You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Remote terminal multiplexer socket is dropped on last-stream close, so every cold-park reveal pays a fresh WebSocket + E2EE handshake #19871
When the last remote terminal stream for a paired-runtime environment closes, closeIfIdle() in src/renderer/src/runtime/remote-runtime-terminal-multiplexer-base.ts:201 tears down the environment's entireterminal.multiplex WebSocket subscription, not just the stream. Because tabs cold-park at 30s and terminal.multiplex is explicitly excluded from shared-control routing (shouldRouteSubscriptionBySupport in src/main/ipc/runtime-environment-support-routing.ts returns false for terminal.multiplex), re-entering any parked tab for that environment always pays a fresh WebSocket connect plus a full E2EE handshake before the pane can attach.
Impact
Routine tab switching against a remote Orca server is slower than it needs to be: park every pane for an environment (or let them cold-park), then reveal one, and the multiplexer must be re-established from scratch. This is a latency/UX cost on the exact "switch tabs on a remote host" flow, and it compounds with the reconnect-churn work (#19647).
Evidence
Measured in the tab-switch e2e rig (tests/e2e/paired-remote-terminal-tab-switch-reconnect.spec.ts): after parking all panes and closing the last stream, the census shows transportUnsubscribeCount incrementing — the multiplexer socket is released — and the next reveal pays a fresh transportSubscribeCount.
Suggested direction
Consider keeping the terminal.multiplex transport warm for a bounded idle window (or routing it through shared control) so a cold-park → reveal cycle re-attaches without a new socket + E2EE handshake. Not fixed here — filing separately from the #19647 writer fix.
Closing — this was filed by an automated review pass that should not have opened GitHub issues. The finding is real and is retained in our internal report; it is not lost by closing here.
No action needed from anyone. Reopen if you want it tracked publicly.
Activity
nwparker commentedon Sep 10, 2026
Closing — this was filed by an automated review pass that should not have opened GitHub issues. The finding is real and is retained in our internal report; it is not lost by closing here.
No action needed from anyone. Reopen if you want it tracked publicly.
fix(runtime): keep a live host verdict when a status probe can't reac…
fix(runtime): keep a live host verdict when a status probe can't reac…