This is a hangover from the Python scripts, but I agree that it would make sense to start all the nodes simultaneously and run a post configure script (which, incidentally, is what ZooKeeper and Cassandra do).
Do you want to open a JIRA? Cheers, Tom On Sat, Aug 14, 2010 at 1:04 PM, Adrian Cole <[email protected]> wrote: > Hi, team. > > As you know, a large part of the startup time in a cloud is spent spinning > up "blank" machines, then installing the base software. I've noticed that > we start the hadoop master and configure it, then the slaves. I'm pretty > sure we can eliminate a few minutes from this by starting the masters and > slaves up simultaneously, then configuring them after the software is on. > Initially, this can be done via runScriptOnNodes... later, this could be > deferred to another system like chef. > > What do you think? > -Adrian >
