Hi Alberto, On Tue, Sep 30, 2008 at 04:14:57PM -0500, Alberto Passalacqua wrote:
The machine was never under high load during openSUSE release. It had a remarkably low load during peak times, less than 1.0 accounting for Apache and 1.something accounting for other services running on the machine (mostly rsync, but rsync is mostly busy before release, not after. Once we have released, everything is already synced.).
Do you maybe interpret something else as an assumed load problem? It seems an assumption of yours.
I interpret as overload a timeout while connecting to download.opensuse.org, and I thought to overload because it happened exactly during the release days. We had some complaint on IRC I can remember of.
Okay, then it most likely it was a mirror.
Is it possible that you were sent to a *mirror* which was overloaded?
I don't know. I remember I could not install patches or add installations sources because the URL could not be found.
Against overloaded mirrors, there are two things that can be done: 1) when downloading an ISO image, use a metalink client. download.opensuse.org will give it the list of best mirrors and the metalink client will automatically cope with failing mirrors. 2) for zypp/YaST, there is early support for metalinks as well, which would help for the download of repository metadata and packages (which this discussion refers to).
That's my point. Adding a "download.opensuse.org" was not possible immediately after there release. I still remember I could not install patches for that reason.
On a related note, I have been working with someone from the US during the last days to gather data on performance of individual mirrors, and I actually disabled 4 US mirrors today, which means that the remaining ones should give a much better service to you. Yes, the mirror situation in the US isn't very good. Hm, are you from the US at all?? Email address looks more like Italy. (I may blindly have assumed you were from the US because I saw quite some complaints from there recently.) Well, in Italy there is only garr.it, but that one seems to work pretty fine most of the time, does it? It has some problems sometimes, but every mirror has. Of course it is possible that it was overloaded on the release day. However: I actually remember now, while browsing the logs of the time, that our mirror surveillance had a bug during 11.0 release time :-( which caused failing mirrors to go unnoticed for a few days. A bug that was already fixed, sneaked back in by updating a package from the buildservice, that still hadn't been rebuilt after a day :( This might be the cause of what you have been seeing. Such a bug hits harder when there is exactly one mirror in a country. Anyway, this rather proves your point about online accessess during installation. There will always be something wrong, in a complex system of dozens of mirrors. This is exactly why I mandate download failover for libzypp: See http://en.opensuse.org/Libzypp/Failover My point is that we need to deal with this kind of problem *not only* during install time. The problem is largely the same when anyone wants to update their running system. However, a failure during install time is clearly more fatal.
I have no doubts there are a lot of possibilities to reduce the problem. My main message was that I don't think we need this at all, because the current situation makes a very limited number of users unhappy, and satisfy a lot more users. I don't think we should listen to "bad press" for every detail. OpenSUSE, as any other big distribution, will always receive "bad press".
Indeed, I agree insofar that it it isn't useful to try to make the loudest crying minority happy. Peter -- Contact: admin@opensuse.org (a.k.a. ftpadmin@suse.com) #opensuse-mirrors on freenode.net Info: http://en.opensuse.org/Mirror_Infrastructure SUSE LINUX Products GmbH Research & Development