[pjmlp]

As do almost every microservice out there, by storing credentials in environment variables, an exploit that manages to read container's memory is enough.

I keep looking for frameworks that do it the right way, holding critical data encrypted all time, but it isn't a thing most people worry about.

[kenniskrag]

What's the threat model. Where do you store the decryption key?

E.g. if my app needs a db connection I can ask a vault service but I need creds for that. The vault service can rotate the creds very fast but is it addition security.

[pjmlp]

The treat model is that your container gets owned.

The password should only exist in the process memory for the few lines of code to open that database connection, and then wiped after you got the handle.

Ideally, homomorphic encryption should be used instead.

https://en.wikipedia.org/wiki/Homomorphic_encryption

[zbentley]

> What's the threat model

Malicious code can read some/all memory in your container, but not necessarily execute. Plenty of such vulns exist.

> Where do you store the decryption key?

Not in memory. Either nowhere after use, on the filesystem, or otherwise accessible on-demand by performing IO.