3 ms·
I've never had any of these issues, so I'm surprised by how spectacularly the author's setup broke. I can understand some Python library failing during the auto
by jeroenhd 6y ago
I've never had any of these issues, so I'm surprised by how spectacularly the author's setup broke. I can understand some Python library failing during the auto update process needing a rollback, but needing a complete reimage to fix the situation? How does that even happen?
> new protocol for the letsencrypt challenges was released, announced in some obscure corner of the web and bit by bit the old method of issuing new certificates was deprecated. You would not have known this unless you interacted with Letsencrypt on a daily basis, but like most of these things their whole value lies in NOT having to interact with them all the time.
Let's Encrypt also sent out emails to people still using the old method at least two months before the old method was phased out. You have to provide an email address when setting up LE exactly for this reason. The only explanation I can think of for why the author wouldn't have received an email is that they did not execute a renew check for their certificate in the three months leading up to the change of service, which is going straight against LE's guidelines (you should check more often, exactly for this reason). This is the
> which is odd because after all the situation hasn’t changed and if it worked last time it should still work this time
The situation obviously _has changed_ otherwise the venv wouldn't have been updated. Libraries get bugfixes even if those don't apply to you specifically; deal with it. You're not the only customer here.
> I understand that Letsencrypt needs to stay with the times. But the Letsencrypt folks should also understand that in the greater scheme of things certificates are not the most important thing in life and that causing your end users this level of grief is in the longer term not in the projects’ best interest.
Web security is not a "set and forget" thing. It's not a toggle button from "insecure" to "secure". If you want a zero-maintenance certificate solution, get a managed server and get someone else to handle the annoying stuff for you.
> The machine that this is all running on will need a much more recent version of its OS. That means I’m going to have to migrate a whole bunch of data to another machine in order to be able to do this without downtime.
This is unsurprising coming from someone who wrote a blog about why they don't like to update. However, the assumption that the entire host OS needs to update just to use a new piece of software is simply wrong. Things like Docker and LXC make it trivial to run newer software without upgrading the entire system. It also forces the software into a contained space to prevent the catastrophic failure the author had to deal with before.
From what I can tell from a Shodan check for this domain, the author still seems to be running Ubuntu 14 (which went EOL last year). That probably explains why the author is afraid that they will need to upgrade their server. However, if you're running on a software stack whose LTS support has already been dropped, you're just postponing the inevitable by not planning an upgrade plan already. Blaming Let's Encrypt for potentially not supporting a platform nobody else supports is a silly excuse.
I agree that Let's Encrypt has its issues. Sure, their client can use some thinning, their decisions are sometimes confusing, but the normal situation is not nearly as bad as the author describes here. If you're bothered by the official client updating things, use a different client; there's plenty of them.
- jacquesm 6y agoThe machine was installed two years ago with what was then the current version of Mailinabox, which in turn apparently relied on Ubuntu 14.04. So, yes, that is outdated now but it wasn't as if there was much choice at the time and two years seems a bit short - maybe I'm wrong about that - on the expected lifespan of a server. But setting up a mailserver, migrating all of the data back on and then testing it is a boatload of work and I have plenty to do as it is. And yes, I did eventually manage to get it to work but in a pretty roundabout way and that should just simply not have been necessary unless there was a huge security issue with the way LE was working before. I'm not aware of any such issues. Essentially you are arguing that something like Mailinabox can not or should not exist because it falls somewhere between 'managed hosting' and something that can be set up for a few years in order to serve a purpose. That's not a very encouraging future, maintaining basic internet infra such as mail, web and DNS servers should not be so hard that you can only do it if you want to pay third parties big bucks to do so.
- jeroenhd 6y agoMailinabox's use of such an old OS is the exact reason I skipped it and went for Mailcow instead. It also placed very strange limitations on the stuff you could do with the system once it was set up and running according to the documentation, presumably to make sure the project maintainers had only a few types of configurations to support. I have nothing against the concept of Mailinabox but when you create a "hands-off" project like that, your project should also include stable major version upgrades to LTS versions. From what I can tell, MIAB switched from 14.04 directly to 18.04 (skipping 16.04). Their goals and ideas are noble, but I don't think the project is very reliable if they skip major OS releases just like that (and are very late to the party when it comes to support for newer versions, releasing a version compatible with 18.04 only months before 14.04 support was dropped entirely). I appreciate the enormous amount of work it takes to achieve something like this, but marketing your product as something "you don’t need to be a technology expert to set up" brings with it certain expectations (or even, responsibilities). Your issue with an outdated LE client should have never happened if the project had achieved its goals, for example. A modern setup with MIAB based on Ubuntu 18.04 will have OS support until 2028, so the situation has improved; Ubuntu has switched to 10 years of support instead of 5. > Essentially you are arguing that something like Mailinabox can not or should not exist because it falls somewhere between 'managed hosting' and something that can be set up for a few years in order to serve a purpose. The problem is that software doesn't work like that, unfortunately. Bugs and security issues are found and need to be patched. Volunteers don't want to fix five years of releases of their software and for companies it doesn't make financial sense to do such work for free. A Windows server box with regular reboots for security updates will work fine for a bunch of years, but that guarantee is hard to make when your software runs on volunteers and occasional corporate sponsors. > That's not a very encouraging future, maintaining basic internet infra such as mail, web and DNS servers should not be so hard that you can only do it if you want to pay third parties big bucks to do so. There's a reason companies have been providing paid services for this since the dawn of the web. Hosting software on the internet isn't maintenance free and it never was. If you desire a setup that keeps working with minimum maintenance without the risks that come with volunteer work, you'll have to pick a stable OS (like Ubuntu or its competitors) and software stack from the LTS repositories and configure it. You can apt-upgrade anything from the official Debian repositories without reconfiguring anything. You can upgrade and It Just Works as long as you stick to the packages provided by the people maintaining your OS. This works until your OS reaches end-of-life, which is the next time you should do intensive system admin effort. Luckily, this is announced long beforehand, so you can plan accordingly. This still means you'll run into trouble when (free) external parties decide to drop your OS because of the growing amount of software stacks and configurations they have to support over the years. However, if you practice running external stuff in a contained environment (with tools like Docker) you'll get to keep the best of both worlds for as long as possible. Alternatively, you can use setups that don't hide behind magical one-command-fixes-all setups, like manually entering certificates instead of relying on Some Cloud System to do it for you. If all that is too much for you, which I can completely understand, then maybe running your own servers might not be for you. The premise of bringing self-hosted, federated services to the every-day consumer is nice, but I'm afraid it'll always be little more than a dream. The modern web is either part of the Cloud (TM) or a bunch of projects glued together by random chance and best effort; running your own stuff unfortunately requires either money to be paid or effort to be made.