Appearance

Startup Scripts, Presets & Cloud-Init Automation ​

Automate node initialization, dependency installation, and service startup with pre-tested installation presets and custom cloud-init bash scripts. This extensive guide covers everything from built-in FOTOhub presets to advanced idempotent startup scripting techniques for complex infrastructure deployments.

When launching instances via POST /compute/v1/instances, you can inject one or more Installation Presets (install_presets) alongside your custom bash scripts (startup_script). These are executed in the eu-central-1 region with complete root access to the provisioned virtual machine.

PRICING REFERENCE

As you automate infrastructure, keep in mind our highly competitive GPU spot pricing (all USD):

  • G5.xlarge: $0.38/hr (spot) - Ideal for intensive SDXL and LLM inference.
  • G4dn.xlarge: $0.20/hr (spot) - Excellent cost-to-performance ratio for general ML workloads.

Architecture & Boot Sequence ​

Before diving into presets and scripts, it is crucial to understand the lifecycle of a FOTOhub instance upon boot.

mermaid
sequenceDiagram
    participant User as API Client
    participant API as FOTOhub API (eu-central-1)
    participant Hypervisor as EC2 Hypervisor
    participant VM as FOTOhub Instance (Ubuntu 22.04)
    
    User->>API: POST /compute/v1/instances
    API-->>User: Returns Instance ID (inst_xxx)
    API->>Hypervisor: Provision VM Resources
    Hypervisor->>VM: Power On & Mount EBS
    VM->>VM: OS Boot & Network Initialization
    VM->>VM: Cloud-Init Phase 1 (Local)
    VM->>VM: Cloud-Init Phase 2 (Network)
    VM->>VM: Fetch FOTOhub Install Presets
    VM->>VM: Execute Install Presets (Sequential)
    VM->>VM: Execute User `startup_script`
    VM->>API: Signal Ready State

SCRIPT EXECUTION CONTEXT

All startup_script and install_presets run as the root user (uid=0, gid=0). The scripts execute in a non-interactive shell environment. Environment variables like $HOME are set to /root, and $PATH includes standard system binary paths. If you need to execute commands as the ubuntu user, you must use sudo -u ubuntu or su - ubuntu -c.

Comprehensive Built-In Installation Presets Reference ​

FOTOhub provides heavily optimized, pre-tested installation scripts that execute automatically during initial instance boot. Specifying these presets offloads the burden of writing robust bash installation logic for common dependencies.

1. docker ​

The docker preset installs the latest stable version of Docker CE and Docker Compose V2.

  • Components Installed: docker-ce, docker-ce-cli, containerd.io, docker-buildx-plugin, docker-compose-plugin.
  • Post-Boot State: The Docker daemon is running and enabled via systemd. The default ubuntu user is added to the docker group, allowing passwordless execution of Docker commands.
  • Log Rotation: Automatically configures Docker daemon with json-file log driver, capping logs at 50MB and 3 files max to prevent disk exhaustion.
  • Boot Time Impact: ~18s

2. python-ml ​

The ultimate preset for Python-based machine learning workloads.

  • Components Installed: Python 3.10+, pip, virtualenv. Pre-installs massive binaries to save bandwidth: PyTorch 2.4 (CUDA 12.x compatible), TorchVision, NumPy, Pandas, Scikit-Learn, FastAPI, Uvicorn, and Hugging Face transformers.
  • Post-Boot State: A global virtual environment is instantiated in /opt/ml-env. The .bashrc for the ubuntu user is modified to activate this environment automatically upon SSH login.
  • Boot Time Impact: ~45s (due to large package extraction)

3. cuda ​

The foundational layer for any GPU-accelerated computing.

  • Components Installed: NVIDIA Proprietary Drivers (v535 or latest stable), CUDA Toolkit 12.2, cuDNN 8.x, and the NVIDIA Container Toolkit.
  • Post-Boot State: The nvidia-smi command is immediately available. Docker is configured with the nvidia runtime as the default, allowing docker run --gpus all out of the box without manual daemon configuration.
  • Boot Time Impact: ~90s (Requires DKMS module compilation)

4. monitoring ​

Essential for production observability and fleet management.

