Support JS symbolication from non-Sentry sources

[erickzhao]

My team was looking to deploy a self-hosted instance of the Symbolicator service to replace something we built internally with Mozilla's source-map library.

However, I saw that the /symbolicate-js docs seem to have a hard dependency on having a Sentry source (as opposed to native stack traces, which can fetch symbols from cloud storage). We were instead hoping of using source maps from our own S3 buckets without routing them through sentry.io.

Is my understanding of this limitation accurate, and would it be feasible to extend this service's capability to reading from S3 sourcemaps? With some pointers, I'd be willing to make a PR if this kind of change is amenable. 🙇

[linear]

SYMBOLI-37

[loewenheim]

Hi! Yes, unfortunately that limitation is exactly as you described. This would be a nice feature to have, but it does take a significant amount of work.

I'm assuming that we would only support sourcemaps with debug IDs. I think supporting any other lookup method is a nonstarter.

If we want to allow custom symbol sources we have to decide the structure in which sourcemaps should be stored there. The obvious choice here is the "unified" layout described in https://docs.sentry.io/platforms/apple/guides/ios/data-management/debug-files/symbol-servers/#directory-layouts. This layout currently doesn't interact with sourcemaps at all, but there's no reason it couldn't. The symsorter tool would also have to be adapted to be able to sort sourcemaps into this layout, but that's not too much of a problem. We'd have to do something similar for minified source files.

Then the debug file fetching logic would have to be extended to different source types (e.g. S3, HTTP, …), similarly to how it currently works for native. The current fetching logic would then just become the logic for Sentry sources.

We would welcome a contribution for this and I can assist with reviews, but it's not gonna be a quick fix.