Fix hash index redo reading a torn primary bucket page.

Replay of XLOG_HASH_DELETE, XLOG_HASH_MOVE_PAGE_CONTENTS and
XLOG_HASH_SQUEEZE_PAGE reads the primary bucket page only to take a
cleanup lock on it, so that no hot standby scan runs in the bucket while
tuples are removed or moved. VACUUM holds only a pin on that page, so
these records do not modify it and carry no full-page image of it.

If the on-disk page is torn and the full-page image that repairs it comes
later in the WAL, this read fails with "invalid page". That can happen in
crash recovery, and on a standby started from a base backup whose copy of
the page is torn.

The lock is needed only to serialize against hot standby scans, which can
run only once recovery has reached consistency. Skip the read and lock
while HotStandbyActiveInReplay() is false. This also avoids reading the
page before its repairing image has been replayed, since that image always
precedes the consistency point.

Reported-by: shihao zhong <[email protected]>
Author: Greg Burd <[email protected]>
Co-authored-by: Vignesh C <[email protected]>
Suggested-by: Amit Kapila <[email protected]>
Reviewed-by: Joao Rito <[email protected]>
Reviewed-by: Amit Kapila <[email protected]>
Discussion: 
https://postgr.es/m/CAGRkXqT_AiwcsWZg5hnR1L2oWO8UpQpkL0iAjoZR+bey=go...@mail.gmail.com
Discussion: 
https://postgr.es/m/caa4ek1lwzk0mak0lnczz6hsqchnamfjb47na8xuwrp4cylf...@mail.gmail.com
Backpatch-through: 15

Branch
------
REL_18_STABLE

Details
-------
https://git.postgresql.org/pg/commitdiff/ed2cc619ac74733c7d20b48131714550e4ce1a08

Modified Files
--------------
src/backend/access/hash/hash_xlog.c       | 52 +++++++++++++++++++++++++------
src/backend/access/transam/xlogrecovery.c |  3 +-
src/include/access/xlogrecovery.h         |  1 +
3 files changed, 45 insertions(+), 11 deletions(-)

Reply via email to