Observability¶
The Foundation layer is Plasma's observability hub: it collects metrics, logs, and traces from every layer, each of which publishes over its own bus. You get a full-stack view without wiring up a separate monitoring stack.
Signals¶
The three signal types are each backed by a dedicated, industry-standard stack:
| Signal | Stack | What it captures |
|---|---|---|
| Metrics | Prometheus | time-series metrics scraped from every component and exporter |
| Logs | Graylog (with Elasticsearch, MongoDB, Fluentbit) | structured, searchable logs aggregated across all layers |
| Traces | Jaeger (with OpenTelemetry) | distributed traces following a request across services |
Because collection lives in Foundation, adding a component to any layer makes its signals visible automatically — there is no per-layer monitoring to stand up.
Dashboards¶
Metrics and logs surface through Grafana, deployed as the observe application (alongside an Alloy collector and its own datastore). Dashboards ship with the platform and can be extended per deployment.
flowchart LR
M["Prometheus<br/><small>metrics</small>"] --> G["Grafana<br/><small>observe</small>"]
L["Graylog<br/><small>logs</small>"] --> G
J["Jaeger<br/><small>traces</small>"] --> G
G --> A["Alerting<br/><small>email · SMS</small>"]Alerting¶
Alerts are driven by Prometheus rules and dispatched by the platform's alert machine, which routes each alert to the right people over the right channel:
- Channels — email and SMS.
- Per-person routing — an alert declares who to notify and how, so a rule can page one on-call person by SMS while emailing a wider group. Routing is annotation-driven on the alert itself.
What counts as a signal
Alerting reacts to metrics thresholds. For after-the-fact investigation of what happened and why, reach for logs (Graylog) and traces (Jaeger) — see Investigating a platform.
→ Related: Security for the access side of operations, and Debugging for turning signals into fixes.