question about the sync of follower with older epoch

[Ax Jiang <ji...@gmail.com>]

HI there

I am new to ZAB and very interested in the ZAB.
My question is that in the paper "Zookeeper's atomic broadcast protocol:
Theory and Practice", in the pseudo code of phase two: sync, in the line
19. If the follower's epoch is not the prospective leader, which means the
epoch of this follower is pretty old, then the state of this follower will
go to the re-election. In the following reelection, since the epoch of this
follower is older it seems data in this follower will never will sync-ed?

I guess I miss something in the reading and any help is appreciated in
advance.

-regards

Re: question about the sync of follower with older epoch

[Michael Han <ha...@apache.org>]

I took a look at this paper, looks like it was talking about ZAB pre 1.0
(the old ZAB protocol). In ZAB 1.0, follower will first sync with leader
and then atomically update its currentEpoch when receiving a NEWLEADER
message from leader. Some documentations about ZAB 1.0 and old ZAB might
help:

https://cwiki.apache.org/confluence/display/ZOOKEEPER/Zab+Pre+1.0
https://cwiki.apache.org/confluence/display/ZOOKEEPER/Zab1.0

On Thu, Oct 4, 2018 at 10:58 AM Ax Jiang <ji...@gmail.com> wrote:

> HI there
>
> I am new to ZAB and very interested in the ZAB.
> My question is that in the paper "Zookeeper's atomic broadcast protocol:
> Theory and Practice", in the pseudo code of phase two: sync, in the line
> 19. If the follower's epoch is not the prospective leader, which means the
> epoch of this follower is pretty old, then the state of this follower will
> go to the re-election. In the following reelection, since the epoch of this
> follower is older it seems data in this follower will never will sync-ed?
>
> I guess I miss something in the reading and any help is appreciated in
> advance.
>
> -regards
>