Matt Reiferson

Hmmm, I do vaguely remember discussing this and being frustrated about the "backwards compatibility".

This is a good sign so far: 

Updated description, let me know if any of this sounds insane @jehiah. /cc @stephensearles @judwhite

**depth, `FIN` accounting, retention windows** The current plan is for each topic to maintain a `RangeSet` (thanks @rolyatmax). A `RangeSet` maintains a list of contiguous ranges of message IDs. For...

@Dieterbe working on correctness right now :grin:. Performance should be better.

@Dieterbe that already is the case in both the current production implementation (IF the message overflowed to disk) and this PR. If it were not the case there would be...

Sorry, I misunderstood what you meant by "sync" - no, there is no mechanism to deliver fsync acks to producers. The complexity doesn't feel like something that would be worth...

NOTE: there is still back-pressure from the periodic fsync because `DiskQueue` cannot make progress during a sync and individual writes are synchronized with the `PUB`.

1. No, not exactly. On a per-node basis, despite the WAL itself being FIFO, there are still requeues that will be redelivered based on their expiration. And, across a cluster...

(2) It's already required to be able to seek to a certain index