Remote env: "explicitly killed" ghost terminal tabs accumulate after host runtime restart (not prunable)

Version: daemon-host 1.4.146
Platform: Windows 11 client + Windows 11 remote host
Setup: Remote environment over Tailscale WebSocket (client machine connects to a remote Orca runtime on another machine).

Symptom

Terminal tabs showing Session "<worktreeId>::<path>@@<hash>" was explicitly killed persist in the workspace, reappear on reconnect, and cannot be removed by closing the tab. They accumulate over time (started with ~40; new ones keep appearing).

Root cause (observed)

The workspace session is persisted keyed by the stable environment id (workspaceSessionsByHostId["runtime:<envId>"]), but individual tabs bind to pty handles (<worktreeId>@@<hash>, serve-<uuid>).
Each time the remote runtime restarts, a new runtimeId is issued (observed across a single debugging session: c4cf3d77… → 9072f8c6…, etc.). The prior runtime's ptys are marked "explicitly killed".
Orca does not prune the tab/pane bindings whose backing pty was killed, and the terminal daemon (orca-terminal-daemon.exe) retains the dead pty sessions in memory.
Result: unclosable "explicitly killed" ghost tabs that come back on every reconnect.

Evidence

The ghost pty ids are not present in either machine's orca-data.json (full recursive search = 0 hits on both client and host) → they live only in daemon memory / runtime state, not persisted config.
Closing the tab does not clear it; it reappears after reconnect.
Restarting orca-terminal-daemon.exe on the host purges all of them at once (but kills all live terminals, and they recur on the next host runtime restart).
The runtimeId changes on every host runtime restart.

Repro

Set up a remote environment; open terminals in it from the client.
Trigger a host runtime restart: reboot / sleep→wake / Orca app restart / network (Tailscale) drop.
Reconnect → prior ptys are "explicitly killed"; their tabs remain as ghosts.
Repeat → ghosts accumulate. Closing tabs does not remove them.

Expected

On reconnect / runtime restart, prune tabs whose backing pty is dead (or transparently re-establish them), and evict dead pty sessions from the daemon — instead of leaving unclosable "explicitly killed" ghost tabs.

Workaround

Restart the terminal daemon on the host (kills all live terminals). Recurs on the next host runtime restart.

Reproduction status: env_blocked (remote env ghost terminals)

Explicitly-killed ghost tabs after remote runtime restart need Windows remote env + Tailscale WebSocket host restarts.

Not labeling cannot_repro.

Still reproduces on 1.4.167, and not only with a Windows remote host — the same symptom appears with a Linux (WSL2) remote runtime.

Environment

Client: Orca desktop 1.4.167 (Windows)
Coordinator: macOS host running orca serve 1.4.167
Remote runtime: orca-ide serve 1.4.167 on WSL2 Ubuntu, run as a systemd user service, reached over Tailscale WebSocket

Observed

The sidebar lists 31 agents, all Done, all stamped with the same relative time — the moment the remote runtime was restarted. Meanwhile the runtime is idle:

orca worktree ps --environment <env>   ->  live:0  pty:no
orca terminal list --environment <env> ->  0 terminals
remote host                            ->  no agent processes, load average 0.00

So these are purely stale list entries, not live PTYs. The CLI and the sidebar disagree: terminal list says 0, the sidebar says 31.

Where the entries actually persist

On the remote runtime, in ~/.config/orca/profiles/local-default/orca-data.json:

workspaceSession
  tabsByWorktree                    dict{1}
  terminalLayoutsByTabId            dict{33}   <-- stale
  terminalPtyIncarnationsByPaneKey  dict{33}   <-- stale
workspaceSessionsByHostId           dict{0}    <-- empty

Worth noting against the root-cause analysis in the description: workspaceSessionsByHostId is empty on this host. The stale tab state lives under the singular workspaceSession instead.

`terminal-history` is a derived artifact, not the source

~/.config/orca/terminal-history/ held 33 directories named <worktreeId>::<path>@@<hash>, 59 MB total. Deleting all of them while the runtime was running:

directory count dropped to 0 (verified)
it returned to 33 within minutes, now 400 KB — scrollback gone, directory structure restored

The directories are regenerated from the two workspaceSession maps above, so clearing history on disk does not clear the sidebar.

Restarting the runtime does not help

systemctl --user restart on the remote runtime leaves all 33 entries in place, matching the "reappear on reconnect" behaviour in the report. No CLI command or UI control removes them (see also #10718 for the missing dismiss affordance).

fix(terminal): a close the host never heard must survive the reconnect

fix(terminal): a close the host never heard must survive the reconnect (

fix(remote): stop absence and rotation from latching terminal UI state (

feat(serve-stats): surface runtime id, browser pages and unverifiable…

Cross-referencing: PR #21277 improves the dismissal of orphaned tabs when a host runtime or worktree restarts/tears down. When closing a tab whose underlying worktree selector is gone, the host resolves idempotently with selector_not_found and the renderer lifecycle locks the close intent into durable state, allowing orphaned tabs on deleted or restarted worktrees to be dismissed without entering an infinite flicker/resurrection loop.

fix(terminal): a close the host never heard must survive the reconnect (

fix(remote): stop absence and rotation from latching terminal UI state (

Fixed by #17945 (shipped in v1.4.197), which stops reconnects and rotations from holding on to tabs whose sessions are gone, plus #16752. Closing. If killed-session ghost tabs still pile up on a current build, please reopen with your Orca version.

Navigation Menu

Repository navigation

Remote env: "explicitly killed" ghost terminal tabs accumulate after host runtime restart (not prunable) #9585

Symptom

Root cause (observed)

Evidence

Repro

Expected

Workaround

Reproduction status: env_blocked (remote env ghost terminals)

Environment

Observed

Where the entries actually persist

`terminal-history` is a derived artifact, not the source

Restarting the runtime does not help

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Remote env: "explicitly killed" ghost terminal tabs accumulate after host runtime restart (not prunable) #9585

Description

Symptom

Root cause (observed)

Evidence

Repro

Expected

Workaround

Activity

nwparker commented on Jul 27, 2026

Reproduction status: env_blocked (remote env ghost terminals)

Karnian commented on Aug 4, 2026

Environment

Observed

Where the entries actually persist

terminal-history is a derived artifact, not the source

Restarting the runtime does not help

LesleyMurfin commented on Sep 17, 2026

nwparker commented on Oct 26, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

`terminal-history` is a derived artifact, not the source