-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Just a heads-up: thanks to recent development fixes in rsync 3.0.0-cvs, we can actually update our internal backups of the upload fileserver on a reasonable schedule.
The incremental recursion mode, new in 3.0, means that the millions of files can be copied as the directories are scanned, instead of building a file list first that's so big that rsync uses all memory on the server and dies before copying anything.
More importantly, a recent fix keeps rsync from segfaulting every few hours, so it can actually get to the end. ;)
Currently uploads are being replicated from amane, the main fileserver, to storage2 in the Tampa facility. We're planning to also start copying them over to some space on the disk array for the toolserver in Amsterdam, which should allow us to:
a) Have a clean offsite backup b) Allow handy access to the public files for toolserver users
That's currently waiting on the array to be fully set up.
It might also be feasible to set up a public rsync server to allow public fetches, but we'll have to see about load and configuration issues.
- -- brion vibber (brion @ wikimedia.org)