[infinispan-issues] [JBoss JIRA] Commented: (ISPN-1035) Servers should avoid reading key/values multiple times per request

[Galder Zamarreño (JIRA)]

    [ https://issues.jboss.org/browse/ISPN-1035?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12594678#comment-12594678 ] 

Galder Zamarreño commented on ISPN-1035:
----------------------------------------

To fix this I had to implement checkpointing in Netty that fixes the problem by putting a marker after key or value is read. This guarantees that key/value pairs which are potentially lengthy are only read once.

To implement this I had rewrite a big chunk of the decoding part and I got rid of the transport abstraction and instead used Netty directly, cos it was gonna make the code much simpler and easier to maintain in the future.

The issue is almost fixed and after that I need to do some tidying up.

> Servers should avoid reading key/values multiple times per request
> ------------------------------------------------------------------
>
>                 Key: ISPN-1035
>                 URL: https://issues.jboss.org/browse/ISPN-1035
>             Project: Infinispan
>          Issue Type: Bug
>          Components: Cache Server
>    Affects Versions: 4.2.1.FINAL, 5.0.0.BETA1
>            Reporter: Galder Zamarreño
>            Assignee: Galder Zamarreño
>            Priority: Critical
>              Labels: hotrod, memcached
>             Fix For: 4.2.2.BETA1, 5.0.0.BETA2, 5.0.0.FINAL
>
>
> When Infinispan servers need to decode keys and values only once per request, otherwise it can generate a lot of garbage byte[]s instances. This is particularly important when bigger key/values are used. The way to do that is tell Netty about intermediate stages of reading a request so that the entire request is not re-read if the entire request has not been read in one go.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira