Updated to Cloud SQL Proxy v2

[arueth]

Background/Change Summary

• Updated to Cloud SQL Proxy v2 in preparation for some upcoming fleet workload identity work

Testing Procedure

Tested in my project using https://github.com/GoogleCloudPlatform/bank-of-anthos/blob/main/docs/ci-cd-pipeline.md

• dev
• staging
• prod

[bourgeoisor]

@arueth currently (or at least last I tried), Bank of Anthos could juuuust fit in a 3-node default config GKE Standard cluster. This PR bumps a lot of requests upwards (sometimes 2x or 3x). What do you think is the new requirement if deploying in a GKE Standard cluster with these changes?

[arueth]

So this is the new configuration running in my environment

I don't have development deployed.

[arueth]

We could create a "minimal" skaffold profile if we wanted to get really lean.

[bourgeoisor]

The reason I mention this is because there's a lot of learning content that leverages Bank of Anthos' default configs which still uses GKE Standard with a limited # of nodes (e.g. 3 or 4). If we bump the default base requests that may break that learning content.

For example, this tutorial deploys default Bank of Anthos on a 4-node cluster: https://cloud.google.com/migrate/containers/docs/migrating-monolith-vm-overview-setup#deploy

You can see a more complete list here: go/boa-usage (internal)

What do you think?

[arueth]

Are the skaffold changes and kubernetes-manifests folders kept synchronized? I believe the vast majority of tutorials outside of the docs in the repo use the kubernetes-manifests folder, is that correct?

[bourgeoisor]

Correct! 90+% of tutorials don't use kustomize / skaffold, they just kubectl apply -f the kubernetes-manifests/ directory.

The sync is made at release time (new release of Bank of Anthos) where we skaffold render

https://github.com/GoogleCloudPlatform/bank-of-anthos/blob/main/docs/releasing/make-release.sh#L58-L72

[arueth]

Would you prefer I move these changes out to a separate autopilot profile or only apply them to the staging/production profiles? It looks like the make-release.sh uses the development profile.

[bourgeoisor]

Yes, kubernetes-manifests/ is essentially the barebone development profile (no Cloud SQL, low resources needed, etc)

How about we make the higher-resources-requests changes on the staging/production profiles like you're suggesting? That way the development profile (and by extension, the generated "barebones" k8s manifests) stay minimal

[arueth]

Since the Autopilot minimums are going to change, I'm changing this to only include the Cloud SQL Proxy v2 changes.

[bourgeoisor]

LGTM!