On Wed, Feb 3, 2010 at 10:53, Florian Pritz <bluewind@xxxxxxxxxxxxxxxx> wrote: > On 02/03/2010 03:12 PM, Lee Burton wrote: >> As for push mirroring, http://www.debian.org/mirror/push_server is a >> decent example >> An identity file with >> no-port-forwarding,no-X11-forwarding,no-agent-forwarding,no-pty,command="/path/to/mirror/script",from="IPADDRESS" >> &" >> Is fairly decent.. > I've talked with someone working on the data distribution system (big > webcluster) for some big company and he said they haven't had good > experience with pushing. Polling (often) has yet been the best solution. > Actually the patch I posted here is quite similar to their system. > > PS: Please don't toppost. > -- > Florian Pritz -- {flo,bluewind}@server-speed.net > In fact Debian does force commands.. just as you suggested earlier. I agree that polling is probably a better solution. To make it "multi-tiered" and to reduce load on the primary mirror could have slightly more intelligent polling than just checking one upstream machine. In this example Let: Primary = Arch Primary Mirror/Mirrors (updated directly by the dbscripts). Tier-1 = Large High-Bandwidth/Traffic mirrors that other mirrors mirror off of Tier-2 = Smaller mirrors It would then go something like: A tier-1 mirror would check against the Primaries once a minute (for the md5sum). A tier-2 mirror would check against two tier-1 mirrors and see if they agree, if they don't it would ask a primary for a tie-break. It would then could notify (via an automated email?, perhaps one in a 24-hour period? if it's been out of date for XX hours) the mirror owner of the out of date mirror? Also I forget, does archlinux/pacman do any sort of GPG checks/signing with packages? Apologies on topposting, I hadn't responded to very much list traffic with this mail client, and have now changed the client's behavior. -- Lee Burton lburton@xxxxxxxx 301 910 0246
