#Erik Sipsma3294 Ugh That actually makes
1 messages · Page 1 of 1 (latest)
Ah yeah, it's the same issue as when I tried to make FSID be a hash... I guess in this case it might just mean we need the ability to mark something as no-cache
Good catch though
I think there's something "fundamental" with secrets though
We basically have a system that works on the basis that everything is cached, written to caching layers
And then we have secrets that fundamentally work by NEVER hitting a caching layer, being ephemeral / in-memory, etc
The two are polar opposites
I'm still thinking there ought to be a first class "secret provider" support that provides entirely different guarantees than regular extensions:
- promise we never cache, never end up in persistent storage
- JIT resolving (needed for rotation, etc) // this is less important but it's something vault etc will assume
Oh I completely agree that we need secret providers, my only attachment to addSecret was that it is possible today whereas secret providers need more design and time. Use of no-cache when an extension needs to call addSecret would just be a stopgap.
However, given that it's not exactly clear how to support no-cache today, it could just be worth jumping right to secret providers
I have some philosophical questions around this
- Should any extension be able to add secrets? Or should they be given some sort of privileged permission (or scope) to do so?
- Should any extension be able to read all secrets? Or should the be given access to (or request access to) specific secrets