C4W2 - GRADED LAB ISSUE: gcp auth plugin is deprecated - ConfigMap unchanged

Course Q&A Machine Learning Engineering for Production(MLOps) Deploying Machine Learning Models in Production
1

Hi,

After updating the configmap.yaml and executing the command “kubectl apply -f tf-serving/configmap.yaml” i receive the following error message suggesting the gcp auth plugin is deprecated.

Although the error appears as a warning, it is not updating the configmap and therefore stops the graded lab.

configmap
configmap2260×164 120 KB

Any help would be appreciated

2

I have been facing the same issue and posted a message similar to yours. Haven’t gotten any response from anyone yet.

3

Hi everyone! Thank you for reporting! We will be looking into this issue today. Will update this thread asap.

4

Hello again! I just tried the lab and was able to complete it successfully. Please don’t mind the warning message about using gcloud. That looks related to another aspect in the backend that does not impact this lab.

From the posted screenshot, it seems the configmap.yaml wasn’t updated as mentioned in the instructions. Please use the Cloud Editor to navigate to that file. The button to switch to the Editor is at the top right of the Cloud Shell.

Screen Shot 2022-07-07 at 11.16.39 AM
Screen Shot 2022-07-07 at 11.16.39 AM1554×120 17.9 KB



It should look something like this after editing:

Screen Shot 2022-07-07 at 11.16.20 AM
Screen Shot 2022-07-07 at 11.16.20 AM2168×136 59 KB

Screen Shot 2022-07-07 at 11.15.24 AM
Screen Shot 2022-07-07 at 11.15.24 AM2130×1076 387 KB


Then you will go back to Cloud Shell and apply the changes as mentioned in the instructions. From there, it should now run as expected.

Just a few gotchas as you go through the rest of the notebook:

  • Please be mindful of code that does not need to be run. The instructions show some parts of the YAML files just to highlight what it is for. If you paste it in your terminal, you will get an error because those are not meant to be run there.

  • The last checkpoint will not turn green unless the cluster creates 3 replicas. With that, you need to wait till locust simulates more users so a single pod gets overloaded and the node creates more pods. That is the main demo in this lab. It shows how you can avoid application failures by implementing autoscaling.

Hope these help!

1 Like
5

For anyone seeing this in the future, here’s a suggestion to build upon on Chris’ suggested steps for configmap.yaml, where if you already accidentally skipped Task 4 step 1, below I show a “delete deployment” step. To prevent any confusion, here are the full detailed steps to fix it:

  1. Follow the steps Chris Favila suggested in his post here, specifically, you should go to Cloud Shell Editor → tf-serving (folder) → configmap.yaml, then update the line 21 “MODEL_PATH:” to be gs://your_quik_labs_project_id-bucket/resnet_101
  2. Then delete your old deployment using the command “kubectl delete deploy image-classifier” in terminal. You MUST delete this old deployment or it’ll run forever and re-running the deployment.yaml in the next step won’t work.
  3. Then, re-run the command “kubectl apply -f tf-serving/deployment.yaml”. This applies your updated MODEL_PATH from deployment.yaml.
  4. Wait 2-3 minutes and then run command “kubectl get deployments”. If it is not 1/1 then it’s not yet ready.
  5. Skip down to the “curl” command step and re-run it (the previous steps don’t need to be re-run). You should now see a prediction output for the image.
1 Like