Node Race Condition | Multiple Schedulers

[Rishabh Mishra]

Hi Team, 

In a multiple scheduler instance setup in a cluster, what is the current behaviour of the race conditions on the nodes among the schedulers i.e. how is the scenario of binding the same node to multiple pods handled where the node eventually can not accomodate all the pods? 

Does it fail in the binding cycle itself or it later fails in the kubelet?

Thanks,

Rishabh

[Aldo Culquicondor]

The kubelet will fail the pods that it observes being bound last.

Aldo

--
You received this message because you are subscribed to the Google Groups "kubernetes-sig-scheduling" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kubernetes-sig-sch...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kubernetes-sig-scheduling/CAG5dug2As9OXX8UJ-Z-jvS%2B1%3D-09_XTm6DtOaNNX4wVxdqs1mQ%40mail.gmail.com.

[Wei Huang]

Moreover, when kubelet fails the pod(s), their .spec.nodeName won't be reset which means you have to manually kill the pod(s).

Regards,
--------------
Wei Huang
hwe...@gmail.com

To view this discussion on the web visit https://groups.google.com/d/msgid/kubernetes-sig-scheduling/CAJi5r2qKs%3DtGFgELqfvKDKyJLX9YcaC6FP6Rv7tRhGv%3D8Rwe%3DQ%40mail.gmail.com.