Re: pg15b3: recovery fails with wal prefetch enabled

On Thu, Sep 01, 2022 at 05:35:23PM +1200, Thomas Munro wrote:
> So it *looks* like it finished early (and without the expected
> error?). But it also looks like it replayed that record, according to
> the page LSN. So which is it? Could you recompile with WAL_DEBUG
> defined in pg_config_manual.h, and run recovery with wal_debug = on,
> and see if it replays 1CAF84B0?

This is with 6672d79139 un-reverted.

$ sudo -u postgres ./tmp_install/usr/local/pgsql/bin/postgres -D /mnt/tmp/15/data -c logging_collector=no -c port=5678 -c wal_debug=on 2>&1 |grep 1CAF84B0 || echo not found
not found

$ sudo -u postgres ./tmp_install/usr/local/pgsql/bin/postgres -D /mnt/tmp/15/data -c logging_collector=no -c port=5678 -c wal_debug=on -c recovery_prefetch=no 2>&1 |grep 1CAF84B0 || echo not found
< 2022-09-01 00:44:55.878 CDT >LOG: REDO @ 1201/1CAF8478; LSN 1201/1CAF84B0: prev 1201/1CAF8438; xid 0; len 2; blkref #0: rel 1663/16881/2840, blk 53 - Heap2/VACUUM: nunused 4
< 2022-09-01 00:44:55.878 CDT >LOG: REDO @ 1201/1CAF84B0; LSN 1201/1CAF84F0: prev 1201/1CAF8478; xid 0; len 5; blkref #0: rel 1663/16881/2840, fork 2, blk 0; blkref #1: rel 1663/16881/2840, blk 53 - Heap2/VISIBLE: cutoff xid 3678741092 flags 0x01
< 2022-09-01 00:44:55.878 CDT >LOG: REDO @ 1201/1CAF84F0; LSN 1201/1CAF8AC0: prev 1201/1CAF84B0; xid 0; len 2; blkref #0: rel 1663/16881/1259, blk 1 FPW, compression method: zstd - Heap/INPLACE: off 33

(Note that "compression method: zstd" is a local change to
xlog_block_info() which I just extracted from my original patch for
wal_compression, after forgetting to compile --with-zstd. I'll mail
about that at a later time...).

--
Justin