feat(core): Add streamGenAiSpans options to stream gen_ai spans#20785
Conversation
andreiborza
force-pushed
the
ab/gen-ai-streaming
branch
from
6299108 to
ec170d6
Compare
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
![sentry[bot]](https://avatars.githubusercontent.com/in/12637?s=60&v=4){kind=small}
size-limit report 📦
|
andreiborza
force-pushed
the
ab/gen-ai-streaming
branch
from
ec170d6 to
239fd1e
Compare
andreiborza
force-pushed
the
ab/gen-ai-streaming
branch
5 times, most recently
from
cf9559a to
5bf30b9
Compare
| const genAiSpans = []; | ||
| const remainingSpans = []; | ||
|
|
||
| for (const span of event.spans) { |
There was a problem hiding this comment.
q: Is this here only for the experimental phase? It seems quite inefficient to loop over all spans everytime.
There was a problem hiding this comment.
We do have a guard that pre-checks if a trace has gen_ai spans so this only happens when we really do need to do it. But yea, this will be gone in the future once we have proper span streaming everywhere.
|
|
||
| Adds a new experimental option `streamGenAiSpans` option that controls how `gen_ai` spans are | ||
| sent to Sentry. When set, the SDK extracts all `gen_ai` spans out of a | ||
| transaction and sends them as v2 envelope items. |
There was a problem hiding this comment.
q: Is it an important change if practically no one knows yet what a v2 envelope item is?
There was a problem hiding this comment.
I think so, the justification for it is in the next line:
This avoids payload size limits of regular transactions.
| sendDefaultPii: true, | ||
| // debug: true, | ||
| integrations: [Sentry.vercelAIIntegration()], | ||
| _experiments: { streamGenAiSpans: true }, |
There was a problem hiding this comment.
l: Is it mandatory to add this option in the test config, even though nothing in the test has changed?
There was a problem hiding this comment.
Good catch, some of these apps don't need this.
Adds a new experimental option `streamGenAiSpans` option that controls how
`gen_ai` spans are sent to Sentry. When set, the SDK extracts all `gen_ai` spans
out of a transaction and sends them as v2 envelope items.
This avoids payload size limits of regular transactions.
```ts
Sentry.init({
dsn: "https://examplePublicKey@o0.ingest.sentry.io/0",
_experiments: {
streamGenAiSpans: true,
},
});
```
Closes: #20170
andreiborza
force-pushed
the
ab/gen-ai-streaming
branch
from
5bf30b9 to
21aff86
Compare
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 623e6f1. Configure here.
andreiborza
force-pushed
the
ab/gen-ai-streaming
branch
from
e9beaa5 to
cc2e598
Compare
nicohrubec
left a comment
nicohrubec
left a comment
There was a problem hiding this comment.
lgtm. we already agreed on the approach before shipping the alpha and the rest is mainly test ports, which should hopefully be pretty much 1:1 ports to the span streaming format? maybe worth double checking on the strategy, but for me it sounds fine to ship this opt-in and then switch the default after a few weeks. thanks for doing this!
| sent to Sentry. When set, the SDK extracts all `gen_ai` spans out of a | ||
| transaction and sends them as v2 envelope items. | ||
|
|
||
| This avoids payload size limits of regular transactions. |
There was a problem hiding this comment.
l: maybe worth it to be a bit more explicit about this? i.e. that this should be enabled if orgs run into the problem of spans being dropped because they are too large
3 participants
We have concluded the experiment of streaming gen_ai spans as v2 envelope items via our 10.50.0-alpha.0 release and are now ready to pull this into stable gated behind a flag.
Adds a new experimental option
streamGenAiSpansoption that controls howgen_aispans are sent to Sentry. When set, the SDK extracts allgen_aispans out of a transaction and sends them as v2 envelope items.This avoids payload size limits of regular transactions.