[Bug]: TestSparse_invalid_insert is flaky

[zhengbuqian]

Is there an existing issue for this?

• I have searched the existing issues

Environment

- Milvus version: master

Current Behavior

TestSparse_invalid_insert is flaky.

The test was originally designed to test against negative sparse vector index, but we later changed to use uint32 instead of int32 as the index type, thus it became impossible to have a negative index: negative int32 is a super large uint32, which is legal.

This test tried to catch illegal negative index by setting the first index of the first row to -10 and verify the insertion to fail. The reason the test is flaky:

1. when we generate more than 1 non zero element in the first row, all indexes are between 0 and 700(as in testutils.GenerateSparseFloatVectors), if we set the first index to -10, after being interpreted as uint32, it is a very large value, violating the indexes in the same row must be sorted rule, causing the insertion to fail and the test to pass.
2. when we generated only 1 non zero element in the first row and we set the first index to -10, this row is a legal row with a super large index, thus the insertion will succeed and the test fails.

Expected Behavior

No response

Steps To Reproduce

No response

Milvus Log

No response

Anything else?

No response