|
| 1 | +#!/bin/bash |
| 2 | +# Buildkite GPU Node Setup - Simplified version |
| 3 | +# Usage: sudo BUILDKITE_AGENT_TOKEN=xxx GPU_TYPE=test ./setup-node-simple.sh |
| 4 | + |
| 5 | +set -euo pipefail |
| 6 | + |
| 7 | +# === CONFIGURATION === |
| 8 | +BUILDKITE_TOKEN="${BUILDKITE_AGENT_TOKEN:?Must set BUILDKITE_AGENT_TOKEN}" |
| 9 | +GPU_TYPE="${GPU_TYPE:?Must set GPU_TYPE (e.g., b200, mi300, h100, test)}" |
| 10 | +NODE_NAME="${NODE_NAME:-$(hostname)}" |
| 11 | + |
| 12 | +# Auto-detect GPU count |
| 13 | +if command -v nvidia-smi &> /dev/null; then |
| 14 | + GPU_COUNT=$(nvidia-smi --query-gpu=count --format=csv,noheader | head -1) |
| 15 | +else |
| 16 | + GPU_COUNT="${GPU_COUNT:-1}" |
| 17 | +fi |
| 18 | + |
| 19 | +echo "=== Buildkite GPU Node Setup ===" |
| 20 | +echo "Node: ${NODE_NAME}" |
| 21 | +echo "GPU Type: ${GPU_TYPE}" |
| 22 | +echo "GPU Count: ${GPU_COUNT}" |
| 23 | +echo "" |
| 24 | + |
| 25 | +# === INSTALL BUILDKITE AGENT === |
| 26 | +if ! command -v buildkite-agent &> /dev/null; then |
| 27 | + echo "Installing Buildkite Agent..." |
| 28 | + apt-get update |
| 29 | + apt-get install -y apt-transport-https gnupg |
| 30 | + curl -fsSL https://keys.openpgp.org/vks/v1/by-fingerprint/32A37959C2FA5C3C99EFBC32A79206696452D198 | \ |
| 31 | + gpg --dearmor -o /usr/share/keyrings/buildkite-agent-archive-keyring.gpg |
| 32 | + echo "deb [signed-by=/usr/share/keyrings/buildkite-agent-archive-keyring.gpg] https://apt.buildkite.com/buildkite-agent stable main" | \ |
| 33 | + tee /etc/apt/sources.list.d/buildkite-agent.list |
| 34 | + apt-get update |
| 35 | + apt-get install -y buildkite-agent |
| 36 | +fi |
| 37 | + |
| 38 | +# === STOP EXISTING AGENTS === |
| 39 | +echo "Stopping existing agents..." |
| 40 | +systemctl stop buildkite-agent 2>/dev/null || true |
| 41 | +for i in $(seq 0 15); do |
| 42 | + systemctl stop "buildkite-agent-gpu${i}" 2>/dev/null || true |
| 43 | + systemctl disable "buildkite-agent-gpu${i}" 2>/dev/null || true |
| 44 | +done |
| 45 | + |
| 46 | +# === CREATE DIRECTORIES === |
| 47 | +mkdir -p /var/lib/buildkite-agent/builds |
| 48 | +chown -R buildkite-agent:buildkite-agent /var/lib/buildkite-agent |
| 49 | + |
| 50 | +# === CONFIGURE GIT TO USE HTTPS === |
| 51 | +sudo -u buildkite-agent git config --global url."https://github.com/".insteadOf "git@github.com:" |
| 52 | + |
| 53 | +# === CREATE AGENT FOR EACH GPU === |
| 54 | +echo "Creating ${GPU_COUNT} agents..." |
| 55 | + |
| 56 | +for gpu_idx in $(seq 0 $((GPU_COUNT - 1))); do |
| 57 | + agent_name="${NODE_NAME}-gpu${gpu_idx}" |
| 58 | + config_file="/etc/buildkite-agent/buildkite-agent-gpu${gpu_idx}.cfg" |
| 59 | + build_dir="/var/lib/buildkite-agent/builds/gpu${gpu_idx}" |
| 60 | + |
| 61 | + mkdir -p "${build_dir}" |
| 62 | + chown buildkite-agent:buildkite-agent "${build_dir}" |
| 63 | + |
| 64 | + # Write config |
| 65 | + cat > "${config_file}" << EOF |
| 66 | +token="${BUILDKITE_TOKEN}" |
| 67 | +name="${agent_name}" |
| 68 | +tags="queue=${GPU_TYPE},gpu=${GPU_TYPE},gpu-index=${gpu_idx},node=${NODE_NAME}" |
| 69 | +build-path="${build_dir}" |
| 70 | +hooks-path="/etc/buildkite-agent/hooks" |
| 71 | +EOF |
| 72 | + |
| 73 | + # Write systemd service |
| 74 | + cat > "/etc/systemd/system/buildkite-agent-gpu${gpu_idx}.service" << EOF |
| 75 | +[Unit] |
| 76 | +Description=Buildkite Agent (GPU ${gpu_idx}) |
| 77 | +Documentation=https://buildkite.com/docs/agent/v3 |
| 78 | +After=network.target |
| 79 | +
|
| 80 | +[Service] |
| 81 | +Type=simple |
| 82 | +User=buildkite-agent |
| 83 | +Environment=NVIDIA_VISIBLE_DEVICES=${gpu_idx} |
| 84 | +Environment=CUDA_VISIBLE_DEVICES=${gpu_idx} |
| 85 | +ExecStart=/usr/bin/buildkite-agent start --config ${config_file} |
| 86 | +RestartSec=5 |
| 87 | +Restart=on-failure |
| 88 | +TimeoutStartSec=10 |
| 89 | +TimeoutStopSec=60 |
| 90 | +
|
| 91 | +[Install] |
| 92 | +WantedBy=multi-user.target |
| 93 | +EOF |
| 94 | + |
| 95 | + echo " Created agent ${gpu_idx}: GPU=${gpu_idx}" |
| 96 | +done |
| 97 | + |
| 98 | +# === START AGENTS === |
| 99 | +echo "Starting agents..." |
| 100 | +systemctl daemon-reload |
| 101 | + |
| 102 | +for gpu_idx in $(seq 0 $((GPU_COUNT - 1))); do |
| 103 | + systemctl enable "buildkite-agent-gpu${gpu_idx}" |
| 104 | + systemctl start "buildkite-agent-gpu${gpu_idx}" |
| 105 | +done |
| 106 | + |
| 107 | +sleep 3 |
| 108 | + |
| 109 | +echo "" |
| 110 | +echo "=== Agent Status ===" |
| 111 | +for gpu_idx in $(seq 0 $((GPU_COUNT - 1))); do |
| 112 | + status=$(systemctl is-active "buildkite-agent-gpu${gpu_idx}" 2>/dev/null || echo "unknown") |
| 113 | + echo " GPU ${gpu_idx}: ${status}" |
| 114 | +done |
| 115 | + |
| 116 | +echo "" |
| 117 | +echo "=== Setup Complete ===" |
| 118 | +echo "Created ${GPU_COUNT} agents for queue: ${GPU_TYPE}" |
| 119 | +echo "Each agent sees only its assigned GPU via NVIDIA_VISIBLE_DEVICES" |
| 120 | +echo "" |
| 121 | +echo "Check agents at: https://buildkite.com/organizations/YOUR_ORG/agents" |
0 commit comments