NotificationsYou must be signed in to change notification settings
Fork28
Star151

Commitd4b34f6

committed

Deduplicate PageIsNew() check in lazy_scan_heap().

The recheck isn't needed anymore, as RelationGetBufferForTuple() nowextends the relation with RBM_ZERO_AND_LOCK. Previously we needed tohandle the fact that relation extension extended the relation and thenseparately acquired a lock on the page - while expecting that the pageis empty.Reported-By: Ranier VilelaDiscussion:https://postgr.es/m/CAEudQArA_=J0D5T258xsCY6Xtf6wiH4b=QDPDgVS+WZUN10WDw@mail.gmail.com

1 parent364bdd0 commitd4b34f6Copy full SHA for d4b34f6

File tree

1 file changed

+11

-21

lines changed

src/backend/access/heap
- vacuumlazy.c

1 file changed

+11

-21

lines changed

`‎src/backend/access/heap/vacuumlazy.c`

Lines changed: 11 additions & 21 deletions

Original file line number	Diff line number	Diff line change
`@@ -1147,45 +1147,35 @@ lazy_scan_heap(Relation onerel, VacuumParams params, LVRelStats vacrelstats,`
`1147`	`1147`
`1148`	`1148`	`if (PageIsNew(page))`
`1149`	`1149`	`{`
`1150`		`-boolstill_new;`
`1151`		`-`
`1152`	`1150`	`/*`
`1153`	`1151`	`* All-zeroes pages can be left over if either a backend extends`
`1154`	`1152`	`* the relation by a single page, but crashes before the newly`
`1155`	`1153`	`* initialized page has been written out, or when bulk-extending`
`1156`	`1154`	`* the relation (which creates a number of empty pages at the tail`
`1157`	`1155`	`* end of the relation, but enters them into the FSM).`
`1158`	`1156`	`*`
`1159`		`- * Make sure these pages are in the FSM, to ensure they can be`
`1160`		`- * reused. Do that by testing if there's any space recorded for`
`1161`		`- * the page. If not, enter it.`
`1162`		`- *`
`1163`	`1157`	`* Note we do not enter the page into the visibilitymap. That has`
`1164`	`1158`	`* the downside that we repeatedly visit this page in subsequent`
`1165`	`1159`	`* vacuums, but otherwise we'll never not discover the space on a`
`1166`	`1160`	`* promoted standby. The harm of repeated checking ought to`
`1167`	`1161`	`* normally not be too bad - the space usually should be used at`
`1168`	`1162`	`* some point, otherwise there wouldn't be any regular vacuums.`
	`1163`	`+ *`
	`1164`	`+ * Make sure these pages are in the FSM, to ensure they can be`
	`1165`	`+ * reused. Do that by testing if there's any space recorded for`
	`1166`	`+ * the page. If not, enter it. We do so after releasing the lock`
	`1167`	`+ * on the heap page, the FSM is approximate, after all.`
`1169`	`1168`	`*/`
`1170`		`-`
`1171`		`-/*`
`1172`		`- * Perform checking of FSM after releasing lock, the fsm is`
`1173`		`- * approximate, after all.`
`1174`		`- */`
`1175`		`-still_new=PageIsNew(page);`
`1176`	`1169`	`UnlockReleaseBuffer(buf);`
`1177`	`1170`
`1178`		`-if (still_new)`
`1179`		`-{`
`1180`		`-empty_pages++;`
	`1171`	`+empty_pages++;`
`1181`	`1172`
`1182`		`-if (GetRecordedFreeSpace(onerel,blkno)==0)`
`1183`		`-{`
`1184`		`-Sizefreespace;`
	`1173`	`+if (GetRecordedFreeSpace(onerel,blkno)==0)`
	`1174`	`+{`
	`1175`	`+Sizefreespace;`
`1185`	`1176`
`1186`		`-freespace=BufferGetPageSize(buf)-SizeOfPageHeaderData;`
`1187`		`-RecordPageWithFreeSpace(onerel,blkno,freespace);`
`1188`		`-}`
	`1177`	`+freespace=BufferGetPageSize(buf)-SizeOfPageHeaderData;`
	`1178`	`+RecordPageWithFreeSpace(onerel,blkno,freespace);`
`1189`	`1179`	`}`
`1190`	`1180`	`continue;`
`1191`	`1181`	`}`

0 commit comments

Comments

(0)

Movatterモバイル変換

Navigation Menu

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

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Commitd4b34f6

File tree

1 file changed

1 file changed

`‎src/backend/access/heap/vacuumlazy.c`

0 commit comments