Review offline support

[bruno-garcia]

A customer who has a MAUI app that goes offline for extended amount of time has reported that events are getting lost.
No client reports indicate data being dropped.

I expected at least:

sentry-dotnet/src/Sentry/Internal/Http/CachingTransport.cs

Line 366 in cb67d21

_options.ClientReportRecorder.RecordDiscardedEvents(DiscardReason.CacheOverflow, envelope);

Reading the code it seems like we don't delete files if the failure reason is network connectivity:

sentry-dotnet/src/Sentry/Internal/Http/CachingTransport.cs

Lines 349 to 357 in cb67d21

	catch (Exception ex) when (IsNetworkError(ex))
	{
	if (_options.NetworkStatusListener is PollingNetworkStatusListener pollingListener)
	{
	pollingListener.Online = false;
	}
	_options.LogError(ex, "Failed to send cached envelope: {0}, retrying after a delay.", file);
	// Let the worker catch, log, wait a bit and retry.
	throw;

But I believe we should run some tests:

1. Put the simulator offline. Generate some events. Let the buffer fill (30 by default), do we get the client reports?
2. If the failure is a timeout, do we delete the issue (aka: review the heuristics for "Network Error")

[bricefriha]

@bruno-garcia, did the user tell you the platforms are affected? (I assume any of them)

[bruno-garcia]

@bruno-garcia, did the user tell you the platforms are affected? (I assume any of them)

Happens on Android/iOS, but offline caching happens in C# so I imagine testing this on Windows/macOS should yield the same result and easier on the devloop

[bruno-garcia]

Confirmed this is known to happen on iOS. Android doesn't have enough usage yet to be sure.
@bricefriha any luck with investigating this?

[bruno-garcia]

While chatting with the customer, an idea to keep track of the count of errors Sentry is capturing, from outside Sentry itself:

// Some field to keep the counter:
int _eventCounter = 0;

// in the Init code of Sentry:
options.SetBeforeSend((evt, hint) => {
  Interlocked.Increment(ref _eventCounter);
});

// At the point where we'd like to report this to our backed:

async Task ReportData() {
// Reset the counter to 0
var lastCountValue = Interlocked.Exchange(ref _eventCounter, 0);

// Send the current count to our backend so we can check against their reports, and againt what's in Sentry
await _apiService.PostAddErrorCount(CurrentUser, lastCountValue);

[bricefriha]

any luck with investigating this?

I tried using Android but couldn't reproduce this. I'll try using iOS with the error count you mentioned above.

Is it possible that it'd related to NativeAOT?

[jamescrosswell]

It might happen in this instance:

sentry-dotnet/src/Sentry/Internal/Http/CachingTransport.cs

Lines 227 to 228 in 7da3cf3

	_options.LogError(ex,
	"Failed to move unprocessed file back to cache (attempt {0}, done.): {1}", attempt, filePath);

Something should show up in the debug logs if that's the case though.

Or alternatively, as Bruno suggested, maybe here:

sentry-dotnet/src/Sentry/Internal/Http/CachingTransport.cs

Lines 366 to 367 in 7da3cf3

	_options.ClientReportRecorder.RecordDiscardedEvents(DiscardReason.CacheOverflow, envelope);
	LogFailureWithDiscard(file, ex);

But there should be client reports in that case.