On Fri, May 6, 2011 at 1:24 PM, Jerome Baum <jer...@jeromebaum.com> wrote: > I am still worried whether cfengine is the right tool for this. It's not > that I want to keep bringing the system into the "database has exactly > <dump> as its content" state -- I just initially want to get it there. So, I > am trying to converge to any of a set of states, and the dump would bring us > to one of those. From that viewpoint, is cfengine "right" for the job? So I > am thinking: > states: A, B1, B2, B3, ... > Initially I am in state A. I want to get into any Bx state. I know a > transition A --> B9, and I will apply it whenever in state A. But when in > state Bx where x != 9, I will not apply it. > Does that make sense?
I'd split this into the following issues: 1. How to ascertain current state X 2. How to ascertain desired state Y 3. How to get from state X to state Y, for some set of (X,Y) options 4. How to pick an appropriate option from #3 5. How to deploy the code for #1, #2, #3, #4 6. How to invoke the code for #1, #2, #3, #4 cfengine offers answers for #5 and #6; You'll almost certainly need something pretty customized to your environment for #1-#4. Whether you use cfengine or something else for #5 or #6 changes very little about how you'd handle #1-#4. Does that seem like an appropriate separation of concerns? I think it does, and I think this decomposition allows you to keep the complexities in your outside-of-cfengine analysis of (1-4), and then just use cfengine to deploy and run the code that happens to do the work. It would be *conceivable* to put "smarts" into cfengine as to what version of a schema is in place, and have the transformation rules directly encoded inside cfengine, but it is by no means obvious that this is the best way to do things, particularly if there are extra semantics surrounding the conversion. (e.g. - suppose there are data transformations that require planning a system outage.) It's still perfectly reasonable to use cfengine to push the code into place, but it might well be that it's the "helper" that merely puts things in convenient spots so that a human can then come in and have the environment pre-prepared to efficiently do some "human activities." I've done that with database version upgrades, for instance. We'd put binaries for a new version of Postgres onto a "golden server," a couple days before the upgrade, and the binaries would get copied into place, but be unused. During an outage, a database administrator would log in, change symlinks (using cfengine, as it happens :-)), and induce the upgrade manually, at the appointed time. The admins only needed to do a pretty minimal amount of "human work." _______________________________________________ Help-cfengine mailing list Help-cfengine@cfengine.org https://cfengine.org/mailman/listinfo/help-cfengine
