On Thu, Apr 20, 2023 at 01:15:40PM +0200, Daniel Gustafsson wrote:
> Maybe. I'm a bit hesitant to add too many smarts to pg_rewind. It's a tool
> for when something has gone wrong with a cluster (albeit probably not at the
> filesystem level), and at that point I feel it's better to put the user fully
> in charge. Perhaps I'm overly cautious, curious to hear from others.
Hmm. pg_rewind is mostly a differential block-level backup tool, so
applying the same rules everywhere across the board would be sensible
here. See that exclude_list_item is able to handle prefixes, and we
may want to extend the same logic for the directory list, as well..
By the way, the patch ought to add some tests? For pg_basebackup,
this would be around "These files should not be copied" in
010_pg_basebackup.pl. pg_checksums has also its own checks in
002_actions.pl.
--
Michael