Remove overzealous _bt_killitems assertion.
authorPeter Geoghegan <[email protected]>
Fri, 10 Oct 2025 18:52:25 +0000 (14:52 -0400)
committerPeter Geoghegan <[email protected]>
Fri, 10 Oct 2025 18:52:25 +0000 (14:52 -0400)
An assertion in _bt_killitems expected the scan's currPos state to
contain a valid LSN, saved from when currPos's page was initially read.
The assertion failed to account for the fact that even logged relations
can have leaf pages with an invalid LSN when built with wal_level set to
"minimal".  Remove the faulty assertion.

Oversight in commit e6eed40e (though note that the assertion was
backpatched to stable branches before 18 by commit 7c319f54).

Author: Peter Geoghegan <[email protected]>
Reported-By: Matthijs van der Vleuten <[email protected]>
Bug: #19082
Discussion: https://postgr.es/m/19082-628e62160dbbc1c1@postgresql.org
Backpatch-through: 13

src/backend/access/nbtree/nbtutils.c

index 41b4fbd1c37e8a301f0ddb4f2f90d84bc9fd3256..7c57f7075dbb7a39b4ec5465c933fb0bc4670ed6 100644 (file)
@@ -3397,7 +3397,6 @@ _bt_killitems(IndexScanDesc scan)
        buf = _bt_getbuf(rel, so->currPos.currPage, BT_READ);
 
        latestlsn = BufferGetLSNAtomic(buf);
-       Assert(!XLogRecPtrIsInvalid(so->currPos.lsn));
        Assert(so->currPos.lsn <= latestlsn);
        if (so->currPos.lsn != latestlsn)
        {