Movatterモバイル変換


[0]ホーム

URL:


Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

Commitd932391

Browse files
committed
Add some more defenses against silly estimates to gincostestimate().
A report from Andy Colson showed that gincostestimate() was not beingnearly paranoid enough about whether to believe the statistics it finds inthe index metapage. The problem is that the metapage stats (other than thepending-pages count) are only updated by VACUUM, and in the worst casecould still reflect the index's original empty state even when it has grownto many entries. We attempted to deal with that by scaling up the stats tomatch the current index size, but if nEntries is zero then scaling it upstill gives zero. Moreover, the proportion of pages that are entry pagesvs. data pages vs. pending pages is unlikely to be estimated very well byscaling if the index is now orders of magnitude larger than before.We can improve matters by expanding the use of the rule-of-thumb estimatesI introduced in commit7fb008c: if the index has grown by morethan a cutoff amount (here set at 4X growth) since VACUUM, then use therule-of-thumb numbers instead of scaling. This might not be exactly rightbut it seems much less likely to produce insane estimates.I also improved both the scaling estimate and the rule-of-thumb estimateto account for numPendingPages, since it's reasonable to expect that thatis accurate in any case, and certainly pages that are in the pending listare not either entry or data pages.As a somewhat separate issue, adjust the estimation equations that areconcerned with extra fetches for partial-match searches. These equationssuppose that a fraction partialEntries / numEntries of the entry and datapages will be visited as a consequence of a partial-match search. Now,it's physically impossible for that fraction to exceed one, but ourestimate of partialEntries is mostly bunk, and our estimate of numEntriesisn't exactly gospel either, so we could arrive at a silly value. In theexample presented by Andy we were coming out with a value of 100, leadingto insane cost estimates. Clamp the fraction to one to avoid that.Like the previous patch, back-patch to all supported branches; thisproblem can be demonstrated in one form or another in all of them.
1 parent2d774aa commitd932391

File tree

1 file changed

+52
-29
lines changed

1 file changed

+52
-29
lines changed

‎src/backend/utils/adt/selfuncs.c

Lines changed: 52 additions & 29 deletions
Original file line numberDiff line numberDiff line change
@@ -7240,6 +7240,7 @@ gincostestimate(PG_FUNCTION_ARGS)
72407240
numEntries;
72417241
GinQualCountscounts;
72427242
boolmatchPossible;
7243+
doublepartialScale;
72437244
doubleentryPagesFetched,
72447245
dataPagesFetched,
72457246
dataPagesFetchedBySel;
@@ -7268,42 +7269,59 @@ gincostestimate(PG_FUNCTION_ARGS)
72687269
memset(&ginStats,0,sizeof(ginStats));
72697270
}
72707271

7271-
if (ginStats.nTotalPages>0&&ginStats.nEntryPages>0&&numPages>0)
7272+
/*
7273+
* Assuming we got valid (nonzero) stats at all, nPendingPages can be
7274+
* trusted, but the other fields are data as of the last VACUUM. We can
7275+
* scale them up to account for growth since then, but that method only
7276+
* goes so far; in the worst case, the stats might be for a completely
7277+
* empty index, and scaling them will produce pretty bogus numbers.
7278+
* Somewhat arbitrarily, set the cutoff for doing scaling at 4X growth; if
7279+
* it's grown more than that, fall back to estimating things only from the
7280+
* assumed-accurate index size. But we'll trust nPendingPages in any case
7281+
* so long as it's not clearly insane, ie, more than the index size.
7282+
*/
7283+
if (ginStats.nPendingPages<numPages)
7284+
numPendingPages=ginStats.nPendingPages;
7285+
else
7286+
numPendingPages=0;
7287+
7288+
if (numPages>0&&ginStats.nTotalPages <=numPages&&
7289+
ginStats.nTotalPages>numPages /4&&
7290+
ginStats.nEntryPages>0&&ginStats.nEntries>0)
72727291
{
72737292
/*
7274-
*We got validstats. nPendingPages canbe trusted, but the other
7275-
*fields are data as of the last VACUUM. Scalethem by the ratio
7276-
*numPages / nTotalPages toaccount for growth sincethen.
7293+
*OK, thestats seem close enough to sane tobe trusted. But we
7294+
*still need to scalethem by the ratio numPages / nTotalPages to
7295+
* account for growth sincethe last VACUUM.
72777296
*/
72787297
doublescale=numPages /ginStats.nTotalPages;
72797298

