perf(h1): drop idle-socket timer floor with a ref'd setImmediate

[anonrig]

This relates to...

• Split out of perf(fetch): skip empty RequestInit work #5701 per review
• Idle-socket validation adds ~1.3ms to every keep-alive request reuse (setTimeout(0) in scheduleIdleSocketValidation) #5493 (idle-socket validation ~1.3ms timer floor)
• fix: run idle socket validation off setImmediate instead of setTimeout #5499 / fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 (setImmediate revert due to unref'd Immediate poll stall)
• GHSA-35p6-xmwp-9g52 (response-queue poisoning; validation must stay)

Rationale

Sequential fetch() / HTTP/1.1 keep-alive reuse was paying Node's setTimeout(0) timer floor (~1.3–1.5ms) on every request. That dominates loopback / low-RTT workloads.

#5499 switched this to setImmediate and was reverted in #5606 because an unref'd Immediate lets poll block for ~500ms when the event loop is otherwise idle.

Changes

• Defer idle-socket validation with a ref'd setImmediate so it still runs after poll (poisoning protection) without the 1ms timer floor or the fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 stall.
• Add a fetch keep-alive reuse stall regression test alongside the existing Pool test.
• Add benchmarks/fetch/sequential-keepalive.mjs.

Features

N/A

Bug Fixes

N/A

Breaking Changes and Deprecations

N/A

Benchmarks

Re-run 2026-08-21 on Node v22.14.0 (linux x64, Intel Xeon, 4 cores). Medians of 5 runs vs main @ 181c293.

Sequential keep-alive fetch().text() on loopback (connections: 1, pipelining: 1, 2000 iterations after 300 warmup):

	p50	p90	p99
main (setTimeout(0))	1.47 ms	1.63 ms	1.84 ms
this PR (setImmediate)	0.20 ms	0.29 ms	0.48 ms

~7.5× faster at p50.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.48%. Comparing base (181c293) to head (b5d45df).

Additional details and impacted files

@@ Coverage Diff @@## main #5707 +/- ##
=======================================
Coverage 93.47% 93.48% =======================================
Files 110 110 Lines 38892 38900 +8 =======================================
+ Hits 36354 36365 +11 + Misses 2538 2535 -3 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

lgtm

[mcollina]

Historical note: #5609 proposed the same central fix as this PR.

• fix: run idle socket validation off setImmediate instead of setTimeout #5499 changed idle-socket validation from an unref'd setTimeout(0) to an unref'd setImmediate.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 then proposed dropping .unref() and keeping that setImmediate ref'd, which is the core approach used here.
• fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 instead reverted validation to setTimeout(0) and was merged.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 was later closed unmerged after the author reran the reproduction against main and concluded the stall no longer reproduced: fix(h1): stop unref'ing the idle socket validation immediate #5609 (comment)

Strictly speaking, #5609 was not formally rejected: it received no reviews and was voluntarily closed as apparently obsolete. Its branch was also rebased after #5606, so its final displayed diff no longer clearly represents the original ref'd-Immediate proposal. Nevertheless, #5609 is the closest prior proposal and belongs in the history of this change.

[anonrig]

This relates to...

• Split out of perf(fetch): skip empty RequestInit work #5701 per review
• Idle-socket validation adds ~1.3ms to every keep-alive request reuse (setTimeout(0) in scheduleIdleSocketValidation) #5493 (idle-socket validation ~1.3ms timer floor)
• fix: run idle socket validation off setImmediate instead of setTimeout #5499 / fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 (setImmediate revert due to unref'd Immediate poll stall)
• GHSA-35p6-xmwp-9g52 (response-queue poisoning; validation must stay)

Rationale

Sequential fetch() / HTTP/1.1 keep-alive reuse was paying Node's setTimeout(0) timer floor (~1.3–1.5ms) on every request. That dominates loopback / low-RTT workloads.

#5499 switched this to setImmediate and was reverted in #5606 because an unref'd Immediate lets poll block for ~500ms when the event loop is otherwise idle.