  • Components Installed: Prometheus Node Exporter, NVIDIA DCGM Exporter (if GPUs are present).
  • Post-Boot State: Systemd services are active. Node exporter listens on 0.0.0.0:9100. DCGM exporter listens on 0.0.0.0:9400. UFW firewall rules are automatically adjusted to allow access from FOTOhub internal monitoring subnets.
  • Boot Time Impact: ~8s

5. jupyter ​

Instantly deploy an interactive data science environment.

  • Components Installed: JupyterLab, Jupyter Notebook, IPython, and common data visualization libraries (Matplotlib, Seaborn).
  • Post-Boot State: JupyterLab runs as a systemd service (jupyter.service) under the ubuntu user. It binds to port 8888 and is accessible via a secure token automatically generated and logged to /var/log/jupyter_token.txt.
  • Boot Time Impact: ~30s

6. comfyui ​

The preferred backend for Stable Diffusion and generative image workflows.

  • Components Installed: ComfyUI core, ComfyUI-Manager, xformers, and standard nodes.
  • Post-Boot State: Cloned into /home/ubuntu/ComfyUI. A systemd service comfyui.service is created to ensure it starts on boot and restarts on failure, bound to port 8188.
  • Boot Time Impact: ~50s

7. vllm ​

High-throughput and memory-efficient LLM inference engine.

  • Components Installed: vllm, ray, xformers.
  • Post-Boot State: The vLLM OpenAI-compatible API server is pre-configured as a systemd service (vllm.service), ready to serve models.
  • Boot Time Impact: ~40s

8. ollama ​

Local LLM execution made incredibly simple.

  • Components Installed: Official Ollama binary and service definitions.
  • Post-Boot State: Ollama daemon runs on port 11434. Models like llama3 or mistral can be pulled immediately in your startup_script.
  • Boot Time Impact: ~12s

Core API Usage: Launching with Presets and Auth ​

All compute API calls require authentication using your Bearer token. Ensure you use the correct regional Base URL.

  • Base URL: https://apis.fotohub.app/compute/v1
  • Region: All compute runs in eu-central-1
  • Auth Header: Authorization: Bearer fh_live_YOUR_API_KEY

Request Example (Multiple Presets) ​

bash
curl -X POST "https://apis.fotohub.app/compute/v1/instances" \
  -H "Authorization: Bearer fh_live_YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "g5-comfy-worker",
    "catalog_id": "g5.xlarge",
    "spot_instance": true,
    "root_volume_size_gb": 150,
    "install_presets": ["cuda", "docker", "comfyui", "monitoring"],
    "startup_script": "#!/bin/bash\necho \"Initialization complete\" > /var/log/custom_boot.log"
  }'
python
import requests
import os

FOTOHUB_API_KEY = os.environ.get("FOTOHUB_API_KEY")

payload = {
    "name": "g4-jupyter-node",
    "catalog_id": "g4dn.xlarge",
    "spot_instance": True,
    "root_volume_size_gb": 100,
    "install_presets": ["cuda", "python-ml", "jupyter", "monitoring"],
    "startup_script": """#!/bin/bash
    echo "Starting custom initialization..."
    # Custom pip packages
    sudo -u ubuntu /opt/ml-env/bin/pip install wandb accelerate bitsandbytes
    """
}

headers = {
    "Authorization": f"Bearer {FOTOHUB_API_KEY}",
    "Content-Type": "application/json"
}

response = requests.post(
    "https://apis.fotohub.app/compute/v1/instances",
    json=payload,
    headers=headers
)

print(response.json())
typescript
import axios from 'axios';

async function launchInstance() {
  const apiKey = process.env.FOTOHUB_API_KEY;
  
  const response = await axios.post('https://apis.fotohub.app/compute/v1/instances', {
    name: 'vllm-inference-endpoint',
    catalog_id: 'g5.xlarge',
    spot_instance: true,
    root_volume_size_gb: 250,
    install_presets: ['cuda', 'docker', 'vllm', 'monitoring'],
    startup_script: '#!/bin/bash\ncurl -X POST http://localhost:8000/v1/models'
  }, {
    headers: {
      'Authorization': `Bearer ${apiKey}`,
      'Content-Type': 'application/json'
    }
  });

  console.log(response.data);
}

launchInstance();

Detailed Cloud-Init Startup Script Scenarios ​

The startup_script property allows you to execute massive, multi-step shell scripts directly on the instance during its first boot.

