local double copy of PyPI via HTTP about bandersnatch

[cooperlees]

This was implemented before I started contributing, but, I personally like to know that I don't have a Man in the Middle Attack as 99.99% of people are syncing from PyPI to local bandersnatch repos. It's also very trivial to add HTTPS to your "DMZ" instance. So I would reccomened that route.

Even tho you're hitting your DMZ mirror, you're also still hitting PyPI's XML RPC API to calculate the differences from where your "internal" mirror is (we are unable to mirror that). You are hitting the JSON API and pulling the packages down locally tho, from your DMZ.

• You could also look at using rsync or some other mirror technologies (e.g. btrfs sends) as well.

This all said, I will accept a PR that defaults to enforcing HTTPS only, as we are today, and allows you to negate that check.

• If you go this route, please in the sample config add a comment stating why this is bad and that it really should only be used internally

Thanks for asking. Feel free to ask any more questions.

[gpcimino]

@cooperlees thanks for the clear answer.

I think the rsync option from DMZ to internal is the way to go.
This is what we have in place now, we are not just quite happy with the performance of rsync with gazillion of files in PyPI.

Due to my lack of knowledge of PEP 381 now i realized the internal bandersnatch needs connection to the (real) PyPI server anyway for the XML RPC API call.
Just for my awareness: what's the best way to have the XML RPC server side part on DMZ? Should i use something like devpi?

Thanks
Giampaolo

[cooperlees]

Yeah the 1000s of files and directories do not help. This is why I also suggested btrfs (or it could be zfs) differential sends. They will be fast as they are at the block level based on snapshots.

devpi also can't replicate the XML RPC API, nothing really can as it's the source of truth for package and mirror serials that PyPI calculates in real time.

I've never used devpi, but for it's PyPI operations, it could run in your DMZ and cache all the PyPI package that your infra needs, but it seems it runs as a proxy. devpi does have a "replication" feature that could possibly satisfy you needs. I don't know all your goals, but it sounds like it could do what you want, especially if the replication also syncs the PyPI cache, which I am not sure it does.

[cooperlees]

Were you able to find a work around for this? If so please share and we'll close this issue.

[gpcimino]

Sorry for late answer.
I gave up and we stick with rsync from DMZ to internal mirror, even if performance are not outstanding.
Thanks for your help