Hello, On Thu, Feb 22, 2018 at 08:17:43AM +0100, deloptes wrote: > Jimmy Johnson wrote: > > I never said that! But I do know what I'm talking about because I do > > what I'm talking about constantly. > > you have said that, because in the official upgrade notes, as Roberto > pointed out, it says you can go only one level up at a time.
I too have done quite a few in-place upgrades where I've skipped releases. In an ideal world I will be deploying a virtual machine straight out of config management so it's quicker to just destroy it and deploy a new one than it is to upgrade an existing one. But occasionally I have to work on some VMs where they're neglected snowflakes and the goal is to bring them up to the current stable release. Since I can just snapshot them and revert to that in the event of catastrophe I am sometimes willing to give skipping a release a try. If it goes terribly wrong I can revert the snapshot and do it one release at a time like you're supposed to. Most of the time it's worked fine. I still wouldn't recommend anyone else to try it though, because some of the time it has gone very badly. IIRC there were some releases where you just couldn't skip through them due to kernel and udev changes. It might have been when trying to go from etch to squeeze without visiting lenny first. Then other times there were problems with specific packages. Since it is explicitly not supported to do this, it's no surprise that some packages break when you try it. I'm surprised it works as well as it does as often as it does. It doesn't really save all that much time. I mean, it's only going to be a reboot or two extra per release. Or course the worst case is that something goes wrong but it's not immediately obvious so you call the job done and only discover the issue after a long period of time, maybe then not even being able to prove it was some issue caused by the way you chose to upgrade as opposed to something else you did in the intervening time. I've never had that happen to me, but it seems like a risk that has to be considered. As I mostly only ever try this on virtual machines they are already quite simple things dedicated to a single task, and I think that helps in spotting aberrant behaviour. So yeah it mostly works, but even speaking as someone who does it, it's not something I'd recommend without fully appreciating the dangers. I've had it break badly! Cheers, Andy -- https://bitfolk.com/ -- No-nonsense VPS hosting
