On Thu, Aug 8, 2013 at 1:55 PM, Michele Tartara <[email protected]> wrote: > On Thu, Aug 8, 2013 at 1:44 PM, Jose A. Lopes <[email protected]> wrote: >> >> > What happens if a job terminates while LuxiD is not available? And >> > what if a job finishes before LuxiD is back? >> >> Am I missing something here or do these questions seem to be same ? > > > They are two parts of the same scenario, but the first one was actually > meant to be "if a job is running while". Sorry, my fault typing. > >> >> What about the following: >> >> * if the job process has no message to deliver or locks to release, >> then it terminates and goes away > > > Possible, but quite unlikely. Most jobs have some kind of lock, if they > modify the state of the cluster in some way. > >> >> >> * if the job process has a message or locks to release, or something >> else to give the master, then it can wait for LuxiD to become >> available and then deliver the message, or release the locks, etc... >> >> What do you think ? > > > It's possible, but keeping processes running seems to me much > resource-heavier than having a bunch of (extremely small) unix sockets on > the disk for some time. >
How about we take this to the whiteboard? :) We have a slot tomorrow morning, where we could discuss freely. thanks, Guido
