GitHub user kyungjunleeme created a discussion: Is the `json` override in `airflow_local_settings.py` still relevant?
Hello. I’m wondering if this approach https://airflow.apache.org/docs/apache-airflow/stable/administration-and-deployment/dag-serialization.html#using-a-different-json-library is still relevant or useful in real-world Airflow deployments. ## Why it might have limited value 1. **Limited scope** – This override only affects Airflow components that import `json` from `airflow_local_settings`. Many operators, hooks, and provider packages import Python’s built-in `json` module directly, so they wouldn’t be impacted. 2. **Minimal performance gains** – In most Airflow environments, performance bottlenecks are related to I/O (e.g., DB queries, network calls, XCom storage) rather than JSON parsing speed. Switching to something like `ujson` may not yield noticeable improvements. 3. **Compatibility risks** – Alternative libraries can behave differently (e.g., handling of `datetime`, `Decimal`, `NaN`), which might cause subtle bugs or inconsistencies, especially in XCom serialization. ## Questions for the community - Is this override mechanism still actively used or tested in Airflow core? - Should we consider clarifying in the docs that it only affects a limited part of the codebase? - Would it make sense to deprecate this recommendation in favor of patching JSON handling directly in custom hooks/operators if needed? I’d love to hear thoughts from the community and maintainers. If the consensus is that this is no longer valuable, perhaps we could update the docs to reflect that. GitHub link: https://github.com/apache/airflow/discussions/54387 ---- This is an automatically sent email for [email protected]. To unsubscribe, please send an email to: [email protected]
