Rishabh Mishra

Exactly, that split keeps the fast check trustworthy. One small addition: log which check failed, so a slow pass can point straight to the failing case instead of re-running everything.

Building all three tiers yourself is the right way to learn this stack. Adding one health check to the Flask service would make the Nginx routing failures much easier to spot during testing.

The bucket-level check is easy to miss when the error sounds like a table policy. Testing one upload as the signed-in user, not as a service key, usually shows which policy is actually blocking it.

Thanks, Albert. One check I would add after the revert is confirming no outbound route to the bastion remains, since that is easy to miss once access works again.

Exactly. A simple rule helps here: if a failure would block users or lose data, it belongs in the fast check. Everything else can wait for the slower review.

The dual-build separation check is the detail I would steal for any demo. Proving the production build has no mock code, and the demo has no real data, removes the two risks people usually worry about with a click-through demo.

The temporary-route approach is the smart part, and the revert to strict isolation is what makes it safe to reuse. I would snapshot the private route table before adding the bastion route, so rolling back is a restore rather than a reconstruction from memory.

The instance-type naming is the part worth slowing down on for a first deploy. I would start with the smallest size that fits the workload, then size up from real CPU and memory use, rather than guessing a large instance up front.

Agreed. I would frame it as a trade-off, not a shortcut: a one-minute gate that always runs beats a thorough one that gets skipped. The fast check should cover the failure that would hurt most, and the rest can live in a slower review.

The saved-plan rule is the part I would keep. Applying exactly what was reviewed removes a whole class of surprises. I would also keep each plan artifact tied to its run, so a re-run cannot quietly apply a newer change.