These are chat archives for atomix/atomix

17th
Nov 2016
Roman Pearah
@neverfox
Nov 17 2016 17:15
@markelliot If a leader becomes segregated, it's not going to be able to commit its log to a majority. Clients connecting to a new leader in the majority partition will succeed normally. So the fact that previous-term leader is marked as a leader is essentially irrelevant since it cannot commit anything. It will step down once the partition heals, and it will rollback all uncommitted changes and apply the new leader's changes.
@markelliot I'm not quite sure what you mean by a "leader-aware filter" since, in Raft, everything already goes through the leader. I don't see why you'd have to do any extra work to get that behavior.
Roman Pearah
@neverfox
Nov 17 2016 17:23
@markelliot I don't know enough to give you a good answer on latency but via CAP theorem and the fact that Atomix is a CP system, you could end up with availability issues under certain conditions like the partition scenario, and it's not going to be as fast as a system that emphasizes A, because it has to guarantee consensus before sending a response to the client.
@markelliot As for elections, there is an election timeout (default 500ms) that causes nodes to become candidates if they don't hear from the leader (250ms heartbeat).
Andrius Dagys
@adagys
Nov 17 2016 17:53
Hey guys, quick Copycat question: is CopycatClient.submit() thread safe? i.e. can two command requests ever end up getting the same sequence no?
Jordan Halterman
@kuujo
Nov 17 2016 17:53
It's thread safe
Andrius Dagys
@adagys
Nov 17 2016 17:54
Ok, thanks
Johno Crawford
@johnou
Nov 17 2016 18:04
i created a prototype project to test consumer / producer but i'm seeing that the producer sometimes blocks when sending and the consumer doesn't receive anything
are there any known issues in 1.0.0-rc9 wrt consumer / producers? I had a look through github issues but may have missed something