See PHI2012. A phage remote adjust ... --timeout X -- --optimize process on the remote survived a client timeout in phage.

Phage timeouts are currently purely on the client. Phage could reasonably attempt to get the remote to shut down before it times out, and emit different result codes for "timed out, killed, kill acknowledged" vs "timed out and abandoned".

The recent ClientAliveInterval change from 208d4a59 also hasn't made to production hosts beyond secure, but seems desirable and to be working well. This would likely fix this without phage changes.

Outside of sshd killing it after a ClientAliveInterval failure, the phage remote client might be able to detect that stdin has closed and terminate.