Server stops responding with Django Channels and granian under low load

[alexiokay]

Environment

• Granian version: 2.6.0
• Python version: 3.11+
• Framework: Django 5.1.5 + Django Channels 4.3.1
• Channel Layer: Redis (channels-redis 4.3.0)
• Deployment: Railway (cloud platform)
• OS: Linux (Railway container)

Granian Start Command

granian --interface asgi --host 0.0.0.0 --port $PORT backend.asgi:application

Problem Description

Server becomes completely unresponsive (freezes) after handling a small number of WebSocket messages through Django Channels. Response time degrades to 30+ seconds and never recovers, requiring container restart.

Symptoms

1. ✅ Server starts normally and handles initial WebSocket connections fine
2. ✅ First ~10-15 messages relay successfully (Mobile → Backend → Desktop via WebSocket)
3. ❌ After ~19 messages in a few seconds, response time spikes to 30+ seconds
4. ❌ Server stops responding to all requests (HTTP and WebSocket)
5. ❌ No error logs, no panic messages - just silent freeze
6. ❌ Container becomes unresponsive until manual restart

Architecture

Django Channels WebSocket consumers using:

• Async Redis connection pool (redis.asyncio, max 50 connections)
• Channel Layer for message routing between WebSocket consumers
• Mobile → Backend → Desktop relay pattern (E2EE message forwarding)

# Simplified consumer example
class RelayConsumer(AsyncWebsocketConsumer):
    async def receive(self, text_data):
        data = json.loads(text_data)
        
        # Forward message to desktop via channel layer
        await self.channel_layer.send(
            desktop_channel,
            {
                'type': 'relay_from_mobile',
                'message': data,
            }
        )

What We Tried (Did NOT Fix)

• ❌ Adjusting --runtime-threads (1, 4, 8, 16)
• ❌ Changing --workers count
• ❌ Optimizing code (cached channels, removed Redis writes from hot path)
• ❌ Reducing message volume (only 19 messages before freeze)
• ❌ Different Granian flags

What FIXED It

✅ Switching to Uvicorn 0.38.0 completely resolved the issue - same code, same load, zero freezes.

# This works perfectly:
uvicorn backend.asgi:application --host 0.0.0.0 --port $PORT --workers 2 --log-level warning

Logs Before Freeze

📨 Message relayed: Mobile → Desktop | User: user@example.com | ⏱️ Total: 15.2ms (cached channel)
📨 Message relayed: Mobile → Desktop | User: user@example.com | ⏱️ Total: 12.8ms (cached channel)
📨 Message relayed: Mobile → Desktop | User: user@example.com | ⏱️ Total: 14.1ms (cached channel)
...
[19 messages total, then complete silence - no more logs]

Railway metrics at freeze time:

• Memory: around 300-400MB relatively stable
• CPU: Normal
• Network: Normal
• Response time: 50ms → 30,000ms+ (never recovers)

System Info

requirements.txt (relevant packages)

Django==5.1.5
channels==4.3.1
channels-redis==4.3.0
redis==5.2.1
granian==2.6.0
asgiref==3.10.0
uvicorn==0.38.0
websockets==15.0.1

Reproduction Steps

1. Set up Django Channels with Redis backend
2. Create WebSocket relay consumer (mobile → backend → desktop)
3. Deploy with Granian using ASGI interface
4. Send 15-20 rapid WebSocket messages through relay
5. Observe server freeze after ~19 messages

Expected Behavior

Server should handle WebSocket messages continuously without freezing (like Uvicorn does).

Actual Behavior

Server freezes silently after ~19 messages with no error logs, requiring restart.

What I have noticed additionally is that it keeps working relatively fine until i start sending websocket messages slowly in faster intervals, but it's stil very small intervals.

Funny that if I run it on may windows machine, this issue doesn't persist or at list I haven't notice.

---

Additional Context
We were very excited to use Granian for its performance benefits, but had to switch to Uvicorn for stability.