On Wed, Sep 07, 2022 at 03:57:29AM -0500, Justin Pryzby wrote:
> But that's also what'll happen when attempting to replay WAL using a postgres
> build which doesn't support the necessary compression method. I ran into this
> while compiling postgres locally while reporting the recovery_prefetch bug,
> when I failed to compile --with-zstd.
Well, I am aware of that as that's how I have tested my change. I
think that the case you are mentioning is really different than this
change, though. The case you are mentioning gets triggered with the
server-side compression of pg_basebackup with a client application,
while the case of a block image restored can only happen when using
inconsistent build options between a primary and a standby (at least
in core, for the code paths touched by this patch). Before posting my
previous patch, I have considered a few options:
- Extend errormsg_buf with an error code, but the frontend does not
care about that.
- Make RestoreBlockImage() a backend-only routine and provide a better
error control without filling in errormsg_buf, but I'd like to believe
that this code is useful for some frontend code even if core does not
use it yet in a frontend context.
- Change the signature of RestoreBlockImage() to return an enum with
at least a tri state instead of a boolean. For this one I could not
convince myself that this is worth the complexity, as we are talking
about inconsistent build options between nodes doing physical
replication, and the error message is the useful piece to know what's
happening (frontends are only going to consume the error message
anyway).
--
Michael
