[robobun]

Reproduction

// idleTimeout: 4, handler sleeps 150 ms, 32 requests spaced 130 ms so
// their arrival phase sweeps one full 4 s wheel period.
const server = Bun.serve({
  port: 0, idleTimeout: 4,
  async fetch(req) { await Bun.sleep(150); return new Response("done"); },
});

On 1.4.0 one of those requests is closed with zero response bytes and req.signal aborts ~80 ms after the request arrived, even though nothing was idle for anywhere near 4 s:

requests killed with NO response: 1/32
earliest req.signal abort after request arrival: 83 ms (configured idleTimeout: 4000 ms)

Cause

us_socket_timeout stores s->timeout = (group->timestamp + ceil(seconds/4)) % 240 and the sweep in loop.c fires when timestamp == s->timeout. But timestamp advances on an absolute LIBUS_TIMEOUT_GRANULARITY (4 s) grid that is unrelated to when the socket was armed, so a socket armed t ms before a grid boundary gets its first tick after only t ms. For idleTimeout: N the actual fire window was (4*(k-1), 4*k] with k = ceil(N/4), i.e. idleTimeout: 4 could fire after ~0 ms and every value N could fire up to 4 s early. server.timeout(req, N), Bun.connect socket timeouts and the WebSocket maxLifetime (minute wheel) arm through the same helpers.

Fix

Arm with one extra tick, clamped to the 240-slot wheel, so a timeout of N seconds fires in (ceil(N/4)*4, ceil(N/4)*4 + 4] s and never before N (packages/bun-usockets/src/socket.c). The same adjustment is applied to us_connecting_socket_timeout and the minute-granularity us_socket_long_timeout / us_connecting_socket_long_timeout.

The existing should allow use of custom timeout test stalled for 10 s under an 8 s idleTimeout; that now fires in (8, 12] s so the stall is bumped to 14 s, and the .resolves / .rejects assertions in that helper are now awaited.

Verification

# without packages/ change
(fail) idleTimeout never fires before the configured number of seconds
  { earliestAbort: 31.72, killed: 1 }

# with packages/ change
(pass) idleTimeout never fires before the configured number of seconds [4848ms]

test/js/bun/http/serve.test.ts: 239 pass / 4 pre-existing container-env failures (localhost/IPv6/proxy) that reproduce on main.

[coderabbitai]

Walkthrough

This PR changes how uSockets computes timeout wheel slots, introducing a bounded tick-rounding helper used by socket and connecting-socket timeout/long-timeout setters, updates the related documentation comment, and adjusts/extends HTTP server idle-timeout tests to match the new timing behavior.

Changes

Timeout Wheel Rounding Fix

Layer / File(s)	Summary
Timeout tick helper and doc comment packages/bun-usockets/src/libusockets.h, packages/bun-usockets/src/socket.c	Adds us_internal_timeout_ticks, a shift-based rounding/capping helper for wheel tick offsets, and updates the header comment to describe the exact firing window using ceil(N/4)*4.
Apply helper across timeout setters packages/bun-usockets/src/socket.c	Updates us_socket_timeout, us_connecting_socket_timeout, us_socket_long_timeout, and us_connecting_socket_long_timeout to compute stored slots as (group timestamp + ticks) % 240 via the new helper, replacing prior direct arithmetic.
HTTP server timeout test updates test/js/bun/http/serve.test.ts	Increases handler sleep and test timeouts, tightens response assertions for success/failure paths, and adds a new concurrent test confirming idleTimeout never fires before the configured threshold.

Possibly related PRs

• oven-sh/bun#30376: Modifies HTTP client logic that arms/re-arms idle timers using the same uSockets timeout scheduling mechanism changed in this PR.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	Clearly states the main behavior change: idleTimeout now acts as a floor and won't fire early.
Description check	✅ Passed	It covers the bug, cause, fix, and verification, though it uses different headings than the template.
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.

---

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

[robobun]

^{Updated 4:40 AM PT - Jul 7th, 2026}

❌ @robobun, your commit f1ba741 has 3 failures in Build #69692 (All Failures):

• test/js/bun/webview/webview.test.ts - code 1 on 🍎 14 aarch64
• test/integration/next-pages/test/dev-server.test.ts - code 1 on 🍎 14 aarch64
• test/js/bun/terminal/terminal.test.ts - code 1 on 🍎 14 x64

---

🧪   To try this PR locally:

bunx bun-pr 33610

That installs a local version of the PR into your bun-33610 executable, so you can run:

bun-33610 --bun

[coderabbitai]

Actionable comments posted: 2

🤖 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 `@packages/bun-usockets/src/socket.c`:
- Around line 104-111: The timeout helper in us_internal_timeout_ticks currently
clamps the computed value to 239, which makes the documented maxLifetime value
of 240 fire one slot early. Update the logic so the public cap in App.h and the
wheel-slot handling stay consistent: either reduce the exposed maxLifetime limit
to 239 or allow the 240th slot to be represented explicitly without being
clamped away.

In `@test/js/bun/http/serve.test.ts`:
- Around line 2514-2568: This timing-sensitive idleTimeout test is flaky under
it.concurrent because sibling test contention can delay the event loop and make
a valid request exceed the 4s idle window, causing false failures in the killed
assertion. Move the test to sequential execution with it(...) (or equivalent
non-concurrent form) and set an explicit timeout aligned with the rest of the
serve tests, while keeping the existing idleTimeout/earliestAbort checks in the
same test body.

🪄 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: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 1ab98b36-3aa5-406c-a666-1425b469a89d

📥 Commits

Reviewing files that changed from the base of the PR and between 3f67971 and f1ba741.

📒 Files selected for processing (3)

• packages/bun-usockets/src/libusockets.h
• packages/bun-usockets/src/socket.c
• test/js/bun/http/serve.test.ts

[robobun]

The diff is green on every lane that ran tests; serve.test.ts (which carries the new idleTimeout never fires before test) passes everywhere.

CI is red only on infra and unrelated flakes across two consecutive builds (69673, 69692):

• darwin 26 aarch64 x2: buildkite-agent artifact download timed out after 120s for step 'darwin-aarch64-build-bun' on both builds. The test shard never downloads the binary, so no tests run.
• Flaky-annotated 1-retry failures on Windows / misc lanes (hot.test.ts, bun-install.test.ts, s3.test.ts, postgres-*, bundler_compile.test.ts), none touching uSockets timeouts or serve.test.ts.

Ready for review; needs a maintainer to merge past the darwin artifact-download infra timeout.

[github-actions]

Closing this PR because it has been inactive for more than 90 days.