Skip to content
Abada Platform

Telemetry

Telemetry is off by default. Workflow readiness and transactions never depend on a collector. OpenTelemetry is optional diagnostics only: it is not the Insight Engine’s cursor and cannot authorize a proposal (see The Insight Engine).

Disabled

Section titled “Disabled”
ABADA_TELEMETRY_ENABLED=false
OTEL_SDK_DISABLED=true

Start only the core profile. No exporter or collector is created; structured container logs and local rolling JSON logs remain available.

Bundled stack

Section titled “Bundled stack”
Terminal window
./release/abada-platform up dev --telemetry

The overlay starts OpenTelemetry Collector, Prometheus, Jaeger, Loki, Alloy and Grafana. Grafana is loopback-only at http://127.0.0.1:3000; production requires a non-placeholder GRAFANA_ADMIN_PASSWORD of at least 16 characters. Datasources and dashboards are provisioned automatically. The internal telemetry-health container checks every bundled backend; it does not gate the engine health endpoint.

Alloy tails only the named engine_logs volume, persists offsets, and runs with anonymous usage reporting disabled. It has no Docker socket access.

The one-shot jaeger-volume-init container prepares the persistent Badger volume for Jaeger’s non-root UID and exits before Jaeger starts. It has no network access and is expected to show as Exited (0) after initialization.

External OTLP

Section titled “External OTLP”

Do not add compose.telemetry.yaml. Set:

ABADA_TELEMETRY_ENABLED=true
ABADA_TELEMETRY_OTLP_ENDPOINT=https://collector.example.net:4318
OTEL_SDK_DISABLED=false
ABADA_TRACING_SAMPLING_PROBABILITY=0.1

The endpoint must be an HTTP(S) OTLP base URL; Abada appends /v1/traces and /v1/metrics. Export uses bounded batches, queueing and five-second timeouts. Collector failure may lose telemetry but cannot roll back a workflow command.

Verify engine readiness separately from telemetry configuration:

Terminal window
curl -fsS http://api.localhost/api/actuator/health/readiness
curl -fsS http://api.localhost/api/actuator/health/telemetryExport

For a failure drill, stop otel-collector, complete a task, confirm the task commits, then restart the collector. At-least-once workflow behavior does not imply lossless telemetry delivery.