On Mon, Oct 14, 2019 at 08:57:16AM +0900, Michael Paquier wrote: > I need to think about that, but shouldn't we have a way to reproduce > that case rather reliably with an isolation test? The patch looks to > good to me, these are also the two places I spotted yesterday after a > quick lookup. The only other caller is isTempNamespaceInUse() which > does its thing correctly.
Actually, reindex-concurrently.spec stresses that, except that in order to reproduce the failure we need to close the connection exactly in the waiting loop before sending the progress report but after looking at VirtualTransactionIdIsValid. Using a debugger and a simple checkpoint I can easily reproduce the crash, but we'd need more to make that test case deterministic, like a termination with the correct timing. So, Alvaro, your patch looks good to me. Could you apply it? -- Michael
signature.asc
Description: PGP signature
