On Tue, 2009-09-01 at 11:52 -0700, [email protected] wrote:
> On Tue, Sep 01, 2009 at 02:20:04AM -0400, Joanmarie Diggs wrote:
> > 
> > On Mon, 2009-08-31 at 20:16 -0700, [email protected] wrote:
> > > On Mon, Aug 31, 2009 at 06:46:37PM -0400, Joanmarie Diggs wrote:
> > > > Hey Padraig.
> > > > 
> > > > With respect to the issue below, __main_application_quit ain't quitting;
> > > > it's getting hung up on the sys.exit(0).
> > > > 
> > > > Looking at some DTrace output, I *think* that the failure of
> > > > __main_application_quit to quit is allowing the transport engine to
> > > > continue running -- and with my connection to the outside world severed
> > > > (or the repo down as was the case this weekend), it just keeps
> > > > trying....
> > > 
> > > Can you attach the DTrace output?  I had thought that the transport was
> > > able to cope with having its consumers closed. I'm in seeing what it is
> > > that continues to run.
> > 
> > Done. Not sure if the attachment will get scrubbed, but since I'm
> > including you and Padraig directly I assume that will suffice.
> 
> The attachment came through just fine.  Thanks for including that
> information.
> 
> > I trimmed it so that the first line begins with my closing the window.
> > The output ends at the point where I killed the process approximately 2
> > minutes later.
> 
> I'm going to revise my hypothesis a bit.  I previously said that this
> problem occurred because the cleanup thread was trying to access the
> transport.  That doesn't appear to be the case.  It looks like in the
> standard case, we wait for all threads to finish, which is why it
> doesn't exit immediately.
> 
> I'm still curious, though, what conditions you're running this under.
> It appears that every network operation is returning an error. 

As I stated above, I'd severed my network connection. This was to
replicate what I saw this weekend when the dev repo went down at a time
coinciding with my need to install a package. Then and/or in the
subsequent tests:

* If the catalog loads from cache, marking an item for install and 
  clicking Install/Update hangs PM (button stayed in Pressed state)

* If the catalog doesn't load from cache and/or you do a refresh,
  Packagemanager hangs (unless the connection to the repo is restored)

* Same deal it seems with doing a search

What I'm curious about is does this have anything to do with the timeout
prevention work? During one of my trials, I didn't kill the PM process
and restored my network connection after around half an hour. Shortly
thereafter the process completed (or at least gracefully exited -- I
didn't bother to check which).

--joanie

_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to