Baseline Profiles and Curious Case of Permissions
Recently I came across an article https://android-developers.googleblog.com/2022/01/improving-app-performance-with-baseline.html with the headline having a mention of “App Performance” and with the subtitle “Or how to improve startup time by up to 40%” 😲. As I focus a lot on app performance this immediately piqued my interested and I thought about giving it a try.
Generally whenever app performance is concerned you always end up spending a decent amount of time in profiling the app using tools like Systrace, Android Studio Profiler or Perfetto to understand the issues/bottlenecks. But reading the article it seemed like I don’t need to spend any time at all with the profiling tools and just need to generate a so called Baseline Profile file and bundle it with the apk. Plus the cherry on the top was that this profiles can be generated and tested using Macrobenchmark with just a few lines of code 🤯.
NOTE: I highly suggest reading the above mentioned article before proceeding further as I will omit many of the details already present in it.
History
Above mentioned article also gave details about the history of how Android optimizes the runtime performance of the app and how the profiles are generated and distributed. Timeline might look something like below:
JIT (all API Levels) -> Fully Compiled (API 21 & 23) -> On Device Profiling (API 24+) -> Cloud Profiles (API 29+).
To understand the limitation of Cloud Profiles I’m taking an excerpt from the article:
While Cloud Profiles are great when they are available, they aren’t always ready to be used when an app is installed. Collecting and aggregating the profiles usually takes several days, which is a problem when many apps update on a weekly basis. Many users will install an update before the Cloud Profile is available.
Apart from the above mentioned limitation there is one more limitation is for the companies who can’t distribute their app via Play Store like my current company GetMega which is into RMG (Real Money Gaming). This was another reason which made me interested into trying out Baseline Profiles.
What are Baseline Profiles?
Baseline Profiles are a new mechanism to provide profiles which can be used on Android 7 (API level 24) and higher. A baseline profile is an ART profile generated by the Android Gradle plugin using a human readable profile format that can be provided by apps and libraries. Example might look like this:
You can read more about the rules and it’s syntax here https://developer.android.com/topic/performance/baselineprofiles#rule_syntax
Generating Baseline Profile
To generate the baseline profile I just followed the official doc https://developer.android.com/topic/performance/baselineprofiles. The process itself was very straightforward with only one thing to keep in mind is that it requires a rooted/userdebug device running Android 9 or higher. For this you can create an emulators without Google APIs, as you cannot use adb root on emulators with Google Play.
A very basic code to generate the baseline profile can be seen below:
Once the file is generated you just need to rename it to baseline-prof.txt and place it in the root directory of your main source set i.e. src/main or YOUR_BUILD_TYPE/main in case you’re having different source sets for different build types (in my case it was benchmark/main).
EDIT: After trying to do this on multiple devices I got similar permission issue as described in the Measuring the improvements section present below on both Oppo & OnePlus (Now owned by Oppo) where the issue got resolved with enabling similar kind of setting but under different name Disable Permissions Monitoring which needs to be toggled on as mentioned here https://stackoverflow.com/a/62431145
Measuring the improvements
Now since the Baseline Profiles are generated and placed in main source set, let’s get to the measuring the improvements that we got from including the baseline profile. The testing part is also very straightforward where you need to create a test using MacrobenchmarkRule with CompilationMode = CompilationMode.Partial(baselineProfileMode = BaselineProfileMode.Require). If you’ve already profiled your app’s startup time in the past then this should be very simple. Official sample code can be seen below.
After the rule was added I ran the test on my POCO F1 and it failed with the following error printed in the console.
Error logs pointed to android.permission.WRITE_EXTERNAL_STORAGE cannot be granted! line. To get more insights I ran the test again with — debug compiler option and I saw something interesting (Line 4 & 6 in the below gist). Even though I already read somewhere that the UI tests involves installing 2 apks. i.e. actual app under test and an app which is responsible for sending the commands to execute things in the actual app. But looking at the debug logs made it more clear. But there was no new info and the debug logs showed the same thing about failure in granting the WRITE_EXTERNAL_STORAGE permission.
After this I spent a lot of time in trying to understand what this means but had no success. Finally I did a google search using GrantPermissionCallable: Permission: android.permission.WRITE_EXTERNAL_STORAGE cannot be granted! and found the solution on https://stackoverflow.com/a/54809466. The solution was very simple where I just needed to enable USB debugging (Security settings), but coming from a world where I didn’t do any UI testing, this option was very new to me 😅.
After enabling the USB debugging (Security settings) option I ran the test again and the tests passed 🙌. The results looked very promising (TTID time reduction was ~30%) as I didn’t had to spent a lot a time into profiling the app in the first place.
Reference Material Links and Credits:
- https://android-developers.googleblog.com/2022/01/improving-app-performance-with-baseline.html
- https://developer.android.com/topic/performance/baselineprofiles
- https://stackoverflow.com/a/54809466
- https://medium.com/androiddevelopers/improving-performance-with-baseline-profiles-fdd0db0d8cc6
- https://github.com/android/nowinandroid
Hope you liked reading the article. Please feel free to reach me on Twitter or LinkedIn. Thank You!
