Fast profiling and benchmarking for Bun.
ostia times commands and functions on Bun,
then tells you whether a change made them slower:
it compares two runs with statistics that account for how noisy your
machine is, and ostia ab confirms a regression in fresh processes
before it reports one.
The CLI and the library do the same things,
and every result is one JSON document.
// suite.ts
import { group, task } from "ostia"
const input = Array.from({ length: 2_000 }, (_, i) => i % 500)
group("dedupe", () => {
task("naive (indexOf scan, O(n²))", () => dedupeNaive(input))
task("Set-based (O(n))", () => [...new Set(input)])
})$ ostia bench suite.ts
Apple M2 · 8 cores · load 3.5 · noise floor 0.4%
Task Median Spread Range Relative
-----------------------------------------------------------------------------------------
dedupe:
naive (indexOf scan, O(n²)) 173.9 µs 180.4 µs…205.8 µs 170.1 µs…635.5 µs 7.40× slower
Set-based (O(n)) 23.5 µs 24.9 µs…72.5 µs 18.7 µs…208.3 µs 1.00×It tells you when a change made something slower.
Change the code, and ostia ab runs your suite against HEAD
and the working tree in alternating batches in one process:
$ ostia ab suite.ts
A/B: working tree vs HEAD (3945768) · 15 rounds · threshold 10% · geomean threshold 1.5%
Task Base Candidate Change p25…p75 Verdict
------------------------------------------------------------------------------------------
dedupe:
naive (indexOf scan, O(n²)) 183.5 µs 178.5 µs -1.9% -3.5%…-0.3%
Set-based (O(n)) 25.2 µs 181.2 µs +618.1% +554.2%…+659.2% regressed, confirmed (repeats: +630.6%, +584.5%)
Geomean +165.4% (threshold 1.5%) · 1 regressed, 0 improved, 1 unchanged of 2 · failExit 1 on a regression, so CI can gate on it. ostia has no dependencies and needs Bun ≥ 1.4.
bun add -d ostiaWrite a suite with group() and task(), as above, then:
ostia bench suite.ts # time each task
ostia ab suite.ts # compare the working tree with HEADExit status: 0 pass, 1 a regression (ab, compare, ci), 2 an error.
# Time two commands against each other
ostia time "bun a.ts" "bun b.ts"
# Compare two saved runs
ostia compare before.json after.json
# Save a baseline on known-good code, then gate CI on it
ostia baseline save
ostia ci
# Capture a CPU profile while timing
ostia time "bun a.ts" --cpu
# One JSON object per line, for a script or an agent
ostia bench suite.ts --format minimalThese docs ship in the package, at node_modules/ostia/docs/.
const doc = await time({
commands: ["bun a.ts", { command: "bun b.ts", label: "b", prepare: "rm -rf dist" }],
samples: 20,
cpu: true,
})
group("parse", () => {
sweep({ size: range(100, 10_000) }, ({ size }) => {
const input = buildInput(size) // unmeasured setup, once per point
task("parse", () => parse(input), { isolate: true })
})
})
const result = compareDocuments(await loadDocument("before.json"), doc)
if (result.summary.verdict === "fail") process.exitCode = 1
const { text } = await renderers.markdown.render(doc, {})- CLI: every command and flag
- Library:
time(),bench(),ab(),profile(), and documents - Config:
ostia.configworkloads, thresholds and baselines - Machine-readable output:
--format minimal, exit codes and error codes, for scripts and agents - Statistics: how a comparison decides, and the noise floor
- Document schema: the JSON a run writes
- Preload recipes: DOM, Svelte and other setups a suite needs