[Jarred-Sumner]

Summary

Fixes the 41/200 timeout flake of test/js/web/fetch/fetch-proxy-tls-intern-race.test.ts on Linux release targets (🐧 25.04 x64-baseline, 🐧 3.23 aarch64, 🐧 13 x64-baseline).

Root cause: the test's CONNECT proxy only handled client.on("error"), not "close". The probe loop aborts each fetch immediately, so the client→proxy socket usually closes with FIN (not RST). The proxy→backend upstream socket was never destroyed in that path. On release builds the setImmediate probe loop fires thousands of times per second, so orphaned TLS connections accumulate on the backend Bun.serve until its accept queue / FD table saturates and the driver's fetch stalls. The driver fetch had no abort signal, so await driverDone hung past the 30s test timeout.

Fix:

• Proxy: destroy upstream on client "close" (and vice versa) so aborted probes don't leak backend connections.
• Fixture: start the hard cap as a setTimeout up front so it's authoritative regardless of what the loops are blocked on; it flips stop and aborts the in-flight driver fetch.
• Fixture: let the driver complete one request against an idle proxy before the probe flood so the driverOk > 0 sanity check holds on loaded CI. The warmup wait is bounded by the same hard-cap timer (addresses the unbounded-loop review feedback on deflake: fix 7 flaky tests with proper root-cause fixes #28873).

Supersedes the fetch-proxy-tls-intern-race-fixture.ts portion of #28873.

Test plan

• 20× sequential runs with release build (bun-pr 29163): all pass at ~5.4s
• 18× parallel runs (6 concurrent × 3 rounds) under load: all pass
• 4× bun bd test (debug+ASAN): all pass

🤖 Generated with Claude Code

[robobun]

^{Updated 7:05 AM PT - Apr 12th, 2026}

❌ Your commit c427de9a has 3 failures in Build #45291 (All Failures):

• test/js/web/fetch/fetch.upgrade.test.ts - code 1 on 🐧 13 aarch64
• test/js/web/fetch/chunked-trailing.test.js - code 1 on 🪟 2019 x64-baseline
• test/bake/dev/stress.test.ts - code 1 on 🪟 2019 x64-baseline

---

🧪   To try this PR locally:

bunx bun-pr 29212

That installs a local version of the PR into your bun-29212 executable, so you can run:

bun-29212 --bun

[github-actions]

This PR may be a duplicate of:

1. deflake: fix 7 flaky tests with proper root-cause fixes #28873 - Both modify the fetch-proxy-tls-intern-race fixture to fix the same flaky proxy test timeout; this PR supersedes the fixture portion of deflake: fix 7 flaky tests with proper root-cause fixes #28873

🤖 Generated with Claude Code

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 533b110e-9ba8-4e88-9341-d876456f13fe

📥 Commits

Reviewing files that changed from the base of the PR and between 9fdbf16 and c427de9.

📒 Files selected for processing (1)

• test/js/web/fetch/fetch-proxy-tls-intern-race-fixture.ts

---

Walkthrough

Added an AbortController-driven hard cap and warmup before probe flood in the fixture; adjusted driver to pass the abort signal. Updated test proxy to track and destroy upstream sockets on client errors/close and destroy client when upstream closes to prevent socket leaks.

Changes

Cohort / File(s)	Summary
Hard-cap timer & warmup test/js/web/fetch/fetch-proxy-tls-intern-race-fixture.ts	Added shared AbortController (hard cap) that aborts the driver's in-flight fetch() after HARD_CAP_MS. Reordered flow to start the hard-cap timer, run driver, perform a warmup loop waiting for driverOk > 0, then launch probe(). Removed prior sleep-based tail termination and added signal: driverAbort.signal to driver fetch() so await driverDone can finish when aborted.
Proxy socket lifecycle fixes test/js/web/fetch/fetch-proxy-tls-intern-race.test.ts	Replaced shadowed local upstream socket with an outer upstream variable. Ensure upstream is destroyed when client emits error/close, and destroy client when upstream emits close, preventing proxy→backend socket leaks during early probe aborts.

🚥 Pre-merge checks | ✅ 2

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title directly and specifically addresses the main fix: deflaking a test timeout issue on Linux release builds by implementing a hard-cap mechanism and proxy socket teardown improvements.
Description check	✅ Passed	The description comprehensively covers the template sections (What does this PR do? and How did you verify?) with detailed root cause analysis, specific fixes, and thorough test plan results.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

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

[claude]

LGTM — test-only deflake with clear root cause and correct bidirectional socket teardown.

Extended reasoning...

Overview

This PR modifies two test files only: the CONNECT proxy fixture and its test harness. No production code is touched.

Changes

• Proxy socket leak fix: adds → and → for proper bidirectional teardown. Previously only events were handled, so FIN-closed sockets from aborted probes leaked upstream connections until the backend's FD table saturated.
• Hard cap as immediate timer: replaces with a started before the async loops, ensuring the cap fires even when the event loop is blocked on stalled I/O.
• AbortController on driver fetch: prevents from hanging when the hard cap fires.
• Warmup loop: lets the driver complete one successful request before the probe flood, making the assertion reliable on loaded CI without introducing an unbounded wait (bounded by the hard cap).

Security risks

None — test-only changes with no user-facing or production code modifications.

Level of scrutiny

Low. These are test infrastructure fixes following standard Node.js networking patterns (bidirectional / handling on piped sockets). The logic is straightforward and the root cause is well-documented.

Other factors

The fix has been validated with 20× sequential and 18× parallel runs on release builds plus debug+ASAN, all passing. No CODEOWNER files cover test helpers. No outstanding reviewer comments.

[claude]

LGTM — test-only fix with correct socket lifecycle management and bounded hard-cap timer.

Extended reasoning...

Overview

This PR modifies two test files only: fetch-proxy-tls-intern-race-fixture.ts and fetch-proxy-tls-intern-race.test.ts. No production code is touched. The changes fix a 41/200 flake rate on Linux release CI targets.

Security risks

None — both files are test infrastructure. No auth, crypto, or permission-sensitive code is involved.

Level of scrutiny

Low. The root cause is clearly identified (missing close handler on the CONNECT proxy causing upstream socket leaks that saturate the backend's FD table), the fix is mechanically sound, and each individual change is straightforward:

• Promoting upstream to outer scope to allow reference in early-registered handlers
• Adding client.on('close', () => upstream?.destroy()) with correct optional chaining (upstream may not be assigned yet)
• Adding upstream.on('close', () => client.destroy()) for symmetric teardown
• Replacing await Bun.sleep(HARD_CAP_MS) + post-sleep stop=true with an upfront setTimeout + AbortController so the cap is authoritative even if the await-loop is blocked
• Adding a warmup phase bounded by the same hard-cap timer

Other factors

No bugs were found by the automated system. CodeRabbitAI found no actionable issues. The author ran 20× sequential, 18× parallel, and 4× debug+ASAN test runs — all passing. The build failures in the robobun comment are against the initial commit (9fdbf165) and are unrelated Zig/libuv infrastructure issues; the latest commit (c427de9) fixes the unref and drops the dead clearTimeout.