11 KiB
fast-redact
very fast object redaction
Default Usage
By default, fast-redact serializes an object with JSON.stringify, censoring any
data at paths specified:
const fastRedact = require('fast-redact')
const fauxRequest = {
headers: {
host: 'http://example.com',
cookie: `oh oh we don't want this exposed in logs in etc.`,
referer: `if we're cool maybe we'll even redact this`
}
}
const redact = fastRedact({
paths: ['headers.cookie', 'headers.referer']
})
console.log(redact(fauxRequest))
// {"headers":{"host":"http://example.com","cookie":"[REDACTED]","referer":"[REDACTED]"}}
API
require('fast-redact')({paths, censor, serialize}) => Function
When called without any options, or with a zero length paths array,
fast-redact will return JSON.stringify or the serialize option, if set.
paths – Array
An array of strings describing the nested location of a key in an object.
The syntax follows that of the EcmaScript specification, that is any JavaScript
path is accepted – both bracket and dot notation is supported. For instance in
each of the following cases, the c property will be redacted: a.b.c,a['b'].c,
a["b"].c, a[``b``].c. Since bracket notation is supported, array indices are also
supported a[0].b would redact the b key in the first object of the a array.
Leading brackets are also allowed, for instance ["a"].b.c will work.
Wildcards
In addition to static paths, asterisk wildcards are also supported.
When an asterisk is place in the final position it will redact all keys within the
parent object. For instance a.b.* will redact all keys in the b object. Similarly
for arrays a.b[*] will redact all elements of an array (in truth it actually doesn't matter
whether b is in an object or array in either case, both notation styles will work).
When an asterisk is in an intermediate or first position, the paths following the asterisk will be redacted for every object within the parent.
For example:
const fastRedact = require('fast-redact')
const redact = fastRedact({paths: ['*.c.d']})
const obj = {
x: {c: {d: 'hide me', e: 'leave me be'}},
y: {c: {d: 'and me', f: 'I want to live'}},
z: {c: {d: 'and also I', g: 'I want to run in a stream'}}
}
console.log(redact(obj))
// {"x":{"c":{"d":"[REDACTED]","e":"leave me be"}},"y":{"c":{"d":"[REDACTED]","f":"I want to live"}},"z":{"c":{"d":"[REDACTED]","g":"I want to run in a stream"}}}
Another example with a nested array:
const fastRedact = require('..')
const redact = fastRedact({paths: ['a[*].c.d']})
const obj = {
a: [
{c: {d: 'hide me', e: 'leave me be'}},
{c: {d: 'and me', f: 'I want to live'}},
{c: {d: 'and also I', g: 'I want to run in a stream'}}
]
}
console.log(redact(obj))
// {"a":[{"c":{"d":"[REDACTED]","e":"leave me be"}},{"c":{"d":"[REDACTED]","f":"I want to live"}},{"c":{"d":"[REDACTED]","g":"I want to run in a stream"}}]}
remove - Boolean - [false]
The remove option, when set to true will cause keys to be removed from the
serialized output.
Since the implementation exploits the fact that undefined keys are ignored
by JSON.stringify the remove option may only be used when JSON.stringify
is the serializer (this is the default) – otherwise fast-redact will throw.
If supplying a custom serializer that has the same behavior (removing keys
with undefined values), this restriction can be bypassed by explicitly setting
the censor to undefined.
censor – <Any type> – ('[REDACTED]')
This is the value which overwrites redacted properties.
Setting censor to undefined will cause properties to removed as long as this is
the behavior of the serializer – which defaults to JSON.stringify, which does
remove undefined properties.
Setting censor to a function will cause fast-redact to invoke it with the original
value. The output of the censor function sets the redacted value.
Please note that asynchronous functions are not supported.
serialize – Function | Boolean – (JSON.stringify)
The serialize option may be a function of a boolean. If a function is supplied, this
will be used to serialize the redacted object. It's important to understand that for
performance reasons fast-redact mutates the original object, then serializes, then
restores the original values. So the object passed to the serializer is the exact same
object passed to the redacting function.
The serialize option as a function example:
const fastRedact = require('fast-redact')
const redact = fastRedact({
paths: ['a'],
serialize: (o) => JSON.stringify(o, 0, 2)
})
console.log(redact({a: 1, b: 2}))
// {
// "a": "[REDACTED]",
// "b": 2
// }
For advanced usage the serialize option can be set to false. When serialize is set to false,
instead of the serialized object, the output of the redactor function will be the mutated object
itself (this is the exact same as the object passed in). In addition a restore method is supplied
on the redactor function allowing the redacted keys to be restored with the original data.
const fastRedact = require('fast-redact')
const redact = fastRedact({
paths: ['a'],
serialize: false
})
const o = {a: 1, b: 2}
console.log(redact(o) === o) // true
console.log(o) // { a: '[REDACTED]', b: 2 }
console.log(redact.restore(o) === o) // true
console.log(o) // { a: 1, b: 2 }
strict – Boolean - [true]
The strict option, when set to true, will cause the redactor function to throw if instead
of an object it finds a primitive. When strict is set to false, the redactor function
will treat the primitive value as having already been redacted, and return it serialized (with
JSON.stringify or the user's custom serialize function), or as-is if the serialize option
was set to false.
