io: wait for writable sockets before retrying sync writes

[stondo]

Summary

Replace the one-second sleep used by synchronous nonblocking socket writes
after EAGAIN/EWOULDBLOCK with a bounded POLLOUT wait.

Fixes #12115

Root cause

fd_io_write() currently sleeps for one full second whenever send() or
sendto() temporarily fills a nonblocking socket buffer. A large response can
hit that path repeatedly, adding one second per backpressure event even when
the socket becomes writable much sooner.

Changes

• wait for POLLOUT for at most one second before retrying;
• use WSAPoll on Windows, matching the existing network implementation;
• do not count an interrupted POSIX wait as one of the bounded retries;
• preserve the existing 30-attempt limit and partial-write reporting;
• add deterministic internal tests for prompt wakeup and peer-close error
  propagation using small nonblocking socket pairs.

Compatibility

There is no configuration or API change. The one-second upper bound and
30-attempt limit remain unchanged. The writer now resumes promptly when the
socket becomes writable instead of always waiting the full second.

The regression tests are POSIX-only because they use socketpair() and
pthreads. The Windows source path uses the same WSAPoll pattern present in
src/flb_network.c.

Testing

Commands:

cmake -S . -B build-sync-final \
  -DFLB_TESTS_RUNTIME=On \
  -DFLB_TESTS_INTERNAL=On \
  -DFLB_EXAMPLES=Off \
  -DCMAKE_CXX_FLAGS=-mavx2
cmake --build build-sync-final -j8
ctest --test-dir build-sync-final --output-on-failure \
  -R '^flb-(rt-in_tcp|rt-out_tcp|it-network|it-http_client|it-http_server)$'
ctest --test-dir build-sync-final --output-on-failure \
  --repeat until-fail:10 \
  -R '^flb-it-network$'

The tests ran in the documented Debian 12 rootless Podman environment. The
explicit AVX2 flag is a GCC 12 compatibility workaround for bundled simdutf,
not part of this change.

Results:

• full runtime/internal build: passed;
• five focused network, HTTP, and TCP tests: passed;
• flb-it-network: passed 10 consecutive iterations;
• negative control with the test and old writer: failed in 2.02 seconds at the
  elapsed-time assertion, as expected;
• patched test: completes in approximately 20-30 ms;
• peer-close test: preserves EPIPE, ECONNRESET, or ENOTCONN instead of
  returning with stale EAGAIN;
• focused regression under Valgrind: 0 bytes live at exit and 0 errors;
• full-range commit-prefix validation: passed;
• git diff --check: passed;
• DCO: signed.

Live syscall evidence from the original reproduction:

before: 16 EAGAIN results, 16 one-second sleeps, 16-17 s response
after:  23 EAGAIN results, 23 POLLOUT waits, 0 sleeps, 0.252 s response

---

Enter [N/A] in the box, if an item is not applicable to your change.

Testing

• [N/A] No configuration surface is changed
• Runtime and syscall output are included above
• Focused Valgrind run shows no leaks or memory errors

Packaging

• [N/A] No packaging or container-output change

Documentation

• [N/A] No user-facing configuration or API change

Backporting

• Candidate for a separate 5.0 backport after the master change is accepted

Fluent Bit is licensed under Apache 2.0. By submitting this pull request I
understand that this code will be released under the terms of that license.

Summary by CodeRabbit

• Bug Fixes
  • Improved reliability of non-blocking socket writes by waiting for socket writability (poll-based) rather than using fixed sleep retries.
  • Enhanced handling of interrupted waits and improved behavior under peer disconnects.
  • Added Windows-compatible readiness waiting for socket operations.
• Tests
  • Added non-Windows coverage to verify non-blocking write completion within a time bound.
  • Added validation that early peer close correctly propagates expected write failures.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 2c4d162f-402c-47e9-a876-141ee98cfb90

📥 Commits

Reviewing files that changed from the base of the PR and between abced0f and 9b88c34.

📒 Files selected for processing (1)

• tests/internal/network.c

🚧 Files skipped from review as they are similar to previous changes (1)

• tests/internal/network.c

---

📝 Walkthrough

Walkthrough

Synchronous socket writes now use poll/WSAPoll to wait for POLLOUT after EAGAIN or WOULDBLOCK, with bounded retries and poll-error handling. Non-Windows socket tests cover completion under backpressure and peer-close error propagation.

Changes

Synchronous socket write wait

Layer / File(s)	Summary
Poll-based write retry src/flb_io.c	Adds Windows WSAPoll support and replaces one-second sleeps after temporary send failure with bounded POLLOUT polling, including interruption and poll-error handling.
Writability and peer-close regression tests tests/internal/network.c	Adds non-Windows socketpair tests that verify timely completion while a reader drains data and failure propagation when the peer closes early.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Suggested reviewers: edsiper

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main change: synchronous writes now wait for socket writability before retrying.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f56c5c1e48

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/internal/network.c`:
- Around line 324-395: Update
test_nonblocking_socket_write_propagates_peer_close to block or ignore SIGPIPE
around the flb_io_fd_write call, restoring the prior signal state afterward so
the test process survives the peer-close condition and can execute its existing
error assertions.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 60941f93-f271-452b-8cd1-4667ab831957

📥 Commits

Reviewing files that changed from the base of the PR and between f56c5c1 and abced0f.

📒 Files selected for processing (2)

• src/flb_io.c
• tests/internal/network.c