Repository navigation
fix(sdk): fill encrypt segments without re-copying the partial buffer (DSPX-5232) - #1095
Draft
dmihalcik-virtru wants to merge 2 commits into
Draft
dmihalcik-virtru wants to merge 2 commits into
dmihalcik-virtru wants to merge 2 commits into
Conversation
Signed-off-by: Dave Mihalcik <dmihalcik@virtru.com>
… (DSPX-5232) writeStream concatenated every input chunk onto the partial segment, so filling a segment of size S from chunks of size c copied about S^2/2c bytes. With 64 KiB chunks (browser File.stream(), Node file streams) encrypt fell from 311 MiB/s at 2 MiB segments to 17 MiB/s at 64 MiB, with peak RSS of 4.8 GiB, and this blocks raising the default segment size (DSPX-4651). Copy each input byte once, into a segment buffer allocated per segment, and hand full segments to the cipher without the extra toArrayBuffer copy. AesGcmCipher also dropped a redundant copy of its concatenated output. Adds scripts/bench-encrypt.mjs (npm run bench:encrypt), which sweeps writeStream over segment sizes in isolated child processes and reports throughput and peak RSS. Signed-off-by: Dave Mihalcik <dmihalcik@virtru.com>
dmihalcik-virtru
force-pushed
the
DSPX-5232-segment-buffer
branch
from
October 19, 2026 17:27
7a9e2a4 to
f961675
Compare
|
|
Important Draft PR not reviewedDraft PRs are not automatically reviewed by default.
To automatically review draft PRs, update your CodeRabbit configuration: reviews:
auto_review:
drafts: true
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
This branch has not been deployed
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
1 participant
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.
What / why
DSPX-5232, part 1 of 2 (CRC-32 is #1096, stacked on this).
writeStreamconcatenated every input chunk onto the partial segment (concatUint8([currentBuffer, value])) and sliced the remainder off with another copy. Filling a segment of size S from chunks of size c therefore copied about S²/2c bytes. BrowserFile.stream()and Node file streams deliver about 64 KiB chunks, so encrypt collapses as segments grow. That blocks raising the default segment size (DSPX-4651). Found in the DSPX-5124 large-TDF analysis (item #8).toArrayBuffercopy.AesGcmCipher.encryptalso drops a redundant copy of its concatenated output.lib/scripts/bench-encrypt.mjs(npm run bench:encrypt). It runs the SDK's ownwriteStreamwith key access stubbed out, so no KAS is involved, and runs each segment size in its own child process so peak RSS is per cell. It reports throughput and peak RSS. It isn't published (scripts/is not infiles).spec/DSPX-5232.md.Results
npm run bench:encrypt -- --lib <main> --lib <this>, Node 24.21.0, Apple M5 Pro, 1 GiB in 64 KiB chunks:Throughput is now flat in segment size. Peak memory is still linear in segment size, as it was before, but at about 9× the segment instead of 66–88×: 16.5 GiB → 2.2 GiB at 256 MiB. It can't be flat, because the cipher needs one segment's plaintext and ciphertext at once. The remaining per-segment copies (analysis #10) are out of scope.
xtest paired A/B (
js@mainvs. the PR stack, 1 KiB / 1 MiB / 32 MiB, local platform): no regressions in any of the 7 cells; noise floor ±4.5%. Those payloads are dominated by about 1.1 s of CLI startup, and the CLI reads in 8 MiB chunks, so this fix barely shows there.How to test
encrypt-segmentation.spec.tsround-trips these chunking patterns and checks the manifest's segment sizes: empty input, one byte at a time, chunks straddling boundaries, exact-segment chunks, one chunk holding several segments, and a large chunk after a partial segment.Risk
Touches the encrypt hot loop and one line of
AesGcmCipher. The ciphertext format is unchanged. A source that overwrites one buffer for every chunk was already broken with any Web Streams consumer, because the stream pulls again before the reader'sread()continuation runs. This PR doesn't change that.