database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-07-27 12:41:57 +03:00
Code Activity

amcheck: Distinguish interrupted page deletion from corruption.

Browse Source 
This prevents false-positive reports about "the first child of leftmost
target page is not leftmost of its level", "block %u is not leftmost"
and "left link/right link pair".  They appeared if amcheck ran before
VACUUM cleaned things, after a cluster exited recovery between the
first-stage and second-stage WAL records of a deletion.  Back-patch to
v11 (all supported versions).

Reviewed by Peter Geoghegan.

Discussion: https://postgr.es/m/20231005025232.c7.nmisch@google.com
This commit is contained in:
Noah Misch
2023-10-30 14:46:05 -07:00
parent 5f06918399
commit 6f81386a9c
3 changed files with 163 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

83
contrib/amcheck/verify_nbtree.c
View File

@ -147,6 +147,9 @@ static void bt_check_every_level(Relation rel, Relation heaprel,
bool rootdescend);
static BtreeLevel bt_check_level_from_leftmost(BtreeCheckState *state,
BtreeLevel level);
static bool bt_leftmost_ignoring_half_dead(BtreeCheckState *state,
BlockNumber start,
BTPageOpaque start_opaque);
static void bt_recheck_sibling_links(BtreeCheckState *state,
BlockNumber btpo_prev_from_target,
BlockNumber leftcurrent);
@ -775,7 +778,7 @@ bt_check_level_from_leftmost(BtreeCheckState *state, BtreeLevel level)
*/
if (state->readonly)
{
if (!P_LEFTMOST(opaque))
if (!bt_leftmost_ignoring_half_dead(state, current, opaque))
ereport(ERROR,
(errcode(ERRCODE_INDEX_CORRUPTED),
errmsg("block %u is not leftmost in index \"%s\"",
@ -829,8 +832,16 @@ bt_check_level_from_leftmost(BtreeCheckState *state, BtreeLevel level)
*/
}
/* Sibling links should be in mutual agreement */
if (opaque->btpo_prev != leftcurrent)
/*
* Sibling links should be in mutual agreement. There arises
* leftcurrent == P_NONE && btpo_prev != P_NONE when the left sibling
* of the parent's low-key downlink is half-dead. (A half-dead page
* has no downlink from its parent.) Under heavyweight locking, the
* last bt_leftmost_ignoring_half_dead() validated this btpo_prev.
* Without heavyweight locking, validation of the P_NONE case remains
* unimplemented.
*/
if (opaque->btpo_prev != leftcurrent && leftcurrent != P_NONE)
bt_recheck_sibling_links(state, opaque->btpo_prev, leftcurrent);
/* Check level */
@ -911,6 +922,66 @@ nextpage:
return nextleveldown;
}
/*
* Like P_LEFTMOST(start_opaque), but accept an arbitrarily-long chain of
* half-dead, sibling-linked pages to the left. If a half-dead page appears
* under state->readonly, the database exited recovery between the first-stage
* and second-stage WAL records of a deletion.
*/
static bool
bt_leftmost_ignoring_half_dead(BtreeCheckState *state,
BlockNumber start,
BTPageOpaque start_opaque)
{
BlockNumber reached = start_opaque->btpo_prev,
reached_from = start;
bool all_half_dead = true;
/*
* To handle the !readonly case, we'd need to accept BTP_DELETED pages and
* potentially observe nbtree/README "Page deletion and backwards scans".
*/
Assert(state->readonly);
while (reached != P_NONE && all_half_dead)
{
Page page = palloc_btree_page(state, reached);
BTPageOpaque reached_opaque = BTPageGetOpaque(page);
CHECK_FOR_INTERRUPTS();
/*
* Try to detect btpo_prev circular links. _bt_unlink_halfdead_page()
* writes that side-links will continue to point to the siblings.
* Check btpo_next for that property.
*/
all_half_dead = P_ISHALFDEAD(reached_opaque) &&
reached != start &&
reached != reached_from &&
reached_opaque->btpo_next == reached_from;
if (all_half_dead)
{
XLogRecPtr pagelsn = PageGetLSN(page);
/* pagelsn should point to an XLOG_BTREE_MARK_PAGE_HALFDEAD */
ereport(DEBUG1,
(errcode(ERRCODE_NO_DATA),
errmsg_internal("harmless interrupted page deletion detected in index \"%s\"",
RelationGetRelationName(state->rel)),
errdetail_internal("Block=%u right block=%u page lsn=%X/%X.",
reached, reached_from,
LSN_FORMAT_ARGS(pagelsn))));
reached_from = reached;
reached = reached_opaque->btpo_prev;
}
pfree(page);
}
return all_half_dead;
}
/*
* Raise an error when target page's left link does not point back to the
* previous target page, called leftcurrent here. The leftcurrent page's
@ -951,6 +1022,9 @@ bt_recheck_sibling_links(BtreeCheckState *state,
BlockNumber btpo_prev_from_target,
BlockNumber leftcurrent)
{
/* passing metapage to BTPageGetOpaque() would give irrelevant findings */
Assert(leftcurrent != P_NONE);
if (!state->readonly)
{
Buffer lbuf;
@ -1934,7 +2008,8 @@ bt_child_highkey_check(BtreeCheckState *state,
opaque = BTPageGetOpaque(page);
/* The first page we visit at the level should be leftmost */
if (first && !BlockNumberIsValid(state->prevrightlink) && !P_LEFTMOST(opaque))
if (first && !BlockNumberIsValid(state->prevrightlink) &&
!bt_leftmost_ignoring_half_dead(state, blkno, opaque))
ereport(ERROR,
(errcode(ERRCODE_INDEX_CORRUPTED),
errmsg("the first child of leftmost target page is not leftmost of its level in index \"%s\"",