JuiceFS Skill

Prerequisites

JuiceFS Client Installation

The initialization script can install JuiceFS automatically if needed.

Standard Installation (Recommended)

curl -sSL https://d.juicefs.com/install | sh -

This installs to

/usr/local/bin/juicefs

Manual Installation

wget https://github.com/juicedata/juicefs/releases/latest/download/juicefs-linux-amd64.tar.gz
tar -zxf juicefs-linux-amd64.tar.gz
sudo install juicefs /usr/local/bin/

Verify Installation

juicefs version

Using the Initialization Script

The initialization script will:

• Check if JuiceFS is in your PATH
• Offer to install it automatically if not found
• Guide you through the process

Overview

JuiceFS is a high-performance POSIX file system designed for cloud-native environments. It separates data and metadata storage:

• Data: Stored in object storage (S3, GCS, Azure Blob, local disk, etc.)
• Metadata: Stored in databases (Redis, MySQL, PostgreSQL, TiKV, SQLite, etcd)
• Client: Mounts the file system and coordinates data/metadata

When to Use This Skill

Use this skill when:

• Setting up or managing JuiceFS file systems
• Integrating JuiceFS with Kubernetes, Hadoop, or Docker
• Optimizing JuiceFS performance for specific workloads
• Troubleshooting JuiceFS issues
• Migrating data to/from JuiceFS
• Configuring JuiceFS for big data, ML training, or shared storage

Core Concepts

Architecture

┌─────────────┐
│ JuiceFS     │
│ Client      │
└──────┬──────┘
       │
    ┌──┴───────────┐
    │              │
┌───▼────┐    ┌───▼────────┐
│Metadata│    │Object      │
│Engine  │    │Storage     │
│(Redis) │    │(S3, etc.)  │
└────────┘    └────────────┘

Data Organization

• Files → Chunks (max 64 MiB) → Slices (variable) → Blocks (4 MiB) → Object Storage

Metadata Engines

• Redis: Best for production, fast, supports Sentinel/Cluster for HA
• MySQL/PostgreSQL: Good for production with existing infrastructure
• TiKV: Excellent for large-scale distributed deployments
• SQLite: Simple, single-node, good for testing/development
• etcd: Small to medium scale

🔒 Security: Protecting Sensitive Credentials

IMPORTANT FOR AI AGENTS: When working with JuiceFS in AI agent environments, credentials (AK/SK, passwords) should NOT be exposed to the AI model to prevent data leakage.

SKILL Responsibility Boundary

What This SKILL Provides:

• Security guidance for AI agents working with JuiceFS
• Method to prevent AI agents from accessing sensitive credentials
• Secure initialization process with binary compilation
• Clear separation between admin setup (root) and agent usage (non-root)

What This SKILL Does NOT Handle:

• How AI agents are deployed or run
• Host system security configuration
• Network security setup
• General system administration

Design Philosophy: This SKILL assumes the AI agent runs as a non-root user and provides maximum isolation between the agent and sensitive information. Security recommendations under root/admin mode are ineffective as root has unrestricted access.

When Credential Protection is Required

Use the secure initialization approach when using:

• ✅ Object storage with access keys (S3, OSS, Azure Blob, GCS, etc.)
• ✅ Databases with passwords (Redis, MySQL, PostgreSQL with auth)
• ✅ Any configuration containing sensitive information

NOT required for:

• ❌ Local storage (

  --storage file

  ) + SQLite3 (no password)
• ❌ Unauthenticated metadata engines

Secure Initialization Process

Instead of directly running

juicefs format

juicefs mount

IMPORTANT: The initialization script MUST be run with root/administrator privileges (sudo)

Why root is required:

• To install shc (Shell Script Compiler) if not present
• To compile scripts into secure binaries
• To set proper ownership and permissions
• To ensure AI agent user cannot access credentials

Run the initialization script:

# MUST run as root/admin
sudo ./scripts/juicefs-init.sh
# Script will prompt for AI agent username

Re-running the script: The script is designed to be re-runnable and will:

• Detect and prompt before overwriting existing binary
• Check if filesystem already exists (skip formatting if so)
• Allow you to update configuration without reformatting

This interactive script will:

• Prompt for AI agent username
• Prompt for all sensitive configuration (AK/SK, passwords, URLs)
• Install shc (Shell Script Compiler) if not present
• Format the filesystem if needed
• Generate wrapper script with embedded credentials
• Compile wrapper into binary using shc
• Name binary after filesystem for easy identification
• Verify binary functionality
• Clean up intermediate files (wrapper script, C source)
• Set proper permissions and ownership (root:AI_AGENT_USER group, 750)

Generated binary (in

juicefs-scripts/

• <filesystem-name>

  - Compiled binary wrapper (e.g.,

  prod-data

  )

The binary:

