old-raft-rs
old-raft-rs copied to clipboard
Stale Reads
As of today the master will immediately satisfy client queries, which may result in stale reads during failover scenarios. Diego's thesis goes into detail about the cause and possible solutions to this issue in section 6.4. We should come up with a plan for how to tackle this.
There's this one: https://www.cs.cmu.edu/~dga/papers/leases-socc2014.pdf I had looked at it before and it seemed promising after some adjustments.
One cute little hack for now is to have the request respond with the time since a heartbeat. Then we can just check this value to see if this is consistent or merely timeline consistent