amcheck eating most of the runtime with wal_consistency_checking

On Mon, Apr 19, 2021 at 7:50 PM Michael Paquier <michael@paquier.xyz> wrote:

While testing wal_consistency_checking, I have noticed that by far
most of the runtime is spent within the regression test check_btree on
the series of three queries inserting each 100k tuples. This also
eats most of the run time of the test on HEAD. Could we for example
consider inserting less tuples with a lower fillfactor to reduce the
runtime of the test without impacting its coverage in a meaningful
way?

I don't see much point. wal_consistency_checking is intrinsically a
tool that increases the volume of WAL by a large multiple. Plus you
yourself only run it once a year.

I run it much more often than once a year (maybe once every 2 - 3
months), but I haven't noticed this at all.

--
Peter Geoghegan