On 25/03/09 09:19 AM, Michael Schuster wrote: > On 03/25/09 08:47, Darren Reed wrote: >> On 24/03/09 02:54 PM, Michael Schuster wrote: >>> On 03/24/09 13:57, Peter Tribble wrote: >>>> 2009/3/23 <Zhenghui.Xie at sun.com>: >>>>> Attached is a writeup based on our discussion of the SCF part. >>>>> Comments? >>>>> Send them out by COB Weds, or silence is gold afterward :-) >>>>> >>>>> Thanks. >>>>> >>>>> -Jan >>>>> >>>>> >>>>> 1. ilb will have one single instance in SMF framework: >>>>> svc:/network/loadbalancer/ilb:default >>>>> 2. persistent configuration of ilb is saved in SCF. >>>> >>>> Why SCF? I wouldn't regard SCF as suitable for holding complex >>>> configuration >>>> data of the type required for something like ilb. >>> >>> how complex are you thinking? >> >> How do you intend to show a dozen services being load balanced, >> along with comments about how and why? > > you still have the ability to "ilbadm import <yourfile>", if you > prefer (and while I just realised that this doesn't handle comments > yet, that's trivial to implement) > >> Text configuration files are useful in many ways, for example, you >> can use RCS or SCCS or whatever and manage them with version >> control software. Not only that, you get a history of who did what, >> when and why (well, hopefully.) > > you can do that with <yourfile>.
Well at least that's something. >> How do I get all that with ilbadm? > > see above. > >> Same can be said for ipadm, dladm and all of the other "wonderful" >> new *adm commands we're dreaming up. > > I don't think I'm the person to answer that ;-) > >>>>> 7. user can use svcprop(1) to get ilb configuration. But should >>>>> NOT use >>>>> svccfg(1M) to change ilb configuration. >>>> >>>> Why not? If you're using SCF, then svccfg should be a supported way of >>>> modifying the configuration >>> >>> why? just because you *can* edit many files with vi doesn't mean >>> that's a supported way of doing it. >> >> No, but it is easy, informative and much less prone to error. > > I think that's up for debate - making a mistake with vi is too easy, > for my liking. But the command line options being pursued offer *NOTHING* that is anything better. *NOTHING* >> A month or so ago, a few of us sat down with a group that does >> professional system administration and they were quite in favour >> of having configuration files that can be validated before loading >> rather than having to do everything on the command line. Why? > > >> Ever done "kill -9 - 1" by accident? >> Or "kill -9 % 1"? > > actually no :-) (you mean as root? that's the only case where it would > affect ilbd). I don't quite see how that pertains to this discussion, > please elaborate so I can understand the concern. The point is that recovering from finger trouble on the command line can be a whole lot harder than recovering from typing an address wrong and going "oops" as you review the configuration before applying it. The above isn't about whether those "kill" things affect ilb, it's whether or not you reboot the box by accident or kill init or not. Darren
