feat(error-tracking): add debug_id snippet injection for cross-MFE stack trace un-minification by amortemousque · Pull Request #401 · DataDog/build-plugins

amortemousque · 2026-06-10T08:04:59Z

What and why?

Current unminification based on (service, version) breaks with MFEs.

When an error spans multiple micro frontends, its stack trace may contain frames from several bundles (for example, cart.js, header.js, and shell-app.js). Unminification today relies on a single (service, version) pair, typically derived from the originating MFE. The backend applies that pair to every frame in the stack trace. As a result, frames from other MFEs, which were uploaded under their own (service, version) pairs—cannot be matched to the correct source maps and remain minified.

debug_id solves this by attaching a UUID to each JS bundle. The backend can then resolves each frame independently.

How?

Add injection plugin capability to pass the chunk source or hash to hook callbacks.
- Expose a new injection callback: (sourceOrHash) => {}
- Move Rollup injection from banner/footer hooks to renderChunk so the plugin can access and modify the generated chunk source.
Implement debug_id injection in the RUM sourceCodeContext plugin.
- Subscribe to the new injection callback and generate a deterministic UUID from the file content
- Add ddDebugId: uuid to DD_SOURCE_CODE_CONTEXT entries.
Add end-to-end tests

datadog-datadog-prod-us1-2 · 2026-06-11T07:29:49Z

Tests

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

_{This comment will be updated automatically if new data arrives.

🔗 Commit SHA: 88eab93 | Docs | Datadog PR Page | Give us feedback!}

…n plugin to handle debugid generation

amortemousque · 2026-06-12T08:41:05Z

While implementing this, I took the opportunity to simplify the injection logic by:

processInjections + processItem becomes prepareInjections + resolveWithFallback where prepareInjections drives the loop and resolveWithFallback resolves async and file injections

processInjections iterated items one at a time with no-await-in-loop. Now prepareInjections resolves all injections in parallel.

Remove Map<string, ToInjectItem>: generated UUIDs were never used for lookup, replaced with ToInjectItem[] directly.

I think a good potential next step would be to replace the parameterized context.inject API with separate hooks per injection type.

amortemousque · 2026-06-12T08:45:09Z

+const isPerChunk = (
+    item: ToInjectItem,
+): item is ToInjectItem & { value: (chunk: ChunkInfo) => string } =>
+    typeof item.value === 'function' && item.value.length === 1;


I'm not a fan of this logic to detect (chunk) => ... per-chunk functions.
Improving this area would require a larger refactor, which I’d prefer to tackle in a dedicated PR. My plan is to replace the parameterized context.inject API with separate hooks for each injection type, making the implementation simpler and easier to reason about.

yoannmoinet · 2026-06-12T13:53:40Z

-    // Use MD5 hash (128 bits, 32 hex characters) for a compact identifier
-    // MD5 is sufficient for non-cryptographic purposes like creating unique identifiers
-    return createHash('md5').update(plainIdentifier).digest('hex');
+    // SHA-256 truncated to 128 bits (32 hex characters) for a compact identifier.
+    return createHash('sha256').update(plainIdentifier).digest('hex').slice(0, 32);


Was this change intentional?

Yes, intentional. MD5 was only used for deterministic IDs, but it is not supported in FIPS environments, which can cause builds to fail (similar to getsentry/sentry-javascript-bundler-plugins#618). SHA-256 avoids that issue.

Please notify apps folks about this.
EDIT: Uh oh, it seems I AM part of the apps folks... I'll forward this.

yoannmoinet · 2026-06-12T14:00:46Z

-    | { position?: InjectPosition.BEFORE | InjectPosition.AFTER; injectIntoAllChunks?: boolean }
+    | { position: InjectPosition.BEFORE | InjectPosition.AFTER; allChunks?: boolean }


This will need a major bump, IIRC we expose this API through the customPlugins.
Is it really necessary?

Hum, indeed. I can rollback this change to avoid breaking changes

yoannmoinet · 2026-06-12T15:47:34Z

-                // Skip if nothing to inject for this chunk type
-                if (!banner && !footer) {
-                    continue;
-                }


This may have a pretty intense impact on performances, as we will read every single bundle now.
I guess it's necessary for the new injection features, but I would double check the impact on performances still.

If only esbuild would offer us a way to hook ourselves right before writing the bundles.

Yes it's also one of my concern. I'll get some data on the overhead

I re-added a check so we inject only when contentsToInject contains an entry with injectIntoAllChunks: true.
This limits the overhead to cases where the customer uses a plugin that requires it.

yoannmoinet · 2026-06-12T15:50:13Z

 };

-export const getPlugins: GetPlugins = ({ options, context }) => {
+export const getPlugins: GetPlugins = ({ bundler, options, context }) => {


bundler doesn't seem to be used.

yoannmoinet · 2026-06-12T15:53:26Z

nit: This was fixed in rollup with MagicString, would it be a good occasion to do this here too?

