fix: large amounts of stacks slows down synthesis #34480
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
A large amount of stacks leads to a large body of metadata into the `manifest.json` file. The file can grow too large to be written to disk (>512MB), but even if it doesn't fail outright, serializing and writing and loading a large JSON file takes a lot of CPU time. This PR splits the metadata for stacks off into a separate file which is written separately, moving it out of the hot path of working with the `manifest.json` file. On a tree of 256 stacks and ~40,000 constructs, this shaves off 5 seconds off of the total synthesis time.
rix0rrr
force-pushed
the
huijbers/synth-perf
branch
from
27b76dd to
046eb38
Compare
aws-cdk-automation
previously requested changes
May 16, 2025
rix0rrr
added
the
pr-linter/exempt-integ-test
aws-cdk-automation
dismissed
their stale review
✅ Updated pull request passes all PRLinter validations. Dismissing previous PRLinter review.
rix0rrr
changed the title
matboros
reviewed
May 19, 2025
| * calls and accumulates into an array, both of which are much slower | ||
| * than this solution. | ||
| */ | ||
| export function* iterateDfsPreorder(root: IConstruct) { |
Contributor
There was a problem hiding this comment.
[q]: preorder would be visiting left children before right one, which is not what this method does. Was that intended?
Contributor
Author
There was a problem hiding this comment.
Preorder means parent before children, which this method does do.
matboros
reviewed
May 19, 2025
| import { IConstruct } from 'constructs'; | ||
|
|
||
| /** | ||
| * Breadth-first iterator over the construct tree |
Contributor
Author
There was a problem hiding this comment.
Ah yes, copy/paste-o. Thanks.
matboros
reviewed
May 19, 2025
| // Use a specialized queue data structure. Using `Array.shift()` | ||
| // has a huge performance penalty (difference on the order of | ||
| // ~50ms vs ~1s to iterate a large construct tree) | ||
| const queue: IConstruct[] = [root]; |
Contributor
There was a problem hiding this comment.
this should be called stack (filo), a queue would be (fifo)
Comments were outdated
Collaborator
AWS CodeBuild CI Report
Powered by github-codebuild-logs, available on the AWS Serverless Application Repository |
rix0rrr
added a commit
to aws/aws-cdk-cli
that referenced
this pull request
Jun 3, 2025
When CDK apps grow extremely large (think 10-100 stacks, 1000-10000 constructs), all metadata together begins to exceed 512MB, the maximum string size in NodeJS. People usually deal with this by disabling metadata, but they shouldn't have to. In addition, even for manifests that don't exceed 512MB the extremely large size of the single JSON object slows down its writing and reading every time, even if the metadata doesn't need to be accessed. An effective solution is to write the metadata of an artifact to a separate file. This PR introduces the ability for that into the Cloud Assembly schema, and updates the CLI to read from both sources if available. Relates to aws/aws-cdk#34480.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
A large amount of stacks leads to a large body of metadata into the
manifest.jsonfile. The file can grow too large to be written to disk (>512MB), but even if it doesn't fail outright, serializing and writing and loading a large JSON file takes a lot of CPU time.This PR splits the metadata for stacks off into a separate file which is written separately, moving it out of the hot path of working with the
manifest.jsonfile. On a tree of 256 stacks and ~40,000constructs, this shaves off 5 seconds off of the total synthesis time.
Also in this PR: construct tree iteration was being done in a lot of places, with function
recursion and returning arrays. Moving this all into some functions that use generators. This improves performance only slightly, but it should improve memory usage by a whole lot for large construct trees.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache-2.0 license