Description of the feature you would like to add / User story
As a KKP administrator / KKP user cluster owner
I would like to have the ability to override machine-controllers global node eviction timeout (--skip-eviction-after) per MachineDeployment / Machine
in order to have fine-grained control after which time a node / machine is deleted on the MachineDeployment / Machine level
Solution details
- machine-controller provides an option to override the globally configured node drain timeout per
MachineDeployment / Machine (e.g. timeout could be specified with an annotation on the Machine, propagated down from the MachineDeployment)
Alternative approaches
None
Use cases
Right now, the machine-controller allows you to configure the node drain timeout only at a global level. However, there may be situations in which you have multiple MachineDeployments, each with different requirements for the rollout / deletion process.
Imagine you have two MachineDeployments, each intended for a specific type of workload (using taints, labels, and node label selectors), such as stateful and stateless workloads. In this case, the rollout behavior of the MachineDeployment for stateless workloads could be configured to be significantly more aggressive than the rollout behavior for stateful workloads.
Additional information
/label customer-request
Internal reference: 9415
Description of the feature you would like to add / User story
As a KKP administrator / KKP user cluster owner
I would like to have the ability to override machine-controllers global node eviction timeout (
--skip-eviction-after) perMachineDeployment/Machinein order to have fine-grained control after which time a node / machine is deleted on the
MachineDeployment/MachinelevelSolution details
MachineDeployment/Machine(e.g. timeout could be specified with an annotation on theMachine, propagated down from theMachineDeployment)Alternative approaches
None
Use cases
Right now, the machine-controller allows you to configure the node drain timeout only at a global level. However, there may be situations in which you have multiple
MachineDeployments, each with different requirements for the rollout / deletion process.Imagine you have two
MachineDeployments, each intended for a specific type of workload (using taints, labels, and node label selectors), such as stateful and stateless workloads. In this case, the rollout behavior of theMachineDeploymentfor stateless workloads could be configured to be significantly more aggressive than the rollout behavior for stateful workloads.Additional information
/label customer-request
Internal reference: 9415