Currently, the writer batcher check the last entry in the batch and wait that entry to meet secondaryDelaySecs before we return the batch. This could cause some unexpected delay when the timespan in one batch is large, though this will only happen for one batch. Also, when the batcher is interrupted, we can return a batch that is not meeting the secondaryDelaySecs requirement.

We can make the batcher smarter to break a batch and only return the batches satisfying secondaryDelaySecs.

Since two problems are minor, we decided to track it in a separate Improvement ticket.