Introduction: Python framework for automatically executing measurement-based experiments on native and web apps running on Android devices
More: Author   ReportBugs   

Quality Gate Status Build Status Coverage Status

Android Runner (AR) is a tool for automatically executing measurement-based experiments on native and web apps running on Android devices.

The following scientific publication gives an overview about the main components, plugins, and configurations of Android Runner (as of 2020): A-Mobile 2020 publication

A complete tutorial on how to use Android Runner is available in the following YouTube playlist: Android Runner Tutorials (YouTube)

As visualized below, Android Runner consists of the following components:

  • Experiment orchestrator: Is in charge of executing the whole experiment according to the experiment configuration provided by the user.
  • Devices manager: Is responsible for providing a layer of abstraction on the low-level operations involving the Android devices.
  • Progress manager: Keeps track of the execution of each run of the experiment.
  • Plugin handler: Provides a set of facilities for managing the profilers and an extension point that third-party developers can use for integrating their own measurement tools into Android Runner.

Overview of Android Runner

How to cite Android Runner

If Android Runner is helping your research, consider to cite it as follows, thanks!

  title={{A Framework for the Automatic Execution of Measurement-based Experiments on Android Devices}},
  author={Ivano Malavolta and Eoin Martino Grua and Cheng-Yu Lam and Randy de Vries and Franky Tan and Eric Zielinski and Michael Peters and Luuk Kaandorp},
  booktitle={35th IEEE/ACM International Conference on Automated Software Engineering Workshops (ASEW '20)},
  url= {},

Setting up environment, installation and dependencies

Instructions can be found here. Instructions for specific plugins are included in the plugins' READMEs.

Quick start

To run an experiment, run:

python3 android-runner path_to_your_config.json

Example configuration files can be found in the subdirectories of the examples directory.



A JSON config that maps devices names to their adb ids for easy reference in config files.

Experiment Configuration

Below is a reference to the fields for the experiment configuration. It is not always updated.

adb_path string Path to adb. Normally don't need to include. Example path: /opt/platform-tools/adb

monkeyrunner_path string Path to Monkeyrunner. Example path: /opt/platform-tools/bin/monkeyrunner

type string Type of the experiment. Can be web, native or plugintest

devices_spec string Specify this property inside of your config to specify a devices.json outside of the Android Runner repository. For example:

   // ....
   "type": "native",
   "devices_spec": "/home/user/experiments/devices.json",
   "devices": {
     "nexus6p": {}
   // ...

repetitions positive integer Number of times each application is profiled in an experiment.

clear_cache boolean Clears the cache before every run for both web and native experiments. Default is false.

randomization boolean Random order of run execution. Default is false.

duration positive integer The duration of each run in milliseconds, default is 0. Setting a too short duration may lead to missing results when running native experiments, it is advised to set a higher duration time if unexpected results appear.

reset_adb_among_runs boolean Restarts the adb connection after each run. Default is false.

time_between_run positive integer The time that the framework waits between 2 successive experiment runs. Default is 0.

devices JSON A JSON object to describe the devices to be used and their arguments. Below are several examples:

  "devices": {
    "nexus6p": {
      "root_disable_charging": "True",
      "charging_disabled_value": 0,
      "usb_charging_disabled_file": "/sys/class/power_supply/usb/device/charge",
      "device_settings_reqs": {"e.www.gyroscopetest": ["location_high_accuracy", ...], ...}
  "devices": {
    "nexus6p": {
      "root_disable_charging": "False"
  "devices": {
    "nexus6p": {}

Note that the last two examples result in the same behaviour.

The root_disable_charging option specifies if the devices needs to be root charging disabled by writing the charging_disabled_value to the usb_charging_disabled_file. Different devices have different values for the charging_disabled_value and usb_charging_disabled_file, so be careful when using this feature. Also keep an eye out on the battery percentage when using this feature. If the battery dies when the charging is root disabled, it becomes impossible to charge the device via USB.

device_settings_reqs can be set to programmatically enable and disable settings on the test device for native apps. It was added to automate the process of turning on and off location services in a randomized experiment where some applications required it and others that didn't. Two options available currently: location services with the help of Google and one without. location_high_accuracy is the option for location services with Google; location_gps_only is the other. More adb commands are likely to be added in the future that work for other sensors and settings. Turn off location services before the experiment starts.

