Skip to content

fix(http2): do not head-of-line block HEADERS, RST_STREAM and GOAWAY behind window-exhausted DATA - #2013

Merged
mosuem merged 1 commit into
mosum/http2-6b-http2-client-backpressure-and-windowsfrom
mosum/http2-4b-no-hol-blocking-on-exhausted-window
Oct 23, 2026
Merged

mosuem merged 1 commit into
mosum/http2-6b-http2-client-backpressure-and-windowsfrom
mosum/http2-4b-no-hol-blocking-on-exhausted-window

Conversation

@mosuem

@mosuem mosuem commented Oct 8, 2026 •

Copy link
Copy Markdown
Member

Note

This PR was generated by an AI coding agent (Jetski) on behalf of @mosuem.

Summary

ConnectionMessageQueueOut is a strict FIFO. Once the connection-level send window is exhausted, a DataMessage at the head of the queue blocks everything behind it, including frames that are not subject to flow control (RFC 9113 Section 6.9: "Flow control only applies to frames that are identified as being subject to flow control. Of the frame types defined in this document, this includes only DATA frames."). In practice: while one stream was waiting for WINDOW_UPDATE, no new request could be opened (HEADERS), no stream could be cancelled (RST_STREAM), and the connection could not be shut down gracefully (GOAWAY) until the peer granted credit.

Changes

  • lib/src/flowcontrol/connection_queues.dart:
    • _peekSendableMessage() replaces the "first message or nothing" rule. When the window is exhausted and the head of the queue is a DataMessage, it returns the first GOAWAY, or the first HEADERS / RST_STREAM whose stream has no earlier queued message. Skipping a message of a stream blocks every later message of that stream (and, for a PUSH_PROMISE, of the promised stream), so per-stream frame order is preserved: trailers never overtake their own DATA.
    • The scan is linear in the queue length, and StreamMessageQueueOut keeps forwarding DATA into this queue while the connection window is exhausted (it only checks the stream window). To avoid rescanning on every sendData() call, enqueueMessage skips _trySendMessages() when a DataMessage is enqueued while _connectionWindow.positiveWindow.wouldBuffer is true (since the new DataMessage cannot be sent and cannot unblock any earlier queued message). Enqueueing a non-DATA message while blocked still costs one pass over the queue.

Test Verification (Fails Before $\rightarrow$ Passes After)

Added to test/client_test.dart:

  • headers-and-rst-not-blocked-or-reordered-on-exhausted-window: stream 1 exhausts the connection window; stream 3's HEADERS and stream 5's HEADERS + RST_STREAM must still go out, in that order, before the window is replenished.
  • trailers-stay-behind-blocked-data-while-other-headers-bypass: stream 1 has blocked DATA and queued trailers; stream 3's HEADERS bypass them, stream 1's trailers are only written after its DATA.

Before fix:

00:05 +0 -1: client-tests client-errors headers-and-rst-not-blocked-or-reordered-on-exhausted-window [E]
  TimeoutException after 0:00:05.000000: Test timed out after 5 seconds.
00:10 +0 -2: client-tests client-errors trailers-stay-behind-blocked-data-while-other-headers-bypass [E]
  TimeoutException after 0:00:05.000000: Test timed out after 5 seconds.

(stream 3's HEADERS never leave the client while stream 1 waits for credit)

After fix:

00:00 +2: All tests passed!

@mosuem
mosuem added this pull request to stack #2016 October 8, 2026 10:42
@mosuem
mosuem force-pushed the mosum/http2-4b-no-hol-blocking-on-exhausted-window branch from 778da70 to f180b1d Compare October 8, 2026 12:04
@mosuem
mosuem force-pushed the mosum/http2-4b-no-hol-blocking-on-exhausted-window branch from f180b1d to 71eb4a0 Compare October 9, 2026 07:54
@mosuem
mosuem removed this pull request from stack #2016 October 9, 2026 07:55
@mosuem
mosuem added this pull request to stack #2020 October 9, 2026 07:55
@mosuem
mosuem force-pushed the mosum/http2-4b-no-hol-blocking-on-exhausted-window branch from 71eb4a0 to a995620 Compare October 9, 2026 08:18
@github-actions

github-actions Bot commented Oct 9, 2026 •

Copy link
Copy Markdown

PR Health

License Headers ✔️
// Copyright (c) 2026, the Dart project authors. Please see the AUTHORS file
// for details. All rights reserved. Use of this source code is governed by a
// BSD-style license that can be found in the LICENSE file.
Files
no missing headers