Changes

• Defer idle-socket validation with a ref'd setImmediate so it still runs after poll (poisoning protection) without the 1ms timer floor or the fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 stall.
• Add a fetch keep-alive reuse stall regression test alongside the existing Pool test.
• Add benchmarks/fetch/sequential-keepalive.mjs.

Features

N/A

Bug Fixes

N/A

Breaking Changes and Deprecations

N/A

Benchmarks

Re-run 2026-08-21 on Node v22.14.0 (linux x64, Intel Xeon, 4 cores). Medians of 5 runs vs main @ 181c293.

Sequential keep-alive fetch().text() on loopback (connections: 1, pipelining: 1, 2000 iterations after 300 warmup):

	p50	p90	p99
main (setTimeout(0))	1.47 ms	1.63 ms	1.84 ms
this PR (setImmediate)	0.20 ms	0.29 ms	0.48 ms

~7.5× faster at p50.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.48%. Comparing base (181c293) to head (b5d45df).

Additional details and impacted files

@@ Coverage Diff @@## main #5707 +/- ##
=======================================
Coverage 93.47% 93.48% =======================================
Files 110 110 Lines 38892 38900 +8 =======================================
+ Hits 36354 36365 +11 + Misses 2538 2535 -3 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

lgtm

[mcollina]

Historical note: #5609 proposed the same central fix as this PR.

• fix: run idle socket validation off setImmediate instead of setTimeout #5499 changed idle-socket validation from an unref'd setTimeout(0) to an unref'd setImmediate.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 then proposed dropping .unref() and keeping that setImmediate ref'd, which is the core approach used here.
• fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 instead reverted validation to setTimeout(0) and was merged.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 was later closed unmerged after the author reran the reproduction against main and concluded the stall no longer reproduced: fix(h1): stop unref'ing the idle socket validation immediate #5609 (comment)

Strictly speaking, #5609 was not formally rejected: it received no reviews and was voluntarily closed as apparently obsolete. Its branch was also rebased after #5606, so its final displayed diff no longer clearly represents the original ref'd-Immediate proposal. Nevertheless, #5609 is the closest prior proposal and belongs in the history of this change.

[anonrig]

This relates to...

• Split out of perf(fetch): skip empty RequestInit work #5701 per review
• Idle-socket validation adds ~1.3ms to every keep-alive request reuse (setTimeout(0) in scheduleIdleSocketValidation) #5493 (idle-socket validation ~1.3ms timer floor)
• fix: run idle socket validation off setImmediate instead of setTimeout #5499 / fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 (setImmediate revert due to unref'd Immediate poll stall)
• GHSA-35p6-xmwp-9g52 (response-queue poisoning; validation must stay)

Rationale

Sequential fetch() / HTTP/1.1 keep-alive reuse was paying Node's setTimeout(0) timer floor (~1.3–1.5ms) on every request. That dominates loopback / low-RTT workloads.

#5499 switched this to setImmediate and was reverted in #5606 because an unref'd Immediate lets poll block for ~500ms when the event loop is otherwise idle.

Changes

• Defer idle-socket validation with a ref'd setImmediate so it still runs after poll (poisoning protection) without the 1ms timer floor or the fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 stall.
• Add a fetch keep-alive reuse stall regression test alongside the existing Pool test.
• Add benchmarks/fetch/sequential-keepalive.mjs.

Features

N/A

Bug Fixes

N/A

Breaking Changes and Deprecations

N/A

Benchmarks

Re-run 2026-08-21 on Node v22.14.0 (linux x64, Intel Xeon, 4 cores). Medians of 5 runs vs main @ 181c293.

Sequential keep-alive fetch().text() on loopback (connections: 1, pipelining: 1, 2000 iterations after 300 warmup):

	p50	p90	p99
main (setTimeout(0))	1.47 ms	1.63 ms	1.84 ms
this PR (setImmediate)	0.20 ms	0.29 ms	0.48 ms

~7.5× faster at p50.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.48%. Comparing base (181c293) to head (b5d45df).

