[FEATURE]: Session heartbeat for multi-session liveness detection

Feature has not been suggested before.

I have verified this feature I am about to request has not been suggested before.

Describe the enhancement you want to request

OpenCode advertises multi-session ("Start multiple agents in parallel on the same project") but provides no way for sessions to detect each other. If a parallel session crashes or hangs, its resources (temp dirs, ports, files) are never released, and other sessions have no way to know.

Add two fields to the existing session table:

heartbeat_at — timestamp, written every 30s while session is alive
peer_id — unique per process, survives session restarts

Extend the existing compaction/prune to skip sessions with heartbeat_at older than 5 min as stale.

Why core: multi-session is a public feature, but without heartbeat it is unsafe. External modules (MCP, SLM) can read the DB for coordination; core just tracks liveness.

Scope: ~10 lines across 3 existing files (session/schema.ts, session/lifecycle.ts, session/compaction.ts). No new tables, no new config, no new dependencies.

This issue might be a duplicate of existing issues. Please check:

Session heartbeat for multi-session liveness detection #27742: Same title and content — "Session heartbeat for multi-session liveness detection" (closed, same author). If you re-opened this because Session heartbeat for multi-session liveness detection #27742 was closed incorrectly, please note that in the issue or ask a maintainer to reopen the original.

This is not a duplicate — #27742 was auto-closed for title format (missing [FEATURE]: prefix). This is the corrected version with the proper title format. Please reopen #27742 or keep this one. Thank you.

this is the exact problem i ended up solving in repowire, and the heartbeat_at + peer_id shape you propose is almost the same design i landed on, so a couple of notes from running it:

keying on a stable peer_id rather than a path/name is the right call. parallel sessions on the same project collide on anything path-derived, so liveness and routing both need the id
on the heartbeat: i deliberately avoided a polling timer. instead liveness is reconciled lazily, when a request next needs the registry it runs a repair pass and evicts stale peers. nothing runs on a schedule, the work piggybacks on real requests. cheaper, and no background loop to babysit

the crashed-session-leaks-resources case is real and ghost eviction handles it well once you have the id + a reconciliation trigger. happy to share specifics if useful, this is squarely the thing the daemon does.

To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.

Navigation Menu

Repository navigation

[FEATURE]: Session heartbeat for multi-session liveness detection #27759

Feature has not been suggested before.

Describe the enhancement you want to request

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[FEATURE]: Session heartbeat for multi-session liveness detection #27759

Description

Feature has not been suggested before.

Describe the enhancement you want to request

Activity

github-actions commented on May 15, 2026

MaximKh commented on May 15, 2026

prassanna-ravishankar commented on Jun 21, 2026

github-actions commented on Aug 20, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions