Re: [jetty-dev] 8.1.2.v20120308 - rare mixup of requests

[Jesse McConnell <jesse.mcconnell@xxxxxxxxx>]

fair enough, thought I don't see how my response was 'hostile' really
and no need to get defensive...this was my first response on the issue
:)

I have not read all the evidence in detail, just enough to know that
the area of code your talking about has had many many months of
intensive review and refactoring for the 7.6.x and 8.1.x versions.
That experience resulted in the entire IO layer of jetty being
replaced in jetty9.  Could there still be dragons there, sure...and if
your issue exists on 8.1.7, well there is a week or so before we are
shooting for 8.1.8 to go out so hopefully it can be pinned in there.

cheers,
jesse

--
jesse mcconnell
jesse.mcconnell@xxxxxxxxx


On Sat, Oct 27, 2012 at 5:45 AM, Per Steffensen <steff@xxxxxxxxxxxxx> wrote:
> Jesse McConnell skrev:
>
>> "I expect some kind situation where (jetty) threads/processes overlap
>> in very unfortunate ways."
>>
>> it is easy to start waving hands at this sort of general 'multi thread
>> ah!'
>
> No it wasnt easy! I have put a lot of investigation into isolating the
> problem - jetty wasnt the prime suspect to begin with, but investigations
> has led to believe that jetty is the cause of the error. I believe the
> evidence speaks for itself, but if you do not agree please tell me - and
> why. I would love to know if anyone can come up with a theory that fits the
> loggings, but where jetty is not to blame. I believe I have asked for that
> already. I just show my evidence/loggings plus explain how it has been
> produced. And then asked a very open question, if you also agree it is a
> jetty bug.
>
>>  area but this issue really needs to get some specifics going or
>> it isn't going to go anywhere
>>
>
> I believe the loggings makes it very specific. If you believe that all bugs
> in a product can always be easily reproduced in a small
> one-machine-a-few-threads-test, thats your problem.
>
> I have to say you sound a little hostile, and I dont like that when Im
> actually just trying to help jetty. The easiest thing for me would just be
> to switch to tomcat (or something) and then leave the bug alone. But I
> believe that when you want to use all the nice free software out there, at
> least you should put an effort into helping making it better whenever you
> can. This is what I am trying to do, by at least reporting the problem and
> taking my time to collect loggings, evidence n'stuff.
>
>> if we can't reproduce it, we probably can't fix it with any certainty,
>>
>
> Maybe some further investigations will make it reproducable. But as I said
> _I_ believe it to be a situation of very unfortunate overlap of threads that
> isnt reproducable in a consistent way unless you insert artificial synching
> in the code. And you cant do that in a clever way without understanding the
> problem properly. And I certainly do not for now.
>
>> nor can we pour massive amounts of time into something that appears to
>> be completely isolated to a edge case that one use case is seeing.
>
> The amount of time you want to put into it is of course up to you. I just
> did my duty and reported what I see as a bug in jetty. And I also want to
> help isolating and solving the problem, if you want my help.
>
>>   It
>> would be far easier to see this happening do to an issue in whatever
>> is running on top of jetty for this use case then it being with jetty
>> itself.
>
> I believe you havnt looked into the evidence provided. At least to me, the
> loggings show that it has nothing to do with the webapp running on top of
> jetty. It might have something to do with OS, virtualization layer or even
> hardware, and I have also openly put that forward as a suggestion. But I do
> believe it is a jetty bug with 90% certainty, and I believe Thomas Becker
> agreed with me, so that made me 95% certain. But Im 99.9999999999% sure it
> is not due to the webapp (Solr) running on top of Jetty.
>
>>   In otherwords, if it can't be reproduced in a small app on
>> jetty that is dealing with huge requests/responses like your speaking
>> about here
>
> It probably can, but it might not be easy.
>
>>  and it only happens with X other systems are in play on top
>> of jetty...then it is hard to start from the position that jetty is to
>> blame.
>>
>
> I didnt start there. But investigations led to believe that it is jetty. I
> would love to be told that Im wrong. But not just by waving hands, like you
> are doing now, but by looking that the heavy evidence put forward, and
> explain how this still can be a bug outside jetty.
>
>> Now..I am not saying this is not an issue in jetty...but before this
>> issue goes any further I think we would really like to see it happen
>> on the latest jetty release of 8.1.7 since we know there were a number
>> of little tweaking to the networking code in there and perhaps there
>> was a race condition like you mention in the early 8.1.x releases that
>> has already been addressed.
>>
>
> In the very beginning I asked if this was a known bug and if it had been
> fixed. I was given a clear "no" as an answer. Therefore I thought it wasnt
> worth trying out a newer version of jetty - and also because you didnt ask
> me to. But now that you did I certainly will try with newest jetty 8.1.7.
> Will get back to you.
>>
>> cheers,
>> jesse
>>
>
> Remember, Im just a nice guy trying to help.
>
> Regards, Per Steffensen
>>
>> --
>> jesse mcconnell
>> jesse.mcconnell@xxxxxxxxx
>
>
> _______________________________________________
> jetty-dev mailing list
> jetty-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/jetty-dev