Why we are moving identity to Rust
Login and sessions sit in the path of every request. Here is what changes, why it matters and what improvement we expect.
Every time you open the dashboard, the server has to answer one question before anything else: who are you. That question runs on every visit, on every project and on every view that needs permissions. It is the door everything else walks through.
Today that door is written in Python. The next step is to move it to Rust.
What changes
The identity service does three things: verify the password, issue the token and decide whether that token is still valid. In Rust it will do exactly the same, with the same contract, but with less work per request.
- Native verification, with the same bcrypt cost and the same hash upgrade when needed.
- Identical tokens: whatever the new service issues also works for the rest of the API.
- Real sessions: every sign-in records the device, the IP and the browser, and can be revoked.
- Signing out invalidates the token immediately, not when it expires.
That last point matters most to you: today a token can stay valid even after you sign out. With the change, it stops working on the next request.
What improves
Login does not become instant, because most of the time is spent in bcrypt, not in the language. What improves is everything around it: less work per request, less memory and fewer cold starts.
| Measurement (estimate) | With Rust | Today |
|---|---|---|
| Direct sign-in | ~0.49 s | ~0.59 s |
| Sign-in over the internet | ~0.69 s | ~0.94 s |
Between 15% and 25% less, with less variation between requests.
What does not change
Passwords are still stored with bcrypt and the same cost. There are no security shortcuts: this is a platform change, not a policy change.
What you see does not change either. The dashboard, projects and permissions work the same; only the engine behind them differs.
When
This is a gradual change. First identity, then the dashboard reads, and finally writes and background jobs. Each block is validated before the next one starts, and the previous one stays as a fallback until the new one is proven.
No promised dates: we would rather ship it right than ship it early.