All source files should start with a license header.

Unrelated files missing license headers
Files
pkgs/http_multi_server/test/cert.dart

This check can be disabled by tagging the PR with skip-license-check.

Unused Dependencies ✔️
Package Status
http2 ✔️ All dependencies utilized correctly.

For details on how to fix these, see dependency_validator.

This check can be disabled by tagging the PR with skip-unused-dependencies-check.

Breaking changes ✔️
Package Change Current Version New Version Needed Version Looking good?
http2 Non-Breaking 3.1.0 3.2.0-wip 3.2.0-wip ✔️

This check can be disabled by tagging the PR with skip-breaking-check.

API leaks ✔️

The following packages contain symbols visible in the public API, but not exported by the library. Export these symbols or remove them from your publicly visible API.

Package Leaked API symbol Leaking sources

This check can be disabled by tagging the PR with skip-leaking-check.

Coverage ⚠️
File Coverage
pkgs/http2/lib/src/flowcontrol/connection_queues.dart 💔 93 % ⬇️ 0 %

This check for test coverage is informational (issues shown here will not fail the PR).

This check can be disabled by tagging the PR with skip-coverage-check.

Changelog Entry ✔️
Package Changed Files

Changes to files need to be accounted for in their respective changelogs.

This check can be disabled by tagging the PR with skip-changelog-check (for example, for non-user-visible changes when the package version is already at -wip).

@mosuem
mosuem force-pushed the mosum/http2-4b-no-hol-blocking-on-exhausted-window branch from a995620 to b321312 Compare October 9, 2026 08:48
Base automatically changed from mosum/http2-4-control-frame-queue-and-rst-order to master October 22, 2026 04:28
@mosuem
mosuem removed this pull request from stack #2020 October 22, 2026 06:44
@mosuem
mosuem force-pushed the mosum/http2-4b-no-hol-blocking-on-exhausted-window branch from b321312 to 092b1b3 Compare October 22, 2026 06:44
@mosuem
mosuem changed the base branch from master to mosum/http2-6b-http2-client-backpressure-and-windows October 22, 2026 06:45
@mosuem
mosuem added this pull request to stack #2030 October 22, 2026 06:47
@mosuem
mosuem force-pushed the mosum/http2-4b-no-hol-blocking-on-exhausted-window branch 2 times, most recently from e373706 to b5dd5d2 Compare October 22, 2026 07:48
…behind window-exhausted DATA

> [!NOTE]
> This PR was generated by an AI coding agent (Jetski) on behalf of @mosuem.

`ConnectionMessageQueueOut` is a strict FIFO. Once the connection-level send window is exhausted, a `DataMessage` at the head of the queue blocks *everything* behind it, including frames that are not subject to flow control (RFC 9113 Section 6.9: *"Flow control only applies to frames that are identified as being subject to flow control. Of the frame types defined in this document, this includes only DATA frames."*). In practice: while one stream was waiting for `WINDOW_UPDATE`, no new request could be opened (`HEADERS`), no stream could be cancelled (`RST_STREAM`), and the connection could not be shut down gracefully (`GOAWAY`) until the peer granted credit.

