[Yum-devel] keepalives and pain

[Michael Stenner]

On Fri, Oct 08, 2004 at 10:58:19AM -0400, Ryan Tomayko wrote:
> >So I'm wondering how hard it would be to force close all
> >keepalives at a certain time.
> 
> Hmm. Without having a chance to look at the code, I'm pretty sure there 
> is something for that. If I remember correctly, the KeepaliveHandler 
> holds onto a list of what may be active connections. There may even be 
> a close_all method but I don't think you have a clean interface to call 
> it since the Handler is managed by URLGrabber. I can probably take a 
> peak at this later on tonight.

By default, URLGrabber uses a single keepalive handler.  While I
generally prefer to not use globals, this has the advantage that
several grabber instances can use a single connection rather than
having to use separate one.  This ALSO has the effect that the handler
is easy to track down.  There is already a global funcion in
urlgrabber.grabber for closing all connections in the global keepalive
handler.  You can simply do:

urlgrabber.grabber.close_all()

Note: there's nothing that prevents other pieces of code from creating
other keeplive handler instances.  These will not be affected.
However, I find it very unlikely that you're doing this so the above
method should do exactly what you want.

					-Michael

> On Oct 8, 2004, at 10:32 AM, seth vidal wrote:
> >Hi,
> > We've recently discovered, due to a 'feature' in rpmlib that if a
> >keepalive is left around, especially during a transaction and the 
> >remote
> >server kills off the connection, that yum will immediately exit w/o any
> >warning or traceback due to rpmlib catching the SIGPIPE and handling it
> >via exit(). So I'm wondering how hard it would be to force close all
> >keepalives at a certain time. So when yum reaches the point where it's
> >all inside the rpm transaction and no longer downloading anything I can
> >force close all urlgrabber keepalives, and be reasonably sure that we
> >won't get any spurious sigpipe's from a server sloughing off extra
> >connections.
> >
> >any ideas?
> >-sv
> >
> >
> >_______________________________________________
> >Yum-devel mailing list
> >Yum-devel at linux.duke.edu
> >https://lists.dulug.duke.edu/mailman/listinfo/yum-devel
> 
> _______________________________________________
> Yum-devel mailing list
> Yum-devel at linux.duke.edu
> https://lists.dulug.duke.edu/mailman/listinfo/yum-devel

-- 
  Michael D. Stenner                            mstenner at ece.arizona.edu
  ECE Department, the University of Arizona                 520-626-1619
  1230 E. Speedway Blvd., Tucson, AZ 85721-0104                 ECE 524G