[infinispan-issues] [JBoss JIRA] (ISPN-3922) Hotrod client should shun nodes after a connection read-timeout for a while because of better performance with replicated-caches

[Wolf-Dieter Fink (JIRA)]

Wolf-Dieter Fink created ISPN-3922:
--------------------------------------

             Summary: Hotrod client should shun nodes after a connection read-timeout for a while because of better performance with replicated-caches
                 Key: ISPN-3922
                 URL: https://issues.jboss.org/browse/ISPN-3922
             Project: Infinispan
          Issue Type: Feature Request
          Components: Remote protocols
    Affects Versions: 6.0.1.Final
            Reporter: Wolf-Dieter Fink
            Assignee: Galder Zamarreño


If a hotrod client access a clustered replicated cache it will use a RoundRobin policy.
In case of a node failure, i.e. Garbage Collection or network issues, a cache operation will fail with a network timeout.
As this 'suspend' phase might take a bit before the node come back to work or get's dropped from the cluster, and the cluster-view of the client is updated, it will be used several times due to the round-robin according to the total number of existing cluster nodes.

In a two node cluster the effect is that each second call will wait for the configured timeout until the data is provided by the existing node. Which is a huge performance drawback.

To improve the performance in such cases a failed connection should expelled from the round robin for a shunning-period or until a new cluster-view is provided from the server side.
To improve the behaviour automatically there should be a default timeout according to the default JGroups detection (i.e. 10sec).
It should be able to customize this time by a hotrod property.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira