build: enable Python 3.12 support and relax core dependency bounds#7821
build: enable Python 3.12 support and relax core dependency bounds#7821pritamdodeja wants to merge 6 commits intotensorflow:masterfrom
Conversation
This commit modernizes the TFX orchestrator to support Python 3.12, while relaxing dependency constraints to allow the use of locally compiled component wheels (version 1.17.0 and 1.18.0). Specific changes include: * pyproject.toml: Updated `requires-python` to allow versions up to 3.12 (range changed from <3.11 to <3.13). * pyarrow: Removed the `<11` upper bound constraint. This allows modern Arrow versions (e.g., 14+) to be used, avoiding legacy build failures on Python 3.12. * Dependency Alignment: Expanded version ranges in `tfx/dependencies.py` for ml-metadata, tfx-bsl, tfdv, tfma, and tft. These now allow version 1.17.0 to resolve, preventing Pip from backtracking into incompatible legacy distributions during installation.
|
Hi @pritamdodeja, Can you kindly resolve the conflicts? Thank you! |
Hello @keerthanakadiri I will work on it this weekend. Pritam |
|
@keerthanakadiri apologies for the delay, I was finally able to get to this. I had to remove tensorflowjs in order for tests to run. I think that has a dependency on tensorflow decision forests, which doesn't have a release for 3.12. My test summary for components looks like the following: ============ 39 failed, 430 passed, 67 skipped, 9 xfailed, 98 warnings in 864.61s (0:14:24) ============= I'm going to run a broader test and see where I get. I see unit tests failing in GitHub. I need to re-produce the issue locally to understand how I can get clean runs. I'd appreciate any advice you have on this front. |
2 participants
This commit modernizes the TFX orchestrator to support Python 3.12, while relaxing dependency constraints to allow the use of locally compiled component wheels (version 1.17.0 and 1.18.0).
Specific changes include:
requires-pythonto allow versions up to 3.12 (range changed from <3.11 to <3.13).<11upper bound constraint. This allows modern Arrow versions (e.g., 14+) to be used, avoiding legacy build failures on Python 3.12.tfx/dependencies.pyfor ml-metadata, tfx-bsl, tfdv, tfma, and tft. These now allow version 1.17.0 to resolve, preventing Pip from backtracking into incompatible legacy distributions during installation.