#!/usr/bin/env bash
set -euo pipefail

WORKER_PORT="${1:-8001}"

echo "Starting Qwen3-8B-FP8 worker on port ${WORKER_PORT}..."
docker exec -d vllm_node sh -c "
cd /root/.cache/huggingface/hub

exec vllm serve 'Qwen/Qwen3-8B-FP8' \
  --served-model-name 'Qwen3-8B-FP8' \
  --max-model-len 32768 \
  --max-num-seqs 8 \
  --gpu-memory-utilization 0.20 \
  --port '${WORKER_PORT}' \
  --host 0.0.0.0 \
  --enable-chunked-prefill \
  --enable-auto-tool-choice \
  --tool-call-parser hermes \
  --trust-remote-code \
  --kv-cache-dtype fp8 \
  -tp 1 \
  >> /tmp/qwen3-8b-worker.log 2>&1 &
"

echo "Waiting for worker to start..."
sleep 10

for i in $(seq 1 30); do
  if curl -sf "http://localhost:${WORKER_PORT}/v1/models" > /dev/null 2>&1; then
    echo "✓ Worker ready on port ${WORKER_PORT}"
    curl -s "http://localhost:${WORKER_PORT}/v1/models" | python3 -m json.tool 2>/dev/null || true
    exit 0
  fi
  echo "  Waiting... (${i}/30)"
  sleep 2
done

echo "✗ Worker failed to start. Check /tmp/qwen3-8b-worker.log"
exit 1