Re: [ANNOUNCE] perdition 1.18-rc1

On Fri, Sep 04, 2009 at 03:36:22PM +0900, Christian Balzer wrote:
> 
> Hello Simon,
> 
> Oh, 1.18-rc1 made it into sid and thus my test boxes, will give it a whirl
> later and next week then. ^_^
> 
> On Fri, 4 Sep 2009 16:10:52 +1000 Simon Horman wrote:
> > On Fri, Sep 04, 2009 at 01:15:16PM +0900, Christian Balzer wrote:
> > > 
> > > Hello Simon,
> > > 
> > > On Fri, 4 Sep 2009 13:24:41 +1000 Simon Horman wrote:
> > > > On Fri, Sep 04, 2009 at 11:23:15AM +0900, Christian Balzer wrote:
> > > > >
> > > > > Looking at the Changelog I didn't spot anything relating to it, so
> > > > > would there be a chance to include the feature/fix raised in:
> > > > > http://lists.vergenet.net/pipermail/perdition-users/2009-January/002074.html
> > > > > 
> > > > > Or at least adding it to the TODO? 
> > > > > 
> > > > 
> > > > Hi Christian,
> > > > 
> > > > sorry that one missed the release. It seems like good idea to me.
> > > > Could you tell me if in the case where you are seeing the problem
> > > > the client is connecting using plain-text or SSL/TLS ?
> > > > 
> > > The particular case back then (never happened again, btw) was plain
> > > text connects. But again, most likely something in the chinese firewall
> > > caused that behavior. 
> > > The problem that I'm still seeing here as mentioned in:
> > > http://lists.vergenet.net/pipermail/perdition-users/2008-February/001973.html
> > > is entirely SSL (pop3s/imaps) based. Sessions/processes will remain in
> > > the "connect" basically forever.
> > > Fixing the actual issue (which is likely something SSL related) would
> > > of course be nice, but some setting that kills off processes that are
> > > in connect state for a given time has ultimately the same effect and
> > > is more universally useful. ^_^
> > 
> > Agreed on all counts.
> > 
> > At this stage I am thinking in terms of a idle timeout.
> > Though I think the default will need to be infinite as
> > some people like to leave idle IMAP session lying around.
> > 
> > 
> 
> Those IMAP sessions would not be a affected by an idle timeout specific 
> to the CONNECT state. And a timeout that affects processes in any state 
> would not be very helpful for avoiding resource starvation while providing
> "eternal" IMAP sessions (which usually see traffic at least every 30
> minutes, if I remember my dovecot configs and RFC correctly).
> Basically a process has no business being in CONNECT for more than 20-60
> seconds.

Understood. I was a bit confused about when you wanted the time-out to
occur.

It should just really be a matter of digging through a few different
levels of calls to io functions so that a time-out can be passed by
the caller - which will know what stage the connection is at - rather
than they current arrangement of always using a global value.
I noisy but simple change I suspect. I'll see if I can get a patch together
shortly.

______________________________________________
Perdition-users mailing list
[email protected]
http://lists.vergenet.net/listinfo/perdition-users