On Sun, Feb 20, 2011 at 12:42 PM, Ryan Lane rlane32@gmail.com wrote:
I don't think we should encourage people to run trunk in production.
I think we should encourage people to run trunk in production *if* they're planning on keeping up with it, will report or fix problems, and are aware that critical bugs might crop up and that there's no automated downgrade path once they make schema changes. Which is to say, we should discourage it for almost everyone, yes, but not everyone.
trunk should be stable enough that it's *possible* to run in production, and we should encourage people to do so if they know what they're doing and are willing to accept the risks. People who run trunk in production can get all the newest features sooner, and they help us by providing feedback faster. It's for exactly this reason that trunk needs to be runnable at all times, as much as possible.
Occasionally security issues pop up in trunk that get caught in code review. People who run trunk are much more likely to have security problems, so on a production site, it's a problem. Similarly, it's possible that commits may come in that can cause data loss, which will later get caught in code review.
Both of these are possible, but not common enough in practice to be a big reason not to use MediaWiki trunk. Much more likely is random stuff not working properly because of a bug, but that's not going to cause lasting damage. Of course, if your site is important enough that occasional downtime or breakage is a problem, then you should certainly be more cautious, but most production wikis aren't actually very important.
Our trunk usually stays stable enough that developers can have a usable working environment, but I don't think it's stable enough to run on any site where you care about security or your data.
More than a few sites do actually run it under those conditions, so I don't think this is true at all. The plan is that Wikimedia should go back to doing something close to that, in fact, although of course with a formal review process of some sort.