#!/usr/bin/env bash# CRITICAL## Deploy vLLM Studio from this machine to the remote GPU server.## ─── Connection ───────────────────────────────────────────────────────────##   Host:   192.168.1.70  (Linux, AMD EPYC 7443P, 8× RTX 3090)#   User:   ser#   Key:    ~/.ssh/linux-ai#   Path:   /home/ser/workspace/projects/lmvllm##   Test:   ssh -i ~/.ssh/linux-ai ser@192.168.1.70 hostname## ─── What runs where ─────────────────────────────────────────────────────##   Docker (infra only, stays up across deploys):#     postgres:16       :5432   LiteLLM database#     litellm           :4100   API gateway##   Native on host (needs nvidia-smi + host process visibility):#     controller (bun)  :8080   Model lifecycle, GPU stats, chat, recipes#     frontend (next)   :3000   Web UI##   Managed separately:#     vLLM / SGLang     :8000   Inference (launched via controller or manually)## ─── How it works ─────────────────────────────────────────────────────────##   1. rsync  — push controller/src, frontend/src, shared/, config/ to remote#   2. install — bun install (controller), npm install (frontend)#   3. restart — kill old process, start new one via nohup, wait for port#   4. verify  — hit health endpoints, print GPU and model status## ─── Usage ────────────────────────────────────────────────────────────────##   ./scripts/deploy-remote.sh              Deploy everything#   ./scripts/deploy-remote.sh controller   Controller only#   ./scripts/deploy-remote.sh frontend     Frontend only#   ./scripts/deploy-remote.sh infra        Restart Docker infra#   ./scripts/deploy-remote.sh status       Check what's running (no changes) set -euo pipefail # ─── Config ─────────────────────────────────────────────────────────────── SSH_KEY="$HOME/.ssh/linux-ai"REMOTE_USER="ser"REMOTE_HOST="192.168.1.70"REMOTE_DIR="/home/ser/vllm/lmvllm" SSH_OPTS="-T -i $SSH_KEY -o ConnectTimeout=5"REMOTE="$REMOTE_USER@$REMOTE_HOST" # ─── Output ─────────────────────────────────────────────────────────────── _c() { printf '\033[%sm' "$1"; }_r="$(_c 31)" _g="$(_c 32)" _y="$(_c 33)" _b="$(_c 36)" _d="$(_c 2)" _n="$(_c 0)" step() { printf '%s==>%s %s\n' "$_b" "$_n" "$*"; }ok()   { printf '%s  ✓%s %s\n' "$_g" "$_n" "$*"; }warn() { printf '%s  !%s %s\n' "$_y" "$_n" "$*"; }fail() { printf '%s  ✗%s %s\n' "$_r" "$_n" "$*"; }dim()  { printf '%s%s%s\n' "$_d" "$*" "$_n"; } die() { fail "$@"; exit 1; } # ─── Helpers ────────────────────────────────────────────────────────────── remote() { ssh $SSH_OPTS "$REMOTE" "$@"; } # rsync a local directory to remote, excluding node_modules and build artifactssync_dir() {  local src="$1" dst="$2"  rsync -az --delete \    --exclude 'node_modules' \    --exclude '.next' \    --exclude 'bun.lock' \    --exclude '.turbo' \    --exclude '*.test.ts' \    --exclude 'test-output' \    -e "ssh $SSH_OPTS" \    "$src" "$REMOTE:$dst" 2>&1 | grep -v 'cannot delete non-empty directory' || true} # Wait for a port to be listening, or fail after N secondswait_port() {  local port="$1" label="$2" max="${3:-10}"  for i in $(seq 1 "$max"); do    if remote "ss -tlnp | grep -q ':${port}\b'" 2>/dev/null; then      return 0    fi    sleep 1  done  fail "$label not listening on :$port after ${max}s"  remote "tail -20 /tmp/${label}-stdout.log" 2>/dev/null || true  return 1} # ─── Sync ───────────────────────────────────────────────────────────────── sync_controller() {  step "Syncing controller"  sync_dir controller/src/      "$REMOTE_DIR/controller/src/"  sync_dir controller/scripts/  "$REMOTE_DIR/controller/scripts/" 2>/dev/null || true  rsync -az -e "ssh $SSH_OPTS" \    controller/package.json controller/tsconfig.json \    "$REMOTE:$REMOTE_DIR/controller/" 2>/dev/null  ok "controller/src → remote"} sync_frontend() {  step "Syncing frontend"  sync_dir frontend/src/ "$REMOTE_DIR/frontend/src/"  rsync -az -e "ssh $SSH_OPTS" \    frontend/package.json frontend/tsconfig.json \    frontend/next.config.ts frontend/tailwind.config.ts \    frontend/postcss.config.mjs \    "$REMOTE:$REMOTE_DIR/frontend/" 2>/dev/null  ok "frontend/src → remote"} sync_shared() {  step "Syncing shared types"  sync_dir shared/ "$REMOTE_DIR/shared/"  ok "shared/ → remote"} sync_config() {  step "Syncing config"  sync_dir config/ "$REMOTE_DIR/config/"  rsync -az -e "ssh $SSH_OPTS" \    docker-compose.yml .env.example \    "$REMOTE:$REMOTE_DIR/"  ok "config/, docker-compose.yml → remote"} sync_all() {  sync_controller  sync_frontend  sync_shared  sync_config} # ─── Install ────────────────────────────────────────────────────────────── install_controller() {  step "Installing controller deps"  remote "cd $REMOTE_DIR/controller && ~/.bun/bin/bun install --frozen-lockfile 2>&1 | tail -1" || \  remote "cd $REMOTE_DIR/controller && ~/.bun/bin/bun install 2>&1 | tail -1"  ok "bun install"} install_frontend() {  step "Installing frontend deps"  remote "cd $REMOTE_DIR/frontend && npm install --silent 2>&1 | tail -3"  ok "npm install"} # ─── Restart ────────────────────────────────────────────────────────────── restart_controller() {  step "Restarting controller on :8080"  remote bash <<'REMOTE'set -ecd /home/ser/vllm/lmvllmdocker compose stop controller 2>/dev/null || truepkill -f "bun.