#3051
- Created on2013-02-15 10:11
- Reported by
- StatusNew
- PriorityNormal (3)
- Assigned to-
- Milestone
- Plan LevelSubtask
- Due Date-
Support node discovery through multicast
As an alternative to using seed nodes. The only need for seed nodes is to have a solution that works in virtualized envs (EC2 etc.). Using seed nodes is complicated and our impl have some startup rules hard to understand and use.
I think that we should also provide node discovery through standard multicast, as an option. It is a lot simpler to deal with.
I think that we should also provide node discovery through standard multicast, as an option. It is a lot simpler to deal with.
Leave a comment
on 2013-02-16 16:48 *
By Helena Edelson
+1 Jonas, this is an important use case. Pre-configured seed nodes I have no use for, as nodes are deployed to EC2 and exist in various states (joining, joined, downing, etc) and everything must be dynamic (not hard coded). That and a means to partition them as clusters of clusters for security. I've started working on that as an extension.
on 2013-02-16 16:58 *
By Patrik Nordwall
I doubt you can use multicast at EC2, but in other env it is of course great.
on 2013-02-16 17:02 *
By Helena Edelson
We use 0MQ. I have large clusters, some are akka nodes, some python, and all register when they come online via 0mq and a nice lightweight naming service that sits on top of akka cluster, in EC2.
on 2013-02-16 17:03 *
By Helena Edelson
Which works very nicely because we can hardcode the 0mq bind addresses but can't hard code any seed nodes. Using the akka 0mq actors is very clean.
