Node.js v26.13.0 documentation

Benchmark runner#

import { bench, suite } from 'node:bench';
const { bench, suite } = require('node:bench');
javascriptcopy

Example benchmark#

Save the following as benchmark.mjs:

import { bench, suite } from 'node:bench';

suite('URL', () => {
  const input = 'https://example.com/a?b=c';

  bench('construct', {
    samples: 30,
    params: { input: 'short' },
  }, (b) => {
    const operations = 10_000;
    let totalLength = 0;

    b.start();
    for (let i = 0; i < operations; i++) {
      totalLength += new URL(input).href.length;
    }
    b.end(operations);

    if (totalLength !== operations * input.length) {
      throw new Error('Unexpected URL result');
    }
  });
});
mjscopy

Run the benchmark from the command line:

node --experimental-bench --bench benchmark.mjs
consolecopy

Measurement model#

Each warmup and measured sample invokes the benchmark function once with a fresh {BenchContext}. The function must either call context.start() and context.end(operations) exactly once, or call context.record(sample) exactly once to provide an externally measured sample. Setup before start() and cleanup after end() are outside the measured region. Promise-returning functions are awaited.

By default, ten unreported warmup invocations run before the measured samples so that early samples are less likely to be taken before the measured code has been optimized. A fixed warmup count does not guarantee a stable optimization state; inspect raw samples and increase warmup when early samples are consistently slower. Set warmup to 0 to measure from the first invocation.

Measurement integrity#

A statistically consistent result does not prove that a benchmark measured the intended work. An optimizing runtime can remove work whose result is unused or specialize it more narrowly than the workload being modeled. Framework and loop overhead can also dominate operations that are too short. To reduce these risks:

• Make values produced by measured work observable outside the measured interval, for example by validating an aggregate derived from every result. Passing them only through unused local computations is insufficient.
• Perform enough operations in each sample to amortize fixed timer reads and calls to context.start() and context.end(). If loop bookkeeping is material relative to one operation, batch multiple operations per iteration and report the total operation count.
• Inspect raw samples for trends that indicate insufficient warmup or optimization tiering, pauses consistent with garbage collection, and multimodal distributions.
• Validate surprising results with an independent benchmark shape that performs the same intended work differently.

Dynamic sampling and variable batches#

Calling context.done() during a measured sample completes the benchmark after that sample. This allows a higher-level tool to treat samples as a maximum and implement a dynamic sampling policy.

The number of operations can differ between samples. Summary statistics treat each sample's rate as one equally weighted observation. In particular, summary.mean is the arithmetic mean of the per-sample rates. It is not the pooled throughput calculated as:

1_000_000_000 * sum(sample.operations) / sum(sample.duration_ns)
textcopy

Comparing benchmark results#

node:bench does not designate a benchmark as a baseline or produce a pass/fail comparison between runs. It exposes raw samples, stable benchmark identities, parameters, and tags so that comparison policy can remain in higher-level tools. A tool can use benchId to match the same declaration and parameters across compatible source layouts, and use a tag or its own metadata to identify a baseline.

Reusable runners#

The module-level declaration functions use a shared runner and schedule it automatically. Higher-level tools can create isolated, explicitly started runners instead:

import { createRunner } from 'node:bench';

const runner = createRunner({ yieldBetweenSamples: false });

runner.bench('example', { samples: 100 }, (b) => {
  const operations = chooseOperationCount();
  b.start();
  runOperations(operations);
  const sample = b.end(operations);

  if (hasEnoughData(sample)) b.done();
});

for await (const record of runner.run()) {
  // Consume structured benchmark records.
}
mjscopy

Command-line runner#

The --bench flag runs one or more explicit benchmark files or glob patterns:

node --experimental-bench --bench benchmark.mjs
node --experimental-bench --bench --bench-reporter=json 'benchmarks/**/*.js'
consolecopy

Files are sorted and executed serially. The default --bench-isolation=process mode runs each file in a separate child process and emits one aggregate summary. Structured events are transferred to the parent without JSON conversion, preserving BigInt durations, errors, and parameter values. Child writes to stdout and stderr are emitted as diagnostic records so they do not corrupt reporter output.