Additional details and impacted files

@@ Coverage Diff @@## main #5707 +/- ##
=======================================
Coverage 93.47% 93.48% =======================================
Files 110 110 Lines 38892 38900 +8 =======================================
+ Hits 36354 36365 +11 + Misses 2538 2535 -3 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

lgtm

[mcollina]

Historical note: #5609 proposed the same central fix as this PR.

• fix: run idle socket validation off setImmediate instead of setTimeout #5499 changed idle-socket validation from an unref'd setTimeout(0) to an unref'd setImmediate.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 then proposed dropping .unref() and keeping that setImmediate ref'd, which is the core approach used here.
• fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 instead reverted validation to setTimeout(0) and was merged.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 was later closed unmerged after the author reran the reproduction against main and concluded the stall no longer reproduced: fix(h1): stop unref'ing the idle socket validation immediate #5609 (comment)

Strictly speaking, #5609 was not formally rejected: it received no reviews and was voluntarily closed as apparently obsolete. Its branch was also rebased after #5606, so its final displayed diff no longer clearly represents the original ref'd-Immediate proposal. Nevertheless, #5609 is the closest prior proposal and belongs in the history of this change.

[anonrig]

This relates to...

• Split out of perf(fetch): skip empty RequestInit work #5701 per review
• Idle-socket validation adds ~1.3ms to every keep-alive request reuse (setTimeout(0) in scheduleIdleSocketValidation) #5493 (idle-socket validation ~1.3ms timer floor)
• fix: run idle socket validation off setImmediate instead of setTimeout #5499 / fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 (setImmediate revert due to unref'd Immediate poll stall)
• GHSA-35p6-xmwp-9g52 (response-queue poisoning; validation must stay)

Rationale

Sequential fetch() / HTTP/1.1 keep-alive reuse was paying Node's setTimeout(0) timer floor (~1.3–1.5ms) on every request. That dominates loopback / low-RTT workloads.

#5499 switched this to setImmediate and was reverted in #5606 because an unref'd Immediate lets poll block for ~500ms when the event loop is otherwise idle.

Changes

• Defer idle-socket validation with a ref'd setImmediate so it still runs after poll (poisoning protection) without the 1ms timer floor or the fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 stall.
• Add a fetch keep-alive reuse stall regression test alongside the existing Pool test.
• Add benchmarks/fetch/sequential-keepalive.mjs.

Features

N/A

Bug Fixes

N/A

Breaking Changes and Deprecations

N/A

Benchmarks

Re-run 2026-08-21 on Node v22.14.0 (linux x64, Intel Xeon, 4 cores). Medians of 5 runs vs main @ 181c293.

Sequential keep-alive fetch().text() on loopback (connections: 1, pipelining: 1, 2000 iterations after 300 warmup):

	p50	p90	p99
main (setTimeout(0))	1.47 ms	1.63 ms	1.84 ms
this PR (setImmediate)	0.20 ms	0.29 ms	0.48 ms

~7.5× faster at p50.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.48%. Comparing base (181c293) to head (b5d45df).

Additional details and impacted files

@@ Coverage Diff @@## main #5707 +/- ##
=======================================
Coverage 93.47% 93.48% =======================================
Files 110 110 Lines 38892 38900 +8 =======================================
+ Hits 36354 36365 +11 + Misses 2538 2535 -3 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

lgtm

[mcollina]

Historical note: #5609 proposed the same central fix as this PR.

• fix: run idle socket validation off setImmediate instead of setTimeout #5499 changed idle-socket validation from an unref'd setTimeout(0) to an unref'd setImmediate.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 then proposed dropping .unref() and keeping that setImmediate ref'd, which is the core approach used here.
• fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 instead reverted validation to setTimeout(0) and was merged.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 was later closed unmerged after the author reran the reproduction against main and concluded the stall no longer reproduced: fix(h1): stop unref'ing the idle socket validation immediate #5609 (comment)

Strictly speaking, #5609 was not formally rejected: it received no reviews and was voluntarily closed as apparently obsolete. Its branch was also rebased after #5606, so its final displayed diff no longer clearly represents the original ref'd-Immediate proposal. Nevertheless, #5609 is the closest prior proposal and belongs in the history of this change.

[anonrig]

This relates to...

