Epic: complete self-host observability — host/container/Redis/Qdrant/stack self-monitoring + remaining exporter gaps

[JSONbored]

Problem: Following the Reviews & PRs dashboard / AI-usage dashboard consolidation audit (#5234, #5247, #5259, #5267, #5302, #5310, #5342, #5349), a full sweep of every integrated backend found several pieces of the self-host stack have zero Prometheus/Grafana visibility today:

• No host-level metrics (CPU/memory/disk/network of the machine itself) — docker stats-only, invisible in Grafana.
• No per-container resource metrics (which container is actually using the CPU/memory quota it's been given).
• Redis — always-on, required — has zero metrics despite every other core backend (Postgres, GPU, backups) already having one.
• Qdrant already exposes a native Prometheus /metrics endpoint (:6333/metrics, confirmed live) that is simply never scraped — a free win.
• Prometheus, Alertmanager, Loki, Tempo, and Grafana all expose their own native /metrics (confirmed live against edge-nl-01) but none are self-scraped — no visibility into whether the observability stack itself is healthy, ingesting, or falling behind.
• The OTEL collector's own internal telemetry port (:8888) isn't enabled/scraped — no visibility into dropped spans/metrics or collector health.
• REES (review-enrichment/, first-party TypeScript) has no /metrics endpoint at all (confirmed: GET /metrics → 404) — analyzer success/failure/duration/timeout rates are invisible.
• browserless (visual-review screenshot capture) exposes a /metrics endpoint, but it returns JSON, not Prometheus text format (confirmed live) — not scrapeable as-is.
• Ollama has no native Prometheus endpoint and no official community exporter exists (confirmed: GET /metrics → 404) — this stays a documented, unactionable limitation, not a sub-issue below.
• Sentry (sentry.io) is currently only reachable via a plain hyperlink from the Grafana resource-hub dashboard — no in-Grafana visualization of error/issue data (distinct from Epic: consolidate Sentry + Alertmanager alerting into one real-time notification path #5007, which is about alerting/notification routing, not visualization).

Area: Self-host observability (docker-compose.yml, prometheus/, grafana/dashboards/, review-enrichment/)

Goal (verbatim from the requesting conversation): "The whole stack should be able to be properly observed/logged/metrics and analysis directly from grafana, so observability is built in and complete from end to end, ideally."

Deliverables — tracked as independently-shippable sub-issues:

• feat(observability): add node-exporter, cAdvisor, redis_exporter, scrape Qdrant + self-monitor the observability stack #5366 — Add node-exporter + cAdvisor + redis_exporter, scrape Qdrant's existing endpoint, self-monitor Prometheus/Alertmanager/Loki/Tempo/Grafana, enable + scrape the OTEL collector's internal telemetry — one new "Infra & stack health" Grafana dashboard. Done via feat(observability): host/container/Redis/Qdrant metrics + stack self-monitoring #5380.
• feat(observability): instrument REES (review-enrichment service) with real Prometheus metrics #5367 — Instrument REES with real Prometheus metrics (request/analyzer success/failure/duration/timeout counters). Done via feat(observability): instrument REES + its main-app client with real Prometheus metrics #5398 + feat(observability): scrape REES metrics + add a dedicated REES Grafana dashboard #5412.
• feat(observability): translate browserless's JSON /metrics into real Prometheus text format #5368 — Translate browserless's JSON /metrics into real Prometheus text (a small sidecar, mirroring the existing backup-exporter pattern). Done via feat(observability): translate browserless's JSON metrics into Prometheus text #5436.
• feat(observability): add a Grafana Sentry datasource for in-Grafana error/issue visibility #5369 — Add a Grafana Sentry datasource so error/issue data is queryable/visualizable directly in Grafana, not just linked out to. Done via feat(observability): add a Grafana Sentry datasource for in-Grafana issue visualization #5453.

Acceptance criteria (epic-level):

• Every backend this stack runs (app, Postgres, Redis, Qdrant, GPU, backups, REES, browserless, Ollama-documented-exception) has either real Prometheus metrics in Grafana or an explicit, documented reason it can't.
• The observability stack itself (Prometheus/Alertmanager/Loki/Tempo/Grafana/OTEL collector) is self-monitored, not just monitoring everything else.
• A self-hoster can answer "is any part of my stack using resources sub-optimally or behaving abnormally" from Grafana alone, without shelling into the host.

Boundaries:

• Ollama gets no exporter built (none exists upstream) — documented as a known, permanent gap, not tracked as a sub-issue.
• Does not include Sentry/Alertmanager notification-routing consolidation — that's Epic: consolidate Sentry + Alertmanager alerting into one real-time notification path #5007, a distinct and complementary effort (alerting vs. visualization).

[linear-code]

JSO-3802