--bench-isolation=none imports all files into the runner process. This mode has lower startup overhead, but module, heap, and process state carry between files, and user writes share stdout and stderr with reporters.

Worker-thread isolation is not a CLI mode. Each newly constructed <Worker> has a separate V8 isolate, JavaScript heap, and event loop, typically with lower startup cost than a child process. Reusing a worker preserves its module and heap state. Workers also share libuv's process-wide thread pool and can share process-global native or addon state, so they do not provide the same boundary as process isolation.

Higher-level tools can experiment with worker isolation by loading benchmark code inside a worker, measuring there, transferring structured sample data, and passing it to context.record(). The reported duration_ns can exclude message transport when the worker captures both timestamps. Tools should identify worker modules and workloads explicitly. They should not stringify arbitrary functions or closures to move them between isolates, because closures cannot be reconstructed with their original lexical environment.

Benchmark files passed to --bench should declare benchmarks but must not call run(). The CLI supports --bench-name-pattern, --bench-samples, --bench-warmup, --bench-reporter, and --bench-reporter-destination. See the command-line options documentation for details.

Benchmark reporters#

The built-in reporters are available from the scheme-only node:bench/reporters module:

import { json, spec } from 'node:bench/reporters';
const { json, spec } = require('node:bench/reporters');
javascriptcopy

Reporter values can be passed directly to stream.compose():

import { bench, run } from 'node:bench';
import { spec } from 'node:bench/reporters';
import process from 'node:process';

bench('example', (b) => {
  b.start();
  doWork();
  b.end(1);
});

run().compose(spec).pipe(process.stdout);
mjscopy

The spec reporter buffers results and outputs a concise table containing the sample count, mean rate, 95% confidence interval for the mean, median rate, and warnings. A coefficient of variation above 5% is reported as noisy, and an absolute skewness above 1 is reported as skewed. The exact human-readable format is subject to change.

The json reporter emits every lifecycle record as newline-delimited JSON. BigInt values, including duration_ns, are encoded as decimal strings. Errors are represented using their name, message, stack, code, cause, and errors properties. As required by JSON, non-finite numbers are encoded as null.

import { run } from 'node:bench';
import process from 'node:process';

async function* names(source) {
  for await (const { type, data } of source) {
    if (type === 'bench:complete') {
      yield `${data.name}\n`;
    }
  }
}

run().compose(names).pipe(process.stdout);
mjscopy

createRunner([options])#

• options <Object>
  • yieldBetweenSamples <boolean> Schedule an event loop turn between sample callbacks. Disabling this also prevents timer-based abort signals from firing between synchronous callbacks. Benchmark timeouts continue to be checked against a monotonic deadline. Default: true.
• Returns: <Object> An isolated benchmark runner with bound after, afterEach, before, beforeEach, bench, describe, run, and suite functions.

Creates an explicitly started benchmark runner. Declarations made through one runner do not interact with declarations made through another runner or through the module-level functions. Call the returned run() function to start the runner and obtain its {BenchmarksStream}.

bench([name][, options], fn)#

• name <string> The benchmark name. Default: The name property of fn, or '<anonymous>' when fn has no name.
• options <Object>
  • diagnosticChannels <Array> String diagnostics channel names, deduplicated and inherited from containing suites by union. Symbol values in the array are silently ignored. Default: [].
  • only <boolean> When any benchmark or containing suite has only set, benchmarks without only in their hierarchy are skipped. Default: false.
  • params <Object> String, finite number, or boolean metadata identifying this benchmark configuration. Parameter keys are sorted when constructing the stable benchmark identity. Default: An empty object.
  • samples <number> The maximum number of measured callback invocations. Must be a positive 32-bit unsigned integer. The benchmark may finish earlier by calling context.done(). Default: 30.
  • signal <AbortSignal> Allows aborting this benchmark.
  • skip <boolean> | <string> If truthy, the benchmark is skipped. A string is included in the result as the skip reason. Default: false.
  • tags <string>[] Labels associated with the benchmark. Tags are lowercased, deduplicated, and inherited from containing suites by union. Default: [].
  • timeout <number> The number of milliseconds after which the benchmark fails. Default: Infinity.
  • warmup <number> The number of unreported callback invocations before measured samples. Must be a 32-bit unsigned integer. Default: 10.
• fn <Function> | <AsyncFunction> The benchmark function. It receives a {BenchContext}.
• Returns: <Promise> Fulfilled with the benchmark result after a top-level benchmark finishes, or with undefined immediately when declared in a suite.

