Hi Seven,
Callback sounds good, but I'd say try things out as you think is best.
Emmanuel has a point about the thread, so if you still can't replicate
the problem before a fix start another with the patch (or attach to a
JIRA, I suppose) and I'll apply and run it on my machine before you
commit.
Cheers,
Johnathan
On Wed, 2011-09-28 at 10:29 +0200, Séven Le Mesle wrote:
> Hi Johnathan,
>
> Ok I get it now :).
>
> So what about passing a CallBack to the start call so we can call it when
> the loop is running.
> This would be more compliant with the current API and the callback or
> listener will be able to observe the loop lifecycle properly.
>
> The in the test we will be able to pass the CountDownLatch.
> I see other nice features like restarting a failed loop.
>
> What you think about it ? Also tell me what you think about making the loop
> a runnable and pushing it to an ExecutorService ?
>
> Séven.
>
> 2011/9/27 Johnathan Meehan <jmee...@phasevariance.com>
>
> > Hi,
> >
> > > I don't get exactly what you mean...
> >
> > I get that a lot. :)
> >
> > What I mean is that the countdown is hitting zero before the
> > IOLoop#start() is even invoked. In this case, we're being returned to
> > the caller (the test) whilst the setup (thread naming and flag set) is
> > still ongoing. So, sometimes it fails as that thread's work just isn't
> > finished.
> >
> > If I put text to the console when running the test, after "running" is
> > set in IOLoop#start() and just before HttpServer#start(int) exits I see
> > this:
> >
> > IOLoop: STARTED I/O-LOOP1
> > IOLoop: STARTED I/O-LOOP2
> > HttpServer: DONE
> > IOLoop: STARTED I/O-LOOP3
> >
> > What I thought was that we could pass the latch to the #start method
> > (forgetting about what it breaks for now) and invoke #countdown when
> > "running" has been set. Or, we could check for the state of the current
> > IOLoop outside of the Thread created in the loop, and wait for it to be
> > "running" before calling #countdown. Perhaps there's a more correct way
> > of doing things, though.
> >
> > As the threads are started in a loop, I don't know if we need the call
> > to #await either - we can simply drop out of the method after the loop
> > has run.
> >
> > See the patch below - not a solution, but it will communicate better
> > what I'm saying. :)
> >
> > Johnathan
> >
> >
> > ### Eclipse Workspace Patch 1.0
> > #P deft
> > Index: src/main/java/org/apache/deft/web/HttpServer.java
> > ===================================================================
> > --- src/main/java/org/apache/deft/web/HttpServer.java (revision 1176522)
> > +++ src/main/java/org/apache/deft/web/HttpServer.java (working copy)
> > @@ -113,16 +113,13 @@
> > @Override
> > public void run() {
> > registerHandler(ioLoop, protocol);
> > - loopsLatch.countDown();
> > ioLoop.start();
> > }
> > }).start();
> > + while(!ioLoop.isRunning()) {
> > + }
> > + loopsLatch.countDown();
> > }
> > - try {
> > - loopsLatch.await();
> > - } catch (InterruptedException e) {
> > - logger.error("Interrupted while waiting for all IOLoop to
> > start", e);
> > - }
> > }
> >
> > /**
> >
> >
> >
> > On Tue, 2011-09-27 at 14:15 +0200, Séven Le Mesle wrote:
> > > Hi Johnathan,
> > >
> > > I don't get exactly what you mean but calling the countdown when the
> > initial
> > > setup is complete make's sense.
> > > You can have a look on what I did to test multi-thread in HTTPServer. I
> > > tried to make the start and stop calls blocking and returning only when
> > > operation is done failed or successfull.
> > > I'm not so happy with the implementation but it does work.
> > >
> > > We can imagine using a boolean flag in the calls to allow switching
> > between
> > > blocking and not blocking mode.
> > >
> > > Séven.
> > >
> > > 2011/9/27 Johnathan Meehan <jmee...@phasevariance.com>
> > >
> > > >
> > > > Hi Seven,
> > > >
> > > > My machine has the same intermittent problem as the server. From what I
> > > > see, the loop in IOLoop#start() is not always established quickly
> > enough
> > > > after #countdown has been invoked before it[1]. The last thread is
> > > > always the one to fail the assertion, then. What about passing the
> > > > CountdownLatch to a method with the appropriate signature, and calling
> > > > #countdown when the inital setup is complete and the loop is actually
> > > > running?
