Skip to content

Troubleshooting

Jump to bottom
charles edited this page Jul 12, 2026 · 2 revisions

Troubleshooting

Common errors and how to fix them.

Quick diagnostics

computing-provider inference status   # registration + model list on Swan Inference
computing-provider inference config   # active config values
computing-provider info               # node ID, wallet addresses

Connection and authentication

authentication required / invalid provider API key

The sk-prov-* key in config.toml is missing or wrong.

# Check the active key
computing-provider inference config

# Fix — set in config.toml
[Inference]
  ApiKey = "sk-prov-<your-key>"

# Or via env var (takes priority over config.toml)
export INFERENCE_API_KEY=sk-prov-<your-key>

Get a new key by logging into inference.swanchain.io → Provider dashboard → API keys.

WebSocket bad handshake / connection refused

The WebSocketURL has a double /ws suffix or the wrong host/port.

# WRONG — client appends /ws automatically
WebSocketURL = "wss://inference-ws.swanchain.io/ws"

# CORRECT
WebSocketURL = "wss://inference-ws.swanchain.io"

For local dev:

WebSocketURL = "ws://localhost:8081"   # not ws://localhost:8081/ws

Models not healthy

Models stay unknown after startup

The health checker cannot reach the model endpoint.

# Confirm the server is running
curl http://localhost:30000/v1/models

# Check models.json endpoint matches the running server
cat ~/.swan/computing/models.json

Models registered but requests not arriving

The model IDs in config.toml [Inference] Models must exactly match the keys in models.json.

# config.toml
Models = ["meta-llama/Llama-3.2-3B-Instruct"]
// models.json — key must match exactly
{
  "meta-llama/Llama-3.2-3B-Instruct": { ... }
}

Hot-reload not picking up models.json changes

Force a reload:

curl -X POST http://localhost:9085/api/v1/computing/inference/models/reload

Docker / GPU

permission denied while trying to connect to the Docker daemon

sudo usermod -aG docker $USER
newgrp docker    # or log out and back in

container "/resource-exporter" already in use

docker rm -f resource-exporter

GPU not visible in containers

# Verify NVIDIA Container Toolkit
nvidia-container-cli info

# Reinstall if needed
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

# Test
docker run --rm --gpus all nvidia/cuda:12.0-base-ubuntu22.04 nvidia-smi

Config changes not taking effect

You changed config.toml but the behavior didn't change.

If you're running from a binary, rebuild:

make clean && make mainnet && make install

Use go run during development to always pick up the latest code:

go run ./cmd/computing-provider run

Logs

# Tail live logs (if started with nohup)
tail -f cp.log

# Filter for errors
grep -i "error\|failed\|refused" cp.log

# Filter for a specific model
grep "meta-llama" cp.log

Still stuck?

When reporting an issue, include:

computing-provider info
computing-provider inference status
tail -50 cp.log

Clone this wiki locally