[Feature]: Improve performance by using a faster library for parsing json schemas

[mikicz]

Suggested Behavior

Use fastjsonschema instead of jsonschema, or even jsonschema-rs. jsonschema-rs claims to be 30x to 390x faster than jsonschema (3x to 7x faster than fastjsonschema).

Why is this needed?

I have an OpenAPI spec that's moderately big, roughly 1.6M and parsing it with OpenAPI takes about 15s on my computer. I unfortunately can't share it, but I can share the profiling on it.

For one-time use it would be fine, but we use it in tests to validate responses, so this adds quite a bit of runtime to tests.

References

This profiling of just loading the OpenAPI yaml file, using OpenAPI.from_file_path("<path>")

The part in blue is jsonschema.

A bit more detail here: x.txt

Would you like to implement a feature?

None

[p1c2u]

Hi @mikicz , thanks for your comment. I see in your profiling that your bottleneck is specification validation on the OpenAPI initialization. You can turn off the validation if you are sure your specification is valid. Please read the documentation how to disable spec validation.

If your intention is it use specification validation feature but make it faster then that would require changes in openapi-spec-validator and openapi-schema-validator which are entirely based on jsonschema.

[p1c2u]

After little bit more investigation and profiling I found out the issue with suboptimal reference resolution in jsonschema-path library. The fix should have more impact than switching to jsonschema-rs.

p1c2u/jsonschema-path#208

[p1c2u]

Betea version with performance improvements was released. Feel free to test it.

[mikicz]

Hi! Thanks for still looking at this! I have indeed found that just disabling spec validation does help quite a bit. Nonetheless I have tried to profile the current release and the pre-release with our spec file.

With validation:

from openapi_core import OpenAPI

if __name__ == "__main__":
    OpenAPI.from_file_path("schema.yaml")

Without validation:

from openapi_core import OpenAPI, Config

if __name__ == "__main__":
    config = Config(spec_validator_cls=None)
    OpenAPI.from_file_path("schema.yaml", config=config)

Old dependencies:

$ uv pip freeze | grep 'openapi\|jsonschema\|pathable'
fastjsonschema==2.21.2
jsonschema==4.26.0
jsonschema-path==0.3.4
jsonschema-specifications==2025.9.1
openapi-core==0.22.0
openapi-schema-validator==0.6.3
openapi-spec-validator==0.7.2
pathable==0.4.4

New dependencies:

$ uv pip freeze | grep 'openapi\|jsonschema\|pathable'
fastjsonschema==2.21.2
jsonschema==4.24.1
jsonschema-path==0.4.0b8
jsonschema-specifications==2025.9.1
openapi-core==0.23.0b1
openapi-schema-validator==0.6.3
openapi-spec-validator==0.8.0b3
pathable==0.5.0b6

The headline is that I do not see any major time improvement - tho I can imagine my setup being less than ideal - I only ran once per variant instead of running multiple times. See pyinstrument profiling:

old.txt
new.txt
new_without_validation.txt
old_without_validation.txt

[p1c2u]

Thanks for sharing your results. The improvement should only be visible when the spec validation is enabled.

I’ve prepared a PoC for the jsonschema-rs version. It’s definitely worth moving forward, but it requires more time.