Replication State Should Send Requests To Highest Validators

[samliok]

Replication requests need to be sent to validators of the highest epoch(rather than the current epoch). Ex. say we receive a future finalization for our current epoch. We currently distribute requests to the signers of the finalization. However, what if the epoch has advanced and those signers are offline and no longer validators?

[yacovm]

If we're a validator, then replication requests can be sent to the current epoch. Recall that the Epoch object can anyway only handle messages for the same epoch.

If we're a non-validator, then there might be a case where we receive a finalization and then the epoch has advanced and those signers are no longer available. However, this is a very rare corner case, as it means that the validator set has changed by at least f+1 nodes (if it has changed by f nodes or less, then there is at least one node that should be able to respond, right?) right after we have received the finalization.

While this may happen, I don't think this case is so bad, because eventually we will receive a finalization from the last epoch which is stable and will be able to replicate to it.

[samliok]

we will receive a finalization from the last epoch which is stable and will be able to replicate to it.

but we wont have the validator set, so how can we verify it is valid? In the upcoming approach, if a validator receives a finalization from a higher epoch it will request the sealing block that created that epoch. It will keep doing this until it gets to the current epoch.

The issue is, when it receives the last finalization for that epoch the signers may be old. And we will request from those signers

[yacovm]

oh I see what you're saying. So, I think the sensible solution would be for the replication mechanism to check if there are at least f+1 signers among the latest validator set from the P-chain, and if not, then to send the requests to nodes from the latest validator set.