7280-
numEntryPages=ginStats.nEntryPages;
7281-
numDataPages=ginStats.nDataPages;
7282-
numPendingPages=ginStats.nPendingPages;
7283-
numEntries=ginStats.nEntries;
7284-
7285-
numEntryPages=ceil(numEntryPages*scale);
7286-
numDataPages=ceil(numDataPages*scale);
7287-
numEntries=ceil(numEntries*scale);
7299+
numEntryPages=ceil(ginStats.nEntryPages*scale);
7300+
numDataPages=ceil(ginStats.nDataPages*scale);
7301+
numEntries=ceil(ginStats.nEntries*scale);
72887302
/* ensure we didn't round up too much */
7289-
numEntryPages=Min(numEntryPages,numPages);
7290-
numDataPages=Min(numDataPages,numPages-numEntryPages);
7303+
numEntryPages=Min(numEntryPages,numPages-numPendingPages);
7304+
numDataPages=Min(numDataPages,
7305+
numPages-numPendingPages-numEntryPages);
72917306
}
72927307
else
72937308
{
72947309
/*
7295-
* It's a hypothetical index, or perhaps an index created pre-9.1 and
7296-
* never vacuumed since upgrading. Invent some plausible internal
7297-
* statistics based on the index page count. We estimate that 90% of
7298-
* the index is entry pages, and the rest is data pages. Estimate 100
7299-
* entries per entry page; this is rather bogus since it'll depend on
7300-
* the size of the keys, but it's more robust than trying to predict
7301-
* the number of entries per heap tuple.
7310+
* We might get here because it's a hypothetical index, or an index
7311+
* created pre-9.1 and never vacuumed since upgrading (in which case
7312+
* its stats would read as zeroes), or just because it's grown too
7313+
* much since the last VACUUM for us to put our faith in scaling.
7314+
*
7315+
* Invent some plausible internal statistics based on the index page
7316+
* count (and clamp that to at least 10 pages, just in case). We
7317+
* estimate that 90% of the index is entry pages, and the rest is data
7318+
* pages. Estimate 100 entries per entry page; this is rather bogus
7319+
* since it'll depend on the size of the keys, but it's more robust
7320+
* than trying to predict the number of entries per heap tuple.
73027321
*/
73037322
numPages=Max(numPages,10);
7304-
numEntryPages=floor(numPages*0.90);
7305-
numDataPages=numPages-numEntryPages;
7306-
numPendingPages=0;
7323+
numEntryPages=floor((numPages-numPendingPages)*0.90);
7324+
numDataPages=numPages-numPendingPages-numEntryPages;
73077325
numEntries=floor(numEntryPages*100);
73087326
}
73097327

@@ -7429,16 +7447,21 @@ gincostestimate(PG_FUNCTION_ARGS)
74297447
/*
74307448
* Add an estimate of entry pages read by partial match algorithm. It's a
74317449
* scan over leaf pages in entry tree. We haven't any useful stats here,
7432-
* so estimate it as proportion.
7450+
* so estimate it as proportion. Because counts.partialEntries is really
7451+
* pretty bogus (see code above), it's possible that it is more than
7452+
* numEntries; clamp the proportion to ensure sanity.
74337453
*/
7434-
entryPagesFetched+=ceil(numEntryPages*counts.partialEntries /numEntries);
7454+
partialScale=counts.partialEntries /numEntries;
7455+
partialScale=Min(partialScale,1.0);
7456+
7457+
entryPagesFetched+=ceil(numEntryPages*partialScale);
74357458

74367459
/*
74377460
* Partial match algorithm reads all data pages before doing actual scan,
7438-
* so it's a startup cost. Again, we haven't any useful stats here, so,
7439-
* estimate it as proportion
7461+
* so it's a startup cost.Again, we haven't any useful stats here, so
7462+
* estimate it as proportion.
74407463
*/
7441-
dataPagesFetched=ceil(numDataPages*counts.partialEntries /numEntries);
7464+
dataPagesFetched=ceil(numDataPages*partialScale);
74427465

74437466
/*
74447467
* Calculate cache effects if more than one scan due to nestloops or array

0 commit comments

Comments
 (0)

[8]ページ先頭

©2009-2025 Movatter.jp