CAPA node draining - Phase 2

[T-Kukawka]

Following Phase 1, after CAPA GA, we should investigate the usage of aws-node-termination-handler app for node draining once instances are being terminated via e.g. autoscaling. This with combination of "MachinePool Machines" feature in CAPA, CAPA logic should help to drain nodes within a pool, for example during an explicit ASG instance refresh.

Updated tasks list:

• https://github.com/giantswarm/giantswarm/issues/31843
• https://github.com/giantswarm/giantswarm/issues/28006

[T-Kukawka]

We have to investigate what are the solutions upstreams. The MachineDeployemnts has implementation for the handling of shutting down the nodes. There is no such solution for MachinePools. Before node shutdown we would have to receive the events from AWS to act upon them.

We have to follow up with upstream what is their plan for this.

[AndiDog]

Maybe related:

• CAPI Resources representing MachinePool Machines kubernetes-sigs/cluster-api#4063
• CAPA Implement MachinePool Machines clusterAPI proposal kubernetes-sigs/cluster-api-provider-aws#4184

Once the "MachinePool Machines" feature lands in CAPA, there's probably a better chance for CAPA logic to drain nodes within a pool, for example during an explicit ASG instance refresh.

[T-Kukawka]

so it is not too far off with milestone 2.4

[T-Kukawka]

Investigation feedback from older issue:
internal - #2170 (comment)
upstream - kubernetes-sigs/cluster-api#8348

[T-Kukawka]

still blocked by: kubernetes-sigs/cluster-api-provider-aws#4184

[T-Kukawka]

@AndiDog will try to give it a nudge

[T-Kukawka]

additional solution besides the health checks: https://github.com/giantswarm/giantswarm/issues/31843