Skip to content

README.md

Latest commit

 

History

History
292 lines (222 loc) · 9.96 KB

README.md

File metadata and controls

292 lines (222 loc) · 9.96 KB

AgeBot: Towards Efficient Detection of Aging-Related Bugs in Android Apps

AgeBot is an automated black-box testing framework for detecting aging-related bugs in Android applications using resource-aware reinforcement learning and other strategies.

📋 Table of Contents

✨ Features

  • Multiple Testing Strategies: Supports various algorithms for aging related bugs detection:

    • Resource-aware Reinforcement Learning
    • State Coverage-based Reinforcement Learning
    • Random Exploration
    • Ablation experiment variants

  • Comprehensive Resource Monitoring: Tracks various Android system resources including:

    • Views, ViewRootImpl, AppContexts, Activities
    • Assets, AssetManagers
    • Local/Proxy Binders
    • Memory (Java Heap, Native Heap, and RSS)
    • File Descriptors, Database Connections
    • Wake Locks, Sockets
    • CPU

  • Automated Setup: Automatic APK installation, permission handling, and login automation

  • Multi-Device Parallel Testing: Support for parallel testing across multiple devices

🏗️ Architecture

AgeBot/
├── agent/                    # Agent implementations
│   ├── Q.py                 # State coverage Q-Learning agent (supports multiple exploration modes)
│   ├── Random.py            # Random exploration agent
│   ├── ResourceQLearning.py # Resource-sensitive Q-Learning agent
│   └── global_data.py       # Global configuration and state variables
├── env/                      # Testing environment implementations
│   ├── aut_env.py           # Android application testing environment (Gymnasium-based)
│   └── mock_env.py          # Mock environment for testing
├── apk/                      # APK management and analysis tools
│   ├── apk.py               # APK installation and component extraction
│   └── apktool/             # APK decompilation tool
├── resources/                # Resource monitoring utilities
│   └── resource.py          # Resource data collection and analysis
├── tools/                    # Experiment analysis tools
│   ├── RQ1_effectiveness.py # RQ1: Effectiveness analysis
│   ├── RQ2_efficiency.py    # RQ2: Efficiency analysis
│   ├── RQ3_precision.py     # RQ3: Precision analysis
│   ├── RQ4_ablation.py      # RQ4: Ablation experiment analysis
│   └── RQ5_characterization.py # RQ5: Characterization analysis
├── result/                   # Test results storage directory
├── logs/                     # Log storage directory
├── main.py                   # Main entry point (CLI interface)
├── utils.py                  # Utility functions and logging system
├── appium_manager.py         # Appium process manager
└── requirement.txt           # Python dependencies

📦 Prerequisites

  • Python: 3.7 or higher
  • Android Debug Bridge (ADB): Installed and accessible in system PATH
  • Appium Server: Version 2.0+
  • Node.js: Required for running Appium
  • Android Devices/Emulators:
    • Rooted devices or Android 7.0+ emulators
    • USB Debugging enabled

🚀 Installation

1. Clone the Repository

git clone https://github.com/GIST-NJU/AgeBot.git
cd AgeBot

2. Install Python Dependencies

pip install -r requirement.txt

3. Install Android SDK and ADB

Ensure ADB is in your system PATH and accessible from command line:

adb version

4. Install and Configure Appium

npm install -g appium
appium driver install uiautomator2

5. Prepare Android Devices

  • Enable USB Debugging on your Android device(s)
  • For emulators, use Android Virtual Devices (AVD)

6. Prepare Test APKs

Place the APK files to be tested in the apk/ directory.

⚙️ Configuration

1. Configure Test Applications

Edit agent/global_data.py to specify the applications you want to test:

apps = {
    "app_name": "apk/path/to/app.apk",
    # Add more applications here
}
N = 1  # Number of parallel devices

2. Configure Appium Ports

  • Single Device Mode: Default port 4723
  • Multi-Device Mode: Ports 4723, 4725, 4727, ... (incremental ports)

Configure ports in agent/global_data.py:

ports = [4723, 4725, 4727]  # Configure based on number of devices

3. Optional: Configure WeChat Notifications

Create a config.json file in the project root:

{
    "wechat_sendkey": "your_sendkey_here",
    "other_settings": {
        "debug": false,
        "max_retries": 3
    }
}

Or set environment variable:

export WECHAT_SENDKEY="your_sendkey_here"

