json-fingerprint

json-fingerprint

Create consistent and comparable fingerprints with secure hashes from unordered JSON data.

A JSON fingerprint consists of three parts: the version of the underlying canonicalization algorithm, the hash function used and a hexadecimal digest of the hash function output. An example output could look like this: jfpv1$sha256$5815eb0ce6f4e5ab0a771cce2a8c5432f64222f8fd84b4cc2d38e4621fae86af.

Fingerprint element	Description
jfpv1	JSON fingerprint version identifier: json fingerprint version 1
$	JSON fingerprint element separator
sha256	Hash function identifier (sha256, sha384 or sha512)
5815eb0c...1fae86af	The secure hash function output in hexadecimal format

Table of Contents

• v1 release checklist (jfpv1)
• Installation
• Examples
  • Create JSON fingerprints
  • Decode JSON fingerprints
  • Match fingerprints
  • Find matches in fingerprint lists
• JSON normalization
  • Alternative specifications
  • JSON Fingerprint v1 (jfpv1)
• Performance
  • Example 1: flat data structures
  • Example 2: nested data structures
  • Example 3: big JSON objects
• Running tests

v1 release checklist (jfpv1)

This is a list of high-level development and documentation tasks, which need to be completed prior to freezing the API for v1. Before v1, backward-incompatible changes to the API are possible. Since the jfpv1 spec is work in progress, the fingerprints may not be fully comparable between different 0.y.z versions.

NB: JSON fingerprints up until v0.12.2 ignored empty objects and arrays as values. This behavior was changed in v0.13.0 which means that JSON fingerprints created with earlier versions may produce different and incomparable hashes depending on the presence of empty objects or arrays.

• Formalized and complete jfpv1 specification
• JSON type support
  • Primitives and literals
  • Arrays
  • Objects
• Flattened "sibling-aware" internal data structure
• Support nested JSON data structures with mixed types
• Support most common SHA-2 hash functions
  • SHA256
  • SHA384
  • SHA512
• Dynamic jfpv1 fingerprint comparison function (JSON string against a fingerprint)
• Performance characteristics that scale sufficiently
• Extensive verification against potential fingerprint (hash) collisions

Installation

To install the json-fingerprint package, run pip install json-fingerprint.

Examples

The complete working examples below show how to use all core features of the json_fingerprint package.

Create JSON fingerprints

JSON fingerprints can be created with the create() function, which requires three arguments: input (valid JSON string), hash function (SHA256, SHA384 and SHA512 are supported) and JSON fingerprint version (1).

import json

import json_fingerprint
from json_fingerprint import hash_functions

json_1 = json.dumps([3, 2, 1, [True, False], {"foo": "bar"}])
json_2 = json.dumps([2, {"foo": "bar"}, 1, [False, True], 3])  # Different order
fp_1 = json_fingerprint.create(input=json_1, hash_function=hash_functions.SHA256, version=1)
fp_2 = json_fingerprint.create(input=json_2, hash_function=hash_functions.SHA256, version=1)
print(f"Fingerprint 1: {fp_1}")
print(f"Fingerprint 2: {fp_2}")

This will output two identical fingerprints regardless of the different order of the json elements:

Fingerprint 1: jfpv1$sha256$2ecb0c919fcb06024f55380134da3bbaac3879f98adce89a8871706fe50dda03
Fingerprint 2: jfpv1$sha256$2ecb0c919fcb06024f55380134da3bbaac3879f98adce89a8871706fe50dda03

Since JSON objects with identical data content and structure will always produce identical fingerprints, the fingerprints can be used effectively for various purposes. These include finding duplicate JSON data from a larger dataset, JSON data cache validation/invalidation and data integrity checking.

Decode JSON fingerprints

JSON fingerprints can be decoded with the decode() convenience function. It returns the version, hash function and secure hash in a tuple.

import json_fingerprint

fp = "jfpv1$sha256$2ecb0c919fcb06024f55380134da3bbaac3879f98adce89a8871706fe50dda03"
version, hash_function, hash = json_fingerprint.decode(fingerprint=fp)
print(f"JSON fingerprint version: {version}")
print(f"Hash function: {hash_function}")
print(f"Hash hex digest: {hash}")

This will output the individual elements that make up a fingerprint as follows:

JSON fingerprint version: 1
Hash function: sha256
Hash hex digest: 2ecb0c919fcb06024f55380134da3bbaac3879f98adce89a8871706fe50dda03

