fix(apijson): avoid O(n²) array marshaling for large slices#4316
Open
Tonyqu123 wants to merge 1 commit into
Open
fix(apijson): avoid O(n²) array marshaling for large slices#4316Tonyqu123 wants to merge 1 commit into
Tonyqu123 wants to merge 1 commit into
Conversation
newArrayTypeEncoder built JSON arrays by repeatedly calling sjson.SetRawBytes(json, "-1", item) inside a loop. sjson is designed for sparse field updates on existing JSON; appending element-by-element forces it to re-parse and reallocate the entire buffer per call, giving the array encoder O(n²) time and O(n²) memory. Replaced with a single-pass bytes.Buffer concatenation, restoring O(n). Output is byte-identical to the previous implementation. Added a benchmark in internal/apijson/array_bench_test.go covering N = 100..30,000 as a regression guard. Benchmarks (Apple M4 Pro, go1.26): N before B/op after B/op mem before ns/op after ns/op speed 100 92 KB 8 KB 11x 138 µs 24 µs 5.9x 1,000 6.9 MB 66 KB 105x 7.1 ms 0.21 ms 33x 10,000 692 MB 845 KB 819x 437 ms 1.18 ms 369x 30,000 6.2 GB 2.0 MB 3,070x 4,091 ms 3.10 ms 1320x 1k → 30k scaling: ns/op grows 580x, B/op grows 893x — clear O(n²). Real-world impact: an SAP BTP controller syncing ~30,000 IPs to a Cloudflare account-level IP list via Rules.Lists.Items.Update was OOM-killed because a single Update call allocated ~6.2 GB.
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
encoder.newArrayTypeEncoderconstructs JSON arrays by repeatedly callingsjson.SetRawBytes(json, "-1", item)inside a loop.sjsonis designed for sparse field updates on existing JSON; appending element-by-element forces it to re-parse and reallocate the entire buffer per call, giving the array encoder O(n²) time and O(n²) memory.This patch replaces the
sjsonloop with a single-passbytes.Bufferconcatenation, restoring O(n).Fixes #4315.
Impact
We hit this in production while syncing ~30,000 IPs to a Cloudflare account-level IP list via
Rules.Lists.Items.Update. A single Update call allocated ~6.2 GB and ran for ~4 seconds, OOM-killing our 1 GiB controller.Benchmarks on
Marshal([]string{...})(Apple M4 Pro, go1.26):1k → 30k scaling: before grew ns/op 580x, B/op 893x — clear O(n²). After the fix, scaling is linear.
Compatibility
Output is byte-identical to the previous implementation (verified against
encoding/jsonforn = 0, 1, 2, 100). All existing tests ininternal/apijsonpass.Tests
Added
internal/apijson/array_bench_test.gowith a correctness check and benchmarks at N = 100, 1k, 5k, 10k, 30k as a regression guard. Happy to drop the benchmark file if you'd prefer to keep the test directory generated-only.Notes
The same
sjson.SetRawBytespattern is used byencodeMapEntries(encoder.go:339) andnewStructTypeEncoder(encoder.go:299) — also technically O(n²), but typically don't hit the pathological case because struct field counts and map sizes stay small. Out of scope for this PR; happy to address in a follow-up if useful.