Codex App restores stale discovered SSH host state on startup and crashes with "No AppServerManager registered for hostId" #18556

Activity

1. github-actions

   added

   bugSomething isn't workingSomething isn't working

   appIssues related to the Codex desktop appIssues related to the Codex desktop app

   app-serverIssues involving app server protocol or interfacesIssues involving app server protocol or interfaces

   on Apr 19, 2026
2. etraut-openai

   added

   remote

   on May 3, 2026
3. EvenOldridge commented on May 15, 2026

   

   EvenOldridge

   on May 15, 2026 – with ChatGPT Codex Connector

   More actions

   Related stale Remote SSH state/reconnect issue observed on Codex App 26.513.20950 (2816) on macOS.

   This is not the exact stale-host/crash case from the issue body, but it looks like the same class of persisted/in-memory Remote SSH state not reconciling after the underlying host becomes healthy again.

   Scenario

   1. Use Codex Desktop with a discovered Remote SSH project, host id shaped like remote-ssh-discovered:<redacted-ssh-alias>.
   2. Let the laptop sleep / network drop while the remote connection is active.
   3. Wake the laptop.
   4. Refresh the underlying SSH/Teleport config externally and verify SSH is healthy.
   5. Attempt to reconnect/open the remote project in Codex Desktop.

   Observed

   The project/connection stayed greyed out with a timeout/error state even though direct SSH worked and the remote Codex binary was available:

   ssh <redacted-ssh-alias> 'test -x "$HOME/.local/bin/codex" && "$HOME/.local/bin/codex" --version'
   # codex-cli 0.130.0

   Relevant app log lines, with host details redacted:

   app_server_connection.transport_connect_failed errorMessage="SSH: codex path probe timed out after 30000ms" hostId=remote-ssh-discovered:<redacted> remoteConnectionMaxRetryAttempts=2 sshSetupStage=remote_codex_lookup transport=websocket
   app_server_connection.remote_connection_retry_disabled errorMessage="SSH: codex path probe timed out after 30000ms" hostId=remote-ssh-discovered:<redacted> reconnectAttempt=2 remoteConnectionMaxRetryAttempts=2 transportKind=websocket
   remote_connections.manager_state_set hostId=remote-ssh-discovered:<redacted> nextState=error nextError={"code":"connection-failed","message":"SSH: codex path probe timed out after 30000ms"}
   

   At this point, the underlying SSH path was healthy, but the Desktop UI appeared to keep serving the stale retry-exhausted connection error. Selecting/opening the remote project was not enough to force a fresh connection object. Cmd+R was unavailable/disabled in this production window.

   The workaround was to go to Settings > Connections and toggle auto-connect for the SSH host off and back on. That forced reconciliation/recreation of the Remote SSH connection:

   remote_connections.manager_state_set hostId=remote-ssh-discovered:<redacted> nextState=disconnected previousState=error source=host_connection_state_changed
   remote_connections.manager_state_set hostId=remote-ssh-discovered:<redacted> nextState=connecting previousState=disconnected source=host_connection_state_changed
   remote_connections.manager_state_set hostId=remote-ssh-discovered:<redacted> nextState=connected previousState=connecting source=host_connection_state_changed
   

   Additional detail

   There also seems to be a config/state mismatch around retry count. ~/.codex/codex-app/config.json had been set to:

   {"remoteConnectionMaxRetryAttempts": 20}

   but the running app logs still showed remoteConnectionMaxRetryAttempts=2, and the global state key remote-connection-max-retry-attempts was observed back at 2 after the Settings toggle. This suggests the app may be restoring or rewriting stale in-memory/global retry settings independently of the app config, and the retry-exhausted connection can remain stuck until the user toggles auto-connect.

   Expected

   When a Remote SSH connection is in retry-exhausted/error state, Codex Desktop should provide a direct retry/reconnect path from the greyed-out project or status badge. If the underlying SSH host is reachable again, selecting the project or clicking a reconnect control should clear the stale remote_connection_retry_disabled state and create a fresh connection attempt.

   Toggling auto-connect in Settings should not be the only way to recover a healthy remote host after sleep/network loss.

   👍React with 👍1Reacted by JiMMyMatrix
4. etraut-openai

   marked Codex app crashes on stale discovered SSH host after ~/.ssh/config is emptied or host entry is removed #18974 as a duplicate of this issue on Oct 13, 2026