Skip to content

Latest commit

 

History

History
27 lines (17 loc) · 1.86 KB

README.md

File metadata and controls

27 lines (17 loc) · 1.86 KB

nodetaint

Controller to manage taints for nodes in a k8s cluster.

What the problem is

Usually there are some system critical daemonsets (e.g. CNI, DNS, etc.) that needs to be running on a node before it can run any other pods. However k8s doesn't guarantee any ordering for pod scheduling in node startup, meaning your workload pods may start before the critical daemonsets have started!

How it works

The controller solves this problem by removing a pre-configured taint from a node after annotated daemonsets are running on the node. To achieve this, you need to configure your cluster to launch nodes with the desired taint: configure kubelet to start with --register-with-taints option. The controller then determines, through annotation, which daemonsets should be running on a node prior to workload pods. It monitors for these daemonset pods to be Ready before removing the configured taint.

Note: If you are using the k8s cluster autoscaler you should align the ignore-taint option with the same taints as registered on the kubelet. This will ensure the cluster autoscaler considers those nodes as "ready" for the purposes of scaling up/down.

Configuration

Command-line

nodetaint can be configured by the following command-line options:

Flag Environment Variable Type Default Required Description
log-level LOG_LEVEL string info no The level of log detail.
bind-address BIND_ADDRESS string :9797 no The address for binding listener.
node-taint NODE_TAINT string yes The startup taint to put on node.
daemonset-annotation DAEMONSET_ANNOTATION string yes The annotation of required daemonset.