GitHub Actions build refactor by mtsokol · Pull Request #162 · google/array_record

mtsokol · 2025-07-08T14:46:44Z

This PR cleans up the wheel build process for array_record and migrates changes present in https://github.com/iindyk/array_record/tree/main 20 commits ahead of main.

Here's a job that I ran for it: https://github.com/mtsokol/array_record/actions/runs/16147213833

Co-authored-by: Ihor Indyk <iindyk@google.com>

iindyk · 2025-07-08T23:17:41Z

oss/build_whl.sh

+  # Install ArrayRecord from the wheel and run smoke tests.
+  # TF is not available on Python 3.13 and above.
+  if (( "${PYTHON_MINOR_VERSION}" < 13 )); then
+    $PYTHON_BIN -m pip install --find-links=/tmp/grain/all_dist --pre array-record


/tmp/grain/all_dist -> /tmp/array_record/all_dist

I think in the workflow run you shared it still installed AR from pypi: https://github.com/mtsokol/array_record/actions/runs/16147213833/job/45569132362#step:7:7181

Ah right, it took a bit more work but all linux wheels are build now:
https://github.com/mtsokol/array_record/actions/runs/16167375237

for MacOS arm64 it fails on import with:

+ /Users/runner/hostedtoolcache/Python/3.10.11/arm64/bin/python oss/test_with_grain.py libc++abi: terminating due to uncaught exception of type std::__1::system_error: mutex lock failed: Invalid argument /tmp/array_record/oss/build_whl.sh: line 15: 12619 Abort trap: 6 $PYTHON_BIN oss/test_with_grain.py

Is it the error that you meant?

I think this is something similar to: apache/arrow#40088, apache/arrow#40088 (comment)

After some research it looks like the reason for this mutex issue are different protobuf versions that ArrayRecord uses and Grain requires.

Grain requires protobuf 24.4 and ArrayRecord similarly, but the ArrayRecord was missing:

write_to_bazelrc "common --check_direct_dependencies=error"

and the actual protobuf installed by ArrayRecord in the current main is 29.0.rc2 (!). I tried reducing versions in ArrayRecord but unfortunately eigen is holding on protobuf here, and tampering with its version leads to compilation errors.

I'm not sure what is the best option right now. The intuition tells me that ArrayRecord now requires 29.0 and the next release should uphold it and Grain should now also upgrade (But I remember that you said it's a painful process for Grain).
On the other hand dragging an ArrayRecord fork which is 35 commits behind also sounds cumbersome, especially that for conda-forge we need an official release that is available on github, we can't rely on a fork.

How painful would be to upgrade protobuf for Grain?

yeah, I also think this is because of the protobuf version. The problem here is not so much in grain, but tensorflow -- updating protobuf in grain is easy, but if both AR and Grain have higher protobuf version than TF, then import with TF will fail the same way. This is how we found out in the first place about this -- TF is not a dependency, but is often installed in the same ecosystem.

I think protobuf version was updated in TF main: tensorflow/tensorflow#92241. So now, hopefully with the next release of TF we can update it too.

Could we try solve this by using TensorFlow nightly in our test? Some in https://pypi.org/project/tf-nightly/#history should already use newer protobuf, right?

yes for tests, but our users wouldn't be using nightly most likely: https://github.com/google-research/kauldron/blob/8537718c703ab606a545eee6daf89238d8def55f/pyproject.toml#L41

mtsokol · 2025-07-09T19:10:22Z

MODULE.bazel

+    version = "0.7.3",
    repo_name = "com_google_array_record",
 )



Let's use actual bazel_dep(name = "protobuf", version = ...) here and enforce errors if different is picked:

write_to_bazelrc "common --check_direct_dependencies=error"

This PR cleans up the wheel build process for array_record and migrates changes present in https://github.com/iindyk/array_record/tree/main 20 commits ahead of main. Here's a job that I ran for it: https://github.com/mtsokol/array_record/actions/runs/16147213833 Ported from #162 PiperOrigin-RevId: 781211931

dryman

Please add a small document on how to run the new workflows. Thanks.

dryman · 2025-07-10T14:41:43Z

MODULE.bazel

-    ignore_root_user_error = True,  # Required for our containerized CI environments.
-    python_version = PYTHON_VERSION,
+http_archive(
+    name = "pybind11",


Why adding an extra pybind11 rule here?

