On 6/30/06, Matthew Toseland <toad at amphibian.dyndns.org> wrote: > On Fri, Jun 30, 2006 at 01:36:50PM -0400, Evan Daniel wrote: > > On 6/30/06, Florent Daigni?re (NextGen$) <nextgens at freenetproject.org> > > wrote: > > >* Evan Daniel <evanbd at gmail.com> [2006-06-29 19:58:27]: > > > > > >> Now that we have an auto-updater, it seems there is more reason than > > >> normal not have self-mandatory builds. > > > > > >Huh ? How comes you can't update using the auto-updater ? Why do you > > >think we are releasing time-delayed self-mandatory builds ? > > > > Because you only give it 12 hours, and by the time it's mandatory my > > node doesn't have it? (yes, this has happened.) Because my node was > > down for those 12 hours? > > And all your peers had upgraded? The assumption that I made was that the > data will be propagated to pretty much all nodes by the time it becomes > mandatory, and therefore any node which isn't online should be able to > get it from another peer which was but didn't update. > > Maybe this assumption is faulty. If it is then we need > update-over-mandatory support (support for updating your node via your > peers even if they are incompatible).
I've had it happen once, but then I only had a few peers active (I think it was 5, maye only 3 or 4). It seems like it would usually but not always be a vald assumption. > > > Because I installed an old version from a > > distro package (probably more of a future concern)? > > A distro package should be updated regularly through the distro. Very few distros update every day or two, and that's the sort of time frame the new build should be mandatory in... > > > What happens if > > I'm behind an oppresive firewall and can't get to a normal we download > > site, and went to great trouble to have a copy snail-mailed to me, > > only to find it's too out of date? > > This is a legitimate concern. The solution is update-over-mandatory. > > > Why does it matter why it doesn't > > work? There are people who will have trouble, I guarantee it. > > > > >> What if, instead of refusing to talk to old builds, nodes simply only > > >> allowed a very small number of requests from them, and routed a small > > >> (or zero) number to them? It seems to me that would be sufficient for > > >> most of the reasons that mandatory builds happen. > > > > > >No. If it's self-mandatory, there is a reason why it is. > > > > Usually because it hurts the rest of the network by sending too many / > > too few requests out. In either case, I don't see how staying > > connected but only accepting a few requests is worse than > > disconnecting. > > > > >> This would allow people with old builds (if they've been on vacation, > > >> or more importantly if they got the build from a distro package or > > >> some such) to connect at least enough to run the auto-updater. > > >> > > >> It also seems a node could prioritize (local) auto-updater requests > > >> over other requests without hurting security, and that it would want > > >> to if it was being throttled as penalty for being old, so that it > > >> would get the update finished ASAP. > > > > > >The problem is that not-up-to-date nodes AREN'T using the auto-updater. > > >And for nodes wich were off during the delay period, 'too bad' for them > > >:p ... > > > > Perhaps for people who have a 24/7 computer to dedicate to the > > project. Which I suspect strongly intersects with people who don't > > really "need" freenet, just like playing with it. Oh, and my node > > doesn't stayup all the time. It crashes occasionally, and doesn't > > come back automagically, so apparently too bad for me too, even though > > I have a computer that's basically up 24/7? > > The wrapper should restart it when it crashes. If it doesn't then there > is a serious problem. It only started recently, and has happened (IIRC) three times in the last week or so. I'll check that everything is configured correctly wrt the wrapper and such when I get home. > > > > > >What we need is update-over mandatory support. > > >https://bugs.freenetproject.org/view.php?id=434 > > > > > >and for distro packages, maybe > > >https://bugs.freenetproject.org/view.php?id=519 > > > > No, we don't need mandatory automatic updating. That's a Bad Thing, > > whether it's Freenet or Microsoft WGA. > > He's not talking about mandatory automatic updating. He's talking about > it being *possible* to update a node from its peers even if they are an > incompatible build. That would make me happy; I didn't actually read the bug, just the email thread, so I got it wrong. Oops. Feel free to just count al my arguments as a vote in favor of making update over mandatory work sooner rather than later :) I just think it would be easier / simpler / quicker to do my proposal, but I certainly won't be offended if you disagree :) I'm still unconvinced that it's a bad thing to accept a few requests from old nodes and not send them any (or rather that it's worse than disconnecting them, even if we only care about our node / network health and not their use experience). But it doesn't make much difference if UOM gets implemented. Thanks Evan
