/access/hash/hashutil.c
b/src/backend/access/hash/hashutil.c
index 41e2eef..4be814e 100644
--- a/src/backend/access/hash/hashutil.c
+++ b/src/backend/access/hash/hashutil.c
@@ -20,6 +20,8 @@
#include "executor/execdebug.h"
#include "storage/bufmgr.h"
#include "utils/lsyscache.h"
+#include "utils/typcache.h"
+#include "catalog/pg_type.h"
/*
@@ -28,16 +30,29 @@
bool
_hash_checkqual(IndexScanDesc scan, IndexTuple itup)
{
- TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
+ TupleDesc tupdesc = _create_hash_desc();
ScanKey key = scan->keyData;
int scanKeySize = scan->numberOfKeys;
+ Datum datum;
+ bool isNull;
+ HashScanOpaque so = scan->opaque;
IncrIndexProcessed();
+#ifdef HASHVALUE_ONLY
+ datum = index_getattr(itup,
+ key->sk_attno,
+ tupdesc,
+ &isNull);
+ if( so->hashso_sk_hash != DatumGetInt32(datum) )
+ return false;
+
+ key++;
+ scanKeySize--;
+#endif
+
while (scanKeySize > 0)
{
- Datum datum;
- bool isNull;
Datum test;
datum = index_getattr(itup,
@@ -50,7 +65,7 @@ _hash_checkqual(IndexScanDesc scan, IndexTuple itup)
return false;
if (key->sk_flags & SK_ISNULL)
return false;
-
+
test = FunctionCall2(&key->sk_func, datum, key->sk_argument);
if (!DatumGetBool(test))
@@ -222,3 +237,31 @@ hashoptions(PG_FUNCTION_ARGS)
PG_RETURN_BYTEA_P(result);
PG_RETURN_NULL();
}
+
+/*
+ * _create_hash_desc - create a hash TupleDesc
+ *
+ * the TupleDesc is with int32 attribute, not uint32 because we've no
pg_type with UINT4OID
+ */
+TupleDesc _create_hash_desc()
+{
+ TupleDesc tupdesc = CreateTemplateTupleDesc(1, false);
+ TupleDescInitEntry(tupdesc, 1, "hashcode", INT4OID, -1, 0);
+ return tupdesc;
+}
+
+/*
+ * _hash_form_tuple - form a tuple with hash code only
+ */
+IndexTuple _hash_form_tuple(Relation rel, Datum* values, bool* isnull)
+{
+ TupleDesc hashdesc;
+ IndexTuple itup;
+ uint32 hashkey;
+
+ hashdesc = _create_hash_desc();
+ hashkey = _hash_datum2hashkey(index, values[0]);
+ values[0] = Int32GetDatum(hashkey);
+ itup = index_form_tuple(hashdesc, values, isnull);
+ return itup;
+}
diff --git a/src/include/access/hash.h b/src/include/access/hash.h
index ab0824d..0cf5ad8 100644
--- a/src/include/access/hash.h
+++ b/src/include/access/hash.h
@@ -100,6 +100,8 @@ typedef struct HashScanOpaqueData
/* Current and marked position of the scan */
ItemPointerData hashso_curpos;
ItemPointerData hashso_mrkpos;
+ /* Hash value of the scan key */
+ int32 hashso_sk_hash;
} HashScanOpaqueData;
typedef HashScanOpaqueData *HashScanOpaque;
@@ -227,6 +229,10 @@ typedef HashMetaPageData *HashMetaPage;
*/
#define HASHPROC 1
+/*
+ * store hash value only in the bucket
+ */
+#define HASHVALUE_ONLY
/* public routines */
@@ -330,6 +336,8 @@ extern Bucket _hash_hashkey2bucket(uint32 hashkey,
uint32 maxbucket,
uint32 highmask, uint32 lowmask);
extern uint32 _hash_log2(uint32 num);
extern void _hash_checkpage(Relation rel, Buffer buf, int flags);
+extern TupleDesc _create_hash_desc();
+extern IndexTuple _hash_form_tuple(Relation rel, Datum* values, bool* isnull);
/* hash.c */
extern void hash_redo(XLogRecPtr lsn, XLogRecord *record);
--
Best Regards,
Xiao Meng
DKERC, Harbin Institute of Technology, China
Gtalk: mx(dot)cogito(at)gmail(dot)com
MSN: cnEnder(at)live(dot)com
http://xiaomeng.yo2.cn
Attachment | Content-Type | Size |
---|---|---|
hash.diff | text/x-diff | 5.9 KB |
From | Date | Subject | |
---|---|---|---|
Next Message | Tatsuo Ishii | 2008-07-16 00:37:25 | Re: [PATCHES] WITH RECURSIVE updated to CVS TIP |
Previous Message | Tom Lane | 2008-07-15 23:15:14 | Re: Lookup penalty for VARIADIC patch |