On Tue, 9 Jan 2007, Nicolas Williams wrote:
> On Tue, Jan 09, 2007 at 02:13:51PM -0800, Jim Hall wrote: > > What if the requirement was to fetch the file only once when the > > service is started or when the user used the svcadm restart command? > > > > In other words: what if there was no requirement to constantly check > > the state of the configuration file? Only during the conditions above? > > If the file couldn't be fetched that first time, then what? The file > dependency FMRI only checks once and that's been a big dissatisfier. > The same would happen with any other similar FMRI. Polling wouldn't > necessarily be a good solution either. The fact that the FMRI has been a big dissatisfier was unknown to me and is a good point. But to answer your question: The service would have to be designed with reasonable defaults, or the service would run disabled until the first config is downloaded, or the system would be jumpstarted with a starter config file for services where that is a fit. And people would have the option of old fashioned config files via the file:// URL. > In any case, what's wrong with fetching the config file in the start > method of the service? > > Well, I can think of one thing that'd be wrong: it'd be nice to have a > uniform remote configuration scheme for SMF. But that wouldn't be based > on fetching config files via HTTP! So the whole smf_fetch_file() scheme > seems: a) not SMF-specific, b) not a general solution to remotely > configure services. > > Of course, since we wouldn't store things like IP filters in SMF a > remote file download utility might be part of a generic remote > configuration scheme for SMF. But key thing to me is that such a > utility could be generic and that if such things as IP filters should be > downloadable then so should SMF service property groups. Don't have a problem with any of this. I am very flexible about the implementation details. It just seemed that SMF would be the generic framework that would manage the state of trying the configuration of a service that could be fetched from an authoratative source. I don't care about the paticular protocol/subroutines/FMRI vs. start method/ details. As long as it meets the goal of centralized management. JIM