Match fingerprints

The match() is another convenience function that matches JSON data against a fingerprint, and returns either True or False depending on whether the data matches the fingerprint or not. Internally, it will automatically choose the correct version and hash function based on the target_fingerprint argument.

import json

import json_fingerprint

json_1 = json.dumps([3, 2, 1, [True, False], {"foo": "bar"}])
json_2 = json.dumps([3, 2, 1])
# "target_fp" contains the JSON fingerprint of "json_1"
target_fp = "jfpv1$sha256$2ecb0c919fcb06024f55380134da3bbaac3879f98adce89a8871706fe50dda03"
match_1 = json_fingerprint.match(input=json_1, target_fingerprint=target_fp)
match_2 = json_fingerprint.match(input=json_2, target_fingerprint=target_fp)
print(f"Fingerprint matches with json_1: {match_1}")
print(f"Fingerprint matches with json_2: {match_2}")

This will output the following:

Fingerprint matches with json_1: True
Fingerprint matches with json_2: False

Find matches in fingerprint lists

The find_matches() function takes a JSON string and a list of JSON fingerprints as input. It creates a fingerprint of the JSON input string of each different variant in the target list, and looks for matches in the fingerprint list. It can optionally also deduplicate the fingerprint input list, and results list.

import json

import json_fingerprint

# "data" produces jfpv1 (SHA256): jfpv1$sha256$d119f4d8...b1710d9f
# "data produces jfpv1 (SHA384): jfpv1$sha384$9bca46fd...fd0e2e9c
json_data = json.dumps({"foo": "bar"})

fingerprints = [
    # SHA256 match
    "jfpv1$sha256$d119f4d8b802091520162b78f57a995a9ecbc88b20573b0c7e474072b1710d9f",
    # SHA256 match (duplicate)
    "jfpv1$sha256$d119f4d8b802091520162b78f57a995a9ecbc88b20573b0c7e474072b1710d9f",
    # SHA384 match
    ("jfpv1$sha384$9bca46fd7ef7aa2e16e68978b5eb5c294bd5b380780e81bcb1af97d4b339bca"
     "f7f6a622b2f1a955eea2fadb8fd0e2e9c"),
    # SHA256, not a match
    "jfpv1$sha256$73f7bb145f268c033ec22a0b74296cdbab1405415a3d64a1c79223aa9a9f7643",
]

# Output all matches, including the two identical SHA256 fingerprints
matches = json_fingerprint.find_matches(input=json_data, fingerprints=fingerprints)
print(*(f"\nMatch: {match[0:30]}..." for match in matches))

# Output deduplicated matches, removing one redundant SHA256 match
deduplicated_matches = json_fingerprint.find_matches(input=json_data,
                                                     fingerprints=fingerprints,
                                                     deduplicate=True)
print(*(f"\nDeduplicated match: {match[0:30]}..." for match in deduplicated_matches))

This will output the following results, first the list with a duplicate and the latter with deduplicated results:

Match: jfpv1$sha256$d119f4d8b80209152...
Match: jfpv1$sha256$d119f4d8b80209152...
Match: jfpv1$sha384$9bca46fd7ef7aa2e1...

Deduplicated match: jfpv1$sha384$9bca46fd7ef7aa2e1...
Deduplicated match: jfpv1$sha256$d119f4d8b80209152...

JSON normalization

The jfpv1 JSON fingerprint function transforms the data internally into a normalized (canonical) format before hashing the output.

Alternative specifications

Most existing JSON normalization/canonicalization specifications and related implementations operate on three key aspects: data structures, values and data ordering. While the ordering of key-value pairs (objects) is straightforward, issues usually arise from the ordering of arrays.

The JSON specifications, including the most recent RFC 8259, have always considered the order of array elements to be meaningful. As data gets serialized, transferred, deserialized and serialized again throughout various systems, maintaining the order of array elements becomes impractical if not impossible in many cases. As a consequence, this makes the creation and comparison of secure hashes of JSON data across multiple systems a complex process.

JSON Fingerprint v1 (jfpv1)

The jfpv1 specification takes a more value-oriented approach toward JSON normalization and secure hash creation: values and value-related metadata bear most significance when JSON data gets normalized into the jfpv1 format. The original JSON data gets transformed into a flattened list of small objects, which are then hashed and sorted, and ultimately hashed again as a whole.

