Wrong output value of TPUVMs when the cluster have the string tpu in its name
williampispico opened this issue · comments
Hi,
I noticed that when I run python3 xpk.py cluster describe --cluster <MYCLUSTER>
the number that returns in the message can be wrong if you name your cluster with the string tpu in its name (e.g., my-tpu-cluster). The command kubectl get node --no-headers=true | grep '\-tpu\-' | wc -l
would fetch even the default-node and consider it a TPU node and show a wrong value in the output.
Here is the line in xpy.py I am mentioning https://github.com/google/xpk/blob/main/xpk.py#L1186
A workaround for that could be instead kubectl get node --no-headers=true --selector='cloud.google.com/gke-tpu-accelerator' | wc -l
Thank you William for the bug and fix! Will get on this shortly. :)