Limitations and Guidelines ​

  • Size Limit: The startup_script string cannot exceed 65536 characters (64KB). If your script is larger, host it on an S3 bucket or GitHub and use your startup_script to simply curl and bash it.
  • Timeout: The entire cloud-init sequence (including all install_presets and your startup_script) is subject to a 30-minute timeout. If your script downloads massive multi-gigabyte models that take longer than 30 minutes, the boot process will be marked as failed in the API, though the script may continue running in the background.
  • Execution User: Runs as root.
  • Logs: Everything output to stdout and stderr is captured.

Scenario 1: Unattended SDXL Model Download & Serving ​

This script ensures that the SDXL base model and refiner are securely downloaded, verified, and loaded into ComfyUI or a custom pipeline. It uses extensive error handling to retry downloads.

bash
#!/bin/bash
set -eEuo pipefail
trap 'echo "Error occurred at line $LINENO" >&2' ERR

# Define variables
MODEL_DIR="/home/ubuntu/ComfyUI/models/checkpoints"
SDXL_BASE_URL="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors"
SDXL_REFINER_URL="https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0/resolve/main/sd_xl_refiner_1.0.safetensors"
MAX_RETRIES=5

# Function for robust downloading with aria2c
download_file() {
    local url=$1
    local dest_dir=$2
    local filename=$(basename "$url")
    
    echo "Starting download of $filename..."
    for ((i=1; i<=MAX_RETRIES; i++)); do
        if aria2c -x 16 -s 16 --continue=true -d "$dest_dir" "$url"; then
            echo "Successfully downloaded $filename"
            return 0
        else
            echo "Download failed. Attempt $i of $MAX_RETRIES. Retrying in 10s..."
            sleep 10
        fi
    done
    
    echo "Failed to download $filename after $MAX_RETRIES attempts."
    exit 1
}

# Install aria2 if not present
apt-get update && apt-get install -y aria2

# Create directory and set permissions
mkdir -p "$MODEL_DIR"
chown ubuntu:ubuntu "$MODEL_DIR"

# Execute downloads
download_file "$SDXL_BASE_URL" "$MODEL_DIR"
download_file "$SDXL_REFINER_URL" "$MODEL_DIR"

# Fix permissions after download
chown -R ubuntu:ubuntu "$MODEL_DIR"

# Restart ComfyUI service if present to detect new models
if systemctl is-active --quiet comfyui.service; then
    echo "Restarting ComfyUI to load new models..."
    systemctl restart comfyui.service
fi

echo "SDXL Setup Complete."

Scenario 2: Custom Conda Environment Initialization ​

For advanced ML workloads where python-ml preset isn't specific enough, you might want to bootstrap Miniconda from scratch.

bash
#!/bin/bash
set -euxo pipefail

# Variables
MINICONDA_URL="https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh"
CONDA_DIR="/opt/miniconda3"
ENV_NAME="pytorch_env"

echo "Starting Miniconda installation..."

# Download Miniconda installer securely
curl -L -O "$MINICONDA_URL"
bash Miniconda3-latest-Linux-x86_64.sh -b -p "$CONDA_DIR"
rm Miniconda3-latest-Linux-x86_64.sh

# Initialize conda for the ubuntu user
su - ubuntu -c "$CONDA_DIR/bin/conda init bash"

# Create a new environment with PyTorch and CUDA 12.1
su - ubuntu -c "$CONDA_DIR/bin/conda create -y -n $ENV_NAME python=3.10"
su - ubuntu -c "$CONDA_DIR/bin/conda install -y -n $ENV_NAME pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia"

# Install additional pip packages within the conda env
su - ubuntu -c "$CONDA_DIR/envs/$ENV_NAME/bin/pip install transformers accelerate diffusers xformers wandb"

# Automatically activate environment on login
echo "conda activate $ENV_NAME" >> /home/ubuntu/.bashrc

echo "Conda environment $ENV_NAME successfully configured."

