[StarCluster] load balanced nodes accepting jobs before ready

François-Michel L'Heureux fmlheureux at datacratic.com
Mon Apr 14 10:19:05 EDT 2014


Hi Stewart!

I ran into a similar issue. I use a complex value to couter that situation.
In steps:

   1. When creating a cluster, I add the complex value in OGS.
   2. Whenever I run a job, I require that complex value. (See flag "-l" in
   qrsh/qsub)
   3. The last step I do when I initialize a node is add that complex
   value/resource to that node.

Hence, if a job is queued, it cannot run on a uninitialized node because
the complex value is missing. The downside is that you have to alter all
your qsub/qrsh commands to request that parameter, otherwise they will
bypass it. (It might be possible to set it as a default requirement, I
haven't looked.)

Good luck
Mich
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/pipermail/starcluster/attachments/20140414/db26f30f/attachment.htm


More information about the StarCluster mailing list