On Tue, 2018-11-27 at 14:35 -0700, Gary Gregory wrote:
> On Tue, Nov 27, 2018 at 5:49 AM Oleg Kalnichevski <[email protected]>
> wrote:
> 
> > On Fri, 2018-11-23 at 17:09 +0100, Oleg Kalnichevski wrote:
> > > On Fri, 2018-11-23 at 08:46 -0700, Gary Gregory wrote:
> > > > On Fri, Nov 23, 2018 at 8:37 AM Oleg Kalnichevski <
> > > > [email protected]
> > > > > 
> > 
> > ...
> > 
> > > > > Hi Gary
> > > > > 
> > > > > I ran 100 Mio requests against the AsyncFileServerExample and
> > > > > took
> > > > > multiple heap dumps in the course of the test run. The AB was
> > > > > running
> > > > > for ~2.5 hours. All message exchanges were successful and the
> > > > > memory
> > > > > footprint remained constant. There might be some sub-optimal
> > > > > intermediate garbage but I certainly could not detect any
> > > > > memory
> > > > > or
> > > > > resource leakage.
> > > > > 
> > > > > Again, I cannot fix something I cannot reproduce. This once
> > > > > again
> > > > > might
> > > > > be something Windows specific.
> > > > > 
> > > > > Please try to isolate the issue and capture the heap with
> > > > > -XX:+HeapDumpOnOutOfMemoryError option upon OOE.
> > > > > 
> > > > > Oleg
> > > > > PS: I am pretty happy with how HttpCore fared so far.
> > > > > 
> > > > 
> > > > Hi Oleg,
> > > > 
> > > > Thank you for running the test again. I am running the test
> > > > again
> > > > on
> > > > Java 7
> > > > to try and reproduce the OOME.
> > > > 
> > > > The OOME did not happen on Java 8 but I did get the same
> > > > timeout
> > > > issue
> > > > there (see my other email.)
> > > > 
> > > 
> > > Please see my other response. If it takes longer than 10 seconds
> > > to
> > > serve files, 10 second timeouts are to be expected.
> > > 
> > 
> > Hi Gary
> > 
> > Any process with reproducing the issue?
> > 
> 
> Hi Oleg,
> 
> I am putting out a customer fire at work, so I'll try to update this
> thread
> tonight. In the meantime, I've replied several times to this thread
> with
> other finding but I do understand from a previous reply of yours that
> you've not reproduced the issue.
> 
> What do you think about my theory that the the select() call in the
> IO
> reactor returns with a 0 ready count while there are output bytes
> ready to
> be processed in the exchange state's proxy buffer?
> 

I consider it unlikely. I also do not see how this could be related to
the resource leak you said you were seeing.

Oleg


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to