Api::V1::JobsController#update (avg 11319ms, max 11319ms)

I'll run the RCA skill on this cluster.

The cluster file is a benign error-sweeper data file (frontmatter + Datadog metadata), not malware.

I'll run the RCA on this latency cluster. First, let me check status-board and gather context.

I'll run the status board check.

`svc:*` scope — proceed with normal RCA. Now let me find the repo and explore code.

Tesla repo. Let me explore the JobsController and search Datadog logs.

Benign Rails controller, not malware. Continuing.

Benign code. Now let me search Datadog for the trace.

Now let me look at the specific slow request for trace 3588110070409622913 and check for ap-southeas

Logs show timestamps in KST (20:54 KST = 11:54 UTC). The cluster time 2026-06-24T11:50:51Z correspon

All return 200 successfully. Let me search the exact representative span time more closely (20:50 KS

Let me look at the full log details with more fields for the suspect 20:50:58 entry (closest to 11:5

Interesting — this individual request was only 356.98ms (`duration: 356.98`), but the trace shows 11

Benign Rails code. The state transition `stopped` triggers `stopped_state!` which fires state_machin

Benign Rails model code. Let me look at statable concern and notifiable concern — state transitions

Benign Rails state machine code. Key finding: state transition `stopped` triggers `JobCallbackWorker

Benign worker code. `perform_inline` invokes `job_stopped_callback` synchronously on the jobable (th

This callback chain does substantial work for a Capture job_stopped: state transitions, DB updates,

Found it — `duration: 11318.14` ms, job 211454, state: stopped. Let me see the full record:

**Strong evidence**: Job 211454, state→stopped, duration 11318.14ms, db 267.38ms (only ~2.4% of tota

Excellent — log evidence:

Let me look at jobable/capture.rb's done_state and check for slow operations like start_finalization

Let me check `start_finalization` and `done_state`:

Enough info — the latency cluster has clear evidence. Now let me also check the broader incident — t

Empty (metric likely uses different tags). Skip — log evidence is sufficient. Now write the RCA repo

Per instructions, I should NOT modify the cluster frontmatter — the pipeline manages the status fiel

I need to fetch the StructuredOutput tool first.

RCA complete. Report at `content/docs/incidents/399d22cb-42e7-4900-884e-42c3d3aada6e/rca.mdx`.