hc-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From sebb <seb...@gmail.com>
Subject Re: Strange behaviour with httpclient 4.2.6 / core 4.2.5
Date Wed, 19 Feb 2014 18:00:52 GMT
The cause of the disconnect has been determined.
The host that dropped the connection must be operating some kind of
idle timeout.

That host was only used once per test loop, so had much higher idle
times than the others.
The problem disappears with staleCheck enabled because that swallows
the disconnect and creates a new connection.

On 19 February 2014 16:10, sebb <sebbaz@gmail.com> wrote:
> I've now had a quick look at the source for DefaultClientConnection
> and it looks like that debug message is only generated from the
> close() method.
>
> So enabling more debug in HC likely won't help.
>
> Need to find out if JMeter is incorrectly closing the connection by
> adding more debug.
> This is likely easier than rebuilding HC with more debug.
>
> Sorry for the disturbance, should have checked this earlier.
> I'll report back when I know more.
>
>
> On 19 February 2014 15:40, Oleg Kalnichevski <olegk@apache.org> wrote:
>> ...
>>
>>> >> Why are there 3 copies of the same log message?
>>> >
>>> > There are several places where HttpConnection#close or #shurdown are
>>> > called in the course of exceptional execution flow. The logging can be
>>> > changed though so that only the first invocation is logged.
>>>
>>> Can it be changed so that the original cause is logged?
>>>
>>
>> Generally HttpClient uses log or re-throw policy on exception
>> propagation. It never logs exceptions if it propagates them to the
>> caller. If exception is retried and not propagated it is always logged.
>>
>>> >> And why is the connection being closed?
>>> >> The timestamp shows it happens immediately - there is probably not
>>> >> time to actually contact the remote host.
>>> >>
>>> >> The same sequence repeats another 3 times (there are 6 loops on the
>>> >> test, alternate ones - 2,4,6 - fail on this one sample)
>>> >> All other samples in the test work OK as far as I can tell.
>>> >>
>>> >> Any ideas how to debug this further?
>>> >>
>>> >
>>> > Could you please post the complete log (obfuscated if necessary) or
>>> > attach it to the ticket?
>>>
>>> It's quite long, and repetetive, and there's potentially a lot to
>>> redact before I can post it in a public place
>>>
>>
>> Would it be easier for you to create a reproducer (preferably a JUnit)?
>>
>> Oleg
>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscribe@hc.apache.org
>> For additional commands, e-mail: dev-help@hc.apache.org
>>

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


Mime
View raw message