@hackylabs/deep-redact

Deep Redact

Faster than Fast Redact ¹ as well as being safer and more configurable than many other redaction libraries, Deep Redact is a zero-dependency tool that redacts sensitive information from strings and objects. It is designed to be used in a production environment where sensitive information needs to be redacted from logs, error messages, files, and other outputs.

Circular references and other unsupported values are handled gracefully, and the library is designed to be as fast as possible while still being configurable.

Supporting both CommonJS and ESM, with named and default exports, Deep Redact is designed to be versatile and easy to use in any modern JavaScript or TypeScript project in Node or the browser.

Installation

npm install @hackylabs/deep-redact

Usage

In order to maintain a consistent usage throughout your project, it is not advised to call this library outside of your global logging/error-reporting libraries.

// ./src/example.ts
import {DeepRedact} from '@hackylabs/deep-redact'; // If you're using CommonJS, import with require('@hackylabs/deep-redact') instead. Both CommonJS and ESM support named and default imports.

const redaction = new DeepRedact({
  blacklistedKeys: ['sensitive', 'password', /name/i],
  serialise: false,
})

const obj = {
  keepThis: 'This is fine',
  sensitive: 'This is not fine',
  user: {
    id: 1,
    password: '<h1><strong>Password</strong></h1>',
    firstName: 'John',
  }
}

redaction.redact(obj)
// {
//  keepThis: 'This is fine',
//  sensitive: '[REDACTED]',
//  user: {
//    id: 1,
//    password: '[REDACTED]',
//    firstName: '[REDACTED]'
//  }
// }

Configuration

Main Options

key	description	type	options	default	required
blacklistedKeys	Deeply compare names of these keys against the keys in your object.	array	Array<string￨RegExp￨BlacklistKeyConfig>	[]	N
stringTests	Array of regular expressions to perform against string values, whether that value is a flat string or nested within an object.	array	RegExp[]	[]	N
fuzzyKeyMatch	Loosely compare key names by checking if the key name of your unredacted object is included anywhere within the name of your blacklisted key. For example, is "pass" (your key) included in "password" (from config).	boolean		false	N
caseSensitiveKeyMatch	Loosely compare key names by normalising the strings. This involves removing non-word characters and transforms the string to lowercase. This means you never have to worry having to list duplicate keys in different formats such as snake_case, camelCase, PascalCase or any other case.	boolean		true	N
remove	Determines whether or not to remove the key from the object when it is redacted.	boolean		false	N
retainStructure	Determines whether or not keep all nested values of a key that is going to be redacted. Circular references are always removed.	boolean		false	N
replacement	When a value is going to be redacted, what would you like to replace it with?	string ￨ function		[REDACTED]	N
replaceStringByLength	When a string value is going to be replaced, optionally replace it by repeating the replacement to match the length of the value. For example, if replaceStringByLength were set to true and replacement was set to "x", then redacting "secret" would return "xxxxxx". This is sometimes useful for debugging purposes, although it may be less secure as it could give hints to the original value.	boolean		false	N
types	JS types (values of typeof keyword). Only values with a typeof equal to string, number, bigint, boolean, symbol, object, or function will be redacted. Undefined values will never be redacted, although the type undefined is included in this list to keep TypeScript happy.	array	Array<'string'￨'number'￨'bigint'￨'boolean'￨'symbol'￨'undefined'￨'object'￨'function'>	['string']	N
serialise	Determines whether or not to serialise the object after redacting. Typical use cases for this are when you want to send it over the network or save to a file, both of which are common use cases for redacting sensitive information.	boolean		false	N
serialize	Alias of serialise for International-English users.	boolean		false	N

BlacklistKeyConfig

key	type	default	required
key	string￨RegExp		Y
fuzzyKeyMatch	boolean	Main options fuzzyKeyMatch	N
caseSensitiveKeyMatch	boolean	Main options caseSensitiveKeyMatch	N
remove	boolean	Main options remove	N
retainStructure	boolean	Main options retainStructure	N

Benchmark

Comparisons are made against JSON.stringify and Fast Redact as well as different configurations of Deep Redact, using this test object. Fast Redact was configured to redact the same keys on the same object as Deep Redact without using wildcards.

The benchmark is run on a 2021 iMac with an M1 chip with 16GB memory running Sonoma 14.5.

JSON.stringify is included as a benchmark because it is the fastest way to deeply iterate over an object although it doesn't redact any sensitive information. Fast-redact is included as a benchmark because it's the next fastest redaction library available. Neither JSON.stringify nor Fast Redact offer the same level of configurability as deep-redact.

scenario	ops / sec	op duration (ms)	margin of error	sample count
JSON.stringify, large object	295500.62	0.0033840876	0.00002	147751
DeepRedact, remove item, single object	36272.4	0.0275691709	0.00016	18137
DeepRedact, custom replacer function, single object	30314.59	0.0329874115	0.00028	15158
DeepRedact, default config, large object	30028.19	0.0333020395	0.0002	15015
DeepRedact, replace string by length, single object	28756.9	0.0347742688	0.00028	14379
DeepRedact, retain structure, single object	24803.01	0.0403176903	0.00032	12402
DeepRedact, fuzzy matching, single object	22243.3	0.0449573621	0.00038	11122
DeepRedact, config per key, single object	21603.85	0.0462880355	0.0013	10802
fast redact, large object	9529.2	0.1049406557	0.00064	4765
DeepRedact, case insensitive matching, single object	6503.72	0.1537581959	0.00105	3252
DeepRedact, default config, 1000 large objects	5915.05	0.1690602382	0.00296	2958
DeepRedact, fuzzy and case insensitive matching, single object	5591.96	0.1788283015	0.00184	2796
JSON.stringify, 1000 large objects	394.41	2.5354059248	0.01001	198
fast redact, 1000 large objects	172.23	5.8060829174	0.06886	87