[infinispan-issues] [JBoss JIRA] (ISPN-7956) Investigate removing PartitionHandlingInterceptor

[Dan Berindei (JIRA)]

     [ https://issues.jboss.org/browse/ISPN-7956?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dan Berindei updated ISPN-7956:
-------------------------------
    Component/s: Core
                     (was: Test Suite - Core)


> Investigate removing PartitionHandlingInterceptor
> -------------------------------------------------
>
>                 Key: ISPN-7956
>                 URL: https://issues.jboss.org/browse/ISPN-7956
>             Project: Infinispan
>          Issue Type: Task
>          Components: Core
>    Affects Versions: 9.1.0.Beta1
>            Reporter: Dan Berindei
>            Assignee: Dan Berindei
>
> When all the owners of a key leave the cluster, the distribution interceptor doesn't know how to handle that key. Instead, it has to signal that to the {{PartitionHandlingInterceptor}} somehow, and PHI decides whether to return a {{null}} value or throw an {{AvailabilityException}}.
> Single-key read commands and write commands are handled by throwing an {{AllOwnersLostException}} (initially an {{RpcException}}) from the distribution interceptor and catching it in {{PartitionHandlingInterceptor}}. If partitioning handling is disabled, the {{AllOwnersLostException}} is instead caught by {{StateTransferInterceptor}}, which returns the {{null}} value (for read commands) or waits for a new topology and retries (for write commands).
> For multi-key read commands the distribution interceptor can't throw an exception, because that would lose the values that were retrieved successfully. For {{GetAllCommand}} we could get away with storing the values in the context, but that wouldn't work for functional commands because they transform the values on the remote node. ISPN-7884 removing the values with missing owners from the result collection, but that means values that are simply missing have to be treated differently, and {{StateTransferInterceptor}} has to iterate through the list of results again in order to remove them.
> It should be simpler to move all the availability checks to the distribution interceptor, and remove the {{PartitionHandlingInterceptor}}. DI already calls some {{PartitionHandlingManager}} methods directly, in order to keep track of partially committed transactions.



--
This message was sent by Atlassian JIRA
(v7.2.3#72005)