Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Skip to content

Commit 5850b20

Browse files
anarazelanarazel
committed
Add pgstattuple_approx() to the pgstattuple extension.
The new function allows to estimate bloat and other table level statics in a faster, but approximate, way. It does so by using information from the free space map for pages marked as all visible in the visibility map. The rest of the table is actually read and free space/bloat is measured accurately. In many cases that allows to get bloat information much quicker, causing less IO. Author: Abhijit Menon-Sen Reviewed-By: Andres Freund, Amit Kapila and Tomas Vondra Discussion: 20140402214144.GA28681@kea.toroid.org
1 parent dcf5e31 commit 5850b20

File tree

6 files changed

+446
-5
lines changed

6 files changed

+446
-5
lines changed

‎contrib/pgstattuple/Makefile

+2-2
Original file line numberDiff line numberDiff line change
@@ -1,10 +1,10 @@
11
# contrib/pgstattuple/Makefile
22

33
MODULE_big = pgstattuple
4-
OBJS = pgstattuple.o pgstatindex.o $(WIN32RES)
4+
OBJS = pgstattuple.o pgstatindex.o pgstatapprox.o $(WIN32RES)
55

66
EXTENSION = pgstattuple
7-
DATA = pgstattuple--1.2.sql pgstattuple--1.1--1.2.sql pgstattuple--1.0--1.1.sql pgstattuple--unpackaged--1.0.sql
7+
DATA = pgstattuple--1.3.sql pgstattuple--1.2--1.3.sql pgstattuple--1.1--1.2.sql pgstattuple--1.0--1.1.sql pgstattuple--unpackaged--1.0.sql
88
PGFILEDESC = "pgstattuple - tuple-level statistics"
99