Scenario 3: Formatting and Mounting an EBS Data Volume ​

When attaching secondary volumes via FOTOhub block storage, you must format and mount them on boot. This script ensures idempotency—it won't wipe the drive if it's already formatted!

bash
#!/bin/bash
set -exo pipefail

DEVICE_NAME="/dev/nvme1n1"
MOUNT_POINT="/data"

# Check if device exists
if [ ! -b "$DEVICE_NAME" ]; then
    echo "Device $DEVICE_NAME not found! Waiting for block device attachment..."
    for i in {1..12}; do
        sleep 5
        if [ -b "$DEVICE_NAME" ]; then
            echo "Device attached."
            break
        fi
    done
fi

if [ ! -b "$DEVICE_NAME" ]; then
    echo "Fatal: Device $DEVICE_NAME never attached."
    exit 1
fi

# Check if the device is already formatted (idempotency check)
FS_TYPE=$(blkid -s TYPE -o value "$DEVICE_NAME" || true)

if [ -z "$FS_TYPE" ]; then
    echo "Formatting device $DEVICE_NAME as ext4..."
    mkfs.ext4 -F "$DEVICE_NAME"
else
    echo "Device $DEVICE_NAME already contains filesystem: $FS_TYPE"
fi

# Create mount point
mkdir -p "$MOUNT_POINT"

# Get UUID for stable fstab mounting
UUID=$(blkid -s UUID -o value "$DEVICE_NAME")

# Check if already in fstab
if ! grep -q "$UUID" /etc/fstab; then
    echo "Adding $DEVICE_NAME to /etc/fstab..."
    echo "UUID=$UUID $MOUNT_POINT ext4 defaults,nofail,discard 0 2" >> /etc/fstab
fi

# Mount everything
mount -a

# Set permissive rights for the ubuntu user
chown -R ubuntu:ubuntu "$MOUNT_POINT"
chmod 755 "$MOUNT_POINT"

echo "Storage provisioning completed."

Scenario 4: Auto-Syncing FOTOhub S3 Artifacts via AWS CLI ​

Automatically sync datasets or configuration files from your FOTOhub S3-compatible object storage to the local NVMe drive.

bash
#!/bin/bash
set -euo pipefail

export AWS_ACCESS_KEY_ID="your_fotohub_s3_key"
export AWS_SECRET_ACCESS_KEY="your_fotohub_s3_secret"
export AWS_DEFAULT_REGION="eu-central-1"
ENDPOINT_URL="https://s3.eu-central-1.fotohub.app"

BUCKET_NAME="fotohub-training-datasets"
LOCAL_SYNC_DIR="/data/datasets"

# Install AWS CLI v2
curl "https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip" -o "awscliv2.zip"
apt-get update && apt-get install -y unzip
unzip -q awscliv2.zip
./aws/install
rm -rf aws awscliv2.zip

mkdir -p "$LOCAL_SYNC_DIR"

echo "Syncing $BUCKET_NAME to $LOCAL_SYNC_DIR..."
aws --endpoint-url "$ENDPOINT_URL" s3 sync "s3://$BUCKET_NAME/" "$LOCAL_SYNC_DIR/" --exact-timestamps

chown -R ubuntu:ubuntu "$LOCAL_SYNC_DIR"
echo "S3 sync complete."

Scenario 5: Unattended Tailscale Mesh VPN Node Join ​

Secure your instances by joining them immediately to your Tailscale network on boot, avoiding public IP exposure.

bash
#!/bin/bash
set -euo pipefail

# Your pre-authorized Tailscale Auth Key (ephemeral or reusable)
TAILSCALE_AUTH_KEY="tskey-auth-XXXXXXXXXXXXX-YYYYYYYYYYYYYYYYY"
HOSTNAME="fotohub-gpu-node-$(cat /etc/machine-id | head -c 8)"

echo "Installing Tailscale..."
curl -fsSL https://tailscale.com/install.sh | sh

echo "Joining Tailscale network..."
tailscale up --authkey="$TAILSCALE_AUTH_KEY" --hostname="$HOSTNAME" --accept-routes --ssh

# Optional: Restrict UFW to only allow Tailscale traffic
ufw allow in on tailscale0
ufw default deny incoming
ufw default allow outgoing
ufw --force enable

echo "Tailscale configuration finished. Node available as $HOSTNAME."

