Consumer lag spikes during event cutovers are not automatically a stop signal. Backfills, replay storms, and temporary dual-publish windows can create lag that recovers within your service-level budget.
Define lag budgets before the change: how many seconds of lag your consumers tolerate, and for how long. Pair lag charts with publish throughput and error rates so you can distinguish catch-up from stuck apply.
If lag climbs while apply rate falls, pause the migration and investigate. If lag climbs while apply rate stays healthy, you may simply be watching a heavy but finite backlog—still communicate, but do not treat every bump as a rollback.
