Skip to content

Choosing a Cutover Model

The cutover model is chosen on a selector bar at the top of the cutover cockpit (the Monitor tab). It is a two-axis choice: a granularity and an order. Together they determine how applications move and what downtime or data-loss tradeoff you accept. The default is Coordinated stop, Rolling. Changing the cutover model requires the JobCutover permission.

How many applications you switch at once.

GranularityBehaviour
Big bangSwitch all applications at once. The topic selection is locked to the whole set; the sidebar cannot deselect topics.
RollingSwitch in waves, per topic or per application. You choose which topics to include in each wave from the sidebar.

The sequence in which producers and consumers move.

OrderBehaviourTradeoff
Coordinated stopPause source producers, drain source consumers to zero lag, seed destination offsets, then start producers and consumers on the destination.Brief producer downtime for zero data loss.
Producers firstCut producers to the destination first, let existing consumers drain the source tail, and once source consumer lag is zero, cut consumers to the destination.Zero producer downtime, with transient consumer lag while draining.

When you choose Producers first, the cockpit enforces a drain gate: the consumer-cutover step stays locked until source consumer lag reaches zero. This guarantees existing consumers finish reading the source tail before they move, so no records are abandoned.

With the model chosen, proceed to Beginning Cutover.