In practice, the jfpv1 specification purposefully ignores the original order of data elements in an array. The jfpv1 specification focuses instead on verifying that the following aspects of JSON datasets being compared match:

• All values in the compared datasets are identical
• The values exist in identical paths (arrays, object key-value pairs)

In the case of arrays, each array gets a unique hash identifier based on the data elements it holds. This way, each flattened value "knows" to which array it belongs to. This identifier is called a sibling hash because it is derived from each array element's value as well as its neighboring values.

Performance

The JSON fingerprint v1 specification and its first implementation have been designed with a primary focus on functional utility over performance. There are some performance-related characteristics that are good to be aware of:

• Due to the way the internal sibling hashes are computed, highly nested data structures will increase the processing time significantly
• The amount of data in a single data element, or the number of elements in a flat array, is much less meaningful performance-wise than the overall depth of the data structure

Below are some examples of the performance impact when processing different types of data structures.

Example 1: flat data structures

Processing an array of arrays with the maximum depth of 2 levels for each datum:

import json
import time

import json_fingerprint
from json_fingerprint import hash_functions

data = json.dumps(
    [
        [1, 2],
        [3, 4],
        [5, 6],
        [7, 8],
        [9, 10],
        [11, 12],
    ]
)
start_time = time.time_ns()  # Measure time in nanoseconds
iterations = 1000
for i in range(iterations):
    json_fingerprint.create(input=data, hash_function=hash_functions.SHA256, version=1)
end_time = time.time_ns()
duration = round(((end_time - start_time) / iterations / 1000000), 2)  # To milliseconds
print(f"Average processing time per JSON fingerprint: {duration} milliseconds")

Performance test results:

Average processing time per JSON fingerprint: 0.27 milliseconds

As seen in the test results, flat data structures perform well on modern computer hardware.

Example 2: nested data structures

Processing a nested array of arrays with datums 11 and 12 7 levels deep in the data structure:

import json
import time

import json_fingerprint
from json_fingerprint import hash_functions

data = json.dumps(
    [
        [1, 2, [3, 4, [5, 6, [7, 8, [9, 10, [11, 12]]]]]],
    ]
)
start_time = time.time_ns()  # Measure time in nanoseconds
iterations = 1000
for i in range(iterations):
    json_fingerprint.create(input=data, hash_function=hash_functions.SHA256, version=1)
end_time = time.time_ns()
duration = round(((end_time - start_time) / iterations / 1000000), 2)  # To milliseconds
print(f"Average processing time per JSON fingerprint: {duration} milliseconds")

Performance test results:

Average processing time per JSON fingerprint: 2.75 milliseconds

Compared to the flat data structure with the same amount of data, the nesting of arrays increased the processing time tenfold.

Example 3: big JSON objects

Processing a dynamically generated JSON object of three different sizes: ~256KiB, ~512KiB, and ~1MiB:

import json
import time

import json_fingerprint
from json_fingerprint import hash_functions


def test_performance(data: str, size: str) -> None:
    start_time = time.time_ns()  # Measure time in nanoseconds
    iterations = 1000
    for i in range(iterations):
        json_fingerprint.create(input=data, hash_function=hash_functions.SHA256, version=1)
    end_time = time.time_ns()
    duration = round(((end_time - start_time) / iterations / 1000000), 2)  # To milliseconds
    print(f"Average processing time per JSON fingerprint ({size}): {duration} milliseconds")


text = "abcdefg " * 16384  # a single text element (~128KiB)
text_list = ["hijklmn " * 128 for i in range(128)]  # 128 * 1KiB text elements (~128KiB)
test_performance(json.dumps({"text": text, "text_list": text_list}), "~256KiB")
test_performance(json.dumps({"text": text * 2, "text_list": text_list * 2}), "~512KiB")
test_performance(json.dumps({"text": text * 4, "text_list": text_list * 4}), "~1MiB")

Performance test result:

Average processing time per JSON fingerprint (~256KiB): 2.91 milliseconds
Average processing time per JSON fingerprint (~512KiB): 5.42 milliseconds
Average processing time per JSON fingerprint (~1MiB): 11.18 milliseconds

Processing fairly sizeable JSON objects with text content in a flat structure scales linearly.

Running tests

The entire internal test suite of json-fingerprint is included in its distribution package. If you wish to run the internal test suite, install the package and run the following command:

python -m json_fingerprint.tests.run

If all tests ran successfully, this will produce an output similar to the following:

...............................
----------------------------------------------------------------------
Ran 31 tests in 0.008s

OK