Must all cluster nodes create all named caches?

[joe.planisky]

Hi all,

I'm just getting started with Infinispan (5.0.0.CR7).  If this question is answered in any of the documentation, please point me to it.

I have 3 nodes in a cluster, clustering mode is "distribution".  Nodes 1 & 2 need to share data in a cache named "pending_work".  Nodes 1 & 3 need to share different data in a cache named "working".  Node 3 does not need to know anything about the "pending_work" cache, so it doesn't get it from its CacheManager at startup. 

However, when node 1 adds an entry to the "pending_work" cache, it gets a series of exceptions starting with a NamedCacheNotFound exception which indicates it's looking for a cache named "pending_work" on node 3.  If I change node 3 so it gets a cache named "pending_work" at startup (even though it doesn't need it or use it), everything works OK (i.e. no exceptions).

Is this normal, expected behavior? Must every node in a cluster get an instance of all named caches used anywhere in the cluster regardless of whether it needs to use that cache or not? 

Thanks in advance.

--

Joe

[ckulenk]

The ConsistentHash does not know which node called "getCache" or which node configured a cache - it always sees all nodes with the same ClusterName. Because ClusterName (see config.html#ce_global_transport or org/infinispan/config/GlobalConfiguration.html#setClusterName(java.lang.String)) is a global option, I think you will have to use multiple CacheManagers with different ClusterNames for each cache to achieve the behavior you want out of the box.

[sannegrinovero]

Christian is correct, you'll have to use different CacheManagers in your case.

All CacheManagers connected in a cluster must have the same configuration, but we don't enforce this as some details might be different; at least the same set of caches should exist on each node and operate in compatible ways; for example the cluster mode should be the same, but the details of a cacheloader might be different, so you can have it store in a different database.

We call the capability to run a different set of caches on each node "asymmetric clusters" and the feature is much requested, tracked as ISPN-658, but not implemented.

I agree it's a serious flaw in the documentation, I'm opening a ticket right now to correct this: ISPN-1252.

[joe.planisky]

Thank you Sanne and Christian for answering my question. 

--

Joe