Do we need so many Sentry transactions?

[leplatrem]

In the week of 15 to 22nd of August, the hits on the __lbheartbeat__ endpoint led to 2.6 millions Sentry transactions.

Do we need them? Do they cost anything? Should we disable performance reports?

[grahamalama]

What did things look like before then?

We recently upgraded Sentry, and part of that change involved adding an environment variable to control the traces_sample_rate. Maybe we can set this to a lower number in production? That's what Sentry recommends, anyway. (related #212).

[grahamalama]

Another option is to adopt a dynamic sampling approach, as described here and suggested by @mostlygeek here.

This way, we could reduce the sampling of __lbheartbeat__, or remove it from sampling entirely