[IMPALA-11468] Port "Block Bloom filter false positive rate correction" test fix from Kudu

• Type: Task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 4.2.0
• Fix Version/s: Impala 4.2.0
• Component/s: Backend
• Epic Color: ghx-label-3
• Assignee: Joe McDonnell
• Reporter: Joe McDonnell
• Created: 02/Aug/22 17:10
• Updated: 12/Aug/22 03:18
• Resolved: 12/Aug/22 03:18

Description

Impala's be/src/util/parquet-bloom-filter.* and Kudu's src/kudu/util/block_bloom_filter* are closely related. The tests in parquet-bloom-filter-test correspond closely with the tests in block_bloom_filter-test. Kudu made an improvement to the test to more accurate calculate the bloom filter false positive rate, and it would be useful to port that over to Impala.

Kudu change: https://github.com/apache/kudu/commit/d1190c2b06a6eef91b21fd4a0b5fb76534b4e9f9

It turns out that the parquet-bloom-filter-test is sensitive to the implementation of unordered_set, and it starts to fail with GCC/libstdc++ 10. The changes made for fixing the false positive rate fix the GCC 10 test failure as well.