BuzzbeeSCD/openchamber
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: settle busy sessions after managed OpenCode restart (#3002)

Browse Source 
* fix: reconcile busy sessions after managed OpenCode restart

Forced health-check restarts previously rebound the event stream without
settling in-flight turns, so sessions stayed busy with no terminal state.
Interrupt those sessions, classify health failures, and retain bounded
process diagnostics for post-restart diagnosis.

Fixes #2943

Co-authored-by: serkraser <serkraser@gmail.com>

* fix: surface interrupted chats after OpenCode restart

Complete unfinished assistant turns as aborted once the session is
authoritatively idle, and show a persistent toast so users can continue
instead of remaining silently stranded.

Fixes #2943

Co-authored-by: serkraser <serkraser@gmail.com>

* fix: redact Basic auth credentials in restart diagnostics

The key/value sanitizer stopped at whitespace, so Authorization: Basic
credentials survived in stderr tails and health snapshots. Redact the
scheme token before that rule runs.

Co-authored-by: serkraser <serkraser@gmail.com>
This commit is contained in:
Serhii Dziupin
2026-08-19 11:53:30 +03:00
committed by GitHub
parent 13a45aa5a5
commit 14d7a0ca9b
24 changed files with 806 additions and 80 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/web/server/lib/opencode/DOCUMENTATION.md
+4 -1
View File
@@ -111,12 +111,13 @@ This module provides OpenCode server integration utilities for the web server ru
- `markSessionUnviewed(sessionId, clientId)`
- `markUserMessageSent(sessionId)`
- `resetAllSessionActivityToIdle()`
- `interruptBusySessionsAfterRestart()`: settles every session whose authoritative status is `busy`/`retry` or whose activity phase is still busy, broadcasts `openchamber:session-status` idle plus an OpenCode-shaped `session.error`, resets leftover activity/cooldowns, and returns the interrupted session IDs in stable order.
- `dispose()`
The runtime maintains active-session count incrementally from idempotent activity phase transitions. Upstream stall-timeout and lifecycle health checks read it in O(1); the hourly cleanup removes activity phases older than 24 hours without broadcasting synthetic state transitions. Snapshot generation remains reserved for the session-activity API.
## Public exports (lifecycle.js)
- `createOpenCodeLifecycleRuntime(dependencies)`: creates lifecycle runtime for managed/external OpenCode process orchestration. The optional `onOpenCodeRestarted` dependency (default `null`) is fired after a successful managed restart; `index.js` wires it to `messageStreamRuntime.rebindUpstream()` so event-stream readers rebind to the possibly-new port (a restart can land on a new port while an orphaned process keeps the old one, which would otherwise leave the chat UI silent — issue #2638).
- `createOpenCodeLifecycleRuntime(dependencies)`: creates lifecycle runtime for managed/external OpenCode process orchestration. The optional `onOpenCodeRestarted` dependency (default `null`) is fired after a successful managed restart. `index.js` rebinds event-stream readers to the possibly-new port (#2638), then calls `interruptBusySessionsAfterRestart()` and broadcasts one `opencode-restart-interrupted` UI notification when interrupted turns exist (#2943).
- Returned API:
- `startOpenCode()`
- `restartOpenCode()`
@@ -147,6 +148,8 @@ macOS `say` voice enumeration starts concurrently with server composition. The s
Transport-triggered health checks share the periodic monitor's failure accounting interval. Rapid WS reconnect callbacks therefore cannot exhaust the managed-process restart threshold using one cached unhealthy result; an exited managed process still restarts immediately.
Managed health failures are classified as `timeout`, `connection_refused`, `connection_reset`, `invalid_response`, or `error`. The lifecycle retains the latest counted failure with a bounded detail string and source. Managed process wrappers continue capturing a sanitized, bounded stderr tail after readiness and retain exit code/signal. Before replacing a managed process, lifecycle snapshots the reason, latest health failure, process diagnostics/aliveness, busy-session count, and timestamp into `lastOpenCodeRestartDiagnostics`; successful startup does not clear this snapshot, and `/health` exposes it for post-restart diagnosis without process environment or credentials.
## Public exports (env-runtime.js)
- `createOpenCodeEnvRuntime(dependencies)`: creates runtime that owns OpenCode CLI environment and binary discovery state.
- OpenCode CLI resolution order is persisted settings, environment overrides, bundled Desktop CLI when available, PATH, known install locations, then platform shell discovery.