Improve the robustness of unit cases

[stayrascal]

Hi, I'm trying to learning Daft start with running the unit test cases via make test, but there are a re unit cases failed in my environemnt, I think we can improve them a little bit.

Running Environment

Mac M1 and Debian 5.4.143(CI Agent).

Linux n35-133-210 5.4.143.bsk.8-amd64 #5.4.143.bsk.8 SMP Debian 5.4.143.bsk.8 Wed Jul 20 08:43:36 UTC  x86_64 GNU/Linux

Problems

make test works well in my mac, but some cases failed in my ci agent.

FAILED tests/io/test_s3_credentials_refresh.py::test_s3_credentials_refresh - botocore.errorfactory.BucketAlreadyOwnedByYou: An error occurred (BucketAlreadyOwnedByYou) when calling the CreateBucket operati...

FAILED tests/io/test_url_upload_local.py::test_upload_local_no_write_permissions - Failed: DID NOT RAISE <class 'daft.exceptions.DaftCoreException'>

FAILED tests/recordbatch/numeric/test_numeric.py::test_table_numeric_trigonometry[sinh-False-False] - assert False is True

FAILED tests/recordbatch/numeric/test_numeric.py::test_table_numeric_trigonometry[cosh-False-False] - assert False is True

FAILED tests/recordbatch/numeric/test_numeric.py::test_table_numeric_trigonometry[arccos-True-False] - assert False is True

1. test_url_upload_local.py::test_upload_local_no_write_permissions

The failed reason is that I use root user to running test, so the user should have full permission on any path, which is not meet the purpose of test_upload_local_no_write_permissions.

Suggestion: it's better skip this test when run as root user?

2. test_numeric.py::test_table_numeric_trigonometry

The problem i meet is the precision of float.
Suggestion: using pytest.approx during comparing x and y?

3. test_s3_credentials_refresh.py::test_s3_credentials_refresh

There are some small problem on this test case.

• it's better to use aws_server, aws_server_ip, aws_server_port, aws_credentials pytest fixtures from conftest (move the aws fixtures from delta_lake conftest.py to io conftest.py) instead of re-preparing the configuratios and start moto server, which might not been stopped if something failed.

• The purpose of this test case is refresh the credential if it's expired (1s in this case), the read_parquet(read schema), df.collect(read content) and write_parquet can share same credential within active window, but it's hard to decide how long it will take in different workload, e.g. the write_parquet maybe take 2 or more seconds, which the credential might be refreshed mutilple times, so it's hard to assert the value of count_get_credentials
  

• The is a trick thing that the start_service("moto_server") will launch two process in my ci agent, but the stop_process(process) only stop one process, the mote_server still running after test case finished. So If i rerun the test, the test bucket already exists.

Suggestion:

• for problem 1, resuse the existing code.
• for problem 2, increase the ttl of credential, and using greater than equal during check whether the credential is refrehed
• for problem 3, random bucket name