WARNING: Always check the battery settings of the device for the charging status of the device after using root disable charging. If the device isn't charging after the experiment is finished, reset the charging file yourself via adb su command line using:

adb su -c 'echo <charging enabled value> > <usb_charging_disabled_file>'

paths Array\ The paths to the APKs/URLs to test with. In case of the APKs, this is the path on the local file system.

apps Array\ The package names of the apps to test when the apps are already installed on the device. For example:

  "apps": [

browsers Array\ Dependent on type = web The names of browser(s) to use. Currently supported values are chrome, firefox and opera.

profilers JSON A JSON object to describe the profiler plugins to be used and their arguments. Below, an example is found:

  "profilers": {
    "trepn": {
      "sample_interval": 100,
      "data_points": ["battery_power", "mem_usage"]
    "android": {
      "sample_interval": 100,
      "data_points": ["cpu", "mem"],
      "subject_aggregation": "",
      "experiment_aggregation": ""

Currently, Android Runner contains the plugins listed below, they can immediately be used as for an experiment.

Name (quality attribute) Description
batterystats (Energy) Uses the batterystats utility and estimates energy consumption via the algorithm proposed in this article.
monsoon (Energy) Collects energy consumption via the Monsoon hardware profiler and the Physalia tool.
trepn (mixed) Collects data via the Trepn profiler, e.g., power consumption, battery temperature, CPUs frequency.
mem-CPU (Performance) Collects memory and CPU usage via the cpuinfo and meminfo Android utilities found in ADB's dumpsys.
frametimes (Performance) Collects frame rendering durations and the number of delayed frames with the technique used in this article.
gc (Performance) Collects the number of garbage collections as in this article.
perfume_js (Performance) Collects performance metrics using the Perfume.js library, e.g. FP, FCP, NavigationTiming, storageEstimate and networkInformation.

Did you develop a plugin for Android Runner? You can create a pull request in this repository and we will include it!

The profiler section may accept custom aggregation scripts. If a user specified aggregation script is used then the script should contain a bash main(dummy, data_dir, result_file) method, as this method is used as the entry point to the script. The aggregation options are as follows:

subject_aggregation string Specify which subject aggregation to use. The default is the subject aggregation provided by the profiler.

experiment_aggregation string Specify which experiment aggregation to use. The default is the experiment aggregation provided by the profiler.

scripts JSON A JSON list of types and paths of scripts to run. Below is an example:

"scripts": {
  "before_experiment": "",
  "before_run": "",
  "interaction": "",
  "after_run": "",
  "after_experiment": ""

Below are the supported types:

  • before_experiment executes once before the first run
  • before_run executes before every run
  • after_launch executes after the target app/website is launched, but before profiling starts
  • interaction executes between the start and end of a run
  • before_close executes before the target app/website is closed
  • after_run executes after a run completes
  • after_experiment executes once after the last run

Instead of a path to string it is also possible to provide a JSON object in the following form:

    "interaction": [
        "type": "python3",
        "path": "Scripts/",
        "timeout": 500,
        "logcat_regex": "<expr>"

Within the JSON object you can use "type" to "python3", "monkeyrunner" or, "monkeyreplay" depending on the type of script. "python3" can be used for a standard python script, "monkeyreplay" for running a Monkeyrunner script with the use of the Monkeyrunner framework and "monkeyrunner" can be used to run a Monkeyrunner directly without the entire Monkeyrunner framework. The "timeout" option is to set a maximum run time in miliseconds for the specified script. The optional option "logcat_regex" filters the logcat messages such that it only keeps lines where the log message matches "\" where "\" is a regular expression.

Plugin profilers

It is possible to write your own profiler and use this with Android Runner. To do so write your profiler in such a way that it uses this class as parent class. The device object that is mentioned within the class is based on the of this repo. To see what can be done with this object, see the source code here.

You can use your own profiler in the same way as the default profilers, you just need to make sure that:

  • The profiler name is the same as your python file and class name.
  • Your python file isn't called '' as this file will be overwritten.
  • The python file is placed in its own directory inside the directory called 'Plugins'; the name of the directory must be in lowercase

To test your own profiler, you can make use of the 'plugintest' experiment type which can be seen here

Experiment continuation

In case of an error or a user abort during experiment execution, it is possible to continue the experiment if desired. This is possible by using a --progress tag with the starting command. For example:

python3 android_runner your_config.json --progress path/to/progress.xml

Detailed documentation

The original thesis can be found here:

About Me
GitHub: Trinea
Facebook: Dev Tools