Follow-up to the no-linger sink. At low-to-moderate throughput the sink
can produce very small batches, increasing the request rate and reducing
the scheduler's ability to build large, conflict-free batches.
Allow users to specify a minimum linger duration. The implementation
should live on the batcher side. One open question is whether batches
need to track the age of the oldest pending key to decide when to flush.
Epic: CRDB-62737
Jira issue: CRDB-64551
Follow-up to the no-linger sink. At low-to-moderate throughput the sink
can produce very small batches, increasing the request rate and reducing
the scheduler's ability to build large, conflict-free batches.
Allow users to specify a minimum linger duration. The implementation
should live on the batcher side. One open question is whether batches
need to track the age of the oldest pending key to decide when to flush.
Epic: CRDB-62737
Jira issue: CRDB-64551