On Thu, Nov 19, 2009 at 05:43:48PM -0600, Shawn Walker wrote:
> [email protected] wrote:
> >On Tue, Nov 17, 2009 at 04:12:11PM -0600, Shawn Walker wrote:
> >>Greetings,
> >>
> >>The following webrev contains fixes and changes for the following issues:
> >>
> >>  12130 install/update operations should always refresh publisher
> >>metadata (when allowed)
> >>  4419 misleading error message when pkg not found because refresh failed
> >>  10976 operations should report when refresh failed if appropriate
> >>
> >>Overview
> >>========
> >>Since the catalog v1 work has been putback, the actual check for new
> >>publisher catalog data is relatively cheap now (this was not
> >>previously the case).  For example, on my system, it takes ~1s real
> >>for the api to check three publisher repositories for new catalog
> >>data.  Previously, this could have taken several seconds.
> >>
> >>As a result, and in light of the continuing confusion when packages
> >>don't seem "immediately available" after publication, the added cost
> >>here was deemed a worthy tradeoff in favour of attempting to
> >>eliminate user confusion for install and update operations.
> >>
> >>Please note that pkg 'info' 'contents' and 'list' remain unchanged
> >>as performance of these operations has been a consistent concern
> >>based on user feedback.
> >>
> >>webrev
> >>======
> >>http://cr.opensolaris.org/~swalker/pkg-12130/
> >
> >The code changes look fine to me, but I'm not sure this is really the
> >right way to go.  I had thought we explictly put a refresh interval into
> >the publisher/0 information.  My understanding was that publishers that
> >expected to publish new content frequently could set this to a low
> >number (like 30 seconds or 5 minutes) and that we would implicitly
> >refresh then.
> >
> >Is is really a good idea to always refresh, even when some repositories
> >will often go a long time without updates.  (p.o.o/release, would be a
> >good example.)
> 
> Even when that functionality is in place, current user feedback has
> indicated that every time an image-update occurs (at the very least)
> it is expected to use the most current software available.
> 
> The alternative would be to limit this change to image-update only,
> and then hope that the future publisher metadata changes to allow
> the publisher to control refresh intervals would be sufficient.
> 
> I had previously asked if it was instead acceptable to simply tell
> the user when we last updated the catalog for each publisher, but
> the consensus seemed to be that was not the right path.

I didn't realize that the clients don't currently pay attention to the
depot's refresh interval.  Knowing that, I think the approach you've
taken in this change makes a lot more sense.

Sorry for my confusion.

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

Reply via email to