Cassandra 2.1
@jeffjirsa how does this look?
As discussed, I think it's a reasonable change. It's unusual to write data to a table with varying TTLs, but it seems relatively low risk.
My primary concern is that many people are going to expect the behavior to match other compaction strategies - for that reason, I'm interested in perhaps creating a toggle to allow either this behavior OR the old/established behavior.
Do you have a strong opinion on that?
Sounds completely reasonable. I'll make the change.
is this irrelevant now?
@jippi - This is a patch that optimizes TWCS for using varied TTLs in the same Table, and specifically is useful only if you're pushing C*/TWCS so hard that you need to aggressively reclaim disk space as you cross various TTL boundaries. It's not that it's irrelevant, but it's an edge case that doesn't generally apply to most people. I'll merge it when it's cleaned up a bit (or when I have time to clean it up a bit).
@jeffjirsa i guess this can be closed?