Warmup invocations use the same callback and timing contract as measured samples, but their samples are discarded. An exception, rejection, timeout, abort, missing timing call, or duplicate timing call stops the current benchmark. Later benchmarks continue to run.

After a timeout or abort, the runner briefly waits for asynchronous benchmark work to settle before continuing. If it remains pending, all later benchmarks that were selected to run fail without running so that their measurements cannot overlap with that work.

For each warmup and measured callback, the runner subscribes to the configured diagnostics channels. Each publication queues a context diagnostic whose message is { name, message }, containing the string channel name and the published message. Subscriptions are removed when the callback settles or is aborted.

A timeout or abort cannot interrupt synchronous JavaScript and does not forcibly cancel asynchronous work that ignores context.signal.

The benchId is based on the declaration source file, hierarchical suite and benchmark names, and canonicalized parameters. It is stable for repeated runs from the same source location, but the embedded source value is not normalized across checkout roots, module formats, operating systems, or path casing.

bench.skip([name][, options], fn)#

bench.only([name][, options], fn)#

suite([name][, options], fn)#

• name <string> The suite name. Default: The name property of fn, or '<anonymous>' when fn has no name.
• options <Object>
  • diagnosticChannels <Array> String diagnostics channel names inherited by nested suites and benchmarks. Symbol values in the array are silently ignored. Default: [].
  • only <boolean> Selects all benchmarks nested in this suite. Default: false.
  • skip <boolean> | <string> Skips all benchmarks nested in this suite. Default: false.
  • tags <string>[] Labels inherited by nested suites and benchmarks. Default: [].
• fn <Function> | <AsyncFunction> A function that declares nested suites, benchmarks, and hooks.
• Returns: <Promise> Fulfilled when a top-level suite finishes, or with undefined immediately when declared in another suite.

describe([name][, options], fn)#

before(fn)#

• fn <Function> | <AsyncFunction> The hook function.

after(fn)#

• fn <Function> | <AsyncFunction> The hook function.

beforeEach(fn)#

• fn <Function> | <AsyncFunction> The hook function. It receives an object with the benchmark's name, params, and signal.

afterEach(fn)#

• fn <Function> | <AsyncFunction> The hook function. It receives an object with the benchmark's name, params, and signal.

run([options])#

• options <Object>
  • namePattern <string> | <RegExp> Only runs benchmarks whose full hierarchical name matches the pattern. String values are interpreted as JavaScript regular expressions.
  • samples <number> Overrides the maximum number of measured callback invocations for every benchmark. Must be a positive 32-bit unsigned integer.
  • signal <AbortSignal> Allows aborting in-progress benchmark execution.
  • warmup <number> Overrides the number of unreported warmup callback invocations for every benchmark. Must be a 32-bit unsigned integer.
  • yieldBetweenSamples <boolean> Schedule an event loop turn between sample callbacks. Default: true, or the value passed to createRunner() for an explicit runner.
• Returns: {BenchmarksStream}

Returns the object-mode event stream for the in-process benchmark run. Call run() during the same turn in which benchmarks are declared, before automatic execution begins. Calling run() is optional when the returned stream is not needed. An explicit runner created by createRunner() does not run automatically, so its run() function may be called later.