• Split out of perf(fetch): skip empty RequestInit work #5701 per review
• Idle-socket validation adds ~1.3ms to every keep-alive request reuse (setTimeout(0) in scheduleIdleSocketValidation) #5493 (idle-socket validation ~1.3ms timer floor)
• fix: run idle socket validation off setImmediate instead of setTimeout #5499 / fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 (setImmediate revert due to unref'd Immediate poll stall)
• GHSA-35p6-xmwp-9g52 (response-queue poisoning; validation must stay)

Rationale

Sequential fetch() / HTTP/1.1 keep-alive reuse was paying Node's setTimeout(0) timer floor (~1.3–1.5ms) on every request. That dominates loopback / low-RTT workloads.

#5499 switched this to setImmediate and was reverted in #5606 because an unref'd Immediate lets poll block for ~500ms when the event loop is otherwise idle.

Changes

• Defer idle-socket validation with a ref'd setImmediate so it still runs after poll (poisoning protection) without the 1ms timer floor or the fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 stall.
• Add a fetch keep-alive reuse stall regression test alongside the existing Pool test.
• Add benchmarks/fetch/sequential-keepalive.mjs.

Features

N/A

Bug Fixes

N/A

Breaking Changes and Deprecations

N/A

Benchmarks

Re-run 2026-08-21 on Node v22.14.0 (linux x64, Intel Xeon, 4 cores). Medians of 5 runs vs main @ 181c293.

Sequential keep-alive fetch().text() on loopback (connections: 1, pipelining: 1, 2000 iterations after 300 warmup):

	p50	p90	p99
main (setTimeout(0))	1.47 ms	1.63 ms	1.84 ms
this PR (setImmediate)	0.20 ms	0.29 ms	0.48 ms

~7.5× faster at p50.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.48%. Comparing base (181c293) to head (b5d45df).

Additional details and impacted files

@@ Coverage Diff @@## main #5707 +/- ##
=======================================
Coverage 93.47% 93.48% =======================================
Files 110 110 Lines 38892 38900 +8 =======================================
+ Hits 36354 36365 +11 + Misses 2538 2535 -3 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

lgtm

[mcollina]

Historical note: #5609 proposed the same central fix as this PR.

• fix: run idle socket validation off setImmediate instead of setTimeout #5499 changed idle-socket validation from an unref'd setTimeout(0) to an unref'd setImmediate.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 then proposed dropping .unref() and keeping that setImmediate ref'd, which is the core approach used here.
• fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 instead reverted validation to setTimeout(0) and was merged.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 was later closed unmerged after the author reran the reproduction against main and concluded the stall no longer reproduced: fix(h1): stop unref'ing the idle socket validation immediate #5609 (comment)

Strictly speaking, #5609 was not formally rejected: it received no reviews and was voluntarily closed as apparently obsolete. Its branch was also rebased after #5606, so its final displayed diff no longer clearly represents the original ref'd-Immediate proposal. Nevertheless, #5609 is the closest prior proposal and belongs in the history of this change.

[anonrig]

This relates to...

• Split out of perf(fetch): skip empty RequestInit work #5701 per review
• Idle-socket validation adds ~1.3ms to every keep-alive request reuse (setTimeout(0) in scheduleIdleSocketValidation) #5493 (idle-socket validation ~1.3ms timer floor)
• fix: run idle socket validation off setImmediate instead of setTimeout #5499 / fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 (setImmediate revert due to unref'd Immediate poll stall)
• GHSA-35p6-xmwp-9g52 (response-queue poisoning; validation must stay)

Rationale

Sequential fetch() / HTTP/1.1 keep-alive reuse was paying Node's setTimeout(0) timer floor (~1.3–1.5ms) on every request. That dominates loopback / low-RTT workloads.

#5499 switched this to setImmediate and was reverted in #5606 because an unref'd Immediate lets poll block for ~500ms when the event loop is otherwise idle.

Changes

• Defer idle-socket validation with a ref'd setImmediate so it still runs after poll (poisoning protection) without the 1ms timer floor or the fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 stall.
• Add a fetch keep-alive reuse stall regression test alongside the existing Pool test.
• Add benchmarks/fetch/sequential-keepalive.mjs.

