[Vm-dev] Squeak socket problem ... help!

[Levente Uzonyi]

On Sat, 11 Oct 2014, David T. Lewis wrote:

>
> On Fri, Oct 10, 2014 at 01:15:21PM +0200, G?ran Krampe wrote:
>>
>> On 10/10/2014 11:46 AM, G?ran Krampe wrote:
>>> Hi guys!
>>>
>>> Long post but an interesting one. Summary for the impatient:
>>>
>>> - Ron seems to prove that:
>>>     - We never reach 1024 fds so no leak.
>>>     - But select() still fails, so we probably have an obscure close()
>>> bug with fds.
>>>
>>> - And I prove that select() is indeed broken above 1024.
>>
>> Aha!
>>
>> It seems Ragnar Hojland Espinosa has already fixed this 11 years ago:
>>
>> http://lists.squeakfoundation.org/pipermail/squeak-dev/2003-May/057750.html
>
> I entered Mantis 7822: "Unix aio VM support for 1024+ descriptors" to track
> this issue:
>
>  http://bugs.squeak.org/view.php?id=7822

I think this is a great summary of the current status on unix systems: 
http://daniel.haxx.se/docs/poll-vs-select.html
I think the best would be to use libevent or libev, but adding support for 
epoll (on linux only) is probably a lot easier thing to do.

>
>>
>> I also found code from David Shaffer that fixes the listening to not
>> miss stuff, also interesting:
>>
>> http://lists.squeakfoundation.org/pipermail/squeak-dev/2005-August/093575.html
>>
>> ...and the patch he mentions:
>>
>> http://lists.squeakfoundation.org/pipermail/squeak-dev/2005-August/093638.html
>>
>
> I entered Mantis 7823: "Patch for socket accept problem on unix VM" to track
> this issue:
>
>  http://bugs.squeak.org/view.php?id=7823

This is pretty interesting. Now I know where those spurious 504 responses 
come from.


Levente

>
> Both of these issues will require some follow up work in order to get them
> included in the VMs. If further progress is made, please keep the Mantis issues
> updated and/or report progress to the vm-dev list.
>
> Dave
>
>