1010
REGRESS = pgstattuple

‎contrib/pgstattuple/pgstatapprox.c

+273
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,273 @@
1+
/*-------------------------------------------------------------------------
2+
*
3+
* pgstatapproc.c
4+
* Bloat estimation functions
5+
*
6+
* Copyright (c) 2014-2015, PostgreSQL Global Development Group
7+
*
8+
* IDENTIFICATION
9+
* contrib/pgstattuple/pgstatapprox.c
10+
*
11+
*-------------------------------------------------------------------------
12+
*/
13+
#include "postgres.h"
14+
15+
#include "access/visibilitymap.h"
16+
#include "access/transam.h"
17+
#include "access/xact.h"
18+
#include "access/multixact.h"
19+
#include "access/htup_details.h"
20+
#include "catalog/namespace.h"
21+
#include "funcapi.h"
22+
#include "miscadmin.h"
23+
#include "storage/bufmgr.h"
24+
#include "storage/freespace.h"
25+
#include "storage/procarray.h"
26+
#include "storage/lmgr.h"
27+
#include "utils/builtins.h"
28+
#include "utils/tqual.h"
29+
#include "commands/vacuum.h"
30+
31+
PG_FUNCTION_INFO_V1(pgstattuple_approx);
32+
33+
typedef struct output_type
34+
{
35+
uint64 table_len;
36+
uint64 scanned_percent;
37+
uint64 tuple_count;
38+
uint64 tuple_len;
39+
double tuple_percent;
40+
uint64 dead_tuple_count;
41+
uint64 dead_tuple_len;
42+
double dead_tuple_percent;
43+
uint64 free_space;
44+
double free_percent;
45+
} output_type;
46+
47+
#define NUM_OUTPUT_COLUMNS 10
48+
49+
/*
50+
* This function takes an already open relation and scans its pages,
51+
* skipping those that have the corresponding visibility map bit set.
52+
* For pages we skip, we find the free space from the free space map
53+
* and approximate tuple_len on that basis. For the others, we count
54+
* the exact number of dead tuples etc.
55+
*
56+
* This scan is loosely based on vacuumlazy.c:lazy_scan_heap(), but
57+
* we do not try to avoid skipping single pages.
58+
*/
59+
static void
60+
statapprox_heap(Relation rel, output_type *stat)
61+
{
62+
BlockNumber scanned,
63+
nblocks,
64+
blkno;
65+
Buffer vmbuffer = InvalidBuffer;
66+
BufferAccessStrategy bstrategy;
67+
TransactionId OldestXmin;
68+
uint64 misc_count = 0;
69+
70+
OldestXmin = GetOldestXmin(rel, true);
71+
bstrategy = GetAccessStrategy(BAS_BULKREAD);
72+
73+
nblocks = RelationGetNumberOfBlocks(rel);
74+
scanned = 0;
75+
76+
for (blkno = 0; blkno < nblocks; blkno++)
77+
{
78+
Buffer buf;
79+
Page page;
80+
OffsetNumber offnum,
81+
maxoff;
82+
Size freespace;
83+
84+
CHECK_FOR_INTERRUPTS();
85+
86+
/*
87+
* If the page has only visible tuples, then we can find out the
88+
* free space from the FSM and move on.
89+
*/
90+
if (visibilitymap_test(rel, blkno, &vmbuffer))
91+
{
92+
freespace = GetRecordedFreeSpace(rel, blkno);
93+
stat->tuple_len += BLCKSZ - freespace;
94+
stat->free_space += freespace;
95+
continue;
96+
}
97+
98+
buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno,
99+
RBM_NORMAL, bstrategy);
100+
101+
LockBuffer(buf, BUFFER_LOCK_SHARE);
102+
103+
page = BufferGetPage(buf);
104+
105+
/*
106+
* It's not safe to call PageGetHeapFreeSpace() on new pages, so
107+
* we treat them as being free space for our purposes.
108+
*/
109+
if (!PageIsNew(page))
110+
stat->free_space += PageGetHeapFreeSpace(page);
111+
else
112+
stat->free_space += BLCKSZ - SizeOfPageHeaderData;
113+
114+
if (PageIsNew(page) || PageIsEmpty(page))
115+
{
116+
UnlockReleaseBuffer(buf);
117+
continue;
118+
}
119+
120+
scanned++;
121+
122+
/*
123+
* Look at each tuple on the page and decide whether it's live
124+
* or dead, then count it and its size. Unlike lazy_scan_heap,
125+
* we can afford to ignore problems and special cases.
126+
*/
127+
maxoff = PageGetMaxOffsetNumber(page);
128+
129+
for (offnum = FirstOffsetNumber;
130+
offnum <= maxoff;
131+
offnum = OffsetNumberNext(offnum))
132+
{
133+
ItemId itemid;
134+
HeapTupleData tuple;
135+
136+
itemid = PageGetItemId(page, offnum);
137+
138+
if (!ItemIdIsUsed(itemid) || ItemIdIsRedirected(itemid) ||
139+
ItemIdIsDead(itemid))
140+
{
141+
continue;
142+
}
143+
144+
Assert(ItemIdIsNormal(itemid));
145+
146+
ItemPointerSet(&(tuple.t_self), blkno, offnum);
