Frequent unexpected full page reloads after upgrading Reflex from 0.8.17 → 0.8.20 (likely caused by new reconnect logic) #6000
Description
Describe the bug
After upgrading Reflex from 0.8.17 to 0.8.20, our deployed apps randomly and frequently trigger a full page reload / full re-hydrate.
- This happens only in deployments (prod + other deployed environments).
- The exact same codebase behaves normally locally (no random reloads).
- There is no obvious error in the Reflex logs that explains the reloads.
- It makes our UI (especially forms) very hard to use because the page suddenly reloads while users are interacting with it.
Downgrading Reflex back to 0.8.17 immediately stops the behavior, using the same code and infrastructure.
To Reproduce
Unfortunately we don’t have a minimal public reproduction, but the behavior is consistently reproducible in our deployed environments under these conditions:
- Upgrade Reflex from 0.8.17 → 0.8.20.
- Deploy the app (behind our existing reverse proxy; in our case Caddy – config unchanged between versions).
- Open a page with interactive state (e.g. a form where you can create a project).
- Start filling out the form / interacting with the page and keep it open for some time.
- Observe that, at seemingly random intervals, the page reloads / does a full re-hydrate by itself.
Expected behavior
- The app should not randomly reload or fully re-hydrate while the user is interacting with it.
- WebSocket reconnects (if any) should be transparent and should not cause a full page reload that wipes client-side state.
- Behavior in deployed environments should match the stability we see locally.
Specifics (please complete the following information):
- Python Version: 3.13.2
- Reflex Version:
- Works: 0.8.17
- Broken: 0.8.20
- OS: Locally MacOS, Deployment Linux
- Browser (Optional): Tested with Chrome and Safari