Closed HaloFour closed 5 years ago
Oof, both PRs have changes to the StateSup. In the other, it was changing the foldLeft
to a foreach
, and this one completely subsumes that but is in conflict. You'll need to resolve it, I can make the easy fix but can't push it to your branch. :|
I expected this, I'll merge from master
We have a number of XtvAPI nodes that don't populate an in-memory brain but instead exist to facilitate Sirius. This includes the ingest cluster which handle PAXOS consensus as well as nodes that are used strictly for Sirius catchup. We pass no-op implementations of
RequestHandler
but Sirius still goes through a full replay of the Uberstore which can add 20-30 minutes to the time it takes to bootstrap the service. As far as I could tell that replay only serves to populate the brain and other housekeeping such as index repair happen beforehand, so it should be safe to skip the log replay.I propose a special
case object
implementation ofRequestHandler
that Sirius would recognize and if that is the instance of theRequestHandler
passed to SiriusFactory then log replay will be skipped.