Approach
In order to achieve lowest cost/highest performance redaction fast-redact
creates and compiles a function (using the Function constructor) on initialization.
It's important to distinguish this from the dangers of a runtime eval, no user input
is involved in creating the string that compiles into the function. This is as safe
as writing code normally and having it compiled by V8 in the usual way.
Thanks to changes in V8 in recent years, state can be injected into compiled functions
using bind at very low cost (whereas bind used to be expensive, and getting state
into a compiled function by any means was difficult without a performance penalty).
For static paths, this function simply checks that the path exists and then overwrites with the censor. Wildcard paths are processed with normal functions that iterate over the object redacting values as necessary.
It's important to note, that the original object is mutated – for performance reasons a copy is not made. See rfdc (Really Fast Deep Clone) for the fastest known way to clone – it's not nearly close enough in speed to editing the original object, serializing and then restoring values.
A restore function is also created and compiled to put the original state back on
to the object after redaction. This means that in the default usage case, the operation
is essentially atomic - the object is mutated, serialized and restored internally which
avoids any state management issues.
Caveat
As mentioned in approach, the paths array input is dynamically compiled into a function
at initialization time. While the paths array is vigourously tested for any developer
errors, it's strongly recommended against allowing user input to directly supply any
paths to redact. It can't be guaranteed that allowing user input for paths couldn't
feasibly expose an attack vector.
Benchmarks
The fastest known predecessor to fast-redact is the non-generic pino-noir
library (which was also written by myself).
In the direct calling case, fast-redact is ~30x faster than pino-noir, however a more realistic
comparison is overhead on JSON.stringify.
For a static redaction case (no wildcards) pino-noir adds ~25% overhead on top of JSON.stringify
whereas fast-redact adds ~1% overhead.
In the basic last-position wildcard case,fast-redact is ~12% faster than pino-noir.
The pino-noir module does not support intermediate wildcards, but fast-redact does,
the cost of an intermediate wildcard that results in two keys over two nested objects
being redacted is about 25% overhead on JSON.stringify. The cost of an intermediate
wildcard that results in four keys across two objects being redacted is about 55% overhead
on JSON.stringify and ~50% more expensive that explicitly declaring the keys.
npm run bench
benchNoirV2*500: 59.108ms
benchFastRedact*500: 2.483ms
benchFastRedactRestore*500: 10.904ms
benchNoirV2Wild*500: 91.399ms
benchFastRedactWild*500: 21.200ms
benchFastRedactWildRestore*500: 27.304ms
benchFastRedactIntermediateWild*500: 92.304ms
benchFastRedactIntermediateWildRestore*500: 107.047ms
benchJSONStringify*500: 210.573ms
benchNoirV2Serialize*500: 281.148ms
benchFastRedactSerialize*500: 215.845ms
benchNoirV2WildSerialize*500: 281.168ms
benchFastRedactWildSerialize*500: 247.140ms
benchFastRedactIntermediateWildSerialize*500: 333.722ms
benchFastRedactIntermediateWildMatchWildOutcomeSerialize*500: 463.667ms
benchFastRedactStaticMatchWildOutcomeSerialize*500: 239.293ms
Tests
npm test
224 passing (499.544ms)
Coverage
npm run cov
-----------------|----------|----------|----------|----------|-------------------|
File | % Stmts | % Branch | % Funcs | % Lines | Uncovered Line #s |
-----------------|----------|----------|----------|----------|-------------------|
All files | 100 | 100 | 100 | 100 | |
fast-redact | 100 | 100 | 100 | 100 | |
index.js | 100 | 100 | 100 | 100 | |
fast-redact/lib | 100 | 100 | 100 | 100 | |
modifiers.js | 100 | 100 | 100 | 100 | |
parse.js | 100 | 100 | 100 | 100 | |
redactor.js | 100 | 100 | 100 | 100 | |
restorer.js | 100 | 100 | 100 | 100 | |
rx.js | 100 | 100 | 100 | 100 | |
state.js | 100 | 100 | 100 | 100 | |
validator.js | 100 | 100 | 100 | 100 | |
-----------------|----------|----------|----------|----------|-------------------|
License
MIT
Acknowledgements
Sponsored by nearForm