[
https://issues.apache.org/jira/browse/GIRAPH-839?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Pavan Kumar updated GIRAPH-839:
-------------------------------
Description:
I was working on upgrading netty to 4.0.14.Final. It is mostly done, after some
polishing I will submit a patch for that soon.
While working on the above, I discovered a bug in
NettyWorkerAggregatorRequestProcessor. The distributeAggregators method is
reusing SendAggregatorsToWorkerRequest objects, which is incorrect since each
task has a different task id & (potentially a different) request id. It does
not cause any problem with current netty but in the new version I guess due to
delayed serialization the object cannot be reused.
Anyway, moving it inside the for loop does not cause any issues with current
netty as well.
was:
I was working on upgrading netty to 4.0.14.Final. It is mostly done, after some
polishing I will submit a patch for that soon.
While working on the above, I discovered a bug in
NettyWorkerAggregatorRequestProcessor. The distributeAggregators method is
reusing SendAggregatorsToWorkerRequest objects, which is incorrect since each
task has a different task id. It does not cause any problem with current netty
but in the new version I guess due to delayed serialization the object cannot
be reused.
Anyway, moving it inside the for loop does not cause any issues with current
netty as well.
> NettyWorkerAggregatorRequestProcessor tries to reuse request objects
> --------------------------------------------------------------------
>
> Key: GIRAPH-839
> URL: https://issues.apache.org/jira/browse/GIRAPH-839
> Project: Giraph
> Issue Type: Bug
> Reporter: Pavan Kumar
> Assignee: Pavan Kumar
> Attachments: GIRAPH-839.patch
>
>
> I was working on upgrading netty to 4.0.14.Final. It is mostly done, after
> some polishing I will submit a patch for that soon.
> While working on the above, I discovered a bug in
> NettyWorkerAggregatorRequestProcessor. The distributeAggregators method is
> reusing SendAggregatorsToWorkerRequest objects, which is incorrect since each
> task has a different task id & (potentially a different) request id. It does
> not cause any problem with current netty but in the new version I guess due
> to delayed serialization the object cannot be reused.
> Anyway, moving it inside the for loop does not cause any issues with current
> netty as well.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
