Hello,
Fly machine supervisor is logging reader error: read ptm: input/output error — which I believe is a host-level pty/console I/O failure. When it occurs the machine’s Postgres becomes unreachable and the node falls out of the cluster.
Confirmed on two separate machines.
Samples from logs:
2026-05-19T07:54:50Z app[78410e5a217678] nrt [info]reader error: read ptm: input/output error 2026-05-19T07:36:36Z app[e82ed30ceed508] nrt [info]reader error: read ptm: input/output error
Cluster is recovering, but wanted to confirm if this is a host issue.
Thank you.
We post about host issues that affect your apps in your personalized status page: https://fly.io/status. That’s a good first place to check for any known issues.
That I/O error you mentioned happens when a process exits and closes its end of a PTY. It could be a symptom of a cluster that’s degraded due to a host issue, but your cluster’s issues could also be independent of infrastructure. The reader error: read ptm: input/output error message on its own doesn’t tell us what the problem is, so I’d generally suggest looking at surrounding logs, metrics from this point in time, and health check output from fly checks list --app <postgres-app-name> to better understand what’s happening cluster-side.
For your Postgres app in particular, 78410e5a217678 was being CPU throttled leading up to these errors. Scaling up from shared-1x to shared-2x (or larger) would give you a higher baseline CPU quota, which should reduce throttling and help with stability.