[erlang-questions] yaws mixing up requests?

[Francesco Cesarini (Erlang Training and Consulting)]

We finally got to the bottom of this one and wanted to post it here for 
the record. It was not a YAWS or a socket problem, and instead a bug in 
the F5 load balancers. When the clients opened http 1.1 connections, the 
load balancers forwarded the requests using existing 1.1 connections 
towards a limited number of front end nodes. With the least used load 
balancing algorithm, requests were being swapped. By changing algorithm, 
the problem was resolved.

Francesco

Ulf Wiger wrote:
> It's 60, which seems to be a fairly normal value.
> Thanks anyway for the suggestion.
>
> BR,
> Ulf W
> ----- "Valentin Micic" <v@REDACTED> wrote:
>
>   
>> I think this would be highly unlikely, but just to get it out of the
>> way...
>> What is the OS TIME_WAIT value -- if too short, wires may be crossed
>> at the
>> kernel level? 
>>
>> V.
>>
>> -----Original Message-----
>> From: erlang-questions@REDACTED [mailto:erlang-questions@REDACTED]
>> On
>> Behalf Of Francesco Cesarini (Erlang Training and Consulting)
>> Sent: 30 May 2009 03:07 PM
>> To: Christian
>> Cc: Ulf Wiger; erlang-questions
>> Subject: Re: [erlang-questions] yaws mixing up requests?
>>
>>
>>     
>>> Another source of failure to inspect for you guys is that yaws
>>> actually reuses processes. Maybe things can leak between requests
>>>       
>> in
>>     
>>> the process registry, or in the process mailbox.
>>>   
>>>       
>> We've done that already. What we've seen is that the error occurs in 
>> conjunction with a socket error and that the requested HTTP request is
>>
>> sent to the wrong user about 500ms after the original request comes
>> in, 
>> so we thought that might be the reason. But when the socket error 
>> occurs, there is nothing suspicious stored in the process state / 
>> dictionary. Gut feeling is that there is some form of overflow /
>> memory 
>> corruption in the socket layer / inets driver (or even better, in the
>>
>> load balancers). We are now trying to reproduce the error on R13B.
>>
>> Thanks for your input,
>> Francesco
>>
>> ________________________________________________________________
>> erlang-questions mailing list. See http://www.erlang.org/faq.html
>> erlang-questions (at) erlang.org
>>
>>
>> ________________________________________________________________
>> erlang-questions mailing list. See http://www.erlang.org/faq.html
>> erlang-questions (at) erlang.org
>>     
>
>