Conversation
## Why I cover cancellation after a real socket write but before the fiber scheduler reports the byte count. Closing the stream then flushes the same buffered bytes a second time. The regression exercises explicit flush, single-argument write, and multi-argument write through Ruby's existing scheduler fixture. I observe bytes at the socket peer rather than private buffer fields, and use no third-party scheduler dependency. ## Verification Local runs on Ruby 4.0.6 and Ruby 4.1.0dev at 55562be each report 3 tests, 18 assertions, and 3 expected failures, with no errors or skips. Each failure receives the buffered message again after cancellation. This commit intentionally contains only the regression tests, not the fix. Assisted-By: devx/2cdbdf48-1a60-4b23-b35e-42bee58a4326
gmalette
commented
Sep 17, 2026
|
|
||
| private | ||
|
|
||
| def assert_no_replayed_buffer |
Author
There was a problem hiding this comment.
This is backported from Shopify/dalli#81
The goal is to show that if the fiber gets interrupted after being blocked in a write, the internal buffer's accounting doesn't account for the bytes written to the kernel's buffer, closing the Ruby socket will flush the same bytes again.
## Why A scheduler can transfer bytes and then unwind without returning their count. Ruby cannot safely retry that pending output buffer: close may send bytes that already reached the peer a second time. I discard the uncertain buffer and propagate the original interruption instead. I apply the guard to explicit buffer flushes and to the buffered prefix of vector writes, which also covers buffer-overflow writes. Successful results, ordinary errno returns, and the fallback for schedulers without a write hook retain their existing handling. This does not roll back bytes already written or promise delivery of the remaining buffered data. ## Verification On Ruby 4.1.0dev, the rebuilt source passes the three deterministic real-socket regressions with 3 tests and 18 assertions, and the complete `test/fiber/test_io.rb` with 13 tests and 53 assertions. Both runs have no failures, errors, or skips. Runtime identity confirms that the executable and loaded shared library come from the patched build. On Ruby 4.0.6, the same regressions pass, as do the native io_uring reproduction and its no-cancellation control. The IO/fiber selection passes 347 tests with one portability skip, the IO bootstrap selection passes 18 tests, and the IO specs pass 1517 examples.
jhawthorn
reviewed
Sep 17, 2026
Comment on lines
+1401
to
+1402
| fptr->wbuf.off = 0; | ||
| fptr->wbuf.len = 0; |
Member
There was a problem hiding this comment.
How do we know that the bytes have actually been written and the write occurred, without checking the return value (which we can't be cause there was an exception).
Dropping bytes could be as bad as replaying them
jhawthorn
reviewed
Sep 17, 2026
|
|
||
| if (!UNDEF_P(result)) { | ||
| return rb_fiber_scheduler_io_result_apply(result); | ||
| if (fptr->wbuf.len && iov[0].iov_base == fptr->wbuf.ptr + fptr->wbuf.off) { |
Member
There was a problem hiding this comment.
Why do we use this branch only in some cases?
jhawthorn
reviewed
Sep 17, 2026
Comment on lines
1487
to
1488
| fptr->wbuf.off += result; | ||
| fptr->wbuf.len -= result; |
Member
There was a problem hiding this comment.
It feels like this is the code that needs to run, but the challenge is that we're not getting result? (because of the interrupt)
|
Interrupted writes should properly clear the write buffer. I'm working on sorting this out. |
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.
Situation
A fiber scheduler can write buffered bytes to the peer and then be interrupted before returning their count. Ruby's write-buffer position has not advanced, so closing the IO can send those bytes again.
Execution
I add a test-only reproduction covering explicit flush and buffered single- and multi-argument writes. The existing scheduler fixture performs real socket I/O, then yields before reporting the first write's result. The test reads the initial message from the peer, interrupts the writer, and asserts that closing it does not transmit another copy.
This draft intentionally contains failing tests, without a fix or any Async/io-event dependency.
Local verification
Ruby 4.0.6 and Ruby 4.1.0dev at
55562be3deeach report 3 tests, 18 assertions, and 3 expected failures, with no errors or skips. Each failure observesHello Worldafter the initial copy has already been read, instead of the expected empty remainder.