Re: [dev] [surf] many CLOSE_WAIT connections

From: Christoph Lohmann <20h_AT_r-36.net>
Date: Sun, 16 Feb 2014 16:41:38 +0100

Greetings.

On Sun, 16 Feb 2014 16:41:38 +0100 Markus Wichmann <nullplan_AT_gmx.net> wrote:
> On Wed, Dec 25, 2013 at 09:54:51AM +0100, Christoph Lohmann wrote:
> > Greetings.
> >
> > On Wed, 25 Dec 2013 09:54:51 +0100 Markus Teich <markus.teich_AT_stusta.mhn.de> wrote:
> > > Heyho,
> > >
> > > I noticed many CLOSE_WAIT connections from surf in the output of `lsof -i -an`
> > > after browsing some websites with surf. Before blaming someone or trying to fix
> > > it: Is this expected behaviour? Or should it be fixed?
> >
> > Webkit is doing that. Whatever is going wrong there, I won’t dig into
> > the webkit source code. It’s too huge.
> >
> >
> > Sincerely,
> >
> > Christoph Lohmann
> >
> >
>
> There's nothing wrong with this at all. Just that in an environment
> where clients vastly outnumber servers, it is probably a good idea to
> have the clients be the ones to bear the cost of CLOSE_WAIT instead of
> the servers.

That CLOSE_WAIT connection are created when webkit is using its super
optimized loading algorithm to open a connection for all requested re‐
sources. Telling the webkit view to load the next URI does not close
the connections. The only way to close them is to close surf, which ex‐
hibits, that webkit is not closing all sockets properly.

Maybe I will find some way to exhibit some object in webkitgtk to regu‐
late this.


Sincerely,

Christoph Lohmann
Received on Sun Feb 16 2014 - 16:41:38 CET