cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Leo Simons <LSim...@schubergphilis.com>
Subject Re: CloudStack Mirrors
Date Tue, 14 Oct 2014 10:13:46 GMT
On Oct 14, 2014, at 10:44 AM, Erik Weber <terbolous@gmail.com> wrote:
> I'm not familiar with Apache policies in this regard, but is there any way
> we could establish a set of official / supported mirrors that all answer on
> the same DNS address?
> I.e. so that we only have to document one url

Typically the mirroring people use 302 redirects rather than DNS magic.

Basically, to do it right, I think we should choose one route of

1. use existing apache mirror infrastructure
* discuss with infrastructure@ the sanity of hosting systemvms on the existing mirrors, if
ok, @see
  http://www.apache.org/info/how-to-mirror.html
  http://people.apache.org/~bodewig/mirror.html
  http://www.apache.org/dev/mirrors.html
  http://www.apache.org/dev/release-download-pages
* release systemvms as actual/official apache artifacts
* create a custom version of dyn.cgi/download.cgi if/as needed
* update scripts to point at that CGI, i.e.
  http://www.apache.org/dyn/closer.cgi/cloudstack/releases/4.4.1/systemvm/systemvm-4.4.1.xen.bz2

2. use existing sourceforge mirror infrastructure 
* release systemvms as semi-official binaries built from apache source
* create a project on sourceforge
* use the sourceforge mirror infrastructure
* example of apache project doing this
  http://www.openoffice.org/download/index.html
  http://sourceforge.net/projects/openofficeorg.mirror/
* update scripts to point at that magic, i.e.
  curl -o systemvm-4.4.1.xen.bz2 -L http://sourceforge.net/projects/cloudstack.mirror/files/4.4.1/systemvm/systemvm-4.4.1.xen.bz2/download
* tell infrastructure@ we are doing this just so they’re aware of the need and how we addressed
it

The problem with using the apache mirrors is that traditionally apache promises to keep the
total repo somewhat size-limited. Lot of sites mirror apache (and fsf) because its 50GB or
so, as opposed to the 100s of GBs you need to be a source forge or linux mirror.

The OpenOffice approach is interesting because before becoming an apache project they had
huge mirror infrastructure (built with mirrorbrain), some of which actually migrated to apache
hardware IIRC with quite a bit of invested effort, but now it seems they abandoned that, and
are just using sourceforge. So I suggest learning their lessons and doing the same.

All the people around the world providing mirroring for open source projects will probably
appreciate not having another thing to configure :)


cheers,


Leo


Mime
View raw message