This thought may have come up as a side effect of the old way of calculating the current batch size. What I was thinking about is if half way through a batch size, the scheduler fails over and the update has to continue from where it left off when the next leader is chosen.
E.g: We have batch sizes 10 and 30. Half way through 30 (25 instances updated), the scheduler fails over. With the new way of calculating which batch size we're on on the fly, I don't think it's an issue anymore. [ Full content available at: https://github.com/apache/aurora/pull/37 ] This message was relayed via gitbox.apache.org for [email protected]