*controller/src/main.ts" 2>/dev/null || truefuser -k 8080/tcp >/dev/null 2>&1 || truesleep 1set -a; source .env 2>/dev/null || true; set +anohup ~/.bun/bin/bun run controller/src/main.ts > /tmp/controller-stdout.log 2>&1 &REMOTE  wait_port 8080 controller || return 1  ok "controller :8080 (pid $(remote "pgrep -f 'bun.*controller/src/main.ts'" 2>/dev/null || echo '?'))"} restart_frontend() {  step "Building frontend"  remote bash <<'REMOTE'set -euo pipefailcd /home/ser/vllm/lmvllm/frontendexport BACKEND_URL=http://localhost:8080export LITELLM_URL=http://localhost:4100export LITELLM_MASTER_KEY=${LITELLM_MASTER_KEY:-sk-master}npx next build 2>&1 | tail -5REMOTE  ok "next build"   step "Restarting frontend on :3000"  remote bash <<'REMOTE'set -euo pipefailcd /home/ser/vllm/lmvllm/frontenddocker compose -f ../docker-compose.yml stop frontend 2>/dev/null || truepkill -f "next start" 2>/dev/null || truepkill -f "next dev" 2>/dev/null || truefuser -k 3000/tcp >/dev/null 2>&1 || truesleep 1export BACKEND_URL=http://localhost:8080export LITELLM_URL=http://localhost:4100export LITELLM_MASTER_KEY=${LITELLM_MASTER_KEY:-sk-master}nohup npx next start > /tmp/frontend-stdout.log 2>&1 &REMOTE  wait_port 3000 frontend 15 || return 1  ok "frontend :3000 (production)"} # ─── Infra ──────────────────────────────────────────────────────────────── start_infra() {  step "Starting Docker infra (postgres + litellm)"  remote "cd $REMOTE_DIR && docker compose up -d postgres litellm 2>&1 | tail -5"  ok "postgres :5432, litellm :4100"} # ─── Status / diagnostics ──────────────────────────────────────────────── show_status() {  step "Status"  echo ""  remote bash <<'REMOTE'_g='\033[32m' _r='\033[31m' _d='\033[2m' _n='\033[0m' probe() {  local label="$1" url="$2"  local code  code=$(curl -s -m 3 -o /dev/null -w "%{http_code}" "$url" 2>/dev/null || echo 000)  if [[ "$code" =~ ^2 ]]; then    printf "  ${_g}%-22s${_n} %s\n" "$label" ":$3 OK"  else    printf "  ${_r}%-22s${_n} %s\n" "$label" ":$3 ($code)"  fi} probe "controller"      http://localhost:8080/health    8080probe "frontend"        http://localhost:3000            3000probe "frontend→proxy"  http://localhost:3000/api/proxy/health 3000probe "vllm"            http://localhost:8000/v1/models  8000 # Services that need port checks instead of HTTP probesfor pair in "litellm:4100" "postgres:5432"; do  label="${pair%%:*}" port="${pair##*:}"  if ss -tlnp 2>/dev/null | grep -q ":${port}\b"; then    printf "  ${_g}%-22s${_n} %s\n" "$label" ":$port OK"  else    printf "  ${_r}%-22s${_n} %s\n" "$label" ":$port down"  fidoneecho "" # GPU tablegpus=$(curl -s http://localhost:8080/gpus 2>/dev/null)if echo "$gpus" | python3 -c "import sys,jsond=json.load(sys.stdin)if not d.get('gpus'): sys.exit(1)for g in d['gpus']:    pct = g['memory_used_mb'] / g['memory_total_mb'] * 100    print(f'  GPU {g[\"index\"]}  {g[\"name\"]:30s}  {g[\"memory_used_mb\"]:>5d}/{g[\"memory_total_mb\"]}MB ({pct:4.0f}%)  {g[\"temp_c\"]:>2d}°C  {g[\"power_draw\"]:>6.1f}W')" 2>/dev/null; then  echo ""fi # Running modelcurl -s http://localhost:8080/status 2>/dev/null | python3 -c "import sys,jsond=json.load(sys.stdin)if d.get('running'):    p=d['process']    print(f'  Model: {p[\"served_model_name\"]}  ({p[\"backend\"]}, pid {p[\"pid\"]}, :{p[\"port\"]})')else:    print('  Model: (none)')" 2>/dev/null || trueREMOTE} # ─── Commands ───────────────────────────────────────────────────────────── cd "$(dirname "$0")/.." case "${1:-}" in  controller)    sync_controller; sync_shared; install_controller; restart_controller    echo ""; show_status ;;  frontend)    sync_frontend; install_frontend; restart_frontend    echo ""; show_status ;;  infra)    sync_config; start_infra ;;  status)    show_status ;;  ""|all)    sync_all    install_controller; install_frontend    start_infra    restart_controller; restart_frontend    echo ""; show_status ;;  *)    echo "Usage: $(basename "$0") [all|controller|frontend|infra|status]"    exit 1 ;;esac