How to add a new server in a cluster?`

[else]

For messaging-2.0.0.BETA1 document in "37.1.2. Synchronization of live-backup pairs" chapter it's said that we need copy-paste the data for persistent storage between live and backup.

For the load balance cluster, do we also have that requirement? If we want to add a new server to the existing running cluster, do we need to copy the data from the existing cluster member to the new one?

Further more, do we support RDB or share folder for this feature? if true we don't need copy-paste manually.

BTW: if we don't have backup server in a cluster, then the cluster cannot bear server crashed? We provide two or more servers and then we cannot stop anyone? I don't think it makes sense.

[clebert.suconic]

Failover is different than load balancing. Failover is aways an exact copy of the live node, so whenever your server fails clients can fail over to the backup. While another node sharing load on the cluster will be a different server.

That text is saying that you need to keep live and backup equal.. that's all.


We don't support shared folder on it, because we manually control the replication.

I'm not sure what you mean by RDB. (not sure you meant RDBMS or that old Oracle database). But we decided by the journal from JBM 2 on, since it's a much faster approach.

BTW: if we don't have backup server in a cluster, then the cluster cannot bear server crashed? We provide two or more servers and then we cannot stop anyone? I don't think it makes sense.

If you need failover, you could just provide backup nodes.

[else]

Thank you for your reply.
I mean "relational database" for RDB. It means we might need mysql or oracle or whatever to share persistent data between cluster or fail over pair nodes. From your answer I guess we don't support it. we sacrificed the convenience to gain performance, right?

For the original question: when add a new broker to a running cluster, do we need to copy persistent data from other nodes? I guess no, since for load balance cluster we didn't guarantee exactly same, is it correct?

Another question: For the existing broker in a cluster, if somehow we stopped it for a while, could we just restart it and join in to the cluster directly? we don't need to worry the data divergence between load balance cluster nodes, do we? It could happen even if the node has backup, since in the worst case the backup server could also failed, and we didn't provide facility for re-assigning a backup node with a running node.

[clebert.suconic]

"Else" wrote:
we sacrificed the convenience to gain performance, right?

I wouldn't say that. It was way more hassle administering a DB.

For the original question: when add a new broker to a running cluster, do we need to copy persistent data from other nodes? I guess no, since for load balance cluster we didn't guarantee exactly same, is it correct?

Another question: For the existing broker in a cluster, if somehow we stopped it for a while, could we just restart it and join in to the cluster directly? we don't need to worry the data divergence between load balance cluster nodes, do we? It could happen even if the node has backup, since in the worst case the backup server could also failed, and we didn't provide facility for re-assigning a backup node with a running node.

I believe you don't need to do any syncs when adding new nodes.. but I will wait for Tim Fox (the project lead) to answer that question if you don't mind. He's the wizard on this module.. and I don't want to give you a wrong answer.

[clebert.suconic]

"Else" wrote:
we sacrificed the convenience to gain performance, right?

I wouldn't say that. It was way more hassle administering a DB.

I mean.. we got a lot of performance thorugh the journal. But the DB for our usage was far away from being convenient IMHO.

[clebert.suconic]

Were you able to run any of the examples (clustered.. etc)?

You wouldn't need to do any syncs for adding new nodes on the cluster.

Also, we are releasing a Beta2 probably tomorrow, that will include a few standalone-clusters configs and a new standalone clustered example, that will probably help you on your testing.