147+
148+
tuple.t_data = (HeapTupleHeader) PageGetItem(page, itemid);
149+
tuple.t_len = ItemIdGetLength(itemid);
150+
tuple.t_tableOid = RelationGetRelid(rel);
151+
152+
/*
153+
* We count live and dead tuples, but we also need to add up
154+
* others in order to feed vac_estimate_reltuples.
155+
*/
156+
switch (HeapTupleSatisfiesVacuum(&tuple, OldestXmin, buf))
157+
{
158+
case HEAPTUPLE_RECENTLY_DEAD:
159+
misc_count++;
160+
/* Fall through */
161+
case HEAPTUPLE_DEAD:
162+
stat->dead_tuple_len += tuple.t_len;
163+
stat->dead_tuple_count++;
164+
break;
165+
case HEAPTUPLE_LIVE:
166+
stat->tuple_len += tuple.t_len;
167+
stat->tuple_count++;
168+
break;
169+
case HEAPTUPLE_INSERT_IN_PROGRESS:
170+
case HEAPTUPLE_DELETE_IN_PROGRESS:
171+
misc_count++;
172+
break;
173+
default:
174+
elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
175+
break;
176+
}
177+
}
178+
179+
UnlockReleaseBuffer(buf);
180+
}
181+
182+
stat->table_len = (uint64) nblocks * BLCKSZ;
183+
stat->tuple_count = vac_estimate_reltuples(rel, false, nblocks, scanned,
184+
stat->tuple_count+misc_count);
185+
186+
/*
187+
* Calculate percentages if the relation has one or more pages.
188+
*/
189+
if (nblocks != 0)
190+
{
191+
stat->scanned_percent = 100 * scanned / nblocks;
192+
stat->tuple_percent = 100.0 * stat->tuple_len / stat->table_len;
193+
stat->dead_tuple_percent = 100.0 * stat->dead_tuple_len / stat->table_len;
194+
stat->free_percent = 100.0 * stat->free_space / stat->table_len;
195+
}
196+
197+
if (BufferIsValid(vmbuffer))
198+
{
199+
ReleaseBuffer(vmbuffer);
200+
vmbuffer = InvalidBuffer;
201+
}
202+
}
203+
204+
/*
205+
* Returns estimated live/dead tuple statistics for the given relid.
206+
*/
207+
Datum
208+
pgstattuple_approx(PG_FUNCTION_ARGS)
209+
{
210+
Oid relid = PG_GETARG_OID(0);
211+
Relation rel;
212+
output_type stat = {0};
213+
TupleDesc tupdesc;
214+
bool nulls[NUM_OUTPUT_COLUMNS];
215+
Datum values[NUM_OUTPUT_COLUMNS];
216+
HeapTuple ret;
217+
int i = 0;
218+
219+
if (!superuser())
220+
ereport(ERROR,
221+
(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
222+
(errmsg("must be superuser to use pgstattuple functions"))));
223+
224+
if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
225+
elog(ERROR, "return type must be a row type");
226+
227+
if (tupdesc->natts != NUM_OUTPUT_COLUMNS)
228+
elog(ERROR, "incorrect number of output arguments");
229+
230+
rel = relation_open(relid, AccessShareLock);
231+
232+
/*
233+
* Reject attempts to read non-local temporary relations; we would be
234+
* likely to get wrong data since we have no visibility into the owning
235+
* session's local buffers.
236+
*/
237+
if (RELATION_IS_OTHER_TEMP(rel))
238+
ereport(ERROR,
239+
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
240+
errmsg("cannot access temporary tables of other sessions")));
241+
242+
/*
243+
* We support only ordinary relations and materialised views,
244+
* because we depend on the visibility map and free space map
245+
* for our estimates about unscanned pages.
246+
*/
247+
if (!(rel->rd_rel->relkind == RELKIND_RELATION ||
248+
rel->rd_rel->relkind == RELKIND_MATVIEW))
249+
ereport(ERROR,
250+
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
251+
errmsg("\"%s\" is not a table or materialized view",
252+
RelationGetRelationName(rel))));
253+
254+
statapprox_heap(rel, &stat);
255+
256+
relation_close(rel, AccessShareLock);
257+
258+
memset(nulls, 0, sizeof(nulls));
259+
260+
values[i++] = Int64GetDatum(stat.table_len);
261+
values[i++] = Float8GetDatum(stat.scanned_percent);
262+
values[i++] = Int64GetDatum(stat.tuple_count);
263+
values[i++] = Int64GetDatum(stat.tuple_len);
264+
values[i++] = Float8GetDatum(stat.tuple_percent);
265+
values[i++] = Int64GetDatum(stat.dead_tuple_count);
266+
values[i++] = Int64GetDatum(stat.dead_tuple_len);
267+
values[i++] = Float8GetDatum(stat.dead_tuple_percent);
268+
values[i++] = Int64GetDatum(stat.free_space);
269+
values[i++] = Float8GetDatum(stat.free_percent);
270+
271+
ret = heap_form_tuple(tupdesc, values, nulls);
272+
return HeapTupleGetDatum(ret);
273+
}

