How long a mini-segment can stay open. How much data needs replay from Kafka when a fail-over happens. More flushing means minisegments are likely to be smaller. Less frequent flushing means more replay in the event of a crash. See BLOCKS_PER_SEGMENT for an example of this variable and more information.

This value is also added to DELETES_DELAY_MILLIS to determine when a redaction rewrite begins. The rewrite starts after the sum of both values have elapsed since the redaction was requested, to allow any minisegments to be merged first.