Yes indeed! It's not exactly the same, as be need to rewrite the already generated .map but I handled them.

yoannmoinet · 2026-06-12T16:03:41Z

-    // Processing sequentially all the items.
-    for (const [id, item] of toInject.entries()) {


This was required to keep the order of injection the same post processing.

Making it fully parallel may introduce a behavior change worth noting.
This could also lead to indeterministic errors if we have cross injection dependencies.

We could keep the processing parallel, but the final injection should keep the same order to prevent any indeterministic behavior.

wdyt?

FMU, the order is already preserved. Promise.all returns results in input order regardless of resolution timing. Same guarantee as the sequential for...of. The only change is parallel execution. Am I missing something?

yoannmoinet · 2026-06-12T16:11:01Z

-                    // For Vite, we inject MIDDLE content by adding a script tag
-                    // that references the virtual injected file


Why remove the comment?

I don't know why but llms tend to remove them. I'll rewrite them

yoannmoinet · 2026-06-12T16:15:58Z

        "glob": "11.1.0",
        "json-stream-stringify": "3.1.6",
        "jszip": "3.10.1",
+        "magic-string": "0.30.21",


Now that it's not optional anymore (it seems) we could remove it from the peerDependencies.

bcaudan · 2026-06-15T09:26:41Z


 export type ErrorTrackingOptions = {
    enable?: boolean;
+    debugId?: boolean;


❓ question: ‏MFE was previously configured under rum. sourceCodeContext and we are now adding errorTracking.debugId.
Could we limit the number of options that the customers have to enable for MFE supports?

I collocated the options like to

type SourceCodeContext = { service?: string; version?: string; ddDebugId?: string; };

ddDebugId is new attribute within DD_SOURCE_CODE_CONTEXT entries

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

amortemousque requested a review from yoannmoinet as a code owner June 10, 2026 08:04

amortemousque marked this pull request as draft June 10, 2026 08:18

amortemousque force-pushed the aymeric/debug-id branch from cfaabe1 to b06c37e Compare June 11, 2026 19:25

Add debug_id snippet injection in error tracking plugin

566a9f1

amortemousque force-pushed the aymeric/debug-id branch from b06c37e to b431c67 Compare June 12, 2026 07:10

amortemousque marked this pull request as ready for review June 12, 2026 07:10

amortemousque requested review from a team as code owners June 12, 2026 07:10

amortemousque force-pushed the aymeric/debug-id branch from b431c67 to a661990 Compare June 12, 2026 07:14

Source code context and debug Id in a single snippet. Extend injectio…

c244d8d

…n plugin to handle debugid generation

amortemousque force-pushed the aymeric/debug-id branch from a661990 to c244d8d Compare June 12, 2026 07:42

amortemousque commented Jun 12, 2026

View reviewed changes

simplify

cb32c25

yoannmoinet reviewed Jun 12, 2026

View reviewed changes

bcaudan reviewed Jun 15, 2026

View reviewed changes

use same snippet for sourceCodeContext and debugId

f00a73f

gh-worker-ownership-write-b05516 Bot removed the request for review from a team June 16, 2026 15:11

amortemousque and others added 3 commits June 16, 2026 17:45

do not read file in esbuilt if no chunck injection

8087b7f

fix test

81e19b8

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Optimize esbuilt reading and writting files

b93045d

amortemousque force-pushed the aymeric/debug-id branch 2 times, most recently from b5e208b to 0d50a64 Compare June 17, 2026 12:52

fix package dependencies

88eab93

amortemousque force-pushed the aymeric/debug-id branch from 0d50a64 to 88eab93 Compare June 17, 2026 13:03

		\| { position?: InjectPosition.BEFORE \| InjectPosition.AFTER; injectIntoAllChunks?: boolean }
		\| { position: InjectPosition.BEFORE \| InjectPosition.AFTER; allChunks?: boolean }

		// Processing sequentially all the items.
		for (const [id, item] of toInject.entries()) {

		// For Vite, we inject MIDDLE content by adding a script tag
		// that references the virtual injected file

Conversation

amortemousque commented Jun 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What and why?

How?

Uh oh!

datadog-datadog-prod-us1-2 Bot commented Jun 11, 2026 • edited by datadog-datadog-prod-us1 Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yoannmoinet Jun 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

amortemousque Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

amortemousque commented Jun 10, 2026 •

edited

Loading

datadog-datadog-prod-us1-2 Bot commented Jun 11, 2026 •

edited by datadog-datadog-prod-us1 Bot

Loading

yoannmoinet Jun 15, 2026 •

edited

Loading

amortemousque Jun 17, 2026 •

edited

Loading