[Question] Why does this work without calling `view.RegisterExporter` from the OpenCensus library?

[mattwelke]

[Question] Why does this work without calling `view.RegisterExporter` from the OpenCensus library?

[dashpole]

Unfortunately, the change that added the functions (and description), #121, doesn't have any details on why having Start and Stop functions is preferable to registering the exporter with a View. The authors of the change have moved on.

The stackdriver view Exporter does look a bit more complicated than the IntervalReader. The view Exporter has to aggregate multiple ExportView calls with a bundler, whereas the IntervalReader just exports on an interval.

[mattwelke]

I see. I don't know about IntervalReader or bundler or anything like that. I was just using OpenCensus according to its public documentation without worrying about how it's implemented. I was noticing this pattern of registering exporters to the views. I used it in production with Stackdriver and Prometheus exporters for a while.

I'm not sure if this is related but I noticed that when it came to overriding the reporting interval, if I did it on the view, it had no effect. I wanted to change it from its default of every 1 minute to every 5 minutes instead to save costs. I tried setting the reporting interval on the Stackdriver exporter and this did work. I reported this a while ago in census-instrumentation/opencensus-go#1153.

EDIT:

Actually, when I check that issue again, I notice that I wasn't able to reproduce my problem from my app in production. In a minimal example, setting the reporting interval while registering views did work. My takeaway from this is that there might be an issue with setting a reporting interval depending on how you happened to wire everything up, now that it's evident to me that there are multiple ways to wire this up and get data in Google Cloud Monitoring.

If you did into the issues linked to in that issue, you go down an interesting rabbit hole where they talk more about why view.RegisterExporter shouldn't be used, and they talk about knowing that a lot of users still do it this way (#193). I'm not sure what to make of it yet, but at least it means people encountering this can do some reading if they're feeling adventurous. :P