ref(o11y): Support SDK configured with the streaming trace lifecycle

[alexander-alderman-webb]

Use the sentry_sdk.traces API when the streaming lifecycle is enabled. The transaction-based API should no-op and warns when the lifecycle is enabled (and does with an up-to-date sentry-sdk, the fact that it didn't previously no-op contributed to an incident).

Bump the minimum sentry-sdk version to 2.52.0 since that's the first version in which Scope. set_custom_sampling_context() is a classmethod.

Closes #761

[cleptric]

We should bump the minimum to the latest, so Span Streaming is fully available.

[untitaker]

this is a library, we should not arbitrarily restrict its version ranges unless actually required

[cleptric]

It is required, hence I raised it 😄

[alexander-alderman-webb]

Bumped 106634c / 6b478e6

[markstory]

Makes sense to me.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

There are 2 total unresolved issues (including 1 from previous review).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 213edcd. Configure here.}

[alexander-alderman-webb]

args/kwargs can be huge, does the sdk trim/stringify this efficiently?

No it doesn't on the streaming trace lifecycle path.
They eagerly go through repr so memory use may be a concern here (depending on the size of the list and dictionary). The serializer for transactions has various limits, but the new logic walks the whole list for args and dictionary for kwargs.

Would it be possible to remove the attributes? We could also put in itertools.islice() or similar to get the first 10 entries, since that's what effectively happened before.

[untitaker]

Would it be possible to remove the attributes? We could also put in itertools.islice() or similar to get the first 10 entries, since that's what effectively happened before.

😅 yeah that trimming is in the sdk for a reason.

i guess we could remove the attributes? kinda sad. some of them can be 40 MB large

[alexander-alderman-webb]

I've kept the transaction-based logic the same and do not attach the attributes in the streaming trace lifecycle now.

I'd have to take a closer look at what's in args and kwargs if we want to keep the attributes. If that'd be important let me know and I'll add some logic to walk the objects how we did with transaction data.

[untitaker]

well, i would not want to maintain this logic in taskbroker, so i think it's fine