StephanEwen commented on pull request #8751: URL: https://github.com/apache/flink/pull/8751#issuecomment-804335647
@carp84 The checkpointing from FLIP-158 is a specific state backend that is used to wrap another state backend. So with that change, we can use the existing RocksDB incremental snapshots and the RocksDB log-based snapshots. We need to see in practice how much it costs to replay one write buffer from the log. With a default write buffer size, that would be around 60 MB. I would assume that doing it in one batch insert would be quite fast, given that there is no processing or serialization overhead (the log just stored bytes as well). -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
