Creating a model performance issue

[needones]

I used viztracer to analyze the performance of HashModel and BaseModel, and found that the time difference between the two was hundreds of times when the model was generated, and a large number of Deepcopies were generated when the HashModel was generated, but I could not locate the problem, Is the pydantic v1 version causing the problem?

create model cost:
BaseModel ~= 3us
HashModel ~= 700us

python:3.11
viztracer==0.16.3
redis-om==0.3.1
redis == 5.0.6

eg: demo.py

from pydantic import BaseModel
from redis_om import HashModel


class TaskStatisticalModel(BaseModel):
    task_uuid: str = ""
    parallel_task: str = ""
    is_alarm: int = 0


class TaskStatisticalModel2(HashModel):
    task_uuid: str = ""
    parallel_task: str = ""
    is_alarm: int = 0


def add_redis_model():
    TaskStatisticalModel2()


def add_base_model():
    TaskStatisticalModel()


def test_redis_model(number=10):
    for i in range(number):
        add_redis_model()


def test_base_model(number=10):
    for i in range(number):
        add_base_model()


if __name__ == '__main__':
    num = 10
    test_redis_model(num)
    test_base_model(num)

use viztracer

viztracer demo.py

# wait
vizviewer result.json

search:
"add_redis_model" => hash_model
"add_base_model" => base model

[XChikuX]

HashModel, JsonModel, EmbeddedJsonModel have several custom logic to ensure the data is modified for compatability purposes.

Using BaseModel for redis-om will lead to unexpected behaviour and is not supported.

Some latency is expected. However, this many deep copies is concerning @slorello89

We have a good pictoral walkthrough from OP. Can we reduce the number of deepcopies?