Notes were taken by Andrew Bennetts.
- why mirror stuff?
- 1) difficult to find archives scattered around the internet. Supermirror is central.
- 2) generate stats about arch archives in general. Average sizes, how many branches of stuff like baz vs. tla, and so on.
- 3) backups
we support baz & tla archives. Signed and unsigned. We want to mirror it all.
- where do the archives come from?
- 1) james finds out about it on mailing lists or whatever
- 2) our importd-generated archives
- 3) push archives. These will be hooked into launchpad, so users can push directly to the supermirror, even without anywhere else to publish it.
- about 80% james is told directly about archives.
- supermirror scans arch.ubuntu.com for new archives
- launchpad users will be able to push to the supermirror into a private area, which will then be exposed to the public area.
kiko: how much history is in the archive? Same as any other archive, so same info as arch stores.
steve: what are push archives? why would a user want one? some people don't have a place to host a public archive. with push archives + launchpad, they can host their source with us. "push" refers to the arch mirroring term: the author the archive sends the code to the mirror, rather than the mirror "pulling" from the author.In general, Steve is strongly in favour of "push" and "pull" as terminology for arch.
- how often are updates? by default 12 hours +/- a bit.
- supermirror is currently using tla (+ extensions to deal with baz archives). All archives on it are in tla format, and will remain that way for the moment.
- as a prospective user, what do you need to do to set up. Two possiblities we could provide:
- 1) just a directory
- easier for uploading existing archives
- lets users store in whatever format they want (inc. baz format)
- 2) we set up an archive for them
we set the format -> we force choice of client
- Which do we use?
- 1) just a directory
- steve: what is the business goal?
- we're already pulling in many archives for importd
- people will want to access and branch from this stuff, and this provides space for people to do that.
we can support accepting in one format, and publishing in another (at least up to a point -- we're not at that point yet, we can convert between baz<-->tla atm).
steve: if we provide "blame" functionality on the supermirror, that's would be a big incentive for people to use the supermirror.
james: the supermirror could also do web-of-trust (steve: web-of-fork) stuff, pretty graphs and things.
- the supermirror predates james being employed by canonical -- it already had160 archives.
- so we already have lots of people's tla archives, so we don't want to force people to use baz to use the supermirror, because we will upset existing users.
- people use the supermirror to find the primary source of an archive atm.
- james expects it will become a big bandwith drain in a few years.
- the supermirror was rewritten to use ZPT in Oxford. Steve would like the code easily accessible to other staff for easy review. Steve should ping James about this in a few days. The supermirror code has never been released; it's owned by Canonical, and there's been no decisions regarding open sourcing the code for it. The data it mirrors is of course open.
- we're mirroring non-arch controlled repositories that we've converted (via importd) as arch at the supermirror. These exist alongside individuals private archives.