If you can swap containers within a pod, it follows that data can stay in one place. Instead of moving data from compute location to compute location, move the compute to the data. Since you can switch containers you can have less Pods, and there is no transfer of data between Stages from data source back and forth to pod.
You can think of this like an assembly line. When a car manufacturer builds a car, it does not assemble stages in different factories moving the car from factory to factory. It uses one factory and moves the car from stage to stage.
Empty Dir sits in RAM and by this documentation it says it survives containers. Thinking that it should be available for stateful set. Now we are not pushing data to disk - its even faster.
Related Resources - the logic to create these mappings could be contained in an Operator responsible for creating compute environments.
https://www.alibabacloud.com/blog/kubernetes-volume-basics-emptydir-and-persistentvolume_594834
https://openkruise.io/en-us/docs/advanced_statefulset.html
https://docs.solo.io/gloo/latest/guides/traffic_management/destination_types/static_upstream/
https://github.com/openkruise/kruise/blob/master/pkg/client/clientset/versioned/typed/apps/v1alpha1/statefulset.go
https://github.com/solo-io/gloo/blob/master/projects/gateway/pkg/api/v1/virtual_service_client.sk.go
No comments:
Post a Comment