Because going back to non-packaged non-vetted flavor-of-the-month code is a retrograde step back to 1993. You lose consistency, you lose the ability to reliably recreate a same environment, you lose tested and low-friction security updates, you lose dependency management, you lose the security of a crypto web-of-trust, and you lose the google-fu of being on the exact same versions of software as thousands of other people.
I'd tend to agree, with one exception: it's sometimes even more desirable to construct own /usr/local version of some critical packages (database, languages, crucial libs) instead of distro versions; then recreate this on your all production nodes and never upgrade it without a carefully thought reason and pre-prod test.
Two main reasons for this:
- sometimes the version provided by the distro doesn't have set of features which you need and careful admin can provide it for you,
- sometimes distro packagers or upstream folks violate semantic versioning (incidentally or not) and upon upgrade from distro, some of your sub-features ("sub-" doesn't have to mean "not critical") break.
Of course you lose the google-fu you've mentioned, but to be honest, many production environments are so distinct that there's no sane way to compare one to "typical Debian X.Y server".
I personally find FreeBSD approach very convenient wrt to the issues above, because ports evolve independently from the base system; but YMMV.
I'll take your word for it. For me, it mostly saves me a bunch of time.
Edit: Just in case I'm coming off as snarky here, I don't mean to. My projects tend to be on the simpler end, and rvm saves me a lot of time. I'm sure it's a little cowboyish for bigger projects - I was just wondering if there was something about rvm that made it universally inappropriate for production use.
Double Edit: "a bunch" is probably an overstatement. rvm was a great help transitioning to 1.9.2. I'm not sure if it will provide much utility going forward.
This is kind of a classic "fast, correct - choose one" debate. If saving time is more important, perhaps you should use RVM in production, if correctness is more important, you probably shouldn't. It comes down to how expensive your mistakes are.
If you only have one server and can script most of the install process it's not a problem. When you have more than one server, rvm is not going to save you time.
I have my gripes with RVM, but this isn't one of them. I work on a project for provisioning and deploying EC2 servers. It's a capistrano plugin called rubber.
In any event, the immediate benefit is we can provide a tool that lets our users set up whatever Ruby they want to use, as we shouldn't be forcing that on them. Moreover, it makes it dead simple to deploy different Rubies out to different parts of the cluster, since everything is role-based. That's far from the common use case -- most people use the same Ruby everywhere. But if you have more complex needs, RVM affords a lot.
http://qa.debian.org/popcon.php?package=ruby1.8