‎contrib/pgstattuple/pgstattuple--1.2--1.3.sql

+18
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,18 @@
1+
/* contrib/pgstattuple/pgstattuple--1.2--1.3.sql */
2+
3+
-- complain if script is sourced in psql, rather than via ALTER EXTENSION
4+
\echo Use "ALTER EXTENSION pgstattuple UPDATE TO '1.3'" to load this file. \quit
5+
6+
CREATE FUNCTION pgstattuple_approx(IN reloid regclass,
7+
OUT table_len BIGINT, -- physical table length in bytes
8+
OUT scanned_percent FLOAT8, -- what percentage of the table's pages was scanned
9+
OUT approx_tuple_count BIGINT, -- estimated number of live tuples
10+
OUT approx_tuple_len BIGINT, -- estimated total length in bytes of live tuples
11+
OUT approx_tuple_percent FLOAT8, -- live tuples in % (based on estimate)
12+
OUT dead_tuple_count BIGINT, -- exact number of dead tuples
13+
OUT dead_tuple_len BIGINT, -- exact total length in bytes of dead tuples
14+
OUT dead_tuple_percent FLOAT8, -- dead tuples in % (based on estimate)
15+
OUT approx_free_space BIGINT, -- estimated free space in bytes
16+
OUT approx_free_percent FLOAT8) -- free space in % (based on estimate)
17+
AS 'MODULE_PATHNAME', 'pgstattuple_approx'
18+
LANGUAGE C STRICT;

‎contrib/pgstattuple/pgstattuple--1.2.sql renamed to ‎contrib/pgstattuple/pgstattuple--1.3.sql

+17-1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
/* contrib/pgstattuple/pgstattuple--1.2.sql */
1+
/* contrib/pgstattuple/pgstattuple--1.3.sql */
22

33
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
44
\echo Use "CREATE EXTENSION pgstattuple" to load this file. \quit
@@ -77,3 +77,19 @@ CREATE FUNCTION pg_relpages(IN relname regclass)
7777
RETURNS BIGINT
7878
AS 'MODULE_PATHNAME', 'pg_relpagesbyid'
7979
LANGUAGE C STRICT;
80+
81+
/* New stuff in 1.3 begins here */
82+
83+
CREATE FUNCTION pgstattuple_approx(IN reloid regclass,
84+
OUT table_len BIGINT, -- physical table length in bytes
85+
OUT scanned_percent FLOAT8, -- what percentage of the table's pages was scanned
86+
OUT approx_tuple_count BIGINT, -- estimated number of live tuples
87+
OUT approx_tuple_len BIGINT, -- estimated total length in bytes of live tuples
88+
OUT approx_tuple_percent FLOAT8, -- live tuples in % (based on estimate)
89+
OUT dead_tuple_count BIGINT, -- exact number of dead tuples
90+
OUT dead_tuple_len BIGINT, -- exact total length in bytes of dead tuples
91+
OUT dead_tuple_percent FLOAT8, -- dead tuples in % (based on estimate)
92+
OUT approx_free_space BIGINT, -- estimated free space in bytes
93+
OUT approx_free_percent FLOAT8) -- free space in % (based on estimate)
94+
AS 'MODULE_PATHNAME', 'pgstattuple_approx'
95+
LANGUAGE C STRICT;

‎contrib/pgstattuple/pgstattuple.control

+1-1
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
# pgstattuple extension
22
comment = 'show tuple-level statistics'
3-
default_version = '1.2'
3+
default_version = '1.3'
44
module_pathname = '$libdir/pgstattuple'
55
relocatable = true

0 commit comments

Comments
 (0)