>>>>> "Luke" == Luke S Crawford <[EMAIL PROTECTED]> writes:
Luke> What concerns me is that while I see some powerful automation, Luke> I don't see any safety nets. "Don't screw it up" is the order Luke> of the day. That's fine if you have people that are good Luke> enough; but bay area labor is tight right now; (speaking of Luke> which; I get referral bonuses) they've hired me, and I see no Luke> evidence that I am an anomaly. I need safety nets. and we're Luke> loosing the graybeards, we really need to put more effort into Luke> preventing a fat finger from doing an incredible amount of Luke> damage. That, and I want to do some development on our tools, Luke> and I am a little frightened of cowboying it on this scale. Luke> Now, my background is in virtualization; I've been running a Luke> vps provider for the last two years, and I've spent the last Luke> year doing Xen consulting. I agree that it is a big problem. I think that it actually affects a large number of the issues that we have in the configuration tool adoption space. I can only speak for myself here, but I think that this is one of the more important issues facing tool developers at the moment. Luke> so this might be partly a "I've got a hammer, all problems Luke> look like nails" thing, but it is a pretty nice hammer, so I Luke> think it's at least worth looking into. Luke> does anyone else use virtualization to simulate massive Luke> systems for the purpose of testing configs? What other Luke> approaches do other people use to install 'safety nets' that Luke> prevent an undercaffinated admin from taking out a huge number Luke> of servers? what is the state of the art here? We aren't using virtualization for testing yet, but we've built linkage between testing, bcfg2, and its reporting system; from my perspective, use of virtualized resources for programmatic testing is a special case of a generic testing process that needs to be much more widespread than it currently is. You're highlighting something that I think is pretty important here. As a community, we don't have shared notions of tool safety. Our tools don't, in general, provide a way for administrators to describe outcomes in meaningful ways. We've done a bunch of work to improve bcfg2 along these lines, and submitted a paper to this year's LISA on the subject. -nld _______________________________________________ lssconf-discuss mailing list lssconf-discuss@inf.ed.ac.uk http://lists.inf.ed.ac.uk/mailman/listinfo/lssconf-discuss
