Mesos autoscale EC2

Question

I was looking for a way to autoscale a mesos or DCOS EC2 cluster dynamically. An example scenario would be if cluster CPU usage is above x % for x minutes spin up new instances, if memory is above X % for x minutes, spin up new instances.

Ideally the instance type should be dynamically determined by the type and amount of resources needed. I saw this projects:

https://github.com/thefactory/autoscale-python

which I suppose can be run as a mesos marathon task itself to handle that, but I was wondering if there is a built in utility in mesos or a generic way to do that on EC2 or GCE. Thanks!

Answer 1

As recently as last week, I had a talk with Mesosphere people, and can confirm that this functionality is not yet built-in to DCOS. I didn't get a feeling that this is something on their cards anytime soon as they kept referring me to Application autoscaling instead of Infra autoscaling (which is what we want here).

The python script from thefactory (Tendril) seems to be not maintained anymore as evidenced by a PR hanging in there for 12months now.

I'm currently looking at Netflix Fenzo which also seems to have been written to enable autoscaling of the infrastructure, among other things.(http://techblog.netflix.com/2015/08/fenzo-oss-scheduler-for-apache-mesos.html)>

I will try to post back once I get a better idea of Fenzo and how it integrated to DC/OS.