Avoid unnecessary empty data frames at end of stream #140
Conversation
This file contains 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
Currently, if a stream terminates with an empty data frame, we will send that empty data frame even if it is succeeded by trailers that are marked end of stream. That empty data frame is unnecessary. We should only send it if the payload length is non-zero and we are not sending trailers.
|
I cannot reproduce the current test failure locally after running the test on a loop for a while, on my branch or on |
|
It seems to me that your logic is correct. |
|
@kazu-yamamoto Thanks for that! Also, we have another PR incoming soon, so don't worry about releasing these changes just yet. |
|
The previous release was for me. |
netbsd-srcmastr
pushed a commit
to NetBSD/pkgsrc
that referenced
this pull request
Jan 30, 2025
# ChangeLog for http2 ## 5.3.9 * Using `ThreadManager` of `time-manager`. ## 5.3.8 * `forkManagedTimeout` ensures that only one asynchronous exception is thrown. Fixing the thread leak via `Weak ThreadId` and `modifyTVar'`. [#156](kazu-yamamoto/http2#156) ## 5.3.7 * Using `withHandle` of time-manager. * Getting `Handle` for each thread. * Providing allocSimpleConfig' to enable customizing WAI tiemout manager. * Monitor option (-m) for h2c-client and h2c-server. ## 5.3.6 * Making `runIO` friendly with the new synchronism mechanism. [#152](kazu-yamamoto/http2#152) * Re-throwing asynchronous exceptions to prevent thread leak. * Simplifying the synchronism mechanism between workers and the sender. [#148](kazu-yamamoto/http2#148) ## 5.3.5 * Using `http-semantics` v0.3. * Deprecating `numberOfWorkers`. * Removing `unliftio`. * Avoid `undefined` in client. [#146](kazu-yamamoto/http2#146) ## 5.3.4 * Support stream cancellation [#142](kazu-yamamoto/http2#142) ## 5.3.3 * Enclosing IPv6 literal authority with square brackets. [#143](kazu-yamamoto/http2#143) ## 5.3.2 * Avoid unnecessary empty data frames at end of stream [#140](kazu-yamamoto/http2#140) * Removing unnecessary API from ServerIO ## 5.3.1 * Fix treatment of async exceptions [#138](kazu-yamamoto/http2#138) * Avoid race condition [#137](kazu-yamamoto/http2#137) ## 5.3.0 * New server architecture: spawning worker on demand instead of the worker pool. This reduce huge numbers of threads for streaming into only 2. No API changes but workers do not terminate quicly. Rather workers collaborate with the sender after queuing a response and finish after all response data are sent. * All threads are labeled with `labelThread`. You can see them by `listThreas` if necessary. ## 5.2.6 * Recover rxflow on closing. [#126](kazu-yamamoto/http2#126) * Fixing ClientSpec for stream errors. * Allowing negative window. (h2spec http2/6.9.2) * Update for latest http-semantics [#122](kazu-yamamoto/http2#124) ## 5.2.5 * Setting peer initial window size properly. [#123](kazu-yamamoto/http2#123) ## 5.2.4 * Update for latest http-semantics [#122](kazu-yamamoto/http2#122) * Measuring performance concurrently for h2c-client ## 5.2.3 * Update for latest http-semantics [#120](kazu-yamamoto/http2#120) * Enable containers 0.7 (ghc 9.10) [#117](kazu-yamamoto/http2#117) ## 5.2.2 * Mark final chunk as final [#116](kazu-yamamoto/http2#116)
# for free
to join this conversation on GitHub.
Already have an account?
# to comment
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.
If a stream terminates with an empty data frame, we will send that empty data frame even if it is succeeded by trailers that are marked end of stream. That empty data frame is unnecessary. We should only send it if the payload length is non-zero and we are not sending trailers.
I discovered this after seeing
too many empty dataerrors from terminating a server handling several concurrent streams with an exception.