Miles Fidelman wrote: > I'm getting ready to rebuild a small cluster (4 nodes, xen > virtualization) that I'm using to support a combination of hosted > services (mostly email lists, a few web sites) and some > experimentation. > > To date, I've installed stuff manually, relied on a > high-availability stack (drbd, crm) to provide failover when things > go south, and some logwatch/alert scripts for monitoring everything. > This time around, I'm thinking about automating some of the install, > configuration management, and monitoring work.
For truly one-off installations you probably are okay managing them manually. However for more than a few machines they are never truly one-off anymore. There will always be a large amount of commonality among the set of them. This commonality is repetitive. And humans are prone to making errors in any repetitive situation. For all of those cases it is very useful to have a system for configuring the entire set of machines. So the most important thing from my perspective is a system for maintaining system configuration across the network of machines. In mindset you maintain the collection of machines as a single entity and not as a bag of individual servers all different. > Which leads me to wonder: Those of you who run server farms on > Debian, what are you using for: > - initial o/s install and configuration (e.g., FAI, other things?) > - software install/update/configuration (chef, puppet, ?) > - virtual machine management > - user management (control panels, ...) > - overall management (nagios, webmin, ...) > > I'm sort of trying to get sense of what people actually use, and in > what combinations, rather than what's got the buzz this week. I think everyone scratches their own itches and then runs out of time. I think everyone does 90% and then wants to finish the last 10% but never gets the cycles to do it. Fortunately that is a different 10% across for everyone. For example I do deployments by a network install from a PXE boot. I do a fully automated installation using preseed files. But this is unrelated to the FAI project. I looked at FAI. It is often referenced. I don't want to say anything bad about it but for my needs it had way too steep of a learning curve to it. It didn't seem necessary since a normal debian-installer and preseeds does everything (and for me just fine). It seemed quite dirty and crufty. I would have needed to clean it up a lot before I would have felt comfortable using it. Although in principle doing a live boot to do the install seems perfectly reasonble. I have considered speeding up my installation by doing something similar. But I recommend standard preseeding over FAI. That is just my opinion because you asked it. For configuration management I use my own tool set which uses portable shell scripts. It is similar to Puppet, Chef, cfengine. But because it is mine I know it best. I can't recommend one over the other but only that having some systematic configuration system is a must. From my exposure I would start with Puppet as a recommendation of something in wide use and with good capability. Although nothing is perfect. But using something is hugely better than using nothing. > Thanks for anything you might share. You should browse through the articles here: http://www.infrastructures.org/ Bob
signature.asc
Description: Digital signature
