0

I have a managed k8S cluster with 1GPU (Tesla K80) activated in west1-b and west1-d (Each zone has this GPU model enabled, and my quota is ok).

Each time that I create a node pool with 2 nodes only one node is up and I have an error message. I have tried to create a node pool without GPU and the error is the same.

Here is the error message : All cluster resources were brought up, but the cluster API is reporting that: only 1 nodes out of 2 have registered; cluster may be unhealthy.

An other cluster on west1-b and west1-c without GPU activated on cluster creation works well.

Thx a lot. Baptiste.

  • 1
    Could you please provide a minimal reproducible example for the issue? (https://stackoverflow.com/help/minimal-reproducible-example) – Vit Dec 06 '19 at 15:01
  • - Create a kubernetes cluster on eu-west-1b and eu-west1-d with a GPU Tesla K80. - Add a node pool with a GPU K80 (2 nodes : 1 on b, one on d) The d zone failed. – Baptiste GAILLET Dec 09 '19 at 08:43
  • - The cluster is created via UI. - The node pool via UI or command line : gcloud container node-pools create pool-1-13-12-gke8-gpu --machine-type=n1-standard-2 --accelerator type=nvidia-tesla-k80,count=1 --cluster cluster_name --num-nodes 1 --enable-autorepair --shielded-integrity-monitoring --no-enable-autoupgrade – Baptiste GAILLET Dec 09 '19 at 08:50
  • - Adding a node pool (via UI) without GPU as the same error : The D zone failed to instanciate a node. – Baptiste GAILLET Dec 09 '19 at 08:54
  • Okay, new test with new cluster creation. GCP has changed the creation method, now you don't have to set if the cluster will need GPU (before was a checkbox during the cluster creation). My new cluster works well, and my old cluster still can't have 2 nodes this morning... New option during the creation you can now set a regional cluster and limit zone where you will deploy your nodes. – Baptiste GAILLET Dec 09 '19 at 11:30

1 Answers1

1

Okay, new test with new cluster creation.

GCP has changed the creation method, now you don't have to set if the cluster will need GPU (before a checkbox was needed during the cluster creation).

My new cluster works well, and my old cluster still can't have 2 nodes this morning...

Ps : New option during the creation, you can now set a regional cluster and limit zone where you will deploy your nodes.