giraph-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Pavan Kumar (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (GIRAPH-872) Minor inconsistencies with netty handler logic after netty 4 upgrade
Date Tue, 18 Mar 2014 16:46:43 GMT

    [ https://issues.apache.org/jira/browse/GIRAPH-872?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13939458#comment-13939458
] 

Pavan Kumar commented on GIRAPH-872:
------------------------------------

I agree with 1 & 2. Minor changes but my bad.
However in 2, it is okay to do ctx.fireChannelInactive(); though not required

Because these are the handler on client side

clientInboundByteCounter -in (in byte counter)
clientOutboundByteCounter -out (out byte counter)
fixed-length-frame-decoder -in (breaks the stream to 13 bytes)
request-encoder -out (convert writable request to bytebuf)
response-handler  -in (handler of responses, 13 bytes received from fixed length)

And in netty inbound handlers are handled top-bottom while outbound are handled bottom to
top.
So, reponse-handler i.e., ResponseClientHandler is the last inbound handler, and there is
no downstream handler that requires the information being passed.

However, since it is harmless. I am okay with the change.

+1 from myside. Please do mvn clean verify

> Minor inconsistencies with netty handler logic after netty 4 upgrade
> --------------------------------------------------------------------
>
>                 Key: GIRAPH-872
>                 URL: https://issues.apache.org/jira/browse/GIRAPH-872
>             Project: Giraph
>          Issue Type: Bug
>          Components: bsp
>    Affects Versions: 1.1.0
>            Reporter: Craig Muchinsky
>            Priority: Minor
>             Fix For: 1.1.0
>
>         Attachments: GIRAPH-872.patch
>
>
> While debugging some other issues I noticed a few minor inconsistencies and suspected
logic bugs in the netty handler changes related to netty 4:
> 1.) Classes (RequestServerHandler, ResponseClientHandler) overriding ChannelInboundHandlerAdapter.exceptionCaught()
are still calling getCause() on the passed Throwable even though what's being passed in now
is the actual cause instead of a wrapper.
> 2.) ResponseClientHandler.channelInactive() isn't calling ChannelHandlerContext.fireChannelInactive()
like RequestServerHandler does. I believe this is an oversight and not by design.
> 3.) ResponseClientHandler.exceptionCaught() is gating the LOG.warn() call with a LOG.isDebugEnabled()
check. I suspect this gate should be removed so the warning gets logged properly.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message