On Fri, Jan 08, 2010 at 04:45:56PM -0600, Shawn Walker wrote: > On 01/ 8/10 04:42 PM, [email protected] wrote: > >On Fri, Jan 08, 2010 at 04:23:11PM -0600, Shawn Walker wrote: > >>* If an exception gets raised during plan execution, doesn't this > >>mean that the connections won't get closed in the GUI? (See line > >>788 of imageplan.py). > > > >What problem are you worried about here specifically? The GUI calls > >api.reset() prior to performing an InstallUpdate, so if the first > >install fails the old transport will get shutdown before a new transport > >is created. > > I was just concerned that the connections would just sit around for > a failed plan prepare/evaluate longer than they needed to. I guess > it isn't that big of a deal since libcurl will timeout for them > anyway.
Libcurl's multi interface is asynchrounous, so the close will happen the next time we invoke the transport. The question is whether it's reasonable to keep a connection open to a server. I had thought we decided that this was okay, because the depot can always implement its own policy about how long it will keep idle connections too. If the GUI gets an exception during pre-execute, chances are that the operation will be re-tried since it didn't succeed. In that case, it seems like it would be better to leave the connections open until we know we're not going to need them again. -j _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