🎯 Usage

Basic Usage

# Test a single application with default algorithm
python main.py --app newpipe --algo q_res --duration 3600

# Test with all algorithms
python main.py --app newpipe --algo all --duration 3600

# Specify multiple algorithms
python main.py --app newpipe --algo random,q_res --duration 3600

Command-line Arguments

Argument Description Default
--app Application name to test (must be defined in global_data.py) Required
--algo Testing algorithm (use all to run all algorithms) q_res
--duration Duration of each test run in seconds 3600
--rounds Number of rounds per algorithm 1
--device Specify device name Auto-detected
--port Appium port 4723

Single Device Mode

  1. Connect one rooted Android device or start one emulator
  2. Run the test:
python main.py --app newpipe --algo q_res_v2 --duration 3600

Multi-Device Parallel Mode

  1. Connect multiple devices or start multiple emulators
  2. Configure global_data.py: 
    N = 3  # Number of devices
ports = [4723, 4725, 4727]
  3. Run parallel tests:
python main.py --app all --algo all --duration 3600 --rounds 3

🧮 Supported Algorithms

Core Algorithms

Algorithm Description
random Random exploration (baseline comparison)
q_res_v2 Resource-aware Q-Learning
q_cov State coverage Q-Learning (maximize state space exploration)

Ablation Experiment Algorithms

Algorithm Description
agebot_r Removes smoothing, normalization and scaling from resource sensitivity reward
agebot_b Removes bug reward component
agebot_e Fixed ε-greedy (no decay)

📊 Analysis Tools

The tools/ directory contains experiment analysis scripts:

File Function
RQ1_effectiveness.py Analyze effectiveness of each algorithm in detecting resource leaks
RQ2_efficiency.py Analyze testing efficiency (time, state coverage rate)
RQ3_precision.py Analyze detection precision and false positive rate
RQ4_ablation.py Ablation experiment results analysis
RQ5_characterization.py Resource leak characterization analysis

📱 Tested Applications

The framework has been tested on 26 real-world open-source Android applications. The table below shows the details of each application under test.

Name Category Ver. #Ins. #Stars #LOC #Act.
Squircle CE Development 2025.1.3 10K+ 1.6K+ 192,527 1
Easter Eggs Development 3.3.0 50K+ 979 32,771 44
Gadgetbridge Development 0.84.0 10K+ 4.4K+ 221,736 87
SelfPrivacy Multimedia 0.12.2 100+ 12 6,043 2
Souvenirs Multimedia 2.9.5 500+ 14 9,206 4
Tuner Multimedia 1.56 1M+ 186 9,502 3
Recurring Expense Tracker Finance Manager 0.19.5 1K+ 326 24,601 4
Money Manager Ex Finance Manager 1.8.1 10K+ 1.9K+ 41,537 41
oinkoin Finance Manager 1.2.1 1K+ 328 39,893 2
Butterfly Note 2.2.3 5K+ 1.5K+ 6,043 5
Saber Note 1.29.5 10K+ 4K+ 71,434 3
NeuroLab Sports & Health 1.0.0 10K+ 163 8,284 15
Broccoli Sports & Health 1.4.3 1K+ 3.3K+ 9,452 6
Food You Sports & Health 3.1.0 3.3K+ 185 78,894 4
Praktika Vortaro Translation & Dictionary 0.26 5K+ 8 3,503 5
PReVo Translation & Dictionary 0.27 10K+ 36 4,365 5
Spell4Wiki Translation & Dictionary 3.3 10K+ 27 53,587 16
News Reader News 1.0.10 10K+ 21 7,358 7
AntennaPod News 3.6.0 1M+ 7.3K+ 23,793 12
taz News 1.9.3 100K+ 44 43,892 13
bilimiao Internet 2.4.6 10K+ 2.5K+ 151,911 9
RedReader Internet 1.24.1 100K+ 2.1K+ 62,750 23
NewPipe Internet 0.27.4 5M+ 35.3K+ 57,363 12
Privacy Browser Browser 3.19.3 5K+ N/A 236,260 11
DuckDuckGo Browser 5.248.0 50M+ 4.4K+ 3,690 95
Firefox Browser 139.0.3 100M+ 10.2K+ 1,189,903 20
FairEmail Email 1.2310 1M+ 4.1K+ 1,322,763 23
K-9 Mail Email 9.0b3 5M+ 12.6K+ 162,720 26
Delta Chat Email 2.35.0 500K+ 1.5K+ 124,811 38

Note:

  • #Ins. and #Stars denote the number of installs on Google Play and the number of stars on GitHub, respectively.
  • #LOC and #Act. represent the number of code lines and Activities, respectively (K=1,000, M=1,000,000).
  • Applications marked as no need to login in global_data.py can be tested without login automation.

📧 Contact

For questions or support, please open an issue on GitHub.

Note: This tool is for research and testing purposes only. Always ensure you have permission to test applications and follow responsible disclosure practices when reporting bugs.