Optimise isAssetReferenced in packager #858
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
🦋 Changeset detectedLatest commit: 33cd346 The changes in this PR will be included in the next version bump. This PR includes changesets to release 108 packages
Not sure what this means? Click here to learn what changesets are. Click here if you're a maintainer who wants to add another changeset to this PR |
Contributor
📊 Type Coverage ReportCoverage Comparison
Files with Most Type Issues (Top 15)
This report was generated by the Type Coverage GitHub Action |
marcins
force-pushed
the
optimise-isAssetReferenced-calls
branch
from
8b00720 to
793231e
Compare
marcins
changed the title
marcins
changed the title
Implements a three-tier optimization strategy for isAssetReferenced calls: 1. **Cache-first lookup** (~0.001ms) - Check expensive computation cache first 2. **Fast checks fallback** (~0.01ms) - Only when cache misses, try quick checks 3. **Expensive computation** (~20-2000ms) - Only when both above fail, with caching **Key Changes:** - Add `isAssetReferencedFastCheck()` method to BundleGraph for fast-only checks - Update ScopeHoistingPackager to use cache-first strategy - Separate fast operations from expensive ones for optimal performance **Technical Details:** - Only expensive bundle traversals get cached results - Cache hit rates: Jira ~10%, Confluence ~95-100% - Reduces cache overhead for operations that are cheaper to compute than cache This resolves performance regressions in Jira while maintaining Confluence optimization benefits through adaptive caching strategy.
marcins
force-pushed
the
optimise-isAssetReferenced-calls
branch
from
fe4be40 to
33cd346
Compare
Contributor
📊 Benchmark Results🎉 Performance improvements detected! 📊 Benchmark ResultsOverall Performance
🔍 Detailed Phase AnalysisThree.js Real Repository (JS)
Three.js Real Repository (Native)
💾 Unified Memory AnalysisThree.js Real Repository (JS) Memory Statistics
Sample Counts: JS: 6, Native: 271 Three.js Real Repository (Native) Memory Statistics
Sample Counts: JS: 6, Native: 517 🖥️ Environment
|
| asset.meta.shouldWrap || | ||
| this.bundle.env.sourceType === 'script' || | ||
| this.bundleGraph.isAssetReferenced(this.bundle, asset) || | ||
| this.isAssetReferencedInBundle(this.bundle, asset) || |
Contributor
There was a problem hiding this comment.
Nit: Passing this.bundle is a bit odd. I'd just reference it directly via this in isAssetReferencedInBundle
Contributor
There was a problem hiding this comment.
Ahh I see the bundle changes in some cases. Ignore me 😅
JakeLane
approved these changes
Nov 13, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation
Looking at some product SSR bundling builds shows that a significant amount of time is spent in
bundleGraph.isAssetReferencedduring packaging. This is inefficient as there are overlapping repeated traversals for different assets during the build. In particular this results in a lot of GC, andclear-ing of the visited BitSet used in the graph traversal.This isn't true across all products and all build types though, it seems to be endemic to some builds only. It's still slightly faster for the builds where this wasn't really an issue just not to the same extent (e.g. below - Confluence is 18%±5% faster, Jira is 4%±2% faster)
Also worth noting that for local development the impact will only be seen for SSR, as the ScopeHoistingPackager is not used for FE builds in dev.
Benchmark results:
Confluence SSR:
ENABLE_OPTIMISATION=falseENABLE_OPTIMISATION=trueConfluence FE:
ENABLE_OPTIMISATION=falseENABLE_OPTIMISATION=trueJira SSR:
ENABLE_OPTIMISATION=falseENABLE_OPTIMISATION=trueJira FE:
ENABLE_OPTIMISATION=falseENABLE_OPTIMISATION=trueChanges
Instead of traversing per asset on demand, we take a multi phase approach:
This is behind a new
precomputeReferencedAssetsfeature flag.This approach is faster for all builds that were tested.
Checklist
docs/folder