Uh oh!
There was an error while loading. Please reload this page.
GH-40079: [CI][Packaging] Enable Azure in more tests and builds - #40080
Conversation
Tom-Newton
commented
Feb 14, 2024
Most of the CI failures are actually reproducible on I will need to investigate the AVX2 build failure. |
| # Azurite requires npm | ||
| RUN export DEBIAN_FRONTEND=noninteractive && \ | ||
| apt-get update -y -q && \ | ||
| apt-get install -y -q npm \ | ||
| && apt-get clean \ | ||
| && rm -rf /var/lib/apt/lists/* |
kou
commented
Feb 14, 2024
Test is https://github.com/apache/arrow/actions/runs/7902101938/job/21566998998?pr=40080#step:6:6540 arrow/cpp/src/arrow/filesystem/azurefs_test.cc Lines 286 to 291 in 2422994 Important backtrace: https://github.com/apache/arrow/actions/runs/7902101938/job/21566998998?pr=40080#step:6:6568 arrow/cpp/src/arrow/filesystem/azurefs.cc Lines 184 to 185 in 2422994 |
Tom-Newton
commented
Feb 17, 2024
Pinning to same versions we get from vcpkg solves some of the CI failures. I think the only remaining failures are ones that also exist on |
kou
commented
Feb 17, 2024
Does it mean that we may need to improve our Azure filesystem implementation to use more recent azure-sdk-for-cpp? (If it's needed, we should work on it as a separated task.) Anyway, could you add a comment why we need to pin azure-sdk-for-cpp?
Could you open an issue for it so that we can work on it as a separated task? |
Created an issue for the python 3.9 GCS testbench issue #40112 For the segfault it looks like going from and it looks like 1.11.0 has been marked as broken on the conda website If I understand this correctly it seems to be a conda issue not an Azure SDK or an Arrow problem. |
Tom-Newton
commented
Feb 18, 2024
I'm going to call this ready for review but I'm not sure if we can merge it while #40112 is un-resolved. |
Tom-Newton
commented
Feb 18, 2024
Hmm.... it looks like the azure tests are slightly flaky. I just changed a comment and the new CI run has an azure failure. It looks like azurite returned an invalid response. I've run this multiple times locally but so far have not been able to reproduce. |
69ed3a5 to
7ed0910CompareRetrying got the same error in a different test case on a different build https://github.com/apache/arrow/actions/runs/7951594516/job/21705210845?pr=40080 |
Tom-Newton
commented
Feb 18, 2024
I think #40120 should resolve the Python 3.9 and GCS testbench problem. I also found an example of the azure tests flaking on I created a separate issue to fix this flakiness #40121 |
7ed0910 to
7c27978Compare31e4920 to
5e3f846Compare| # under the License. | ||
| aws-sdk-cpp=1.11.68 | ||
| # There is a problem with the 1.11.0 conda release of azure-core-cpp https://github.com/conda-forge/admin-requests/pull/911 |
There was a problem hiding this comment.
I am not sure that is a relevant link. AFAIU there is not really a problem with the 1.11.0 conda package itself. It's that conda did not pin azure-core-cpp in other packages (eg building tiledb with azure-core-cpp against 1.10, and then running that in an env with 1.11 installed, then we got segfaults).
But for the purpose of our CI, we are both building and running against the same azure-core-cpp version, right? In that case, I would expect that building+running with azure-core-cpp would work just fine. Or if not, that means our azurefs implementation might not yet be compatible with the latest azure-core-cpp?
There was a problem hiding this comment.
Hmm... I tried using 1.11.0 building from source with cmake and everything worked. With 1.10.0 1.11.0 from conda we got segfaults.
There was a problem hiding this comment.
You mean 1.11 from conda? (not 1.10) Do you remember if there was more information in the logs about the crash?
There was a problem hiding this comment.
#40080 (comment) summarises the info we have.
It looks like 1.11.1 is now available from conda though and that works, so I can remove this version restriction.
assignUser
commented
Feb 22, 2024
@github-actions crossbow submit -g python |
Revision: 53b9d2c Submitted crossbow builds: ursacomputing/crossbow @ actions-47658c1ea8 |
Remaining CI failures: I don't think any of these are related to my changes. |
assignUser
left a comment
There was a problem hiding this comment.
I agree CI failures seem unrelated but I'll let @raulcd have the last word on that. Thanks :)
Uh oh!
There was an error while loading. Please reload this page.
kou
commented
Feb 22, 2024
Those failures are unrelated because they are happen on main too. |
After merging your PR, Conbench analyzed the 5 benchmarking runs that have been run so far on merge-commit b089c6a. There were no benchmark performance regressions. 🎉 The full Conbench report has more details. It also includes information about 3 possible false positives for unstable benchmarks that are known to sometimes produce them. |

Rationale for this change
We want python side tests of
AzureFileSystemto run in CI.What changes are included in this PR?
exportto enable Azure pyarrow testsmain#40121Probably all of this should have been included in #39971
Are these changes tested?
There is no new functionality to test
Are there any user-facing changes?
No