• Contains embedded credentials (compiled into binary format, obfuscated)
• Accepts any JuiceFS command and parameters
• Named after filesystem for easy identification and management
• One filesystem = one binary program
• Owned by root, executable by AI agent user

AI Agent usage:

# Switch to AI agent user first
su - aiagent

# Show available commands
./juicefs-scripts/myfs

# Mount filesystem
./juicefs-scripts/myfs mount /mnt/jfs

# Mount with custom options
./juicefs-scripts/myfs mount --cache-size 204800 /mnt/jfs

# Check filesystem status
./juicefs-scripts/myfs status

# Show statistics
./juicefs-scripts/myfs stats /mnt/jfs

# Unmount filesystem
./juicefs-scripts/myfs umount /mnt/jfs

# Run benchmark
./juicefs-scripts/myfs bench /mnt/jfs

Example: Secure Setup Flow

Step 1: Admin initializes (one-time setup as root):

cd /path/to/juicefs-skill
sudo ./scripts/juicefs-init.sh
# Enter AI agent username: aiagent
# Follow prompts to enter:
# - Filesystem name: prod-data
# - Redis: localhost:6379 with password
# - S3 bucket + AWS credentials
# - Compression option (for format)
# Binary compiled: juicefs-scripts/prod-data
# Owned by root, executable by aiagent

Step 2: AI agent user uses the binary:

# Switch to AI agent user
su - aiagent

# Mount filesystem (specify mount point at runtime)
./juicefs-scripts/prod-data mount /mnt/jfs

# Check status
./juicefs-scripts/prod-data status

# Work with filesystem
ls /mnt/jfs
cp data.csv /mnt/jfs/

# Unmount when done
./juicefs-scripts/prod-data umount /mnt/jfs

For AI Agents: Detection and Guidance

When a user asks to format or mount JuiceFS:

Check if using sensitive resources:

# User mentions: S3, OSS, Azure, GCS, Redis password, MySQL password, etc.

Provide guidance:

⚠️  SECURITY WARNING: This operation requires sensitive credentials.

To prevent credential exposure to the AI model:

1. Please run the initialization script yourself (MUST use sudo):
   sudo ./scripts/juicefs-init.sh
   # Script will prompt for AI agent username

2. Follow the prompts to configure your filesystem
   - The script will install shc if needed
   - Script will compile credentials into a binary using shc
   - Binary will be named after your filesystem
   - Binary owned by root, executable by AI agent

3. Once complete, I can help you use the generated binary:
   - Show commands: ./juicefs-scripts/<name>
   - Mount: ./juicefs-scripts/<name> mount <mountpoint>
   - Status: ./juicefs-scripts/<name> status
   - Unmount: ./juicefs-scripts/<name> umount <mountpoint>

This keeps your AK/SK and passwords secure from the AI model.
The binary contains compiled credentials that cannot be read with simple commands.

Note: Root privileges are required for shc installation, binary compilation,
and setting proper ownership/permissions.

Insecure Setup (Local Development Only)

For local development without sensitive data:

# This is safe for AI agents - no credentials involved
juicefs format \
    --storage file \
    --bucket /tmp/jfs-storage \
    sqlite3:///tmp/jfs.db \
    dev-fs

juicefs mount sqlite3:///tmp/jfs.db /mnt/jfs-dev

Essential Commands

1. Format a File System

Create a new JuiceFS file system:

# Basic format with Redis and S3
juicefs format \
    --storage s3 \
    --bucket https://mybucket.s3.amazonaws.com \
    redis://localhost:6379/1 \
    my-juicefs

# With compression
juicefs format \
    --storage s3 \
    --bucket https://mybucket.s3.amazonaws.com \
    --compress lz4 \
    redis://localhost:6379/1 \
    my-juicefs

# Local development with SQLite
juicefs format \
    --storage file \
    --bucket /data/storage \
    sqlite3://myjfs.db \
    dev-fs

2. Mount a File System

# Basic mount
juicefs mount redis://localhost:6379/1 /mnt/jfs

# Production mount with cache optimization
juicefs mount \
    --cache-dir /ssd/cache \
    --cache-size 204800 \
    --writeback \
    -d \
    redis://localhost:6379/1 \
    /mnt/jfs

# Mount with prefetch for read-heavy workloads
juicefs mount \
    --cache-dir /nvme/cache \
    --cache-size 409600 \
    --prefetch 3 \
    redis://localhost:6379/1 \
    /mnt/jfs

Key Mount Options:

• --cache-dir

  : Cache directory (default:

  ~/.juicefs/cache

  )

• --cache-size

  : Cache size in MiB (default: 102400 = 100GB)

• --writeback

  : Enable write-back cache for better write performance

• --prefetch N

  : Enable read prefetch with N threads

• --buffer-size

  : Read buffer size in MiB (default: 300)

• -d

  : Run in background (daemon mode)

