this way we wouldn't need to get a sudden rise in read latency after flush.
or is something similar is already i there?

Reply via email to