TBH I mostly followed Grain setup as that's the repo we're mainly working with: https://github.com/google/grain/blob/3bc1d2582dffdddf00b0d436a24b6313b69c617d/MODULE.bazel#L31
I can check it without it that one - and remove it if it also builds successfully.

AR is using pybind11 through pybind_bazel already. So it should work without http_archive pulled pybind11.

Done! Removed pybind11 rule from here.

mtsokol · 2025-07-11T10:18:15Z

Please add a small document on how to run the new workflows. Thanks.

Done! I also enforced direct dependencies versions and fixed protobuf version to the one used by TF nightly: #162 (comment)

Here's a workflow with passing jobs: https://github.com/mtsokol/array_record/actions/runs/16217754006

This PR cleans up the wheel build process for array_record and migrates changes present in https://github.com/iindyk/array_record/tree/main 20 commits ahead of main. Here's a job that I ran for it: https://github.com/mtsokol/array_record/actions/runs/16147213833 Ported from #162 PiperOrigin-RevId: 781211931

This PR cleans up the wheel build process for array_record and migrates changes present in https://github.com/iindyk/array_record/tree/main 20 commits ahead of main. Here's a job that I ran for it: https://github.com/mtsokol/array_record/actions/runs/16147213833 Ported from #162 PiperOrigin-RevId: 783082302

iindyk · 2025-07-14T23:25:30Z

submitted in #163

mtsokol mentioned this pull request Jul 8, 2025

Add Windows build #161

Draft

mtsokol force-pushed the build-refactor branch from 145a009 to b7d6cb7 Compare July 8, 2025 15:07

GitHub Actions build refactor

d68a705

Co-authored-by: Ihor Indyk <iindyk@google.com>

mtsokol force-pushed the build-refactor branch from b7d6cb7 to d68a705 Compare July 8, 2025 15:12

iindyk reviewed Jul 8, 2025

View reviewed changes

mtsokol force-pushed the build-refactor branch 3 times, most recently from 33913a7 to c9d4a11 Compare July 9, 2025 08:04

Fix --find-links option

9b9979a

mtsokol force-pushed the build-refactor branch 2 times, most recently from 928465e to 36c467e Compare July 9, 2025 09:41

Python versions support in the Bazel config

8bd05e5

mtsokol force-pushed the build-refactor branch from 36c467e to 8bd05e5 Compare July 9, 2025 09:49

Config update

fce714d

mtsokol commented Jul 9, 2025

View reviewed changes

copybara-service bot mentioned this pull request Jul 9, 2025

GitHub Actions build refactor #163

Merged

dryman suggested changes Jul 10, 2025

View reviewed changes

Enforce direct dependencies

772dd32

mtsokol requested a review from dryman July 11, 2025 10:18

Skip TF import test on Darwin

49893bc

Fix python-tests.yml workflow

01e57f7

mtsokol force-pushed the build-refactor branch from e24cb8f to 01e57f7 Compare July 14, 2025 17:57

iindyk closed this Jul 14, 2025

mtsokol deleted the build-refactor branch July 15, 2025 08:41

Conversation

mtsokol commented Jul 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

iindyk Jul 8, 2025

Choose a reason for hiding this comment

Uh oh!

mtsokol Jul 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mtsokol Jul 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mtsokol Jul 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

iindyk Jul 9, 2025

Choose a reason for hiding this comment

Uh oh!

mtsokol Jul 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

iindyk Jul 9, 2025

Choose a reason for hiding this comment

Uh oh!

mtsokol Jul 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dryman left a comment

Choose a reason for hiding this comment

Uh oh!

dryman Jul 10, 2025

Choose a reason for hiding this comment

Uh oh!

mtsokol Jul 10, 2025

Choose a reason for hiding this comment

Uh oh!

dryman Jul 10, 2025

Choose a reason for hiding this comment

Uh oh!

mtsokol Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

mtsokol commented Jul 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

iindyk commented Jul 14, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

mtsokol commented Jul 8, 2025 •

edited

Loading

mtsokol Jul 9, 2025 •

edited

Loading

mtsokol Jul 9, 2025 •

edited

Loading

mtsokol Jul 9, 2025 •

edited

Loading

mtsokol Jul 9, 2025 •

edited

Loading

mtsokol Jul 9, 2025 •

edited

Loading

mtsokol commented Jul 11, 2025 •

edited

Loading