Implement extension for AWS Lambda for automatic integration.

[marshall-lee]

This PR aims to improve the experience for users who want to integrate Sentry to their AWS Lambda functions.

For now, users have to upload @sentry/serverless package and its every dependency into a node_modules directory of their lambda function. It's especially inconvenient for people who just write the code in the AWS Console and don't have any external dependencies but want to enable error tracking & tracing.

Thanks to AWS, there Lambda Layers exist. We can bundle all the Sentry packages and distribute it as pluggable layer available by its ARN string. Added a layer to the lambda, one can then load the integration automatically when the lambda starts using a special environment variable NODE_OPTIONS=-r @sentry/serverless/dist/auto. Then, SENTRY_DSN environment variable must be set too and SENTRY_TRACES_SAMPLE_RATE is also available to enable tracing.

TODO:

• Write a publish script that pushes a zip bundle to every region of AWS (it creates a new version of a layer).
• Run this script on every release (???). Don't know how to do it really. cc @kamilogorek

Another problem I don't know what to do with is layer versions. Layers don't have anything like tags, e.g @v5.27.3. Layer versions are just auto-incrementing numbers, e.g. arn:aws:lambda:eu-central-1:123456789:layer:SentrySDK:1, arn:aws:lambda:eu-central-1:123456789:layer:SentrySDK:2, ...
So we need to figure out a way how to inform users which version is the latest. Maybe the serverless README.md must be somehow updated every time a new layer version is uploaded? Or the README.md should just contain a hyperlink to some page with the latest version numbers? I really don't know.

[lobsterkatie]

@marshall-lee - Thanks for this. Kamil's out until next week, but in the meantime, two things:

SENTRY_TRACES_SAMPLE_RATE is also available to enable tracing

How would you handle the tracesSampler option documented here?

As for running a script on every release, that's controlled by this file.

[marshall-lee]

@lobsterkatie

How would you handle the tracesSampler option documented here?

tracesSampler, beforeSend, beforeBreadcrumb and other options that require writing a js function, cannot be provided when using @sentry/serverless/dist/auto script. We can support more scalar environment variables of config but not the ones that require writing a code. This automated integration is intended for simple use cases where you don't need much Sentry customization. Only SENTRY_DSN variable is really required and you're gtg.

But even when not using @sentry/serverless/dist/auto, having a layer with a full Sentry SDK is still useful. Users still don't have to upload sdk for every function, they just link a layer to it and then require('@sentry/serverless') becomes available so they can manually call Sentry.init({ dsn: '...', tracesSampler: function() { ... } }); like they did before.

[lobsterkatie]

tracesSampler, beforeSend, beforeBreadcrumb and other options that require writing a js function, cannot be provided when using @sentry/serverless/dist/auto script. We can support more scalar environment variables of config but not the ones that require writing a code. This automated integration is intended for simple use cases where you don't need much Sentry customization. Only SENTRY_DSN variable is really required and you're gtg.

But even when not using @sentry/serverless/dist/auto, having a layer with a full Sentry SDK is still useful. Users still don't have to upload sdk for every function, they just link a layer to it and then require('@sentry/serverless') becomes available so they can manually call Sentry.init({ dsn: '...', tracesSampler: function() { ... } }); like they did before.

Ah, got it. Thanks for the explanation. Never mind, then! :-)

[ajjindal]

@HazAT could you take a look at this while Kamil is out? We are trying to get this out by end of the week.

Thanks @lobsterkatie

[HazAT]

Is this file empty on purpose?
packages/serverless/index.js

WRT publishing this, we need to upload this publically to our own AWS account to a specific layer?

[marshall-lee]

@HazAT

WRT publishing this, we need to upload this publically to our own AWS account to a specific layer?

Yep. There's also a script for this called packages/serverless/scripts/publish-awslambda-layer.js. It uploads a layer data to a bunch of AWS regions and makes it public.

Note that it must be uploaded to every region user might want because layer region and lambda function region must match.

[marshall-lee]

@kamilogorek you can re-check, i removed all questionnable stuff.

[scefali]

Closing because we merged a PR that implements this: #3165