# Scheduler Worker – Observability & Runbook

## Purpose
Monitor planner and runner health for the Scheduler Worker (Sprint 16 telemetry). The new .NET meters surface queue throughput, latency, backlog, and delta severities so operators can detect stalled runs before rescan SLAs slip.

> **Grafana note:** Import `docs/ops/scheduler-worker-grafana-dashboard.json` into the Prometheus-backed Grafana stack that scrapes the OpenTelemetry Collector.

---

## Key metrics

| Metric | Use case | Suggested query |
| --- | --- | --- |
| `scheduler_planner_runs_total{status}` | Planner throughput & failure ratio | `sum by (status) (rate(scheduler_planner_runs_total[5m]))` |
| `scheduler_planner_latency_seconds_bucket` | Planning latency (p95 / p99) | `histogram_quantile(0.95, sum by (le) (rate(scheduler_planner_latency_seconds_bucket[5m])))` |
| `scheduler_runner_segments_total{status}` | Runner success vs retries | `sum by (status) (rate(scheduler_runner_segments_total[5m]))` |
| `scheduler_runner_delta_{critical,high,total}` | Newly-detected findings | `sum(rate(scheduler_runner_delta_critical_total[5m]))` |
| `scheduler_runner_backlog{scheduleId}` | Remaining digests awaiting runner | `max by (scheduleId) (scheduler_runner_backlog)` |
| `scheduler_runs_active{mode}` | Active runs in-flight | `sum(scheduler_runs_active)` |

Reference queries power the bundled Grafana dashboard panels. Use the `mode` template variable to focus on `analysisOnly` versus `contentRefresh` schedules.

---

## Grafana dashboard

1. Import `docs/ops/scheduler-worker-grafana-dashboard.json` (UID `scheduler-worker-observability`).
2. Point the `datasource` variable to the Prometheus instance scraping the collector. Optional: pin the `mode` variable to a specific schedule mode.
3. Panels included:
   - **Planner Runs per Status** – visualises success vs failure ratio.
   - **Planner Latency P95** – highlights degradations in ImpactIndex or Mongo lookups.
   - **Runner Segments per Status** – shows retry pressure and queue health.
   - **New Findings per Severity** – rolls up delta counters (critical/high/total).
   - **Runner Backlog by Schedule** – tabulates outstanding digests per schedule.
   - **Active Runs** – stat panel showing the current number of in-flight runs.

Capture screenshots once Grafana provisioning completes and store them under `docs/assets/dashboards/` (pending automation ticket OBS-157).

---

## Prometheus alerts

Import `docs/ops/scheduler-worker-prometheus-rules.yaml` into your Prometheus rule configuration. The bundle defines:

- **SchedulerPlannerFailuresHigh** – 5%+ of planner runs failed for 10 minutes. Page SRE.
- **SchedulerPlannerLatencyHigh** – planner p95 latency remains above 45 s for 10 minutes. Investigate ImpactIndex, Mongo, and Feedser/Vexer event queues.
- **SchedulerRunnerBacklogGrowing** – backlog exceeded 500 images for 15 minutes. Inspect runner workers, Scanner availability, and rate limiting.
- **SchedulerRunStuck** – active run count stayed flat for 30 minutes while remaining non-zero. Check stuck segments, expired leases, and scanner retries.

Hook these alerts into the existing Observability notification pathway (`observability-pager` routing key) and ensure `service=scheduler-worker` is mapped to the on-call rotation.

---

## Runbook snapshot

1. **Planner failure/latency:**
   - Check Planner logs for ImpactIndex or Mongo exceptions.
   - Verify Feedser/Vexer webhook health; requeue events if necessary.
   - If planner is overwhelmed, temporarily reduce schedule parallelism via `stella scheduler schedule update`.
2. **Runner backlog spike:**
   - Confirm Scanner WebService health (`/healthz`).
   - Inspect runner queue for stuck segments; consider increasing runner workers or scaling scanner capacity.
   - Review rate limits (schedule limits, ImpactIndex throughput) before changing global throttles.
3. **Stuck runs:**
   - Use `stella scheduler runs list --state running` to identify affected runs.
   - Drill into Grafana panel “Runner Backlog by Schedule” to see offending schedule IDs.
   - If a segment will not progress, use `stella scheduler segments release --segment <id>` to force retry after resolving root cause.
4. **Unexpected critical deltas:**
   - Correlate `scheduler_runner_delta_critical_total` spikes with Notify events (`scheduler.rescan.delta`).
   - Pivot to Scanner report links for impacted digests and confirm they match upstream advisories/policies.

Document incidents and mitigation in `ops/runbooks/INCIDENT_LOG.md` (per SRE policy) and attach Grafana screenshots for post-mortems.

---

## Checklist

- [ ] Grafana dashboard imported and wired to Prometheus datasource.
- [ ] Prometheus alert rules deployed (see above).
- [ ] Runbook linked from on-call rotation portal.
- [ ] Observability Guild sign-off captured for Sprint 16 telemetry (OWNER: @obs-guild).