new: decode json responses with orjson

[joein]

No description provided.

[netlify]

✅ Deploy Preview for poetic-froyo-8baba7 ready!

Name	Link
🔨 Latest commit	c065773
🔍 Latest deploy log	https://app.netlify.com/sites/poetic-froyo-8baba7/deploys/67b32771f9558d0008040d0d
😎 Deploy Preview	https://deploy-preview-902--poetic-froyo-8baba7.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site configuration.

[agourlay]

out of curiosity, how much faster is this?

[joein]

out of curiosity, how much faster is this?

@agourlay
I am about to post a note here on it

But as of now some of the common responses can be decoded up to 3x times faster (it is only fair to those with with_payload=True or with_vectors=True, since it only affects decoding)

[joein]

I benchmarked it on scroll queries with with_vectors=True and with_payload=True for a limit of 5, 10, 20 and 30 records

I used randomly generated vectors with dim = 512 and dim = 1024, generated payload with a fixture from the tests
Benchmarking results include just the json decoding step (response.json()) and the whole scroll request

The results are:

{
    "decode": {
        "512": {
            "5": {
                "json": 4.8874993808567524e-05,
                "orjson": 2.366595435887575e-05
            },
            "10": {
                "json": 7.658399408683181e-05,
                "orjson": 4.395801806822419e-05
            },
            "20": {
                "json": 0.0020279159653000534,
                "orjson": 0.0007863339851610363
            },
            "30": {
                "json": 0.00019866699585691094,
                "orjson": 8.283299393951893e-05
            }
        },
        "1024": {
            "5": {
                "json": 5.583302117884159e-05,
                "orjson": 2.7542002499103546e-05
            },
            "10": {
                "json": 6.133399438112974e-05,
                "orjson": 2.8709007892757654e-05
            },
            "20": {
                "json": 0.00012050004443153739,
                "orjson": 4.754198016598821e-05
            },
            "30": {
                "json": 0.00016829196829348803,
                "orjson": 6.799999391660094e-05
            }
        }
    },
    "scroll": {
        "512": {
            "5": {
                "json": 0.0043065829668194056,
                "orjson": 0.0034446659847162664
            },
            "10": {
                "json": 0.002439292031340301,
                "orjson": 0.0023802079958841205
            },
            "20": {
                "json": 0.0047377090086229146,
                "orjson": 0.0032056660274975
            },
            "30": {
                "json": 0.003125959017779678,
                "orjson": 0.0026631250511854887
            }
        },
        "1024": {
            "5": {
                "json": 0.00181554100709036,
                "orjson": 0.0019507919787429273
            },
            "10": {
                "json": 0.001797415956389159,
                "orjson": 0.0016768340137787163
            },
            "20": {
                "json": 0.0022662909468635917,
                "orjson": 0.0019526249961927533
            },
            "30": {
                "json": 0.0038724589976482093,
                "orjson": 0.003087958029936999
            }
        }
    }
}

I generated the input with:

from typing import Optional, Any

import numpy as np
import json

from tests.fixtures.payload import one_random_payload_please

NUM_RECORDS = 1_000

def generate_numpy(output_file: Optional[str] = None, dim: int = 512) -> np.ndarray:
    arr = np.random.random((NUM_RECORDS, dim))
    if output_file is not None:
        with open(output_file, 'wb') as f:
            np.save(f, arr)
    return arr

def generate_json(output_file: Optional[str] = None) -> list[dict[str, Any]]:
    data = []
    if output_file is not None:
        with open(output_file, 'w') as f:
            for i in range(NUM_RECORDS):
                record = one_random_payload_please(i)
                if 'rand_datetime' in record:
                    record['rand_datetime'] = str(record['rand_datetime'])
                try:
                    f.write(json.dumps(record) + '\n')
                except Exception as e:
                    from pprint import pprint
                    pprint(record)
                    raise e
                data.append(record)
    else:
        data = [one_random_payload_please(i) for i in range(NUM_RECORDS)]
    return data


if __name__ == '__main__':
    generate_json('payload.jsonl')
    for dim in [512, 1024]:
        generate_numpy(f'vectors_{dim}.npy', dim=dim)

[joein]

Going to add benchmarking results with limit = 100, 1_000 and 10_000

[joein]

{
    "decode": {
        "512": {
            "100": {
                "json": 0.002053125004749745,
                "orjson": 0.0008660830208100379
            },
            "1000": {
                "json": 0.016057875007390976,
                "orjson": 0.0042370830196887255
            },
            "10000": {
                "json": 0.23768579197349027,
                "orjson": 0.1923259579925798
            }
        },
        "1024": {
            "100": {
                "json": 0.0010743339662440121,
                "orjson": 0.000690957996994257
            },
            "1000": {
                "json": 0.0073619160102680326,
                "orjson": 0.0038023749948479235
            },
            "10000": {
                "json": 0.26379175001056865,
                "orjson": 0.22926875000121072
            }
        }
    },
    "scroll": {
        "512": {
            "100": {
                "json": 0.00935820903396234,
                "orjson": 0.00594179198378697
            },
            "1000": {
                "json": 0.03989037498831749,
                "orjson": 0.023745999962557107
            },
            "10000": {
                "json": 0.4403287079767324,
                "orjson": 0.37456325005041435
            }
        },
        "1024": {
            "100": {
                "json": 0.005677750043105334,
                "orjson": 0.0063817090122029185
            },
            "1000": {
                "json": 0.026618165953550488,
                "orjson": 0.027344707981683314
            },
            "10000": {
                "json": 0.4460842920234427,
                "orjson": 0.4439006670145318
            }
        }
    }
}

[arthur-st]

Out of curiosity, have you considered evaluating msgspec for this?