On Fri, 2017-03-17 at 10:59 +0100, Luca Beltrame wrote:
Il giorno Fri, 17 Mar 2017 10:16:29 +0100 Dimstar / Dominique Leuenberger <dimstar@opensuse.org> ha scritto:
Thanks for your info - Seems we have an added complexity with Leaseeb (the only mirror currently having the file) in as they do geolocation
For the record, this can occur with many different files in different repos: this time it hit TW (but the KDE: repos were regularly affected). It's something that I already filed a ticket to admin@o.o, but it gets "solved" when a repo is updated and re-published.
I just posted some info on a bug, that might be interested for everybody here in the public too - the situation is 'tricky': for mirrorbrain, we have the server registered as "mirror.leaseweb.com" which 'here' resolves to: mirror.ams1.nl.leaseweb.net. and there comes the problem: LeaseWeb is not 'one server' - so even though mirrorbrain scans the 'registered' one to confirm files are in place when we redirect, that does not have to match how LeaseWeb's internal sync is complete. While debugging this together with a user over IRC, we found for example that he was redirected (by geopi dns resolution) to mirror.sfo12.us.leaseweb.net.http That's generally nice of Leaseweb, but it kills all assumptions and knowledge mirrorbrain has a) the 'location based redirection' for the server is useless: we could offer the SFO server to US users at a better prio than an NL server (this only makes a difference when there are multiple mirrors found - otherwise the NL is 'good enough' for US users) b) MirrorBrain has no information about the sync status internal at leaseweb - it only has the information about the one it can reach: which happens to be the geographically closest one in NL - and that one had the files in place at the time the users got redirected So, seems LeaseWeb found a nice way to circumvent all the logic put into MirrorBrain to not redirect to dead mirrors :( And, of course, there is a chance other CDNs will hit us in similar ways. Cheers, Dominique