On Fri, Jan 09, 2009 at 09:54:02PM +0100, Francisco J Ballesteros wrote: > it's not reliable. > I had to fix several things, but the main problem is that > it's not coherent wrt sources. You might end up with incosistent > files, depending on when lsr runs. > > I'm using a variant of lsr, but I'm looking forward to a more > coherent way of doing it.
I am still missing something (and did not get a response on an earlier
attempt at asking a slightly different question), but... why doesn't sources
1) allow read-only access to its backing venti to the Internet at large
2) nightly run a full vac job against the nightly dump [1] as none [2] and
3) publish the score so derived? [3]
This makes "venti/copy -f" an excellent mirroring mechanism; mirrors would
be only very slightly lagged from the official sources (see [1]), and it
would seem to have only very modest bandwidth requirements relative to full
tree traversal.
If this an acceptable mechanism (I don't see anything wrong with it, but
I've been wrong before), I'd be very happy if somebody seeded the system
with the dumps already taken.
[1] And/or snapshots. Those would work just as well, be just as coherent,
and be more frequent in case nightly crawls aren't good enough.
[2] Running as none means that only world-readable files get included in the
published capabilities. Presumably that's exactly the class that would
be mirrored.
[3] A vac archive generated by merging each run's vac output, say, with the
last run's root score published at the root of sources?
Thanks much.
--nwf;
pgpnibPr4kN13.pgp
Description: PGP signature
