On Fri, Apr 13, 2018 at 11:09 AM, Stephen Frost <sfrost@snowman.net> wrote:
If you can reliably reproduce this with either pg_dump or psql, then the next step would perhaps be jumping in with gdb to see where that error is being hit..
I forgot to follow-up yesterday. It seems like my issue was caused by a bug with the pg_qualstats extension that I was using in conjuction with the POWA extension. There was already a similar issue reported on their tracker at https://github.com/powa-team/powa/issues/104
It seems that high TPS stress testing on Wednesday afternoon triggered the bug. When the stress testing tailed off, the error messages stopped happening and I was able to run pg_dump without errors. I removed those extensions shortly after. The authors have put out a bug fix but I can't use this system to test their fix and my sandbox isn't available for the same HA server stress test that we were doing on Wednesday.