Problem with Hotrod server

[ntsankov]

We are running a cache cluster using the hotrod server and connecting to it with the java hotrod client.

We had the following problem:

Client was running fine at the beginning and started getting errors after some time:

Exception in thread "main" HotRodServerException{messageId=0, errorStatusCode=0} org.infinispan.client.hotrod.exceptions.InvalidResponseException: Invalid magic number. Expected a1 and received 50

While on the server the following error was logged:

ERROR [org.infinispan.server.core.AbstractProtocolDecoder$] (HotRodServerWorker-1-1) Exception reported
          java.lang.NullPointerException
          at org.infinispan.server.hotrod.HotRodEncoder.getTopologyResponse(HotRodEncoder.scala:84)
          at org.infinispan.server.hotrod.HotRodEncoder.encode(HotRodEncoder.scala:32)
...

This error was traced to be caused by having an expiration setting on the default cache in our config, which is apparently used by hotrod server for its topology cache, so the "view" entry expired after 30 seconds and caused this NPE on the server. Which in turn caused an invalid or no header to be written to the response and the resulting error on the client.

Work around was to create a "namedCache" and configure the expiration on it.

Version 4.1.0-Final

[galder.zamarreno]

This is a bug: https://jira.jboss.org/browse/ISPN-669. We need to be more restrictive on the configuration for the topology cache. It should have no eviction and be configured with REPL_SYNC. Alternatively, create a named cache with '___hotRodTopologyCache' name and configure it with sync repl and no eviction. That way you can keep your default cache as is.

[galder.zamarreno]

The issue has now been fixed and will be included in 4.2.0.BETA1

I've also added an faq entry to: http://community.jboss.org/docs/DOC-13439