Exclude optional performance instrumentation by default for AWS Serverless #11991
Comments
|
Thank you for raising this, this is a valid point. We will adjust this so that we do not include all integrations by default for serverless, meaning you'll have to manually add the DB integrations etc. you care about. |
…11998) In order to keep bundle size compact, performance integrations (except for http & fetch) have to be manually added for serverless packages. This means that users will have to do e.g. this if they want to have mysql instrumented: ```js import * as Sentry from '@sentry/aws-serverless'; Sentry.init({ integrations: [Sentry.mysqlIntegration()] }); ``` Closes #11991 --------- Co-authored-by: Lukas Stracke <lukas.stracke@sentry.io>
|
We just shipped v8.0.0, where we do not automatically add all auto Instrumentation anymore. This should help to reduce the default bundle size! |
|
Amazing, thank you for getting this in last minute ❤️ Separately, it would be nice to have a little more detail to the instrumentation customization guide. |
|
@mydea I just tried v8.0.0 🎊 ... and the problem still exists. The core of the issue is that they are "statically" imported and still on a valid code path: Here is the latest screenshot from esbuild output: So there is no difference, despite the changes you included in v8. Would it be possible to create a "node-core" package that is most of the current "node" package? The difference being, using
Then you would be required to add any additional if you wanted. |
|
So just the presence/exporting of certain things should normally not affect the bundle size. As long as the code is not used, it should be tree-shaken away. How do you build your lambda function, if I may ask? Maybe there is some config option missing to ensure tree shaking works as expected 🤔 EDIT: Ignore this, I've seen there is a new issue with newer info. |
|
Actually, I just noticed my "fix" did not really fix this. I will make sure to actually fix the bundle size, sorry about that... |
mydea
changed the title
mydea
changed the title
…11998) In order to keep bundle size compact, performance integrations (except for http & fetch) have to be manually added for serverless packages. This means that users will have to do e.g. this if they want to have mysql instrumented: ```js import * as Sentry from '@sentry/aws-serverless'; Sentry.init({ integrations: [Sentry.mysqlIntegration()] }); ``` Closes #11991 --------- Co-authored-by: Lukas Stracke <lukas.stracke@sentry.io>
|
This is resolved, thank you! ❤️ |
Problem Statement
Opentelemetry adds a massive amount of overhead. The screen shot below represents a minimized / treeshook lambda built with
esbuildthat has very little of its own code.I understand some of the challenges here (ie. nothing is free or magic), but Sentry + Opentelemetry is > 50% of the bundled lambda.
For opentelemetry specifically, it bundles a whole series of instrumentation hooks which will never be consumed.
Of the 16 bundled instrumentation hooks, we will only use three of them:
There is a guide which mentions "customizing" the opentelemetry configuration:
https://docs.sentry.io/platforms/javascript/guides/node/migration/v7-to-v8/v8-opentelemetry/#custom-opentelemetry-setup
But it seems incomplete. It is not clear how or where I would instantiate the three instrumentation providers.
Solution Brainstorm
It would be great if there was a complete example of an aws lambda which configures only three instrumentation providers:
This should be a "gold-standard" for how to configure sentry for aws-lambda, because even with full ESM + treeshaking + minification the size of Sentry + Opentelemetry is a substantial problem.
The text was updated successfully, but these errors were encountered: