> ## Documentation Index
> Fetch the complete documentation index at: https://committ.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# ConnectionWatchdog

> DEPRECATED -- Legacy zombie socket detector replaced by HydrationSync.

# ConnectionWatchdog

**Source:** `components/system/ConnectionWatchdog.tsx` (51 lines -- tombstone file)

<Warning>
  This component has been DEPRECATED and removed from the provider tree. The file is kept as a tombstone to prevent accidental re-creation of the same anti-pattern.
</Warning>

***

## Why It Was Removed (April 2026)

The ConnectionWatchdog was a legacy system that attempted to detect zombie WebSocket connections and fix them by calling convexClient.clearAuth() directly. This approach was fundamentally broken for four reasons:

### 1. Bypassed the ResurrectionProvider

The centralized lifecycle system (ResurrectionProvider to ConvexClientWrapper) creates a brand-new ConvexReactClient on resurrection. The Watchdog instead called clearAuth() on the SAME dead client, leaving it in a half-alive state that corrupted subsequent queries.

### 2. Killed In-Flight Sagas

Calling clearAuth() mid-mutation destroyed the WebSocket transport while Triple-Write Sagas were waiting for Convex responses, causing 15-second timeouts, ROLLBACK cascades, and "Local Sync Failed" modals.

### 3. Panicked During Cold Starts

Convex serverless deployments naturally go to sleep after inactivity. The Watchdog's 16-second probe timeout (2x 8s probes) was far too aggressive for the 10-30 second cold-start window. It would nuke auth, and when the server finally woke up, the first query hit a "Server Error" (auth token cleared) causing a RED SCREEN crash.

### 4. Probed a Non-Existent Function

The **health\_check** query was never deployed to the Convex backend. The probe always failed with "function not found", making it useless as a health signal.

***

## Replacement

All zombie socket detection is now handled by useHydrationSync (inside the HydrationEngine), which already:

* Monitors foreground transitions via AppState
* Sends real Convex queries as health probes (not fake health checks)
* Calls resurrect() from the ResurrectionProvider on failure
* Creates a brand-new ConvexReactClient (the correct fix)
* Implements Smart Blame Logic to distinguish network loss from zombie sockets

***

## Lesson Learned

Never bypass the centralized lifecycle system. Connection recovery must flow through the ResurrectionProvider to ensure proper client teardown and recreation.