I was thinking of the job idle time, where we spin up more instances if jobs sit waiting for, say, 15 seconds instead of 60.

Jim

On Mon, Apr 28, 2014 at 2:05 PM, Dannon Baker <dannon.baker@gmail.com> wrote:

The time after which it culls instances?  It is not, but because of the way Amazon bills for instances, you never want to kill an instance until the end of the hour (since, regardless of when you kill an instance, you're billed for the remainder of the hour).

-Dannon

On Fri, Apr 25, 2014 at 3:01 PM, Jim McCusker <jmccusker@5amsolutions.com> wrote:

Thanks, is the idle time configurable?

On Fri, Apr 25, 2014 at 2:43 PM, Dannon Baker <dannon.baker@gmail.com> wrote:

You can review the exact code here( see 'slow_job_turnover') :  https://bitbucket.org/galaxy/cloudman/src/7b8f04895ad309e0168cb3de66446ae20f3d8b3e/cm/services/autoscale.py?at=default

But, basically, load on any particular node isn't very useful for autoscaling in this context because most jobs cannot be split among multiple nodes.  What we use is a heristic to determine churn and jobs waiting to run.  Generally speaking, if you have jobs waiting to run for a bit, your cluster should scale.

-Dannon

On Fri, Apr 25, 2014 at 2:37 PM, Jim McCusker <jmccusker@5amsolutions.com> wrote:

What are the exact conditions that will trigger autoscaling in a galaxy/cloudman instance? I'm seeing 100% utilization on the head node but no attempts by cloudman at spinning up new instances. What sort of state is supposed to trigger that?

Thanks,

Jim

___________________________________________________________
Please keep all replies on the list by using "reply all"
in your mail client.  To manage your subscriptions to this
and other Galaxy lists, please use the interface at:
  http://lists.bx.psu.edu/

To search Galaxy mailing lists use the unified search at:
  http://galaxyproject.org/search/mailinglists/