> > > >
> > > > Johnathan
> > > >
> > > > [1]
> > > > loopsLatch.countDown() - Count: 2
> > > > start() - I/O-LOOP1
> > > > loopsLatch.countDown() - Count: 1
> > > > start() - I/O-LOOP2
> > > > LOOP: org.apache.deft.io.IOLoop@32955cd5 - true
> > > > LOOP: org.apache.deft.io.IOLoop@68ed662d - true
> > > > LOOP: org.apache.deft.io.IOLoop@3f9f3736 - false
> > > > loopsLatch.countDown() - Count: 0
> > > > start() - I/O-LOOP3
> > > >
> > > >
> > > > On Tue, 2011-09-27 at 01:02 +0200, Séven Le Mesle wrote:
> > > > > Great, so I'm not guilty this time cool.
> > > > > The problem seems to appear every time on Solaris nodes, it might
> > come
> > > > from
> > > > > server load ...
> > > > > But we should fix this.
> > > > >
> > > > > 2011/9/27 Johnathan Meehan <jmee...@phasevariance.com>
> > > > >
> > > > > >
> > > > > > I noticed on the build mailing list that somebody was talking about
> > > > > > Jenkins being a little sluggish, but that seems to depend on the
> > node
> > > > > > being hit. Built okay this time on "ubuntu5" (that is, completed
> > > > without
> > > > > > errors).
> > > > > >
> > > > > > https://builds.apache.org/job/Deft/26/consoled
> > > > > >
> > > > > >
> > > > > > On Tue, 2011-09-27 at 00:22 +0200, Séven Le Mesle wrote:
> > > > > > > I don't believe in luckyness for such things.
> > > > > > > But well we can hope for this time ...
> > > > > > > So I'll ask Niklas directly to get Job administration rights.
> > > > > > >
> > > > > > > Thanks again for your great help.
> > > > > > >
> > > > > > > Séven
> > > > > > >
> > > > > > > 2011/9/27 Johnathan Meehan <jmee...@phasevariance.com>
> > > > > > >
> > > > > > > >
> > > > > > > > That's done but the build is unstable as you remarked on
> > earlier.
> > > > It
> > > > > > > > timed-out first time around whilst recording test results[1],
> > so
> > > > I'll
> > > > > > > > start it again now and see what kind of luck we have.
> > > > > > > >
> > > > > > > > > I don't know how to get the rights, but that would be great.
> > > > > > > >
> > > > > > > > Niklas handled that for me, so perhaps you might ask him
> > directly.
> > > > > > > >
> > > > > > > > [1] https://builds.apache.org/job/Deft/24/console
> > > > > > > >
> > > > > > > >
> > > > > > > > On Mon, 2011-09-26 at 23:56 +0200, Séven Le Mesle wrote:
> > > > > > > > > Hi Johnathan,
> > > > > > > > >
> > > > > > > > > yes you can update the Jenkins because I do not have the
> > rights
> > > > to
> > > > > > update
> > > > > > > > > it.
> > > > > > > > > I don't know how to get the rights, but that would be great.
> > > > > > > > >
> > > > > > > > > Thanks for your help.
> > > > > > > > >
> > > > > > > > > Séven.
> > > > > > > > >
> > > > > > > > > 2011/9/26 Johnathan Meehan <jmee...@phasevariance.com>
> > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Hi Seven,
> > > > > > > > > >
> > > > > > > > > > The build was okay on my machine, and the examples seem to
> > run
> > > > > > > > correctly
> > > > > > > > > > too. Do you have rights to update Jenkins? It didn't sound
> > in
> > > > your
> > > > > > > > > > previous mail like you did, so let me know and I can change
> > it
> > > > for
> > > > > > you.
> > > > > > > > > >
> > > > > > > > > > Johnathan
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On Mon, 2011-09-26 at 23:35 +0200, Séven Le Mesle wrote:
> > > > > > > > > > > Hy again,
> > > > > > > > > > >
> > > > > > > > > > > I just made the copy from sandbox to trunk all seems to
> > be
> > > > right.
