2009/3/29 Tarek Ziadé <[email protected]>:
> RIght I am fixing this right now.
>
> There's something we didn't talk about yet : since there will be other
> package indexes
> out there (not PyPI mirrors) that might have their own mirrors, we do
> need to provide
> somewhere the hostname that holds the mirrors IP for the client software to 
> work
> the same way in all case.
>
> That is, mirrors.pypi.python.org for PyPI, but maybe
> mirrors.packages.plone.org for
> another index.
>
> I'd go for a /mirror-hostname unique page at PyPI (and its mirrors)
> even if this page is unreachable when PyPI is down.
>
> This enforces that the other indexes also use the DNS technique, but I
> think it's fine

Another approach is some way of detecting the mirror index (e.g., a
<link> on the index front page), and strongly suggest that clients
cache that mirror index location.

The most reliable way for a tool like pip to use the mirror, I think,
would be to try the main index always to get metadata, then it could
use a mirror for fetching the actual packages.  There's less
synchronization issues in that case, and only when an index is down
would pip need to fall back entirely on the mirror.  Anyway, this
would fit the mirror index detection pattern.

-- 
Ian Bicking  |  http://blog.ianbicking.org
_______________________________________________
Catalog-SIG mailing list
[email protected]
http://mail.python.org/mailman/listinfo/catalog-sig

Reply via email to