Skip to main content

Fleet

Fleet view is about service liveness and manifest state. It answers "what is deployed and visible to the runtime right now?" rather than "did one run succeed?".

It tells you:

  • which services are online by reading heartbeat-derived liveness
  • which manifest_hash is active for each service
  • which capabilities the runtime currently sees from each manifest

The practical split is:

  • /api/v2/services gives the live service inventory: service name, version, SDK language and version, liveness, action count, service_sources count, and optional full manifest expansion
  • /api/v2/fleet/releases gives the released flow inventory: flow_id, version number, release sequence, release time, and a SHA-256 of the released body

liveness currently resolves to online, offline, or shutting_down from heartbeat records. A service can still exist in manifest storage while being offline, so compare both liveness and manifest state before assuming a rollout is healthy.

When debugging drift, line up three values first: released flow version, live manifest_hash, and the service's advertised capabilities such as actions, service_sources, and node support. That combination tells you whether the graph is asking for something the current fleet does not actually expose.