On 24/03/15 07:34, Eduard Bloch wrote: > Hallo, > * Carlos Maddela [Mon, Mar 23 2015, 06:20:46PM]: >> >> On 22/03/15 21:25, Eduard Bloch wrote: >>> Hallo, >>> * Carlos Maddela [Fri, Mar 20 2015, 01:37:03AM]: >>> >>>>>> https://mirrors.fedoraproject.org/publiclist . Some of the sites you >>>>>> added are already in the list that I compiled a couple of weeks ago but >>>>>> not all; and they are also not on the publiclist page. Could you reveal >>>>>> the source of that mirror urls? >>>>> Those mirrors were automatically generated by the metalink line in the >>>>> standard fedora-repos-rawhide package, which was >>>>> https://mirrors.fedoraproject.org/metalink?repo=rawhide&arch=$basearch. >>>>> I ended up disabling the metalink feature, because just when you thought >>>>> you had all mirrors covered, a new one would be chosen. >>> I am thinking about using this metalink service to create my mirror >>> lists instead of parsing the publiclist page. I hacked something >>> together. Could you please verify the resulting output? Script attached, >>> run it like: >>> DBTMP=/tmp sh fedorasources.sh >>> >>> Thanks. >>> >>> I am not sure how to interpret the urls, and rawhide repositories are >>> maybe something that needs to be requested explicitly?! >>> >> I tried the script with 3 different Australian ISPs multiple times and >> all resulted in the same output attached. I suspect the output will > I have the weird feeling that this web service has some location > based extra handling. When I run the command below, I get a list of > 81 mirrors! (From Europe, Germany) > >> never be a comprehensive list, perhaps just the mirrors closest to your >> location and/or with the least amount of traffic. Some of the mirrors >> included in my patch were not output this time. > I looked at your list and picked one example of the missing ones > (missing in MY longer list, of course). Which has led me to: > http://ftp.jaist.ac.jp/pub/Linux/Fedora/releases/21/Images/ > > and it contains armhfp arch here (at least I hope this is the case). > IMHO it should have been reported when metalink was called with: > >> arch: repo=fedora-21&arch=armhfp > Right? Or maybe this is the nondeterministic behaviour that you > mentioned above? Should it be called 3-10 times maybe? > > I've had a quick look at the mirrormanager code from https://git.fedorahosted.org/cgit/mirrormanager/ and found you can pass these other parameters with your request:
> path={e.g. > /pub/fedora/linux/development/rawhide/x86_64/os/repodata/repomd.xml > instead of repo and arch} > country={e.g. de} > netblock={??? Didn't look into much, but maybe xxx.xxx.xxx.xxx/xxx} > location={??? Didn't look into much either, but maybe just an alias to > country} > ip={Any ip address you want to simulate that the request came from, > but most useful when 0.0.0.0 for global} Also, instead of metalink, the request may be done through mirrorlist. The output is a simple list, which is easier to parse than the XML output by metalink. For example, the following request: > https://mirrors.fedoraproject.org/mirrorlist?repo=fedora-21&arch=armhfp&ip=0.0.0.0 includes ftp.jaist.ac.jp, as you would expect. In the code, I came across some comments that mentioned that some mirrors may be made visible only to other mirrors. So although adding the ip=0.0.0.0 will give you a more complete list, it would never be a truly complete list unless you mirror the archive yourself. A case in point is the Australian mirror that I've always used: mirror.aarnet.edu.au. It doesn't seem to be officially listed as a Fedora mirror, yet you already have it included as a Fedora mirror in apt-cacher-ng. Best regards, Carlos -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org