The recovery-pass investigation reported a slow GET /api/v1/chat/session-activity endpoint that needs separate profiling. This issue tracks that path separately from the 15-second interrupted-turn candidate scan optimization.
Available live observations (reported by the operator): a 210-second sample contained recurring 2–3 second event-loop stalls roughly every 19 seconds and a slow database span. PostgreSQL sampled a bulk chat-turn status read; the telemetry hash has not been conclusively attributed to a caller. The last reported Core revision was 831d0d0c646a31c184bca7c868034034484c9308.
Collect a fresh authenticated endpoint trace and SQL/CPU timing with caller context, distinguish request work from background recovery, and measure latency and loop lag under the same session/transcript workload. Do not assume the shared SQL fingerprint uniquely identifies the endpoint or the recovery scan. Preserve durable activity/recovery semantics and follow AGENTS.md's independent diagnosis, focused test-selection and product-quality gates before a repair.
Live-host access was unavailable in the cloud investigation; no live measurements or endpoint fix are claimed here.
Activity
0No activity yet.