[KUDU-3286] Add special handling for empty strings for Bloom filter predicate push down

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.13.0
• Fix Version/s: 1.15.0
• Target Version/s: 1.15.0

Docs Text

Updated hash computation for empty strings in the FastHash implementation to conform with the
handling in Apache Impala. For Bloom filter predicate pushdown feature that uses FastHash,
this makes the Kudu client older than version 1.15.0 incompatible with Kudu server version 1.15.0
and Kudu client version at or newer than 1.15.0 incompatible with Kudu server version earlier than
1.15.0. Both client library and Kudu server need to be updated to version 1.15.0 or above if using
the Bloom filter predicate feature.

Manifestations of this incompatibility are following messages in the logs:

- "Not implemented: call requires unsupported application feature flags: 4".
- "Not implemented: call requires unsupported application feature flags: 5".

• Assignee: Bankim Bhavsar
• Reporter: Bankim Bhavsar
• Created: 18/May/21 17:22
• Updated: 04/Jun/21 13:47
• Resolved: 18/May/21 17:26

Description

Fast hash used with Bloom filter predicate pushdown has special handling for nullptr.

https://github.com/apache/kudu/blob/master/src/kudu/util/hash_util.h#L95

However there isn't any special handling for empty objects/strings. Fast hash for an empty string with seed=0 generates a hash value of 0. This doesn't set any bits in Bloom filter and as a result empty strings are reported as not present.

Impala uses the direct bloom filter approach and includes special handling for empty strings.
https://github.com/apache/impala/blob/master/be/src/runtime/raw-value.inline.h#L352

This leads to discrepancy between Impala and Kudu and returns incorrect join results.