Memtable flushing HUGE logical fallacy

[mrsladoje]

We flush N memtables at once concurrently using a worker pool. Each memtable gets the according index for the sstable to flush into, but a "fresher" memtable might flush before an "older" one, causing data loss.

Instead, we should only flush the oldest memtable when we fill up all N memtables, and do so in a separate goroutine to allow for it to be done in the background.