Hi Aleksey I haven't been burnt by this (yet) so you may want to take this with reservation :)
I'd go for a double approach here. As you say, I'd count root's instances of cf-agent and use abortclasses to give up if the count is higher than 1. To be more effective, this check should happen as early as possible, and will give a long running cf-agent a chance to complete. However, nobody can guarantee that a cf-agent run won't hang forever (for whatever reason). In that case, the approach above would completely block your cfengine instance. So, at some regular interval, e.g.: once every 4 hours, I'd do the same check and actively kill spurious instances. Of course, this check must happen _before_ the one above, and should be the really first thing the agent does. On 07/09/12 05:49, Aleksey Tsalolikhin wrote: > 1. Abort the CFEngine agent run if there is an earlier instance of it > already running. (Do this via abortclasses) Like this, but with the "safety net" as above. > 2. Make a promise that cf-agent will kill earlier instances of > cf-agent. I noticed cf-agent won't signal itself, haven't played > around if it will signal another instance of cf-agent. even if it's > averse to shooting another cf-agent, we can kill it using an external > shell command. Don't like this: if a cf-agent legitimately takes more than 5 mins to complete, that won't give it a chance to complete a run. > 3. Set a timeout on every commands type promise (doesn't really > address the scenario where a complete native cf-agent run takes 5 > minutes and 01 seconds, so runs overlap, thus loading the host server) Never used timeout, so no clue. Ciao -- bronto _______________________________________________ Help-cfengine mailing list Help-cfengine@cfengine.org https://cfengine.org/mailman/listinfo/help-cfengine