Features

N/A

Bug Fixes

N/A

Breaking Changes and Deprecations

N/A

Benchmarks

Re-run 2026-08-21 on Node v22.14.0 (linux x64, Intel Xeon, 4 cores). Medians of 5 runs vs main @ 181c293.

Sequential keep-alive fetch().text() on loopback (connections: 1, pipelining: 1, 2000 iterations after 300 warmup):

	p50	p90	p99
main (setTimeout(0))	1.47 ms	1.63 ms	1.84 ms
this PR (setImmediate)	0.20 ms	0.29 ms	0.48 ms

~7.5× faster at p50.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.48%. Comparing base (181c293) to head (b5d45df).

Additional details and impacted files

@@ Coverage Diff @@## main #5707 +/- ##
=======================================
Coverage 93.47% 93.48% =======================================
Files 110 110 Lines 38892 38900 +8 =======================================
+ Hits 36354 36365 +11 + Misses 2538 2535 -3 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

lgtm

[mcollina]

Historical note: #5609 proposed the same central fix as this PR.

• fix: run idle socket validation off setImmediate instead of setTimeout #5499 changed idle-socket validation from an unref'd setTimeout(0) to an unref'd setImmediate.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 then proposed dropping .unref() and keeping that setImmediate ref'd, which is the core approach used here.
• fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 instead reverted validation to setTimeout(0) and was merged.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 was later closed unmerged after the author reran the reproduction against main and concluded the stall no longer reproduced: fix(h1): stop unref'ing the idle socket validation immediate #5609 (comment)

Strictly speaking, #5609 was not formally rejected: it received no reviews and was voluntarily closed as apparently obsolete. Its branch was also rebased after #5606, so its final displayed diff no longer clearly represents the original ref'd-Immediate proposal. Nevertheless, #5609 is the closest prior proposal and belongs in the history of this change.

[anonrig]

This relates to...

• Split out of perf(fetch): skip empty RequestInit work #5701 per review
• Idle-socket validation adds ~1.3ms to every keep-alive request reuse (setTimeout(0) in scheduleIdleSocketValidation) #5493 (idle-socket validation ~1.3ms timer floor)
• fix: run idle socket validation off setImmediate instead of setTimeout #5499 / fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 (setImmediate revert due to unref'd Immediate poll stall)
• GHSA-35p6-xmwp-9g52 (response-queue poisoning; validation must stay)

Rationale

Sequential fetch() / HTTP/1.1 keep-alive reuse was paying Node's setTimeout(0) timer floor (~1.3–1.5ms) on every request. That dominates loopback / low-RTT workloads.

#5499 switched this to setImmediate and was reverted in #5606 because an unref'd Immediate lets poll block for ~500ms when the event loop is otherwise idle.

Changes

• Defer idle-socket validation with a ref'd setImmediate so it still runs after poll (poisoning protection) without the 1ms timer floor or the fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 stall.
• Add a fetch keep-alive reuse stall regression test alongside the existing Pool test.
• Add benchmarks/fetch/sequential-keepalive.mjs.

Features

N/A

Bug Fixes

N/A

Breaking Changes and Deprecations

N/A

Benchmarks

Re-run 2026-08-21 on Node v22.14.0 (linux x64, Intel Xeon, 4 cores). Medians of 5 runs vs main @ 181c293.

Sequential keep-alive fetch().text() on loopback (connections: 1, pipelining: 1, 2000 iterations after 300 warmup):

	p50	p90	p99
main (setTimeout(0))	1.47 ms	1.63 ms	1.84 ms
this PR (setImmediate)	0.20 ms	0.29 ms	0.48 ms

~7.5× faster at p50.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.48%. Comparing base (181c293) to head (b5d45df).

Additional details and impacted files

@@ Coverage Diff @@## main #5707 +/- ##
=======================================
Coverage 93.47% 93.48% =======================================
Files 110 110 Lines 38892 38900 +8 =======================================
+ Hits 36354 36365 +11 + Misses 2538 2535 -3 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

lgtm

[mcollina]

Historical note: #5609 proposed the same central fix as this PR.

