fix: Do not call before_send for transactions

[rhcarvalho]

This matches the behavior with JS and the specs.

The change in JS was introduced in getsentry/sentry-javascript#2600.
See notes under "Few other fixes".

[rhcarvalho]

Got a test failure in test_aws.py:

https://travis-ci.com/github/getsentry/sentry-python/jobs/352702298#L1612-L1616

Quick glance at the test implementation and I guess it could be flakey.

cc @untitaker

[rhcarvalho]

Rebased to fix merge conflict.

[vmarkovtsev]

Is there a standard way to filter out transactions now? I've got all my backend endpoint handler entry points wrapped in spans and need to silence a few unimportant ones to save $$$ on the events quota.

[rhcarvalho]

Is there a standard way to filter out transactions now? I've got all my backend endpoint handler entry points wrapped in spans and need to silence a few unimportant ones to save $$$ on the events quota.

@vmarkovtsev you can use sentry_sdk.scope.add_global_event_processor and pass in a func to silence transaction events (and in fact do anything before_send could do). Make sure to check for event.get("type") == "transaction".

This is a workaround. Keep in mind that manually sampling transactions will potentially skew the aggregate metrics you see in Sentry.

[rhcarvalho]

@vmarkovtsev also worth noting that event processors run right before the transaction is sent, which means that the sampling decision has already been propagated to other services for the life time of the transaction. It might or might not be of importance to you.

If you're doing manual instrumentation, you should also be able to override the sampling decision early on by starting a transaction with sampled=False.

We're working on a more general solution for controlling transaction sampling from SDKs.

[vmarkovtsev]

In my case, I am using aiohttp integration which automatically starts transactions for each incoming HTTP request. I am not sure how to set sampled=False then... But yeah, canceling the sampling whatsoever is better ofc.

Thanks for the workaround! It should work fine for me 👍