|[p2-dev] Evaluate the usefulness of using Eclipse mirrors|
Hi,All of our builds that run outside Eclipse are now started with -Declipse.p2.mirrors=false. The same is true for my own IDE. Why? Because everything works much better that way even though I'm in Europe. Our builds fail intermittently without this property and updating my IDE seems to take forever.
I'm well aware that the use of this flag will increase the load on eclipse.org and I really wish I could avoid that, but with builds frequently failing and downloads that frequently hit sites that take forever to never respond, I can't.
I think there are many reasons for why the mirrors concept performs badly. The most obvious one is that some update sites have a mirrorsURL that is incorrect. I've seen a lot of sites for one version (say 3.7) that appoints an update site targeted for another (3.6) and thereby sending p2 on a wild goose chase around the world in order to find something that it will never find. I've been guilty of that error myself. It's easy enough to forget. I've seen other sites where this URL is just plain wrong. Site's that use a URL although they don't have mirrors at all, etc.
Perhaps it's time to evaluate the whole concept of using mirrors this way? Does it bring any advantages in real life or is it just a feature that works well in theory? Could it be that it causes a lot of pain and time consuming problems for users of p2 and hence bad will for p2 (this is my own concern) so that we could do better without it? If so, what other options are there?
In my opinion, the whole concept of load balancing and mirrors should be outside p2's concern. How to best distribute the material and ensure delivery is far better catered for using other well-known technologies specially designed to handle this and Eclipse.org already exploits this for read-only CVS.
How do others perceive the use of mirrors in p2? Regards, Thomas Hallgren
Back to the top