On Tue, 2015-08-11 at 15:42 +0200, Ralf Corsepius wrote: > On 08/11/2015 01:32 PM, Patrick O'Callaghan wrote: > > On Tue, 2015-08-11 at 13:13 +0200, Ralf Corsepius wrote: > > > On 08/11/2015 12:51 PM, Patrick O'Callaghan wrote: > > > > On Tue, 2015-08-11 at 10:35 +0200, Heinz Diehl wrote: > > > > > F22, in short: first running "dnf --refresh upgrade" shows > > > > > some > > > > > new > > > > > packets. Then "dnf clean all" followed by "dnf --refresh > > > > > upgrade" > > > > > shows the same packets to be updated, and *some more*. > > > > > > > > So two update commands at different times give different > > > > results? > > > > > > IIUC, you are misunderstanding. > > > > > > The issues behind this are > > > - "dnf --refetch" is refetching different versions of metadata > > > from > > > different (and differently sync'ed and/or broken) mirrors > > > - fedora's mirrorlists are pointing to mirrors being out of sync. > > > > What matters is whether dnf is seeing the same state the two times > > it > > runs. In this case it clearly isn't. > Exactly. With "dnf --refresh" it often does not see the same state. > > However, it should! The fact it does not see the same state, means > https://mirrors.fedoraproject.org pushing bogus information. > > > > In addition to that, https://mirrors.fedoraproject.org seems to > > > have > > > been down and inaccessible for several hours, last weekend, which > > > caused additional issues with dnf (and yum). > > > > Same result. If dnf is run twice you can't guarantee it will give > > the > > same result. > If dnf and mirror-management was functional, then - except in those > rare > situations when the master has just been updated - they must point to > mirrors carrying an identical state. Only if all the mirrors you access are in sync with each other. Without a consensus protocol this cannot be guaranteed. > > That's an inherent feature of loosely distributed systems > > (where there isn't a distributed consensus protocol). Obviously the > > wider apart the two runs, the more differences will tend to appear, > > but > > the presence of differences does not in itself indicate a problem. > My assumption is: mirror-manager is dysfunctional and dnf isn't > sufficiently robust. "Robustness" can mean different things, at least two of which are: 1) dnf always gets the latest updates any reachable mirror has 2) All the mirrors show the same updates so dnf can get any of them These two criteria are not the same, unless there's a consensus protocol, which I'd lay serious money there isn't as it's expensive to do and is fragile in the face of network outages. I'm not saying that dnf couldn't be better than it is (I've had problems similar to what you report), but it cannot be perfect in these conditions. Remember that yum wasn't either, it's just more mature. I suggest you check out the vast literature on distributed synchronization if you want to know more about this stuff. poc -- users mailing list users@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe or change subscription options: https://admin.fedoraproject.org/mailman/listinfo/users Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines Have a question? Ask away: http://ask.fedoraproject.org