Scenario 6: Automated SSL Configuration with Nginx and Certbot ​

If you are exposing an API directly to the internet, use this script to install Nginx, configure a reverse proxy, and obtain a Let's Encrypt SSL certificate automatically.

bash
#!/bin/bash
set -exo pipefail

DOMAIN="api.my-fotohub-app.com"
EMAIL="[email protected]"
BACKEND_PORT=8000

# Install dependencies (redundant if using nginx-certbot preset, but safe)
apt-get update
apt-get install -y nginx certbot python3-certbot-nginx

# Create Nginx server block configuration
cat > /etc/nginx/sites-available/$DOMAIN << EOF
server {
    listen 80;
    server_name $DOMAIN;

    location / {
        proxy_pass http://127.0.0.1:$BACKEND_PORT;
        proxy_http_version 1.1;
        proxy_set_header Upgrade \$http_upgrade;
        proxy_set_header Connection 'upgrade';
        proxy_set_header Host \$host;
        proxy_cache_bypass \$http_upgrade;
        proxy_set_header X-Forwarded-For \$proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto \$scheme;
    }
}
EOF

# Enable the site
ln -sf /etc/nginx/sites-available/$DOMAIN /etc/nginx/sites-enabled/
rm -f /etc/nginx/sites-enabled/default

# Test and reload Nginx
nginx -t
systemctl reload nginx

# Run Certbot non-interactively
certbot --nginx -d $DOMAIN --non-interactive --agree-tos -m $EMAIL --redirect

echo "SSL configuration via Certbot is complete."

Scenario 7: WireGuard Point-to-Point VPN Server ​

Set up a highly performant WireGuard server for direct, low-latency access to your FOTOhub instance.

bash
#!/bin/bash
set -euo pipefail

SERVER_PORT=51820
SERVER_IP="10.8.0.1/24"
CLIENT_IP="10.8.0.2/32"

# Install WireGuard
apt-get update
apt-get install -y wireguard

# Generate keys
cd /etc/wireguard
umask 077
wg genkey | tee server_private_key | wg pubkey > server_public_key
wg genkey | tee client_private_key | wg pubkey > client_public_key

SERVER_PRIV=$(cat server_private_key)
CLIENT_PUB=$(cat client_public_key)

# Configure the WireGuard interface (wg0)
cat > /etc/wireguard/wg0.conf << EOF
[Interface]
Address = $SERVER_IP
SaveConfig = true
PrivateKey = $SERVER_PRIV
ListenPort = $SERVER_PORT
PostUp = iptables -A FORWARD -i wg0 -j ACCEPT; iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE
PostDown = iptables -D FORWARD -i wg0 -j ACCEPT; iptables -t nat -D POSTROUTING -o eth0 -j MASQUERADE

[Peer]
PublicKey = $CLIENT_PUB
AllowedIPs = $CLIENT_IP
EOF

# Enable IP forwarding
sed -i 's/#net.ipv4.ip_forward=1/net.ipv4.ip_forward=1/' /etc/sysctl.conf
sysctl -p

# Start and enable the WireGuard service
systemctl enable wg-quick@wg0
systemctl start wg-quick@wg0

# Output client configuration to a secure location for later retrieval
CLIENT_PRIV=$(cat client_private_key)
SERVER_PUB=$(cat server_public_key)
PUBLIC_IP=$(curl -s ifconfig.me)

cat > /root/client.conf << EOF
[Interface]
PrivateKey = $CLIENT_PRIV
Address = 10.8.0.2/24
DNS = 1.1.1.1

[Peer]
PublicKey = $SERVER_PUB
Endpoint = $PUBLIC_IP:$SERVER_PORT
AllowedIPs = 0.0.0.0/0
PersistentKeepalive = 20
EOF

echo "WireGuard setup complete. Client config generated at /root/client.conf."

Scenario 8: Complex Multi-Container Docker Compose Deployment ​

If your application consists of multiple microservices, deploying via docker-compose.yml dynamically created in the startup script is an excellent pattern.

bash
#!/bin/bash
set -exo pipefail

APP_DIR="/opt/myapp"
mkdir -p "$APP_DIR"

# Generate docker-compose.yml
cat > "$APP_DIR/docker-compose.yml" << 'EOF'
version: '3.8'

services:
  api:
    image: python:3.10-slim
    container_name: api_server
    ports:
      - "8000:8000"
    environment:
      - REDIS_URL=redis://cache:6379/0
      - DB_HOST=database
    volumes:
      - ./src:/app
    working_dir: /app
    command: bash -c "pip install fastapi uvicorn redis asyncpg && uvicorn main:app --host 0.0.0.0"
    depends_on:
      - cache
      - database

  cache:
    image: redis:7-alpine
    container_name: redis_cache
    restart: always
    ports:
      - "6379:6379"

  database:
    image: postgres:15-alpine
    container_name: postgres_db
    restart: always
    environment:
      POSTGRES_USER: admin
      POSTGRES_PASSWORD: supersecretpassword
      POSTGRES_DB: appdb
    volumes:
      - pgdata:/var/lib/postgresql/data

volumes:
  pgdata:
EOF

# Generate dummy source code for demonstration
mkdir -p "$APP_DIR/src"
cat > "$APP_DIR/src/main.py" << 'EOF'
from fastapi import FastAPI
app = FastAPI()

