Ratelimit the number of pods created every second in a deployment
There is no such direct option in K8s 1.18. Here is related github issue. Some possible workarounds are:
use multiple deployments, maybe within multiple namespaces. It may be hard to manage single deployment with 1000s of replicas in therms of any changes.
implement random configurable delay inside your client(or with lightweight wrapper) to spread the load over time, depending on total wall clock startup time of your 1000s clients and server power
- implement custom metrics agent with
pod is readycheck to scale deployment with HorisontalPodAutoscaler as you need and use Scaling Policies to create 100 pods per minute - implement your own pod autoscaler