Open timothysc opened 9 years ago
Tim, would you please elaborate?
On Wed, Dec 17, 2014 at 9:43 AM, Timothy St. Clair <notifications@github.com
wrote:
Folks -
IMHO the best use case for k8's+ mesos integration is to drive feature parity with GKE cluster manager utilities. Perhaps we can breakout the feature details here.
— Reply to this email directly or view it on GitHub https://github.com/mesosphere/kubernetes-mesos/issues/109.
James DeFelice 585.241.9488 (voice) 650.649.6071 (fax)
@timothysc I'd really appreciate your thoughts re: #68.
Mesos can scale the constraints of the executor container up and down depending on the resources assigned to the tasks. However, the tasks end up running as Docker containers in a completely separate process hierarchy. I think what I want is an approach (systemd compatible) that would place the Docker containers in the same hierarchy as the kubelet-executor (perhaps along the lines of systemd-docker).
I believe the references are here: http://youtu.be/c3sGfTQUHcM?t=10m20s & https://cloud.google.com/sdk/gcloud/reference/preview/ namely - https://cloud.google.com/sdk/gcloud/reference/preview/autoscaler/
Basically it allows you to define custom auto scaling rules based on feedback from the cluster metrics. You can define the rules based on a number of metrics.
As load increases on a given service you scale up based on the rules.
looping in @adam-mesos since he mentioned some possible Mesosphere investments in this space
Autoscaling is definitely important, but I think we have other low-hanging fruit to tackle first in the K8s-Mesos framework. In the meantime, Mesosphere is investigating general autoscaling tools for Mesos clusters. Let's wait a month and see what comes out of that. FYI @kozyraki @benh
I've been thinking about something like this w/ respect to autoscaling:
at this point in time, k8s has beta support for autoscaling that requires the heapster add-on to be deployed: http://kubernetes.io/v1.1/docs/user-guide/horizontal-pod-autoscaler.html
TODO: enable heapster in our mesos/docker cluster and start running test cases for the autoscaling feature
Folks -
IMHO the best use case for k8's+ mesos integration is to drive feature parity with GKE cluster manager utilities. Perhaps we can breakout the feature details here.