Skip to content

test(HTMLRewriter): measure kernel peak RSS, not a point sample that straddles a segment decommit #34786

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
robobun wants to merge 2 commits into from
Closed

test(HTMLRewriter): measure kernel peak RSS, not a point sample that straddles a segment decommit #34786

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d4f7c75
test(HTMLRewriter): compare peak RSS, not a point sample that straddl…
robobun Jul 20, 2026
598fcfb
use kernel maxRSS instead of per-pass peak sampling
robobun Jul 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 20 additions & 12 deletions test/js/workerd/html-rewriter-leak.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,11 +47,13 @@ test("onEndTag callbacks are released after the rewrite", () => {
// LOLHTMLContext.deinit() must destroy those allocations. Previously it only
// unprotected the held JSValues and leaked the struct memory.
//
// RSS is a high-water mark — Bun.gc(true) collects every wrapper and its
// lol-html builder, but the allocators don't promptly hand pages back to the
// OS. So warmup runs the *same* workload as the measured phase: the allocator
// footprint is established before the baseline, and any growth past that is
// what's actually retained.
// `process.memoryUsage.rss()` is a point sample; mimalloc hands a ~30 MB
// segment back to the OS on its own cadence between passes, so two point
// samples can straddle that release and read as a +31 MB "delta" with nothing
// retained (observed on the 4-vCPU Windows lane). `resourceUsage().maxRSS` is
// the kernel's monotone peak (VmHWM / PeakWorkingSetSize), so it never dips:
// flat when freed memory is reused, strictly rising (~50 MB over 3 passes)
// when it leaks.
//
// Skipped in debug: at this N a debug pass is ~40s and the extra debug-build
// allocation tracking adds enough RSS noise to drown the signal. CI has no
Expand All @@ -70,19 +72,23 @@ test.skipIf(isDebug)(
}

const N = 4000;
const samples = [];
function pass() {
for (let i = 0; i < N; i++) once();
Bun.gc(true);
return process.memoryUsage.rss();
samples.push(process.memoryUsage.rss());
}

pass(); pass();
const before = pass();
pass(); pass();
const after = pass();
// ru_maxrss is KB on Linux/Windows, bytes on macOS; normalize to bytes.
const maxRSS = () => process.resourceUsage().maxRSS * (process.platform === "darwin" ? 1 : 1024);

for (let i = 0; i < 3; i++) pass();
const before = maxRSS();
for (let i = 0; i < 3; i++) pass();
const after = maxRSS();

process.stdout.write(
JSON.stringify({ before, after, deltaMB: (after - before) / 1024 / 1024 }) + "\\n",
JSON.stringify({ samples, before, after, deltaMB: (after - before) / 1024 / 1024 }) + "\\n",
);
`;

Expand Down Expand Up @@ -111,9 +117,11 @@ test.skipIf(isDebug)(
.trim();
expect(filteredStderr).toBe("");

// Logged so a future red carries the per-pass trace without a repro.
console.log(stdout.trim());
const { deltaMB } = JSON.parse(stdout.trim());

// Unfixed: ~50 MB over 3 measured passes. Fixed: ±1 MB plateau.
// Unfixed: ~50 MB over 3 measured passes. Fixed: <1 MB.
// Threshold sits at ~half the unfixed signal.
expect(deltaMB).toBeLessThan(25);
expect(exitCode).toBe(0);
Expand Down
Loading