Hi all, The configuration thread is going in several directions. It seems as if there are two approaches being discussed here:
1.) Making drizzle manageable from the network. As I understand that was the original point of Mark's comments that started the configuration discussion as well as Jeremy's comment about remote restart etc. 2.) Making it possible for drizzle to read configuration from different locations via plugins so you can support different kinds of configuration repositories and mechanisms. My experience has been that it's better to make service configuration as simple as possible and assume that some management framework will supply additional top-down intelligence and handle errors, many of which cannot be resolved at the level of a single host. If you take this view #1 seems to be the more important problem to solve, which means: A.) Specifying initial configuration values in a simple and intuitive way. B.) Providing the ability to set configuration values persistently while the server is running. C.) Authorizing to prevent accidental or malicious changes. D.) Distinguishing changes that can be applied while online with those that require a restart of some kind. E.) Restoring defaults from disk when things get messed up. F.) Making all important admin operations including restart available to the network. If you assume there is a service manager that starts the server automatically on boot or if it fails you should never have to touch the local box. In fact, I kind of wonder about the wisdom of having multiple configuration plugins. This seems to be a problem where it would be better to have one mechanism that is very robust and simple to use. Cheers, Robert P.s., We did the network manageable approach for Tungsten Replicator. It's hard to get right without a consistent state machine design to ensure configuration changes are only accepted in certain states. On 4/28/09 12:06 PM PDT, "MARK CALLAGHAN" <mdcal...@gmail.com> wrote: > On Tue, Apr 28, 2009 at 11:58 AM, Brian Aker <br...@tangent.org> wrote: >> Hi! >> >> On Apr 28, 2009, at 11:25 AM, MARK CALLAGHAN wrote: >> >>> general problem is that some of the state changes done to a server are >>> not persistent and if the server is restarted, the state change is >>> lost. For example, if skip-slave-start is not in my.cnf and you run >> >> >> On the current list of things to do is to take out the configuration system >> and refactor it out to plugins. What sort of one would you want? A central >> system, or the ability to flush to a file? >> >> Cheers, >> -Brian >> >> > > I don't know much about plugins. I know we have messed around building > tools that parse and edit my.cnf to keep it in sync with the state of > the server. It would be great if that tool wasn't needed and drizzle > had the notion of a transient versus persistent admin command (one > limited to runtime, the other changes runtime and the config file). > > > -- > Mark Callaghan > mdcal...@gmail.com > > _______________________________________________ > Mailing list: https://launchpad.net/~drizzle-discuss > Post to : drizzle-discuss@lists.launchpad.net > Unsubscribe : https://launchpad.net/~drizzle-discuss > More help : https://help.launchpad.net/ListHelp > -- Robert Hodges, CTO, Continuent, Inc. Email: robert.hod...@continuent.com Mobile: +1-510-501-3728 Skype: hodgesrm _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : drizzle-discuss@lists.launchpad.net Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
