The metrics service

The metrics service is the observability plane of a cell. It collects telemetry, computes metrics, raises alerts, synthesizes health from raw signals, and turns metered work into cost analytics. It is one of the two services (with admin) whose whole job is to observe the others.

Feeds: Platform Foundation.

What it owns

• Telemetry: collecting raw signals from every service.
• Metrics: deriving metrics from telemetry.
• Alerting: raising and routing alerts on thresholds.
• Health synthesis: turning many signals into a single health picture.
• Cost analytics: attributing metered work to a cost view.

Sub-modules (public names)

Sub-module	What it does
Collectors	Gather telemetry from every service
Exporters	Expose metrics for scraping and roll-up
Alerting	Threshold evaluation and alert routing
Awareness	Health synthesis from raw signals
FinOps	Cost analytics over metered work

APIs and surfaces

The metrics service exposes an HTTP surface: a loopback clear port for intra-cell callers and a mutual-TLS port advertised on the fabric so a peer can roll up health across cells. Its gRPC surface exists but ships disabled by default, metrics roll-up rides the HTTP surface. External access to any metrics view is through the node front door and the Observability & metering surfaces. See the Ports & protocols reference.

Workers

Background workers cover telemetry collection, metric derivation, alert evaluation on a fixed cadence, health synthesis, and cost roll-up. Each runs per cell.

The observability flow

Figure: the observability flow. Metrics (filled, core) feed alerting (amber : it raises a held/attention state), health synthesis (drawn verified as the single synthesized picture), and cost analytics; synthesized health rolls up to peers over mutual TLS. Ceiling: metering is attributed per cell; this shows the collection path, not the metering ledger's verifiability guarantees.

Table equivalent

From	Relationship	To	State
The eight other services	emit telemetry to	Collectors	neutral
Collectors	derive	Metrics	core
Metrics	raise	Alerting	attention
Metrics	synthesize into	Health	verified picture
Metrics	attribute into	Cost analytics	neutral
Health	rolls up over mTLS to	Peer cells	edge (cross-cell)

In the deployment matrix

metrics runs identically everywhere. Single-cell, roll-up listeners are idle and observability is wholly local. Multi-cell, health synthesizes at the node level across cells. Sovereign / air-gapped, everything is metered, attributed, and alertable locally with no external dependency, and an unreachable peer is an absent edge in the roll-up.

Related topics

• Observability & metering, the customer-facing surfaces.
• Ports & protocols reference.