- **`lib/src/flowcontrol/connection_queues.dart`**:
  - `_peekSendableMessage()` replaces the "first message or nothing" rule. When the window is exhausted and the head of the queue is a `DataMessage`, it returns the first `GOAWAY`, or the first `HEADERS` / `RST_STREAM` whose stream has no earlier queued message. Skipping a message of a stream blocks every later message of that stream (and, for a `PUSH_PROMISE`, of the promised stream), so per-stream frame order is preserved: trailers never overtake their own `DATA`.
  - The scan is linear in the queue length, and `StreamMessageQueueOut` keeps forwarding `DATA` into this queue while the connection window is exhausted (it only checks the stream window). To avoid rescanning on every `sendData()` call, `enqueueMessage` skips `_trySendMessages()` when a `DataMessage` is enqueued while `_connectionWindow.positiveWindow.wouldBuffer` is `true` (since the new `DataMessage` cannot be sent and cannot unblock any earlier queued message). Enqueueing a non-DATA message while blocked still costs one pass over the queue.

Added to `test/client_test.dart`:
- `headers-and-rst-not-blocked-or-reordered-on-exhausted-window`: stream 1 exhausts the connection window; stream 3's `HEADERS` and stream 5's `HEADERS` + `RST_STREAM` must still go out, in that order, before the window is replenished.
- `trailers-stay-behind-blocked-data-while-other-headers-bypass`: stream 1 has blocked `DATA` *and* queued trailers; stream 3's `HEADERS` bypass them, stream 1's trailers are only written after its `DATA`.

**Before fix:**
```text
00:05 +0 -1: client-tests client-errors headers-and-rst-not-blocked-or-reordered-on-exhausted-window [E]
  TimeoutException after 0:00:05.000000: Test timed out after 5 seconds.
00:10 +0 -2: client-tests client-errors trailers-stay-behind-blocked-data-while-other-headers-bypass [E]
  TimeoutException after 0:00:05.000000: Test timed out after 5 seconds.
```
(stream 3's `HEADERS` never leave the client while stream 1 waits for credit)

**After fix:**
```text
00:00 +2: All tests passed!
```
@mosuem
mosuem force-pushed the mosum/http2-4b-no-hol-blocking-on-exhausted-window branch from b5dd5d2 to 655002d Compare October 22, 2026 08:01
@mosuem
mosuem marked this pull request as ready for review October 22, 2026 08:02
@mosuem
mosuem requested a review from brianquinlan October 22, 2026 08:13
@mosuem
mosuem removed this pull request from stack #2030 October 22, 2026 08:19
@mosuem
mosuem added this pull request to stack #2032 October 22, 2026 08:19
@mosuem
mosuem merged commit 3998b87 into master Oct 23, 2026
59 checks passed
@mosuem
mosuem deleted the mosum/http2-4b-no-hol-blocking-on-exhausted-window branch October 23, 2026 05:31
copybara-service Bot pushed a commit to dart-lang/sdk that referenced this pull request Oct 28, 2026
Revisions updated by 19 package dependencies found.
Moving forward revisions for: http.

Commit message:

[deps] rev http

Revisions updated by `dart tools/rev_sdk_deps.dart http`.