import { bench, run } from 'node:bench';

bench('example', { samples: 3 }, (b) => {
  b.start();
  doWork();
  b.end(1);
});

for await (const { type, data } of run()) {
  if (type === 'bench:complete' && data.error === undefined) {
    console.log(data.name, data.summary.mean);
  }
}
mjscopy

runFile(path[, options])#

• path <string> | <Buffer> | <URL> The path of one benchmark module.
• options <Object>
  • env <Object> The child process environment. Property values must be strings or undefined. This replaces, rather than extends, the parent environment. Default: A snapshot of process.env.
  • execArgv <string>[] Node.js command-line options for the child process. This replaces, rather than extends, inherited options. Benchmark runner options, positional arguments, and options that select another execution mode are not allowed. Default: Compatible options inherited from the current process.
  • signal <AbortSignal> Terminates the child process when aborted.
• Returns: {BenchmarksStream}

Runs exactly one benchmark module in a fresh child process and returns its object-mode event stream. A relative path is resolved from the current working directory when runFile() is called. path is not interpreted as a glob. Unless the signal is aborted or the stream is destroyed before startup, every call uses a new child. Input discovery, ordering, concurrency, retries, and multi-file scheduling remain the caller's responsibility.

When the Permission Model is enabled, the caller must have file system read access to path and permission to create child processes.

Records use advanced child process serialization, preserving supported structured values such as bigint and errors. Child writes to stdout and stderr become 'bench:diagnostic' records. A permission failure, module loading error, abnormal child exit, or cancellation also emits an error diagnostic and produces a terminal 'bench:summary' whose success property is false; these execution failures do not error the stream. If module evaluation fails after declaring benchmarks, those declarations still run before the unsuccessful summary.

Class: BenchContext#

context.index#

• <number>

context.name#

• <string>

context.params#

• <Object>

context.phase#

• <string>

context.signal#

• <AbortSignal>

context.start()#

context.end(operations[, options])#

• operations <number> The number of completed operations. Must be a positive safe integer.
• options <Object>
  • detail <any> Additional structured-cloneable sample data. With CLI process isolation, it must also be supported by advanced child process serialization.
• Returns: <Object> The sample's operations, duration_ns, computed rate, and optional cloned detail.

context.record(sample)#

• sample <Object>
  • operations <number> The number of completed operations. Must be a positive safe integer.
  • duration_ns <bigint> An externally measured positive duration in nanoseconds no greater than Number.MAX_SAFE_INTEGER.
  • detail <any> Additional structured-cloneable sample data. With CLI process isolation, it must also be supported by advanced child process serialization.
• Returns: <Object> The normalized sample, including its computed rate and optional cloned detail.

context.diagnostic(message[, options])#

• message <any> A structured-cloneable diagnostic value. With CLI process isolation, it must also be supported by advanced child process serialization.
• options <Object>
  • level <string> Either 'info' or 'warning'. Default: 'info'.
  • detail <any> Additional structured-cloneable diagnostic data. With CLI process isolation, it must also be supported by advanced child process serialization.
• Returns: <undefined>

Queues a diagnostic associated with the current benchmark, phase, and sample index. Multiple diagnostics preserve call order. They are emitted after the sample callback settles and before that sample's 'bench:sample' event. Warmup diagnostics are emitted even though warmup samples are not. Diagnostics queued before a callback failure are emitted before the failed 'bench:complete' event and do not themselves cause the benchmark to fail. If a timeout or abort wins before the callback settles, queued diagnostics might not be emitted.

context.done()#

Class: BenchmarksStream#

BenchmarksStream is an object-mode <stream.Readable>. Each lifecycle record is both emitted as a named event and made available on the stream as { type, data }.

The events are emitted in execution order:

• 'bench:plan'
• 'bench:start'
• 'bench:sample'
• 'bench:complete'
• 'bench:diagnostic'
• 'bench:summary'

Named event payloads, readable records, and benchmark completion values are independent snapshots. Mutating a value received through one delivery mechanism does not change values received through the others. As with other <EventEmitter> events, multiple listeners for the same named event receive the same event payload. Memory referenced through a <SharedArrayBuffer> remains shared, following structured clone semantics.

