Tried out this patch and it applies, compiles, and passes check-world. Also flipped things around in pg_recvlogical.c to exit-on-success to ensure it's actually being called and that worked too. Outside of a more complicated harness that simulates fsync errors not sure how else to test this further.
I did some searching and found a FUSE based on that looks interesting: CharybdeFS[1]. Rather than being fixed at mount time, it has a client/server interface so you can change the handling of syscalls on the fly[2]. For example you can error out fsync calls halfway through a test rather than always or randomly. Haven't tried it out but leaving it here as it seems relevant.
On Wed, Jun 26, 2019 at 12:11 AM Michael Paquier <michael@paquier.xyz> wrote:
Why using a different error code. Using EXIT_FAILURE is a more common practice in the in-core binaries. The patch looks fine to me except that, that's a good first cut.
An error code specific to fsync issues could help with tests as the harness could check it to ensure things died for the right reasons. With a generic "messed up fsync" harness you might even be able to run some existing tests that would otherwise pass and check for the fsync-specific exit code.
