I wrote: > Oh, one more thing: that buildfarm instance has been running this > same test case about once a day for a good long time. We never > saw this type of error before, but now we have two such failures > in its last ten runs. So it seems pretty clear that the issue > started in mid-April. Oh ... I lied. I realized that Postgres' handling of this error has varied in different branches, and when I looked for straight reports of "Input/output error" I found a few more: 2021-03-20 00:48:48.117 MSK [4089174:11] 008_fsm_truncation.pl PANIC: could not fdatasync file "000000010000000000000002": Input/output error 2021-04-06 19:30:54.103 MSK [3355008:11] 008_fsm_truncation.pl PANIC: could not fdatasync file "000000010000000000000002": Input/output error pg_basebackup: could not fsync file "000000010000000000000013": Input/output error The last one is harder to get an exact timestamp for, but it was approximately 2021-04-12 05:47:07 MSK. In any case, it seems this has been going on at least since mid-March. I don't see any other similar failures in our logs, going back six months. regards, tom lane