Once a consumer starts reading, the runner honors the stream's object-mode high-water mark and waits between records when the consumer is slower than the producer. These waits occur after sample timing has ended, and records are not dropped. Snapshot creation and delivery waits are excluded from benchmark timeout accounting. Before readable consumption starts, records accumulate in the standard readable buffer and are included in readableLength. This keeps an unread stream and a consumer using only named events from deadlocking, but the buffer can grow without bound. A named-event-only consumer that does not need readable records should call stream.resume() to discard them. Destroying the stream stops readable delivery but does not cancel benchmark execution, so benchmark completion promises still settle. Automatically scheduled module-level runs drain their stream internally.

With process isolation, each record sent by a child is acknowledged only after the parent has accepted it. A child sends no additional record until it receives that acknowledgement, bounding the IPC relay when a reporter is slow.

Every benchmark-scoped event contains runId, fileRunId, entryFile, benchId, parentId, and namePath. runId and fileRunId are opaque and change between runs. entryFile identifies the top-level benchmark file whose loading caused the declaration, while file identifies the source location of the declaration itself. parentId is based on the containing suite's source file and hierarchical name path.

After asynchronous suite declarations settle, an in-process runner emits one 'bench:plan' event for every benchmark it collected, in declaration order. All plans from that runner are emitted before its suite hooks or benchmark callbacks run. With process isolation, files run in separate children, so plans for a later file are emitted after an earlier child has completed. With no isolation, all files share one runner and their plans are emitted before any benchmark executes. Plan data contains the benchmark-scoped identity, location, tags, and parameters described in benchmark result, together with:

• diagnosticChannels <string>[] The inherited string channel names subscribed to during each callback.
• samples <number> The effective maximum number of measured callback invocations after run-level overrides.
• warmup <number> The effective number of unreported warmup callback invocations after run-level overrides.
• timeout <number> | <null> The timeout in milliseconds, or null when no timeout is configured.
• yieldBetweenSamples <boolean> Whether an event loop turn is scheduled between sample callbacks.
• selected <boolean> Whether the benchmark is eligible to run after applying skip, only, and namePattern selection. Execution can still be prevented by a duplicate declaration, suite build, hook, abort, or other runtime failure.
• skip <boolean> | <string> When selected is false, the explicit skip value or the selection reason, such as 'only' or 'name pattern'.

The plan contains execution settings known to the runner. Runtime version, operating system, processor, and other environment metadata are intentionally left for reporters and higher-level tools to collect.

Sample result#

• operations <number> The positive operation count passed to context.end() or context.record().
• duration_ns <bigint> The measured duration in nanoseconds.
• rate <number> Operations per second.
• detail <any> The optional cloned sample detail.

Benchmark result#

A completed benchmark result contains:

• runId <string> The opaque logical run identity.
• fileRunId <string> The opaque file runner or child execution identity.
• entryFile <string> | <null> The top-level file that caused this declaration.
• benchId <string> The stable declaration identity within the same source layout.
• parentId <string> | <null> The stable containing suite identity.
• name <string> The benchmark name.
• namePath <string>[] The hierarchical suite and benchmark names.
• file <string> The declaration source file.
• line <number> The source line.
• column <number> The source column.
• tags <string>[] The inherited canonical tags.
• params <Object> The canonical parameter metadata.
• samples <Object>[] The exact measured samples in measurement invocation order.
• summary <Object>
  • mean <number> The equally weighted arithmetic mean of per-sample rates, not pooled throughput across all operations and durations.
  • median <number> The median per-sample rate.
  • min <number> The minimum per-sample rate.
  • max <number> The maximum per-sample rate.
  • stddev <number> The population standard deviation of rates.
  • coefficientOfVariation <number> stddev / mean.
  • confidenceInterval <Object> The 95% Student's t confidence interval for the mean rate, with lower and upper properties.
  • medianConfidenceInterval <Object> The 95% nonparametric confidence interval for the median rate, with lower and upper properties.
  • skewness <number> The skewness of the scaled rate histogram.