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
• 🏗️ Architecture
• 📦 Prerequisites
• 🚀 Installation
• ⚙️ Configuration
• 🎯 Usage
• 🧮 Supported Algorithms
• 📊 Analysis Tools
• 📱 Tested Applications
• 📧 Contact

✨ 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.