fetch-hooks

Hook a WhatWG-compatible fetch function with customised behaviour, e.g.:

• Handling data: URIs
• Handling file: URLs
• Handling s3: URLs
• Enforcing https: as your transport protocol

More experimentally:

• Dumping curl command lines
• Logging to rsyslog
• Adding your own custom lifecycle hooks

Being considered:

• OpenZipkin B3 propagation

Usage

const { hook, fetch, hooks } = require('fetch-hooks');
const _fetch = hook(fetch, myHook); // construct a hooked fetch
const response = await _fetch(uri); // and use it as if it were normal
console.log(await response.text());

Hooking data: URLs for methods GET and HEAD

Add hooks.data to support data: URIs:

const _fetch = hook(fetch, hooks.data);
const response = await _fetch('data:text/ascii;base64,TUlORCBCTE9XTg==');
console.log(await response.text());

Hooking file: URLs for methods GET and HEAD

Add a hooks.file return value to support S3 bucket access via s3: URIs:

const _fetch = hook(fetch, hooks.file({ baseURI: process.cwd() }));
const response = await _fetch('file:test/data/smiley.txt');
console.log(await response.text());

File hooks are only active for request URIs within baseURI, which defaults to the process' current working directory at request time.

WARNINGS:

Due to quirks of node-fetch 2.1.1 (2018-03-05):

• If you request a relative file: URI, hook will resolve it against the process' current working directory at request time before passing it to the hooks. This is necessary to survive the Request constructor.

• If you call response.text(), all files are read as if encoded in UTF-8, even if they aren't. I'm jamming charset=UTF-8 into the Content-Type as fair warning. Try the node-fetch extensions response.body.buffer and response.body.textConverted() if this doesn't work for you.

Hooking s3: URLs for methods GET, HEAD, PUT, and DELETE

Add a hooks.s3 return value to support S3 bucket access:

// construct an S3 Service object
const { S3 } = require('aws-sdk');
const s3 = new S3({
    region: 'ap-southeast-2',
    signatureVersion: 'v4',
});
 
// attach it to an S3 hook for your bucket's base URL:
const { hook, fetch, hooks } = require('fetch-hooks');
const s3hook = hooks.s3(s3, { baseURI: 's3://bucket', acl: 'private' });
const _fetch = hook(fetch, s3hook);
 
// fetch content from the bucket
const response = await _fetch('s3://bucket/key');
console.log(await response.text());

S3 URIs give the bucket name where you'd expect the host name, consistent with the aws s3 command line.

S3 hooks are only active for request URIs within baseURI, which defaults to s3:/ to match any bucket.

The acl option specifies a canned ACL, and defaults to private.

Enforcing HTTPS

Add hooks.httpsOnly to enforce that you're only willing to speak over the HTTPS protocol.

const { hook, fetch, hooks } = require('fetch-hooks');
const _fetch = hook(fetch, /* other hooks, */ hooks.httpsOnly);

It's safe to use hooks.httpsOnly last in the chain, as the file: and data: hooks will have already responded, and the s3: hook will have changed the request to a signed https: request.

Troubleshooting with curl

Add hooks.curl to write curl commands to standard error:

const _fetch = hook(fetch, /* other hooks, */ hooks.curl);

You can also set the DEBUG environment variable to have useful information dumped to the console. See the debug documentation for more detail.

WARNINGS:

• This part of the API is not yet stable, the input handling in particular. I reserve the right to make breaking changes with minor or patch level version bumps until I see some sign of third party usage. I welcome PRs and suggestions.

• The commands assume any input for PUT, POST etc are in /tmp/input. The hook does not write any such content to /tmp/input.

Logging to a remote syslog server

Add hooks.rsyslog to send packets to an RFC5424 compliant server.

const _fetch = hook(fetch, /* other hooks, */ hooks.rsyslog({
  target_host: '127.0.0.1',
  target_port: 514,
  elide: url => withPartsRemoved(url),
}));

elide is optional. The default for elide will remove, from the URLs sent to the remote syslog:

• The auth component
• The query component
• The data in the pathname component, if the protocol is data:

Otherwise put, the default elide preserves only:

• protocol
• host, which includes the port number
• pathname, unless protocol is data:

I chose a default this conservative so neither of us have to scramble to remove usernames, passwords, and secrets embedded in queries from our log files.

To make your own choices, override elide with a function returning a string given a URL. The following will pass the full URL:

const _fetch = hook(fetch, /* other hooks, */ hooks.rsyslog({
  target_host: '127.0.0.1',
  target_port: 514,
  elide: url => url,
}));

For full documentation of the rest of the options, see the rsyslog package.

WARNINGS:

• This part of the API is not yet stable, the output format in particular. I reserve the right to make breaking changes with minor or patch level version bumps until I see some sign of third party usage. I welcome PRs and suggestions.

• The len= segment requires some guesswork, and might not match the number of bytes on the wire, especially if the server omits or lies about the content-length.

Adding your own lifecycle hooks

Return a function named prereq, postreq, or error to get called either before a request, after a request, or when a hook fails:

• postreq(req, res, err) will be called after a request is made, with either res or err set to null depending on whether the request crashed out or succeeded.

• error(err) will be called if a call to a hook or lifecycle hook function fails.

WARNINGS:

• This part of the API is not yet stable. I reserve the right to make breaking changes with minor or patch level version bumps until I see some sign of third party usage. I welcome PRs and suggestions.

Under the Covers

const { hook, fetch, hooks } = require('fetch-hooks');
const _fetch = hook(fetch /* , ... hooks */);
const response = await _fetch(uri);
console.log(await response.text());

A hooked fetch will construct a Request using node-fetch and then, for each of its hooks:

• Call the hook
• await its return value
• Ignore the hook if the return value is falsey
• Resolve with the return value's response property if found
• Continue with the return value's request property if found

If there are no more hooks, a hooked fetch will:

• Pass through to its upstream fetch (its first argument) if no hooks are left, or
• Reject with an error if its upstream fetch is null

Typings

I've cloned the Microsoft typings for fetch and related types, adapting them for node-fetch. If the typings don't match the node-fetch reality, please open an issue.

Previous experiments that didn't work:

• Relying on the Microsoft typings directly, i.e. requiring dom in compilerOptions.lib in tsconfig.json. The global namespace clutter from dom made it hard to find undefined variables, fetch in particular.

• Relying on @types/node-fetch. I didn't enjoy the mismatch with Microsoft's typings for fetch, and didn't have the time to encourage the breaking changes required to track closer. They're looking better, now, so it might be worth revisiting this in the future.

Background

A few small things bother me about using WhatWG fetch for back end programming:

• I can't use fetch for protocols other than http: and https:, making it hard to use when I'm receiving a trusted URI that could reasonably have a file: or data: protocol.

• If my packages take a fetch as part of their configuration, I need to mock fetch during tests. I'd like that to be easier.

• If they don't, I have to have them take an agent as part of their configuration if they expect HTTPS. I also have to stand up an HTTPS server.

• It'd sometimes be handy to apply different outbound headers for different hosts, but I don't want to make the code making the requests responsible for that application.

This repository is an experiment which might end up in proof by contradiction. My premise is:

• An API-compatible fetch with the ability to register hooks could solve all the above, and possibly open some exciting possibilities.