Firebase performance - _app_start metric outliers, metric reports few minutes

[adamszeremeta]

Description

Hello,
we are using Firebase Performance SDK in our few apps. Sometimes _app_start metric reports outstanding app start duration in comparision with normal results (under 1s). This seems like a bug / situation we can't ourself understand.

Normal _app_start session metric looks like this:

where _app_start took 550ms and our custom metric to first view controller (App_Start_until_home_screen) took under 2s.

However we can also see sessions (few of them) which look like this:

In this case first gray trace is our custom metric (9 seconds) and _app_start has shown duration of 1032s.

Any idea what situation is this? How can we eliminate it? Seems to me like some kind of SDK bug.

Edit: Seems like this can be reported if app starts due to background fetch. I guess we have to disable then firebase performance? (but there is no really a good way to do this now). When app launches in background "didBecomeActive" is never called so metric is waiting for user app launch (if still alive) and reports that then.

Reproducing the issue

No response

Firebase SDK Version

FirebasePerformance (10.0.0)

Xcode Version

14

Installation Method

CocoaPods

Firebase Product(s)

Analytics, Crashlytics, In-App Messaging, Performance, Remote Config

Targeted Platforms

iOS

Relevant Log Output

No response

If using Swift Package Manager, the project's Package.resolved

Expand Package.resolved snippet

Replace this line with the contents of your Package.resolved.

If using CocoaPods, the project's Podfile.lock

Expand Podfile.lock snippet

- FirebasePerformance (10.0.0):
    - FirebaseCore (~> 10.0)
    - FirebaseInstallations (~> 10.0)
    - FirebaseRemoteConfig (~> 10.0)
    - GoogleDataTransport (~> 9.2)
    - GoogleUtilities/Environment (~> 7.8)
    - GoogleUtilities/ISASwizzler (~> 7.8)
    - GoogleUtilities/MethodSwizzler (~> 7.8)
    - nanopb (< 2.30910.0, >= 2.30908.0)

[google-oss-bot]

I couldn't figure out how to label this issue, so I've labeled it for a human to triage. Hang tight.

[rizafran]

Thanks for reaching out, @adamszeremeta. Before I proceed with reproducing the issue, could you provide your Podfile.lock/Podfile?

[adamszeremeta]

I can't give you directly Podfile as this is multi layer project. I can give you lock file, but I've redacted our internal private libraries.

Podfile_lock.txt

[rizafran]

Hi @adamszeremeta, could you confirm if the issue is happening to any iOS versions?

[adamszeremeta]

Yes, I saw it on multiple iOS 15 and 16 versions, for example:
16.1, 16.2, 16.3
15.5 15.6, 15.7,
14.8

[visumickey]

Firebase Performance currently eliminates app start traces that are involved in the context of app backgrounding Reference. So, ideally this situation should not occur. The duration of 1032s definitely sounds suspicious.

Do you what percentile of the app start traces are at a higher duration? Is is quite common or some rare app start duration measurements? Secondly, can you shed some light around the custom trace you created to measure the app start duration (Specifically - when did you start and end the trace metric)?