A Node.js and Deno transform stream for throttling bandwidth which distributes available bandwidth evenly between all requests in a "group", accurately simulating the effect of network conditions on simultaneous overlapping requests.
- Idiomatic pipeable Transform API for use in Node.js
- Idiomatic pipeable TransformStream API for use in Deno
- Distributes the desired bandwidth evenly over each second
- Distributes the desired bandwidth evenly between all active requests
- Abortable requests ensure bandwidth is redistributed if a client aborts a request
- Node.js Installation
- Deno Installation
- Usage
- Configuration Options
- Dynamic Configuration
- Aborted Requests
- Repo Structure
Firstly, install the package using your package manager of choice.
npm install bandwidth-throttle-stream
You may then import the createBandwidthThrottleGroup()
factory function into your project.
import {createBandwidthThrottleGroup} from 'bandwidth-throttle-stream';
In Deno, all modules are imported from URLs as ES modules. Versioned releases of bandwidth_throttle_stream
are available from deno.land/x. Note that as per Deno convention, the package name is delineated with underscores (_
).
import {createBandwidthThrottleGroup} from 'https://deno.land/x/bandwidth_throttle_stream/mod.ts';
The above URL will return the latest release, but it is strongly advised to lock your import to a specific version using the following syntax, where the x.y.z
semver can be any published version of the library:
import {createBandwidthThrottleGroup} from 'https://deno.land/x/[email protected]/mod.ts';
Using the imported createBandwidthThrottleGroup
factory function, we must firstly create a "bandwidth throttle group" which will be configured with a specific throughput in bytes (B) per second.
// Create a group with a configured available bandwidth in bytes (B) per second.
const bandwidthThrottleGroup = createBandwidthThrottleGroup({
bytesPerSecond: 500000 // 500KB/s
});
Typically we would create a single group only for a server running a simulation, which all incoming network requests to be throttled are routed through. However, we could also create multiple groups if we wanted to run multiple simulations with different configurations on a single server.
Once we've created a group, we can then attach individual pipeable "throttles" to it, as requests come into our server.
The most simple integration would be to insert the throttle (via .pipe
, or .pipeThrough
) between a readable stream (e.g file system readout, server-side HTTP response), and the response stream of the incoming client request to be throttled.
// Attach a throttle to a group (e.g. in response to an incoming request)
const throttle = bandwidthThrottleGroup.createBandwidthThrottle(contentLength);
// Throttle the response by piping a `stream.Readable` to a `stream.Writable`
// via the throttle
someReadableStream
.pipe(throttle)
.pipe(someWritableStream);
// Attach a throttle to a group (e.g. in response to an incoming request)
const throttle = bandwidthThrottleGroup.createBandwidthThrottle(contentLength);
// Throttle the response by piping a `ReadableStream` to a `ReadableStreamDefaultReader`:
const reader = someReadableStream
.pipeThrough(throttle)
.getReader()
Note that a number value for contentLength
(in "bytes") must be passed when creating an individual throttle. This should be the total size of data for the request being passed through the throttle, and is used to allocate memory upfront in a single Uint8Array
typed array, thus preventing expensive GC calls as backpressure builds up. When throttling HTTP requests, contentLength
can be obtained from the 'content-length'
header, once the headers of the request have arrived:
const contentLength = parseInt(req.get('content-length'))
const { body, headers } = await fetch(destination);
const contentLength = parseInt(headers.get("content-length"));
If no contentLength
value is available (e.g. the underlying server does not implement a content-length
header), then it should be set to a value no smaller than the size of largest expected request. To keep memory usage within reason, arbitrarily high values should be avoided.
We may want to perform some specific logic once a request is complete, and all data has been processed through the throttle.
In Node.js, rather than piping directly to a response, we can use the done
event to manually write data, and the end
event to manually handled completion.
request
.pipe(throttle)
.on('data', chunk => response.write(chunk)
.on('end', () => {
response.end();
// any custom completion logic here
});
In Deno, the call to request.respond()
returns a promise which resolves once the request is completed and all data has been pulled into the body
reader.
import {readerToDenoReader} from 'https://deno.land/x/[email protected]/mod.ts';
const reader = request
.pipeThrough(throttle)
.getReader()
await request.respond({
status: 200
body: readerToDenoReader(reader, contentLength),
});
// any custom completion logic here
Note that in the Deno example above, a reader may be passed directly to request.respond()
allowing real-time streaming of the throttled output. However, the Deno std
server expects a Deno.Reader
as a body
(rather than the standard ReadableStreamDefaultReader
), meaning that conversion is needed between the two.
The readerToDenoReader
util is exposed for this purpose, and must be provided with both a reference to ReadableStreamDefaultReader
(reader
), and the contentLength
of the request.
Each bandwidth throttle group accepts an optional object of configuration options:
const bandwidthThrottleGroup = createBandwidthThrottleGroup({
bytesPerSecond: 500000 // 500KB/s,
ticksPerSecond: 20 // aim to write output 20x per second
});
The following options are available.
interface IConfig {
/**
* The maximum number of bytes allowed to pass through the
* throttle, each second.
*
* @default Infinity
*/
bytesPerSecond?: number;
/**
* Defines how frequently the processing of bytes should be
* distributed across each second. Each time the internal
* scheduler "ticks", data will be processed and written out.
*
* A higher value will ensure smoother throttling for requests
* that complete within a second, but will be more expensive
* computationally and will ultimately be constrained by the
* performance of the JavaScript runtime.
*
* @default 40
*/
ticksPerSecond?: number;
}
A group can be reconfigured at any point after creation via its .configure()
method, which accepts the same configuration interface as the createBandwidthThrottleGroup()
factory.
// Create a group with no throttling
const bandwidthThrottleGroup = createBandwidthThrottleGroup();
// ... after some configuration event:
bandwidthThrottleGroup.configure({
bytesPerSecond: 6000000
})
When a client aborts a requests, its important that we also abort the throttle, ensuring the group can re-balance available bandwidth correctly, and backpressure buffer memory is released.
const throttle = bandwidthThrottleGroup.createBandwidthThrottle(contentLength);
request.on('aborted', () => {
// Client aborted request
throttle.abort();
});
request
.pipe(throttle)
.pipe(response);
const throttle = bandwidthThrottleGroup.createBandwidthThrottle(contentLength);
const reader = request
.pipeThrough(throttle)
.getReader()
try {
await request.respond({
status: 200
body: readerToDenoReader(reader, contentLength),
});
} catch(err) {
// request aborted or failed
throttle.abort();
}
This repository contains shared source code for consumption by both Deno (TypeScript ES modules), and Node.js (JavaScript Common.js modules).
Wherever a Deno or Node.js specific API is needed, a common abstraction is created that can be swapped at build time. Platform specific implementations are denoted with either a .deno.ts
or .node.ts
file extension. Platform specific entry points to these abstractions reside in the lib/Platform/
directory.
The source code (contained in the lib/
directory) is ready for direct consumption by Deno is written in ESNext TypeScript, but requires some modifications to produce Node.js compatible NPM distribution code.
The Node.js build process comprises the following steps:
- Copy all contents of
lib/
tosrc/
(git ignored) - Remove all
.ts
file extensions from modules insrc/
(seescipts/replace.ts
) - Replace any imports from
src/Platform/*
with a@Platform
alias (seescipts/replace.ts
) - Run
tsc
on contents ofsrc/
using thets-transform-paths
plugin to replace@Platform
alias with Node.js entry points. - Output compiled, Common.js code to
dist/
(git ignored), and publishdist/
to NPM.