Skip to content

Commit ca7c816

Browse files
committed
Tweak TOAST code so that columns marked with MAIN storage strategy are
not forced out-of-line unless that is necessary to make the row fit on a page. Previously, they were forced out-of-line if needed to get the row down to the default target size (1/4th page). Kevin Grittner
1 parent a5375bf commit ca7c816

File tree

3 files changed

+33
-18
lines changed

3 files changed

+33
-18
lines changed

doc/src/sgml/storage.sgml

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
<!-- $PostgreSQL: pgsql/doc/src/sgml/storage.sgml,v 1.29 2009/06/17 21:58:49 tgl Exp $ -->
1+
<!-- $PostgreSQL: pgsql/doc/src/sgml/storage.sgml,v 1.30 2009/07/22 01:21:22 tgl Exp $ -->
22

33
<chapter id="storage">
44

@@ -347,7 +347,7 @@ The <acronym>TOAST</> code recognizes four different strategies for storing
347347
<literal>MAIN</literal> allows compression but not out-of-line
348348
storage. (Actually, out-of-line storage will still be performed
349349
for such columns, but only as a last resort when there is no other
350-
way to make the row small enough.)
350+
way to make the row small enough to fit on a page.)
351351
</para>
352352
</listitem>
353353
</itemizedlist>

src/backend/access/heap/tuptoaster.c

Lines changed: 6 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
*
99
*
1010
* IDENTIFICATION
11-
* $PostgreSQL: pgsql/src/backend/access/heap/tuptoaster.c,v 1.93 2009/06/11 14:48:54 momjian Exp $
11+
* $PostgreSQL: pgsql/src/backend/access/heap/tuptoaster.c,v 1.94 2009/07/22 01:21:22 tgl Exp $
1212
*
1313
*
1414
* INTERFACE ROUTINES
@@ -796,8 +796,12 @@ toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup,
796796
}
797797

798798
/*
799-
* Finally we store attributes of type 'm' external, if possible.
799+
* Finally we store attributes of type 'm' externally. At this point
800+
* we increase the target tuple size, so that 'm' attributes aren't
801+
* stored externally unless really necessary.
800802
*/
803+
maxDataLen = TOAST_TUPLE_TARGET_MAIN - hoff;
804+
801805
while (heap_compute_data_size(tupleDesc,
802806
toast_values, toast_isnull) > maxDataLen &&
803807
rel->rd_rel->reltoastrelid != InvalidOid)

src/include/access/tuptoaster.h

Lines changed: 25 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -6,7 +6,7 @@
66
*
77
* Copyright (c) 2000-2009, PostgreSQL Global Development Group
88
*
9-
* $PostgreSQL: pgsql/src/include/access/tuptoaster.h,v 1.43 2009/01/01 17:23:56 momjian Exp $
9+
* $PostgreSQL: pgsql/src/include/access/tuptoaster.h,v 1.44 2009/07/22 01:21:22 tgl Exp $
1010
*
1111
*-------------------------------------------------------------------------
1212
*/
@@ -24,20 +24,26 @@
2424
#define TOAST_INDEX_HACK
2525

2626

27+
/*
28+
* Find the maximum size of a tuple if there are to be N tuples per page.
29+
*/
30+
#define MaximumBytesPerTuple(tuplesPerPage) \
31+
MAXALIGN_DOWN((BLCKSZ - \
32+
MAXALIGN(SizeOfPageHeaderData + (tuplesPerPage) * sizeof(ItemIdData))) \
33+
/ (tuplesPerPage))
34+
2735
/*
2836
* These symbols control toaster activation. If a tuple is larger than
2937
* TOAST_TUPLE_THRESHOLD, we will try to toast it down to no more than
30-
* TOAST_TUPLE_TARGET bytes. Both numbers include all tuple header overhead
31-
* and between-fields alignment padding, but we do *not* consider any
32-
* end-of-tuple alignment padding; hence the values can be compared directly
33-
* to a tuple's t_len field.
38+
* TOAST_TUPLE_TARGET bytes through compressing compressible fields and
39+
* moving EXTENDED and EXTERNAL data out-of-line.
3440
*
3541
* The numbers need not be the same, though they currently are. It doesn't
3642
* make sense for TARGET to exceed THRESHOLD, but it could be useful to make
3743
* it be smaller.
3844
*
3945
* Currently we choose both values to match the largest tuple size for which
40-
* TOAST_TUPLES_PER_PAGE tuples can fit on a disk page.
46+
* TOAST_TUPLES_PER_PAGE tuples can fit on a heap page.
4147
*
4248
* XXX while these can be modified without initdb, some thought needs to be
4349
* given to needs_toast_table() in toasting.c before unleashing random
@@ -46,13 +52,21 @@
4652
*/
4753
#define TOAST_TUPLES_PER_PAGE 4
4854

49-
#define TOAST_TUPLE_THRESHOLD \
50-
MAXALIGN_DOWN((BLCKSZ - \
51-
MAXALIGN(SizeOfPageHeaderData + TOAST_TUPLES_PER_PAGE * sizeof(ItemIdData))) \
52-
/ TOAST_TUPLES_PER_PAGE)
55+
#define TOAST_TUPLE_THRESHOLD MaximumBytesPerTuple(TOAST_TUPLES_PER_PAGE)
5356

5457
#define TOAST_TUPLE_TARGET TOAST_TUPLE_THRESHOLD
5558

59+
/*
60+
* The code will also consider moving MAIN data out-of-line, but only as a
61+
* last resort if the previous steps haven't reached the target tuple size.
62+
* In this phase we use a different target size, currently equal to the
63+
* largest tuple that will fit on a heap page. This is reasonable since
64+
* the user has told us to keep the data in-line if at all possible.
65+
*/
66+
#define TOAST_TUPLES_PER_PAGE_MAIN 1
67+
68+
#define TOAST_TUPLE_TARGET_MAIN MaximumBytesPerTuple(TOAST_TUPLES_PER_PAGE_MAIN)
69+
5670
/*
5771
* If an index value is larger than TOAST_INDEX_TARGET, we will try to
5872
* compress it (we can't move it out-of-line, however). Note that this
@@ -72,10 +86,7 @@
7286
*/
7387
#define EXTERN_TUPLES_PER_PAGE 4 /* tweak only this */
7488

75-
#define EXTERN_TUPLE_MAX_SIZE \
76-
MAXALIGN_DOWN((BLCKSZ - \
77-
MAXALIGN(SizeOfPageHeaderData + EXTERN_TUPLES_PER_PAGE * sizeof(ItemIdData))) \
78-
/ EXTERN_TUPLES_PER_PAGE)
89+
#define EXTERN_TUPLE_MAX_SIZE MaximumBytesPerTuple(EXTERN_TUPLES_PER_PAGE)
7990

8091
#define TOAST_MAX_CHUNK_SIZE \
8192
(EXTERN_TUPLE_MAX_SIZE - \

0 commit comments

Comments
 (0)