@app.get("/")
def read_root():
    return {"Hello": "FOTOhub Docker Compose World"}
EOF

# Set permissions
chown -R ubuntu:ubuntu "$APP_DIR"

# Start the stack (assumes docker preset was used)
cd "$APP_DIR"
docker compose up -d

echo "Multi-container application deployed and running."

Idempotency Patterns ​

Because instances might occasionally reboot, or you might re-run startup scripts via the /run-script API, it is highly recommended to write idempotent bash scripts. This means running the script multiple times results in the same state without causing errors or duplicating data.

Pattern 1: State Files ​

Create a lock file or state file upon successful completion.

bash
#!/bin/bash
if [ -f /var/run/my_setup_complete.lock ]; then
    echo "Setup already completed. Skipping."
    exit 0
fi

# ... do expensive setup ...

touch /var/run/my_setup_complete.lock

Pattern 2: Checking Existence ​

Check if users, directories, or configurations already exist before creating them.

bash
# User creation
if ! id -u customuser > /dev/null 2>&1; then
    useradd -m -s /bin/bash customuser
fi

# Appending to files (preventing duplicate lines)
if ! grep -q "custom_export=true" /etc/environment; then
    echo "custom_export=true" >> /etc/environment
fi

Pattern 3: systemctl Safety ​

Use --now and enable combined safely.

bash
# Restarts if running, starts if stopped, enables on boot
systemctl enable --now my_service || true

Debugging Cloud-Init and Startup Scripts ​

When a script fails, the instance will still boot, but the FOTOhub API may mark the status as degraded. Finding out why a script failed is critical.

Viewing Logs via the FOTOhub API ​

You can stream or fetch the final logs generated during the cloud-init process using the logs endpoint.

bash
curl -X GET "https://apis.fotohub.app/compute/v1/instances/inst_90f23b/logs?lines=200" \
  -H "Authorization: Bearer fh_live_YOUR_API_KEY"

Investigating Directly on the Instance via SSH ​

If you have SSH access to the node, the ultimate source of truth is the cloud-init output log.

bash
# View the entire execution log
cat /var/log/cloud-init-output.log

# Follow the log in real-time during boot
tail -f /var/log/cloud-init-output.log

# Check for specific failure messages
grep -i "error" /var/log/cloud-init-output.log
grep -i "failed" /var/log/cloud-init-output.log

The standard cloud-init framework logs all activities to:

  1. /var/log/cloud-init-output.log: Standard output and standard error from your startup_script.
  2. /var/log/cloud-init.log: Detailed internal operations of the cloud-init system itself.

PRO TIP: SET -X

Always start your bash scripts with set -x. This will print every single command to /var/log/cloud-init-output.log before it is executed, making it trivial to see exactly which line of your script caused a failure.

Dynamic Script Execution via /run-script ​

Execute administrative bash scripts on running instances dynamically without opening manual SSH sessions or restarting the VM. This is incredibly useful for remote diagnostics or triggering updates.

Endpoint: POST /compute/v1/instances/{id}/run-script ​

bash
curl -X POST https://apis.fotohub.app/compute/v1/instances/inst_90f23b/run-script \
  -H "Authorization: Bearer fh_live_YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "script": "nvidia-smi --query-gpu=utilization.gpu,memory.used,memory.total --format=csv,noheader"
  }'

Response Example ​

The API blocks until the script completes and returns the exact exit code, standard output, and standard error.

json
{
  "success": true,
  "exit_code": 0,
  "stdout": "4%, 1824 MiB, 24576 MiB\n",
  "stderr": "",
  "duration_ms": 320
}

This endpoint shares the same execution context as startup_script: it runs as root in a non-interactive shell. For long-running background tasks, wrap your commands in nohup or create temporary systemd units.

© 2026 FOTOhub. All rights reserved.