Lifecycle actions target a pipeline and produce both a new desired-state revision
and a durable convergence operation.
Pause
Pause gracefully stops the supervised engine and retains resumable state. The
deployment stays paused across gateway outages and pod restarts.
Resume
Resume selects running. The supervisor starts the engine with the selected
configuration and waits for local readiness. After a pause, the connector resumes
from retained state, subject to the source’s own retention window.
Drain
Drain stops the engine and invokes connector-specific cleanup:
Returning a drained database-source deployment to running can therefore require a
fresh bootstrap. Kafka replay starts from the retained group offset unless an
operator resets that group externally.
Drain is not a long pause. It can invalidate the state required for lossless
resume. Confirm source-specific behavior before using it in production.
Reconcile and rebootstrap
Reconcile removes destination documents no longer represented by the source for
connectors that advertise the capability. Rebootstrap clears supported local
cursor/join state and forces a fresh snapshot. Kafka is the exception: its cursor
is the broker-side consumer-group offset, so Fleet reports the limitation and the
operator must reset that group externally. These are explicit, auditable
operations and never generic remote shell commands.
Inspect progress with:
