On Sat, Aug 28, 2004 at 06:23:38PM +0200, Michael Tautschnig wrote: > - start the update-process from one single server, not on the client Update has to be done on the client: how else can it access the local filesystem ? In my setup, the server starts the update processes on the clients via the backdoor-sshd.
> - push the conffiles to the client In my setup, clients do a cvs update as the first step. > - tell the client, what to do with the new files > > For all of that I intend to use cvs-hooks and the rcsinfo-file; there I > specify the command to run on the destination host; I began implementing this, but canceled the automatic-push cvs hook really fast: as every human, an administrator may make mistakes. so before pushing conffiles to clients and killing 100 computer whith just one cvs commit, the new configuration has to be tested. Ideally you have a group of test clients, where you carefully observe the consequences of your changes. only if no problems occur there, you mark them as ready-for-production-use (we do this by "cvs tag"ing them STABLE). The second problem: you can't update too many machines at the same time because of the cvs locking problems and the high network load, so when implementing a cvs hook, you have to do it in a sequence of groups of machines. > BTW: What did you mean by "the FAI-Manual" - the fai-guide? The one from > your packages or the official one? the official one... most of the stuff in the packages published on the website is not up-to-date. -- c u henning
