[EPIC] Android Profiling Improvements and Hardening

[markushi]

Description

This is a meta issue to collect reported issues around Android Profiling and how we can improve profiling overall.

Refer to the individual issues to learn more about the status.

Likely root cause: Third Party code

Beta Give feedback

1. invalid pthread_t 0x7893464cb0 passed to pthread_getcpuclockid #2604
   Platform: Android Platform: Native Sync: Jira Type: Bug +4
   

Likely root cause: Android Tracer

Beta Give feedback

1. pthread_join failed for sampling thread shutdown: Invalid argument #3561 -> https://issuetracker.google.com/issues/362293861
2. SIGSEGV when user interaction instrumentation is enabled #3653 -> https://issuetracker.google.com/issues/361129298

Likely root cause: Sentry SDK

Beta Give feedback

No tasks being tracked yet.

[stefanosiano]

I'm thinking about some workarounds that we can do, until it gets fixed (if it gets fixed in older Android versions)
We can add another flag to guard "dangerous" profiles, and here is some idea for such detection:

• With our Gradle plugin we can check the app code to see if there is any other call to Debug.startMethodTracingSampling and Debug.stopMethodTracing
• Hook into native implementation and workaround the crashing calls
• We can detect if a native library is called (via gradle plugin or at runtime if possible)
• Create a file to check crashes between profile start/stop and disable it on next run
• Cut out the AOSP profiler and bundle it inside our SDK, changing the abort with a warning

If you have any other workaround, feel free to add a comment with it
The best would be create a custom profiler, but that's a big effort thing

[kahest]

Update: the two issues that require fixes on ART side have been fixed by Google:

• pthread_join failed for sampling thread shutdown: Invalid argument #3561 -> https://issuetracker.google.com/issues/362293861
• SIGSEGV when user interaction instrumentation is enabled #3653 -> https://issuetracker.google.com/issues/361129298

We're still keeping the underlying GH issues open until we get some information/confirmation that the fixes are rolled out with Google System/Security upates.