Support for Hazelcast clustering

[Jim Marino]

Hi,

I've added support for Hazelcast-based clustering in the 3.0 line. This can replace the etcd-based clustering for users who don't want to run a separate etcd installation. The advantage of Hazelcast is it allows F3 runtimes to autodiscover the cluster. Configuring Hazelcast (optional) is done by adding a Hazelcast config file at <runtime>/config/hazelcast.xml.

The Hazelcast extension hasn't been extensively tested but leadership election and address propagation should be working. 

I was also thinking about adding a Hazelcast binding in the same extension to support distributed channels.

Comments, suggestions and testing are welcome.

Jim

[Tomáš Fecko]

Hi,

we are testing etcd and now we tried hazelcast. What we have a problem with, is that it doesn't look like the domain nodes knows about each other...

There is no log with etcd, or with hazelcast. It looks like the nodes from 2 domains joined same cluster - it's possible that it's just a multicast group and domain membership is above that?

I tried it with simple setup, 1 server, 2 domains (A, B), 1 node in domain A, 2 nodes in domain B

here are the logs with hazelcast and etcd...

[Jim Marino]

Hi Tomas,

Thanks for such a quick turnaround. I'll test this today and see if I can check in a fix. 

Thanks!

--
You received this message because you are subscribed to the Google Groups "fabric3" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fabric3+u...@googlegroups.com.
To post to this group, send email to fab...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fabric3/509d1e44-42af-48f8-9a81-45fc97be14f7%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Jim Marino]

Hi Tomas,

I've updated the Hazelcast to display more log output (set the runtime monitor to debug) as well as fixed a few problems with leader election.

Can you try Hazelcast and let me know how it works? Once we have that working, I'll also updated the etcd extension.

Jim

[Tomáš Fecko]

Hi Jim,

I tried hazelcast, looks like the domains, that are started are mixed together. 

If there are 2 domains in single multicast reachable network, they will interconnect - which is not desirable. 

For 2 domains, I would expect 2 nodes, where domain scoped composites could be deployed.

I think jgroups were using something they call channels, which allowed asymetric clustering over the same network (like having two separate infinispan clusters on single subnet).

Tomas

[Jim Marino]

Hi Tomas,

I was thinking domains would be run on separate Hazelcast setups, e.g using different multicast addresses. This would mean defining the multicast address in the Hazelcast config.

Is that how you have your setup or were you thinking of using Hazelcast differently?

Jim

--

You received this message because you are subscribed to the Google Groups "fabric3" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fabric3+u...@googlegroups.com.
To post to this group, send email to fab...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/fabric3/7089bd6c-2778-4cfb-b010-7c576fe149ef%40googlegroups.com.

[Tomáš Fecko]

I should have think of that, creation of custom config should suffice...

thanx, I will test that