Am I the only one who misses just SSH-ing into a box to debug things?

Don't get me wrong, I love the abstraction and the promise of serverless. My Vault clusters are, in many ways, a managed abstraction themselves. But lately, I've been hitting a wall that's making me nostalgic for a simple SSH session.

The issue is the black box. When a Lambda function or a Cloud Run instance is failing because of a secret—maybe a Vault token isn't being renewed correctly, or an environment variable sourced from a managed service isn't populated—the debugging loop is brutal. You're left with logs, which are just a symptom, and you have to infer the state of the entire runtime. With a traditional box, I could `ssh` in, check the actual process, `cat` the token file, see what's in `/proc/self/environ`, maybe even run a quick `strace`. The feedback is immediate and complete.

Now, I have to rely on adding more verbose logging, redeploying, and waiting for the cold start. It feels like debugging through a keyhole. For secrets-related issues, this is especially painful. Was the transient failure in the KMS decrypt call? Did the IAM role attach correctly? The managed runtime abstracts away the very layer I often need to inspect.

I know the counter-arguments: immutability, scalability, no SSH maintenance. But there's a real trade-off in operational insight. Sometimes I wonder if we've traded too much control for convenience. Does anyone else have a workflow or tooling that bridges this gap, especially when dealing with secrets injection in these environments? Or do we just accept the opacity as the cost of doing business?