mirror of
https://github.com/n8n-io/n8n.git
synced 2026-10-17 21:49:36 +02:00
One turn per durable agent thread at a time, on every surface: preview chat, published chat integrations, background wake, task runs, HITL resume. - AgentThreadTurnCoordinator: bounded FIFO per thread and main (50 waiters), distributed agent-thread-turn lease, abort-aware waiting, queue-full error. - agent_execution.activeThreadId claim with a unique partial index as the fail-closed fence; heartbeat drops the claim when the row is no longer ours. - Chat SDK runs with concurrency 'concurrent'; the coordinator orders turns. - Preview chat surfaces the queue-full error; channel threads post a notice. Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|---|---|---|
| .. | ||
| generated | ||
| db.md | ||