[infinispan-issues] [JBoss JIRA] (ISPN-4291) Poor REPL state transfer performance with cache stores

[Dennis Reed (JIRA)]

    [ https://issues.jboss.org/browse/ISPN-4291?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12975752#comment-12975752 ] 

Dennis Reed commented on ISPN-4291:
-----------------------------------

Yeah, there may still be room for optimization here.

Two use cases:
- node joining the cluster
  It cannot be skipped here, as there may be stale data in the cache store which can only be determined by looking through it all.
  (this was the main use case I was looking at).
- later state transfers
  Here it could be skipped if no segments are lost (I.E. repl)

> Poor REPL state transfer performance with cache stores
> ------------------------------------------------------
>
>                 Key: ISPN-4291
>                 URL: https://issues.jboss.org/browse/ISPN-4291
>             Project: Infinispan
>          Issue Type: Bug
>      Security Level: Public(Everyone can see) 
>          Components: State Transfer
>    Affects Versions: 6.0.2.Final
>            Reporter: Dennis Reed
>            Assignee: Dan Berindei
>
> During a state transfer every single key is loaded from the cache store to determine whether it's still owned by the local node.
> This is an extremely slow operation, and doesn't even make sense for a REPL cache.



--
This message was sent by Atlassian JIRA
(v6.2.6#6264)