> > > > > > > > > > > Review needed :)
> > > > > > > > > > > http://svn.apache.org/viewvc/incubator/deft/trunk/
> > > > > > > > > > >
> > > > > > > > > > > Tell me,
> > > > > > > > > > >
> > > > > > > > > > > Séven
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > 2011/9/26 Séven Le Mesle <sleme...@xebia.fr>
> > > > > > > > > > >
> > > > > > > > > > > > Well it's mainly a copy of the branch to the Trunk or a
> > > > merge
> > > > > > if
> > > > > > > > you
> > > > > > > > > > prefer
> > > > > > > > > > > > it.
> > > > > > > > > > > >
> > > > > > > > > > > > Don't forget to update Jenkins configuration though.
> > > > > > > > > > > >
> > > > > > > > > > > > Another question:
> > > > > > > > > > > > After some commit, the DeftSystem test remains unstable
> > > > while
> > > > > > > > working
> > > > > > > > > > on my
> > > > > > > > > > > > mac :p.
> > > > > > > > > > > > I suspect a Selector bug or so can someone reproduce
> > the
> > > > error
> > > > > > of
> > > > > > > > the
> > > > > > > > > > > > Jenkins running on Solaris ?
> > > > > > > > > > > >
> > > > > > > > > > > > Thanks.
> > > > > > > > > > > >
> > > > > > > > > > > > Séven.
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > 2011/9/26 Roger Schildmeijer <schildmei...@gmail.com>
> > > > > > > > > > > >
> > > > > > > > > > > >>
> > > > > > > > > > > >> On Sep 26, 2011, at 6:19 PM, Emmanuel Lecharny wrote:
> > > > > > > > > > > >>
> > > > > > > > > > > >> > On 9/26/11 6:03 PM, Niklas Gustavsson wrote:
> > > > > > > > > > > >> >> On Mon, Sep 26, 2011 at 5:38 PM, Roger Schildmeijer
> > > > > > > > > > > >> >> <schildmei...@gmail.com> wrote:
> > > > > > > > > > > >> >>> Good question. I think we have fixed all the
> > issues
> > > > that
> > > > > > were
> > > > > > > > > > > >> "prerequisites" before we could leave the sandbox.
> > Anyone
> > > > > > > > disagree?
> > > > > > > > > > > >> >> Go for it.
> > > > > > > > > > > >> > just do it (tm) :)
> > > > > > > > > > > >>
> > > > > > > > > > > >> Anyone done it before and feel comfortable about doing
> > it
> > > > > > again :)
> > > > > > > > ?
> > > > > > > > > > > >>
> > > > > > > > > > > >>
> > > > > > > > > > > >> >
> > > > > > > > > > > >> >
> > > > > > > > > > > >> > --
> > > > > > > > > > > >> > Regards,
> > > > > > > > > > > >> > Cordialement,
> > > > > > > > > > > >> > Emmanuel Lécharny
> > > > > > > > > > > >> > www.iktek.com
> > > > > > > > > > > >> >
> > > > > > > > > > > >>
> > > > > > > > > > > >>
> > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > --
> > > > > > > > > > > > Séven Le Mesle
> > > > > > > > > > > > Xebia IT Architects
> > > > > > > > > > > >
> > > > > > > > > > > > Email : sleme...@xebia.fr
> > > > > > > > > > > > Mobile : +33(0)1 46 91 76 16
> > > > > > > > > > > >
> > > > > > > > > > > > http://www.xebia.fr
> > > > > > > > > > > > http://blog.xebia.fr
> > > > > > > > > > > >
> > > > > > > > > > > > Siège Social
> > > > > > > > > > > > La Défense Colisée
> > > > > > > > > > > > 10 / 12 Avenue de l'arche
> > > > > > > > > > > > Faubourg de l'Arche
> > > > > > > > > > > > 92419 Courbevoie Cedex
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > >
> > > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > >
> > > > >
> > > >
> > > >
> > > >
> > > >
> > >
> > >
> >
> >
> >
>
>
