These are chat archives for atomix/atomix

9th
Jun 2016
David Moravek
@dmvk
Jun 09 2016 06:22
@madjam yes, that may be possible, but when it comes to ephemeral members, how would you know when to remove the listeners from the specific state machine?
i reckon it may be enough to watch copycatClient for onStateChangeevents and once it transitions from SUSPENDED to OPEN state, call the sync() method again
plus, we would have to send to send the Join command again, if it was sent by that client before
andreas-gilbert
@andreas-gilbert
Jun 09 2016 07:06
@madjam I need a simple leader election using 2 nodes to execute some work by only one node. What about the BalancingClusterManager.withQuorumHint value - it is possible to set this to 1 ?
Jordan Halterman
@kuujo
Jun 09 2016 16:33
hmm
sorry I’ve been at Spark Summit in SF
@andreas-gilbert if you do that you effectively have a one node cluster (the other node is just a listener). That means if the one node dies then the cluster stops working altogether, thus you have no fault tolerance. The only other possibility if you need a two-node cluster is using the standalone cluster and using AtomixClient on those two nodes
in that case, your two nodes can crash independently without effecting the cluster since consensus is being done in a separate cluster of e.g. 3 nodes
will catch up on all the other conversations throughout the day
Jonathan Halterman
@jhalterman
Jun 09 2016 17:55
@kuujo re: the problem that @taobaorun hit the other day, it would actually be pretty sweet if we could detect inconsistencies between the a server's configuration and the log it's reading and give a nice warning.
Jonathan Halterman
@jhalterman
Jun 09 2016 18:08
@andreas-gilbert The doc page on fault tolerance (http://atomix.io/atomix/docs/fault-tolerance/) talks a bit about how many failures can be tolerated by different node configurations. The way this works is pretty typical for any strongly consistent system.
Jordan Halterman
@kuujo
Jun 09 2016 18:10
I think it’s easy… servers just need to store their server address in their logs and validate that the logs in the log directory are for the same server by address. Currently it’s done by server name, but servers usually have the same name on each node
Jordan Halterman
@kuujo
Jun 09 2016 23:46
releasing Copycat 1.1.1 with a critical bug fix and I’ll be pushing some changes to Atomix in preparations for a release there too