• fix: run idle socket validation off setImmediate instead of setTimeout #5499 changed idle-socket validation from an unref'd setTimeout(0) to an unref'd setImmediate.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 then proposed dropping .unref() and keeping that setImmediate ref'd, which is the core approach used here.
• fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 instead reverted validation to setTimeout(0) and was merged.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 was later closed unmerged after the author reran the reproduction against main and concluded the stall no longer reproduced: fix(h1): stop unref'ing the idle socket validation immediate #5609 (comment)

Strictly speaking, #5609 was not formally rejected: it received no reviews and was voluntarily closed as apparently obsolete. Its branch was also rebased after #5606, so its final displayed diff no longer clearly represents the original ref'd-Immediate proposal. Nevertheless, #5609 is the closest prior proposal and belongs in the history of this change.

[anonrig]

This relates to...

• Split out of perf(fetch): skip empty RequestInit work #5701 per review
• Idle-socket validation adds ~1.3ms to every keep-alive request reuse (setTimeout(0) in scheduleIdleSocketValidation) #5493 (idle-socket validation ~1.3ms timer floor)
• fix: run idle socket validation off setImmediate instead of setTimeout #5499 / fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 (setImmediate revert due to unref'd Immediate poll stall)
• GHSA-35p6-xmwp-9g52 (response-queue poisoning; validation must stay)

Rationale

Sequential fetch() / HTTP/1.1 keep-alive reuse was paying Node's setTimeout(0) timer floor (~1.3–1.5ms) on every request. That dominates loopback / low-RTT workloads.

#5499 switched this to setImmediate and was reverted in #5606 because an unref'd Immediate lets poll block for ~500ms when the event loop is otherwise idle.

Changes

• Defer idle-socket validation with a ref'd setImmediate so it still runs after poll (poisoning protection) without the 1ms timer floor or the fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 stall.
• Add a fetch keep-alive reuse stall regression test alongside the existing Pool test.
• Add benchmarks/fetch/sequential-keepalive.mjs.

Features

N/A

Bug Fixes

N/A

Breaking Changes and Deprecations

N/A

Benchmarks

Re-run 2026-08-21 on Node v22.14.0 (linux x64, Intel Xeon, 4 cores). Medians of 5 runs vs main @ 181c293.

Sequential keep-alive fetch().text() on loopback (connections: 1, pipelining: 1, 2000 iterations after 300 warmup):

	p50	p90	p99
main (setTimeout(0))	1.47 ms	1.63 ms	1.84 ms
this PR (setImmediate)	0.20 ms	0.29 ms	0.48 ms

~7.5× faster at p50.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.48%. Comparing base (181c293) to head (b5d45df).

Additional details and impacted files

@@ Coverage Diff @@## main #5707 +/- ##
=======================================
Coverage 93.47% 93.48% =======================================
Files 110 110 Lines 38892 38900 +8 =======================================
+ Hits 36354 36365 +11 + Misses 2538 2535 -3 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

lgtm

[mcollina]

Historical note: #5609 proposed the same central fix as this PR.

• fix: run idle socket validation off setImmediate instead of setTimeout #5499 changed idle-socket validation from an unref'd setTimeout(0) to an unref'd setImmediate.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 then proposed dropping .unref() and keeping that setImmediate ref'd, which is the core approach used here.
• fix: revert idle socket validation to setTimeout(0) to prevent stall on idle event loop #5606 instead reverted validation to setTimeout(0) and was merged.
• fix(h1): stop unref'ing the idle socket validation immediate #5609 was later closed unmerged after the author reran the reproduction against main and concluded the stall no longer reproduced: fix(h1): stop unref'ing the idle socket validation immediate #5609 (comment)

Strictly speaking, #5609 was not formally rejected: it received no reviews and was voluntarily closed as apparently obsolete. Its branch was also rebased after #5606, so its final displayed diff no longer clearly represents the original ref'd-Immediate proposal. Nevertheless, #5609 is the closest prior proposal and belongs in the history of this change.