Federation behavior
Understand ownership and availability when connecting runtime nodes.
Each runtime owns an independent coordination store and executes its own jobs. A peer can forward requests to a job's owner and retain cached summaries. All workers of a job remain on that owner.
When a peer cannot reach an owner, cached summaries may remain visible, but live mutation, streaming, logs, and artifact requests can fail with 503 MN_NODE_UNAVAILABLE. Reconnect to the owner; a second node does not automatically take over its active jobs.
Use Connect runtime nodes for setup and verification, Monitor for execution status, and Reliability for recovery boundaries.