launchpad-reviewers team mailing list archive

[Colin Watson <cjwatson@xxxxxxxxxxxxx>]

Thanks.  We try!

Package deltas (assuming you mean those rather than index deltas, pdiffs
or whatever) haven't been as much of a pressing issue, so there's
nothing currently in progress on those.  But now that we have much more
scalable librarian space thanks to Swift, there are useful things we
could do that look more like a modern service: prepare deltas in the
database/librarian, expire them some threshold after the packages
they're based on are superseded, and have a service in prodstack that
proxies the paths that debdelta would want to fetch to the appropriate
librarian paths.  We'd then be able to scale much more sensibly than
previous attempts to prototype this.

Would you like us to put this or something like it on the OLS feature
backlog?  I do think that we ought to do diskless PPAs first to relieve
ourselves of ppa.launchpad.net being a single big
un-horizontally-scalable host attached to the SAN.  That project would
involve building a good chunk of the same technology, we can do it
without needing to coordinate with any client-side work, and we
intentionally implemented by-hash in such a way that it would help with
the first stages of diskless archives.  Our best guess on performance is
that we'll need PS5 Swift to be in place for the librarian before we can
deploy either of these, so there's a certain lead time involved anyway.

-- 
https://code.launchpad.net/~cjwatson/launchpad/fewer-by-hash-algorithms/+merge/291871
Your team Launchpad code reviewers is requested to review the proposed merge of lp:~cjwatson/launchpad/fewer-by-hash-algorithms into lp:launchpad.