Re: spurious(?) warnings in archive recovery

Here is a script that reproduces the issue conveniently.

This is targetted for pg11, might also work on pg10 (uses declarative
partitioning as a convenient way to spread traffic over many tables).

What it does is:

- spin up a new server in ./data1
- populate some data
- start a base backup into ./data2 (but no copying files yet)
- generate a bunch of activity
- actually copy the files for the backup
- finish the backup
- spin up the recovery server in ./data2

(there's some dead code in the script for testing torn-page cases, not
used at present and can be ignored)

Looking at the ./postgres2.log file created for the server in ./data2,
what I see is a ton of this:

LOG: redo starts at 0/15000028
WARNING: xlog min recovery request 0/41420F58 is past current point 0/151D1248
CONTEXT: writing block 2 of relation base/12719/16591_fsm
WAL redo at 0/151D01F0 for Heap/HOT_UPDATE: off 76 xmax 772 ; new off 79 xmax 0
WARNING: xlog min recovery request 0/458157D8 is past current point 0/15392030
CONTEXT: writing block 0 of relation base/12719/16387_vm
WAL redo at 0/15391FD0 for Heap/INSERT: off 14
WARNING: xlog min recovery request 0/42B23FA0 is past current point 0/154A8408
CONTEXT: writing block 2 of relation base/12719/16459_fsm
WAL redo at 0/154A7350 for Heap/HOT_UPDATE: off 43 xmax 776 ; new off 79 xmax 0
WARNING: xlog min recovery request 0/41406A60 is past current point 0/155CD2D0
CONTEXT: writing block 2 of relation base/12719/16393_fsm
WAL redo at 0/155CC230 for Heap/HOT_UPDATE: off 27 xmax 778 ; new off 79 xmax 0
[repeat some dozens of times]

The noise is worst with wal_log_hints=on (as given in the script);
turning that off doesn't solve the issue, there are still warnings given
for _vm pages.

--
Andrew (irc:RhodiumToad)