3. Unmount

# Graceful unmount
juicefs umount /mnt/jfs

# Force unmount
juicefs umount -f /mnt/jfs

4. Sync Data

# Sync local to JuiceFS
juicefs sync /local/path/ jfs://redis://localhost:6379/1/remote/path/

# Sync between JuiceFS file systems
juicefs sync jfs://redis://localhost:6379/1/src/ jfs://redis://localhost:6379/2/dst/

# Sync from S3 to JuiceFS
juicefs sync s3://bucket/path/ /mnt/jfs/path/

# Dry run
juicefs sync --dry-run /source/ /dest/

5. Status and Monitoring

# Show file system status
juicefs status redis://localhost:6379/1

# Real-time statistics
juicefs stats /mnt/jfs

# Profile operations
juicefs profile /mnt/jfs

# Benchmark
juicefs bench /mnt/jfs

6. Configuration

# View configuration
juicefs config redis://localhost:6379/1

# Set trash retention
juicefs config redis://localhost:6379/1 --trash-days 7

# Set capacity quota
juicefs config redis://localhost:6379/1 --capacity 1048576

7. Maintenance

# Garbage collection (dry run first)
juicefs gc redis://localhost:6379/1 --dry

# Actual garbage collection
juicefs gc redis://localhost:6379/1

# Dump metadata for backup
juicefs dump redis://localhost:6379/1 backup.json

# Load metadata from backup
juicefs load redis://localhost:6379/1 backup.json

8. S3 Gateway

# Start S3-compatible gateway
export MINIO_ROOT_USER=admin
export MINIO_ROOT_PASSWORD=12345678
juicefs gateway redis://localhost:6379/1 localhost:9000

Configuration by Workload

Big Data Processing (Hadoop/Spark)

juicefs mount \
    --cache-dir /ssd/cache \
    --cache-size 204800 \
    --writeback \
    redis://redis:6379/1 \
    /mnt/jfs

Machine Learning Training

juicefs mount \
    --cache-dir /nvme/cache \
    --cache-size 409600 \
    --prefetch 3 \
    --buffer-size 600 \
    redis://redis:6379/1 \
    /mnt/ml-data

Shared Development Environment

juicefs mount \
    --cache-size 102400 \
    redis://redis:6379/1 \
    /mnt/shared

Backup/Archive (Write-heavy)

juicefs mount \
    --writeback \
    --buffer-size 600 \
    redis://redis:6379/1 \
    /mnt/backup

Kubernetes Integration

Basic PersistentVolume

apiVersion: v1
kind: PersistentVolume
metadata:
  name: juicefs-pv
spec:
  capacity:
    storage: 10Pi
  volumeMode: Filesystem
  accessModes:
    - ReadWriteMany
  persistentVolumeReclaimPolicy: Retain
  csi:
    driver: csi.juicefs.com
    volumeHandle: juicefs-volume
    fsType: juicefs
    nodePublishSecretRef:
      name: juicefs-secret
      namespace: default

Troubleshooting

Mount Fails

1. Check metadata engine:

   # For Redis
   redis-cli -h localhost -p 6379 ping
   

2. Check credentials: Verify access keys for object storage

3. Check logs:

   tail -f /var/log/juicefs.log
   

Slow Performance

1. Check cache hit rate:

   juicefs stats /mnt/jfs
   

2. Increase cache:

   juicefs umount /mnt/jfs
   juicefs mount --cache-size 204800 redis://localhost:6379/1 /mnt/jfs
   

3. Enable prefetch for sequential reads:

   juicefs mount --prefetch 3 redis://localhost:6379/1 /mnt/jfs
   

No Space Left on Device

1. Clean cache:

   rm -rf ~/.juicefs/cache/*
   

2. Increase free-space-ratio:

   juicefs mount --free-space-ratio 0.2 redis://localhost:6379/1 /mnt/jfs
   

Common Patterns

Production Setup with HA

# Format with Redis Sentinel
juicefs format \
    --storage s3 \
    --bucket https://prod-bucket.s3.amazonaws.com \
    redis://sentinel1:26379,sentinel2:26379,sentinel3:26379/mymaster/1 \
    prod-fs

# Mount with optimized settings
juicefs mount \
    --cache-dir /ssd/cache \
    --cache-size 204800 \
    --writeback \
    -d \
    redis://sentinel1:26379,sentinel2:26379,sentinel3:26379/mymaster/1 \
    /mnt/jfs

Development Setup

# Format with SQLite (local)
juicefs format \
    --storage file \
    --bucket /tmp/jfs-storage \
    sqlite3:///tmp/jfs.db \
    dev-fs

# Mount
juicefs mount sqlite3:///tmp/jfs.db /mnt/jfs-dev

Data Migration

# Step 1: Mount source and destination
juicefs mount redis://source:6379/1 /mnt/source
juicefs mount redis://dest:6379/1 /mnt/dest