http (https://github.com/dart-lang/http/compare/c76dfb3..c76dfb3):
  c76dfb3  Wed Oct 28 10:21:55 2026 -0700  Brian Quinlan  fix(http2): accept larger headers (dart-lang/http#2041)
  23495b0  Mon Oct 26 15:17:28 2026 -0700  Brian Quinlan  chore(cronet_http): ready 1.10.1 (dart-lang/http#2039)
  27591d4  Mon Oct 26 15:08:10 2026 -0700  Brian Quinlan  fix(cronet): avoid cronet namespace duplication (dart-lang/http#2034)
  9cdb8da  Fri Oct 23 11:14:21 2026 -0700  Kevin Moore  Tighten pull_request_label workflow permissions (dart-lang/http#2035)
  3998b87  Fri Oct 23 07:31:29 2026 +0200  Moritz  fix(http2): do not head-of-line block HEADERS, RST_STREAM and GOAWAY behind window-exhausted DATA (dart-lang/http#2013)
  be2b51f  Fri Oct 23 07:31:28 2026 +0200  Moritz  fix(http2): Http2Client applies response-body backpressure to the HTTP/2 stream and uses 4 MiB / 16 MiB receive windows (dart-lang/http#2029)
  3b504cf  Thu Oct 22 06:28:12 2026 +0200  Moritz  fix(http2): never write RST_STREAM ahead of the stream's still-queued HEADERS (dart-lang/http#2008)
  a641fac  Thu Oct 22 06:28:12 2026 +0200  Moritz  fix(http2): preserve the complete response on RST_STREAM(NO_ERROR) (dart-lang/http#2012)
  5d896c7  Thu Oct 22 06:28:11 2026 +0200  Moritz  fix(http2): keep flow-controlled end-stream DATA alive until it is flushed (`#1998`) (dart-lang/http#2007)
  c834aa0  Thu Oct 22 06:28:11 2026 +0200  Moritz  fix(http2): ignore PRIORITY and unknown frames on streams and reject non-zero stream control frames (dart-lang/http#2006)
  2d35515  Thu Oct 22 06:15:59 2026 +0200  Moritz  fix(http2): replenish connection receive window when stream closes with unconsumed data (dart-lang/http#2005)
  1217181  Tue Oct 20 19:17:59 2026 -0400  Mouad Debbar  [http_multi_server] Prepare to release 3.2.3 (dart-lang/http#2028).

http (https://github.com/dart-lang/http/compare/a8a24a0..c76dfb3):
  c76dfb3  Wed Oct 28 10:21:55 2026 -0700  Brian Quinlan  fix(http2): accept larger headers (dart-lang/http#2041)
  23495b0  Mon Oct 26 15:17:28 2026 -0700  Brian Quinlan  chore(cronet_http): ready 1.10.1 (dart-lang/http#2039)
  27591d4  Mon Oct 26 15:08:10 2026 -0700  Brian Quinlan  fix(cronet): avoid cronet namespace duplication (dart-lang/http#2034)
  9cdb8da  Fri Oct 23 11:14:21 2026 -0700  Kevin Moore  Tighten pull_request_label workflow permissions (dart-lang/http#2035)
  3998b87  Fri Oct 23 07:31:29 2026 +0200  Moritz  fix(http2): do not head-of-line block HEADERS, RST_STREAM and GOAWAY behind window-exhausted DATA (dart-lang/http#2013)
  be2b51f  Fri Oct 23 07:31:28 2026 +0200  Moritz  fix(http2): Http2Client applies response-body backpressure to the HTTP/2 stream and uses 4 MiB / 16 MiB receive windows (dart-lang/http#2029)
  3b504cf  Thu Oct 22 06:28:12 2026 +0200  Moritz  fix(http2): never write RST_STREAM ahead of the stream's still-queued HEADERS (dart-lang/http#2008)
  a641fac  Thu Oct 22 06:28:12 2026 +0200  Moritz  fix(http2): preserve the complete response on RST_STREAM(NO_ERROR) (dart-lang/http#2012)
  5d896c7  Thu Oct 22 06:28:11 2026 +0200  Moritz  fix(http2): keep flow-controlled end-stream DATA alive until it is flushed () (dart-lang/http#2007)
  c834aa0  Thu Oct 22 06:28:11 2026 +0200  Moritz  fix(http2): ignore PRIORITY and unknown frames on streams and reject non-zero stream control frames (dart-lang/http#2006)
  2d35515  Thu Oct 22 06:15:59 2026 +0200  Moritz  fix(http2): replenish connection receive window when stream closes with unconsumed data (dart-lang/http#2005)
  1217181  Tue Oct 20 19:17:59 2026 -0400  Mouad Debbar  [http_multi_server] Prepare to release 3.2.3 (dart-lang/http#2028)

Change-Id: I259130f06771f4b7ba61c7610847e6677b5c8ed5
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/570641
Reviewed-by: Nate Bosch <[email protected]>
Commit-Queue: Brian Quinlan <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants