axis-java-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andreas Veithen <andreas.veit...@gmail.com>
Subject Re: Sandesha 2 test case failures.
Date Sat, 25 Sep 2010 19:05:28 GMT
On Sat, Sep 25, 2010 at 17:02, Amila Suriarachchi
<amilasuriarachchi@gmail.com> wrote:
>
>
> On Thu, Sep 23, 2010 at 5:10 PM, Glen Daniels <glen@thoughtcraft.com> wrote:
>>
>> Gentlefolk:
>>
>> A few comments from me as well.
>>
>> 1) Clearly it was a serious oversight on my part not to have merged the
>> 1.5.1
>> code over to the trunk. :(
>>
>> 2) Aside from simply sharing the MHCM, there is another important change
>> to
>> the 1.5.1 (and now trunk) code, and that is to clean up, by default, the
>> previous request for a given client each time sendReceive() is called -
>> this
>> allows the user to run up to two threads of many requests to the same host
>> without manually calling cleanup, and without blocking or connection
>> starvation.
>
> if the MHCM is shared then users have to call cleanUpTransport method. These
> kind of workarounds
> won't solve the problem entirely. This may be the case with Sandesha.

The point is that by design (because of support for deferred parsing),
users have to call cleanupTransport anyway. This is not a workaround
and has nothing to do with the choice of a shared MCHM. Even if the
transport opened a new connection for every request, it would still be
necessary to call cleanupTransport to make sure that the connection is
closed properly.

>>
>> 3) The right solution, IMO, is not to simply go back to creating a new
>> MHCM/HTTPClient for each request - the old code had resource leakage
>> issues
>> even when you switched on all the switches that people put in trying to
>> fix
>> problems prior to 1.5.1,
>
> You need to explain this with proper details. For me (I may have miss
> something) your fix
> technically do the same thing as setting a shared MHCM using parameters.
>
>>
>> which is why the CLOSE_WAIT issues kept being
>> reopened.  Resource leakage / failing to close connections correctly is a
>> bug, plain and simple.
>>
>> 4) The current code certainly isn't ideal, and we should figure out how to
>> make it cleaner/easier/better for 1.6.  I opened up a specific thread for
>> this topic: http://markmail.org/thread/wfpsfpxm5kmhf7yx - can we continue
>> the
>> discussion there?
>
> I think we have discussed every thing regarding above issue. Better thing is
> to call a vote
> to decide whether we need to make MHCM sharing by default or not.
> if there are majority votes for that then we need to clean up the existing
> code ( and adding creating a new
> connection as an option). otherwise simply revert the patch.
>
> thanks,
> Amila.
>
>>
>> Thanks,
>> --Glen
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: java-dev-unsubscribe@axis.apache.org
>> For additional commands, e-mail: java-dev-help@axis.apache.org
>>
>
>
>
> --
> Amila Suriarachchi
> WSO2 Inc.
> blog: http://amilachinthaka.blogspot.com/
>

---------------------------------------------------------------------
To unsubscribe, e-mail: java-dev-unsubscribe@axis.apache.org
For additional commands, e-mail: java-dev-help@axis.apache.org


Mime
View raw message