C4W3: Implementing Canary Releases of TensorFlow model deployments with Kubernetes and Istio

Course Q&A Machine Learning Engineering for Production(MLOps) Deploying Machine Learning Models in Production
30

Hi! Welcome to the community and thank you for sharing the workaround. I’ve consulted with our partners and they said we could indeed do this while they’re fixing the lab. The other option is to use an older version of GKE:

gcloud config set compute/zone ${CLUSTER_ZONE}
gcloud beta container clusters create ${CLUSTER_NAME} \
    --machine-type=n1-standard-4 \
    --num-nodes=6 \
    --workload-pool=${WORKLOAD_POOL} \
    --logging=SYSTEM,WORKLOAD \
    --monitoring=SYSTEM \
    --subnetwork=default \
    --cluster-version=1.21 \
    --labels mesh_id=${MESH_ID}

The problem with this is it might stop working when GKE v1.21 eventually becomes unavailable in the Cloud Shell. So, reducing the number of nodes is the better approach.

Thanks again!