Conversation
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## master #246 +/- ##
==========================================
+ Coverage 87.69% 87.93% +0.24%
==========================================
Files 498 500 +2
Lines 71322 72986 +1664
==========================================
+ Hits 62546 64182 +1636
- Misses 8776 8804 +28
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
The size check only ran in SctpHandler::handle_write, where the error is logged and dropped, so send() returned Ok for a message that was never sent and left its bytes charged to outstanding_bytes. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The buffer was only ever read for its length, which is the negotiated max message size. Use that directly instead of allocating and zeroing up to 256 KiB per connection. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Reassembly was bounded by the negotiated send limit, the smaller of the two peers' limits. A peer may send anything up to the max-message-size this endpoint advertises, so a peer that advertised less than it sends (or omitted the attribute, implying 64 KiB) had larger messages dropped with ErrShortBuffer. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The ceiling was a fixed 256 KiB. It is now the SCTP receive buffer size
(1 MiB by default), which a message must fit in to be reassembled.
Unbounded and Bounded(0) resolve to it, and the SDP advertises the same
value that is enforced, instead of 0 ("any size") for Bounded(0).
sctp_receive_buffer_deadlock now has the sender ignore the receiver's
limit, since a compliant peer can no longer send a message larger than
the receive buffer, and checks the association survives it.
BREAKING CHANGE: SctpMaxMessageSize::MAX_MESSAGE_SIZE and
SctpMaxMessageSize::as_usize are removed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
scripts/bench.py lists, checks, runs and compares the workspace's criterion benchmarks. `compare` builds a base revision in a worktree outside the repository and alternates base and head runs on one machine; `upstream` does the same against webrtc-rs/rtc, overlaying this tree's benchmark sources. benchmarks/rtc-bench is an in-memory, virtual-clock PeerPair harness with end-to-end benches for connection setup, data-channel throughput, the RTP path and an allocations-per-packet report. New protocol benches cover rtc-sctp, rtc-shared's replay detector and rtc-ice. CI builds and runs every benchmark once (`bench.py check --providers all`). docs/benchmarking.md documents the suite. ISSUES.md is the memory and performance analysis the framework was built to measure. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
1. Rejected SRTP packets no longer retain stream state.
2. The unused SCTP working buffer is gone.
3. Application backlogs are bounded by bytes as well as message count.
4. Parsed RTP headers are reused and owned buffers encrypted in place.
5. Pipeline scratch queues are reused and redundant traversal avoided.
6. Negotiated routing metadata is cached.
7. Retransmission history is bounded by bytes and age.
8. Sample-builder storage is right-sized.
9. Pre-handshake DTLS record queues are bounded.
10. DTLS fragment metadata is bounded and reassembly is linear.
11. Copies are removed from the DTLS record path.
12. ICE binding requests are pruned in place.
13. NACK retry bookkeeping is linear.
14. Congestion-control feedback is borrowed rather than cloned.
15. DTLS handshake-cache entries are borrowed during transcript
inspection.
Each issue's Resolution section in ISSUES.md records the change,
before/after numbers and what was left open.
Also adds SIMD.md, an analysis of SIMD opportunities, with its
exploratory kernel probes in benchmarks/simd-probes.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Every section of SIMD.md is implemented, using portable abstractions only: `wide` vectors (backend chosen at compile time), memchr's substring search, the crc-fast CRC library, or safe Rust the compiler vectorizes. There are no std::arch intrinsics. Each section's Resolution records the change, how equivalence with the old code was checked, and before/after numbers from `bench.py compare` on an Apple M1 Max (x86-64 was not measured). - H.264: the payloader and the Annex B reader find start codes with a prebuilt memmem finder; the reader copies spans in bulk (71-76x). - Audio: mono, stereo and four-channel layout fast paths (5-17x). - rtc-crypto moves to aes 0.9 / ctr 0.10 / ccm 0.6, so applications get hardware AES on aarch64 without `--cfg aes_armv8`, which they never inherited from this repository's config. In-repository SRTP AES-CM is ~12% and DTLS CBC encryption ~19% slower as a result. `bench.py external` builds benchmarks/external-consumer as an application would and checks the backend. - CBC decryption batches 32 blocks from 512 bytes up. - FlexFEC serializes each protected packet once per block into reused buffers, with a `wide` XOR (encode 1.4-1.7x). - STUN FINGERPRINT and Ogg page checksums use crc-fast (3.8x at 100 B; Ogg pages up to 59x). - RFC 8888 feedback decodes in bulk from a contiguous chunk (3.3-4.3x). - NACK generation and response, receiver reports and the TWCC arrival map work a word or slice at a time (NACK generation 1.2-8.1x). - New Sample slice conversions for PCM, bit-exact with `From`. Fixes an H26xReader panic (usize underflow) when data arrived after end of stream. New criterion targets: rtc-media h26x/ogg/pcm and rtc-interceptor feedback/flexfec, plus H.264, RFC 8888 and STUN fingerprint groups in existing targets. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Brings in the benchmark framework, the ISSUES.md memory and performance fixes, and the SIMD.md implementation. Conflicts, all between ISSUES.md webrtc-rs#2 on the branch and master's SCTP max-message-size work, which removed the same unused working buffer: - transport/sctp, handler/sctp: kept master's direction semantics. Inbound messages are bounded by local_max_message_size(), this endpoint's advertised limit capped at the SCTP receive buffer; sends by the negotiated max_message_size(). Dropped the branch's max_message_len(), which applied the negotiated size to both directions, and its tests. The branch's byte-bounded read backlog (ISSUES.md webrtc-rs#3) is kept. - handler/interceptor: master's rid validation (drop a packet whose rid is not negotiated) with the branch's borrowed &str rid. - peer_connection/mod.rs: comment wording, master's. ISSUES.md webrtc-rs#2 now notes what the merge superseded. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
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
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.
So this issue has come up whilst I am implementing a layer on top of RTC.
I had to use Claude to implement it as it is way beyond what I am comfortable with in terms of RTC internals.
Leaving it here in case it's useful but have not yet tested it in a Production app yet.