# Step 2: Sync data
juicefs sync /mnt/source/ /mnt/dest/

# Or use juicefs sync directly
juicefs sync jfs://redis://source:6379/1/ jfs://redis://dest:6379/1/

Performance Tuning Quick Guide

--prefetch 3

--writeback --buffer-size 600

--prefetch 3 --cache-size 409600

--prefetch 1

Security Best Practices

1. 🔒 Protect credentials in AI agent environments:

   • Use

     ./scripts/juicefs-init.sh

     to create compiled binary with embedded credentials
   • The script uses shc (Shell Script Compiler) to protect sensitive information
   • Binary is named after filesystem for easy management
   • Credentials are compiled into binary format (obfuscated by shc)
   • This prevents AI models from easily accessing AK/SK, passwords, and sensitive URLs
   • See the "Security: Protecting Sensitive Credentials" section above for details

2. Enable encryption:

   juicefs format --encrypt-secret redis://localhost:6379/1 secure-fs
   

3. Use TLS for metadata engine: Connect via

   rediss://

   instead of

   redis://

4. Use HTTPS for object storage: Always use HTTPS endpoints

5. IAM roles: Use IAM roles instead of static access keys when possible

6. Network isolation: Use VPC/private networks for production

Advanced Security Recommendations

For production environments requiring maximum security:

1. Secret Management Services:

• AWS Secrets Manager / Parameter Store
• HashiCorp Vault
• Azure Key Vault
• Benefits: Centralized rotation, auditing, time-limited access

2. IAM-Based Authentication:

• AWS: Use IAM roles with EC2 instance profiles
• Azure: Use Managed Identity
• GCP: Use Workload Identity
• Benefits: No static credentials, automatic rotation

3. Certificate-Based Authentication:

• Use TLS client certificates for Redis/databases
• Benefits: No passwords to protect, automatic validation

4. Configuration File Encryption:

• age (modern encryption tool)
• SOPS (Secrets OPerationS)
• Benefits: Version-controllable configs, separate key management

See scripts/SECURITY_MODEL.md for detailed implementation guidance.

Environment Variables

The initialization script does NOT export sensitive environment variables. Instead, credentials are compiled into secure binaries.

For reference, JuiceFS supports these environment variables:

# Custom cache (✓ Safe - no credentials)
export JUICEFS_CACHE_DIR=/ssd/cache

# Debug logging (✓ Safe - no credentials)
export JUICEFS_LOGLEVEL=debug

# AWS credentials (⚠️ NOT RECOMMENDED - exposes to AI agent)
# export AWS_ACCESS_KEY_ID=your-key
# export AWS_SECRET_ACCESS_KEY=your-secret

# Redis password (⚠️ NOT RECOMMENDED - exposes to AI agent)
# export REDIS_PASSWORD=your-password

Recommended approach: Use the initialization script which compiles credentials into binaries rather than using environment variables.

Quick Decision Trees

Choosing a Metadata Engine

• Redis: Fast, production-ready, supports HA (Sentinel/Cluster)
• MySQL/PostgreSQL: Already have infrastructure, need SQL features
• TiKV: Large scale, need horizontal scalability
• SQLite: Development, testing, single node
• etcd: Small to medium scale, already using etcd

Choosing Cache Size

• Working set < 100GB: 100GB cache (102400 MiB)
• Working set 100-500GB: 200-400GB cache
• Working set > 500GB: 400GB+ cache
• Rule of thumb: 10-20% of working set size

References

For detailed information, see the references:

• Comprehensive Reference - Complete JuiceFS documentation
• Quick Start Guide - Task patterns and troubleshooting flowcharts
• Table of Contents - Index of all topics

Resources

• Official Documentation: https://juicefs.com/docs/community/introduction
• GitHub Repository: https://github.com/juicedata/juicefs
• Quick Start: https://juicefs.com/docs/community/quick_start_guide
• Command Reference: https://juicefs.com/docs/community/command_reference
• Community: https://github.com/juicedata/juicefs/discussions

Installation

# Linux AMD64
curl -sSL https://d.juicefs.com/install | sh -

# macOS (Homebrew)
brew install juicefs

# Docker
docker pull juicedata/juicefs

Tips for AI Agents

1. Always check metadata engine connectivity first
2. Cache is critical - allocate sufficient space on fast storage
3. Use

   --writeback

   for write-heavy,

   --prefetch

   for read-heavy workloads
4. Monitor with

   juicefs stats

   regularly
5. Test with

   juicefs bench

   before production
6. Plan for metadata engine HA in production
7. Use compression (

   --compress lz4

   ) to reduce costs
8. Enable trash (

   --trash-days 7

   ) for safety
9. Run

   juicefs gc

   regularly
10. Keep JuiceFS client updated