search

rootfs / node-fencing

Apache License 2.0
5 stars 6 forks source link

Details #48

Open coolpalani opened 6 years ago

coolpalani commented 6 years ago

Could you give some lighten about this project to quickly understand.

bronhaim commented 6 years ago

I'd love to give more details, but please be more specific what lighten is missing? The readme explains the desire for maintaining bare-metal k8s cluster using fence devices and monitoring for healthiness to recover stateful pods and re-join defected nodes to cluster.

coolpalani commented 6 years ago

I started to read about statefulset from https://kubernetes.io/docs/tutorials/stateful-application/basic-stateful-set/ I am able to understand basic usage of statefulset interms of reserving few resources to pods.

if node goes to not ready state, will automatically reboot it as per this node fencing. correct me if i am wrong.

bronhaim commented 6 years ago

When node goes to not-ready the monitor will start fence flow which consists of 3 steps, each step contains method to run, reboot is one of them. The node probably runs pods, part of them can relate to stateful set deployments - means that the scheduler won't start new instances until the node responsive again. Instead of waiting for the default eviction treatment, after rebooting\fencing the node successfully, the monitor will destroy those pods to trigger the re-scheduling

kunalkushwaha commented 4 years ago

Hi @bronhaim @rootfs , Thanks for creating this project. I wanted to know the current status of this project and if it is still in active development? There is not much activity on this project recently.

I am really interested fencing implementation for Kubernetes as its a big bottleneck to migrate Legacy application architecture.