Baggage creation breaks external APIs

[raphaelm]

How do you use Sentry?

Self-hosted/on-premise

Version

1.10.1

Steps to Reproduce

I just spent literal 5 hours debugging a very weird issue in our production system. A specific API call to an external system started to return an error code 400.

After a long search, I could pinpoint the change to our recent upgrade of sentry-sdk, and specifically to this pull request: #1589

Starting from 1.9.8, Sentry will add a Baggage request to all external HTTP targets we call with requests.get() by default. I'm sure this is really helpful when calling internal systems that are also sentry-instrumented, but in our case we happen to call an external API that happens to break when it sees that header.

Our temporary solution is now to set propagate_traces=False, which is sad, because it also breaks propagation to useful targets (such as celery). We'd very much appreciate a more detailed way to configure this.

Expected Result

Sentry is not modifying requests to external APIs by default.

Actual Result

Sentry is adding a Baggage header to requests to external APIs even though we never asked for it.

[sl0thentr0py]

hey @raphaelm apologies for that, we'll add a trace_propagation_targets config to enable control over which origins we add outgoing headers to.
It's interesting that baggage broke your setup because we also always added sentry-trace before by default.

[raphaelm]

we'll add a trace_propagation_targets config to enable control over which origins we add outgoing headers to.

That'd be perfect!

It's interesting that baggage broke your setup because we also always added sentry-trace before by default.

Yeah… The target origin in our case is a German government entity, not sure what they're doing over there.

[antonpirker]

Related to this Issue #1785

[antonpirker]

I will close this issue in favor of the Issue that adds the solution to this problem to the SDK:
#1785

So @raphaelm please check the linked Issue for progress! Thanks!