1009727 - Experiment install/recheck errors: `this._providerManager is null`

Status: RESOLVED WONTFIX

(Firefox Health Report Graveyard :: Client: Desktop, defect)

Description

[Benjamin Smedberg]

From the experiment monitoring dashboard, I'm seeing some exceptions:

TypeError: this._providerManager is null

This has something to do with us calling into FHR while checking the experiment, but I don't know the full stack. Georg do you know where this would be failing?

Comment 1

[Georg Fritzsche [:gfritzsche]]

Hm, i'm not sure how this can happen. We have two paths for FHR interaction:
* ExperimentsProvider.collectDailyData()
* fire "experiments-changed" -> ExperimentsProvider.observe() -> ExperimentsProvider.recordLastActiveExperiment()

I don't quite see yet where ExperimentsProvider.onShutdown() is triggered from, maybe we hit nulling out _providerManager [1] before it?

[1] http://hg.mozilla.org/mozilla-central/annotate/2f8af55d6e9a/services/healthreport/healthreporter.jsm#l645

Comment 2

[Georg Fritzsche [:gfritzsche]]

(In reply to Georg Fritzsche [:gfritzsche] [away May 17 - June 1] from comment #1)
> I don't quite see yet where ExperimentsProvider.onShutdown() is triggered
> from, maybe we hit nulling out _providerManager [1] before it?
> 
> [1]
> http://hg.mozilla.org/mozilla-central/annotate/2f8af55d6e9a/services/
> healthreport/healthreporter.jsm#l645

Richard, do you have an idea on whether it's possible to hit a null _providerManager before onShutdown() from a HealthReporter?

Comment 3

[Richard Newman [:rnewman]]

This kind of error usually means that an observer is firing after something has been cleaned up, but in that case the value is usually undefined, not null.

_providerManager *starts* as null, so you'll also get this error if you're poking at things prior to _initializeProviderManager being called.

That's called from a task after storage becomes available, so I can quite imagine that an experiment is running before FHR has finished initing the provider manager.

It's also possible that storage init is failing, so FHR hasn't been inited at all.

Comment 4

[Georg Fritzsche [:gfritzsche]]

FHR is going away per bug 1209088, wontfix.