Errata

Uncordon issue when gnsh.service starts

  • Affected version: 1.3.3 and earlier

Gnsh is the Graceful Node Shutdown Helper program. It cordons and drains the node when the node is gracefully powered off and uncordons the node when the node is booted.

When gnsh systemd service starts at boot, /usr/bin/gnsh script tries to uncordon the node only once and can fail if kube-apiserver is not ready yet.

I modified /usr/bin/gnsh script to try to uncordon the node 10 times over 30 seconds if the node is not uncordoned.

Download the modified gnsh script and put it in /usr/bin/ directory on all kubernetes nodes.:

$ chmod +x gnsh
$ sudo cp gnsh /usr/bin/

Then, It will be applied on next boot.