[Catalog-sig] PEP 381 (Was: PyPI down?)

["Martin v. Löwis"]

> FWIW, I don't see the PEP to be completed. The actual mirror
> protocol, how to handle multiple (unsynchronized) indexes and the API
> design are clearly undecided -- and not discussed.

The actual mirror protocol *is* decided, even though it's not explicitly
spelled out in the PEP. It is based entirely on existing API; no new API
on the PyPI side is planned. Basically, you do a lot of HTTP GETs.

As Tarek explains, the handling of unsynchronized indices is also
discussed. Each mirror has a timestamp of last synchronization.

I expect that mirrors won't be behind more than two minutes. If you
have a strong requirement to provide a better quality, you'll need
to propose a PEP change. I would personally prefer to have such a
feature only in a future revision of the protocol, based on practical
experience.

If you would like to propose a push model, I'd be curious what the
advantage would be of one of the complicated-sounding APIs you mentioned
over a simple trigger URL that the mirror might provide.

Regards,
Martin