Sequentially boot up the pods in a deployment with multiple replicas

[tieungao88]

Hi everyone,

I have the following use case: "Sequentially booting up pods in a deployment with multiple replicas."

Specifically, I have a deployment with 6 replicas, and I would like the pods to start sequentially when deploying to Kubernetes. For example, pod1 should fully start and reach the Running state before pod2 starts, and so on. Alternatively, I would be fine with starting 2 pods initially, and after these 2 pods are running, proceeding to start the next ones.

The reason for this requirement is that my application is quite resource-intensive during the boot-up phase. Each time a pod starts, it consumes 2 vCPUs, but once fully running, it uses significantly fewer resources.

My question is: Can scheduler-plugins handle this scenario?

Thanks in advance for your insights!

[Huang-Wei]

SchedulingGates could be a nice fit for this usecase:

• create the workload carrying non-nil .schedulingGates
• craft your own controller to un-gate each pod (or self-defined pods in batch) so as to "rate-limit" the scheduling of the pods

So you don't need to build a custom scheduler plugin - the in-tree SchedulingGates plugin gate pods for you. What you need to is just build a controller.