David Mollitor created HIVE-23829:
-------------------------------------
Summary: Compute Stats Incorrect for Binary Columns
Key: HIVE-23829
URL: https://issues.apache.org/jira/browse/HIVE-23829
Project: Hive
Issue Type: Bug
Reporter: David Mollitor
Assignee: David Mollitor
I came across an issue when working on [HIVE-22674].
The SerDe used for processing binary data tries to auto-detect if the data is
in Base-64. It uses
{{org.apache.commons.codec.binary.Base64#isArrayByteBase64}} which has two
issues:
# It's slow since it will check if the array is compatible,... and then process
the data (examines the array twice)
# More importantly, this method _Tests a given byte array to see if it contains
only valid characters within the Base64 alphabet. Currently the method treats
whitespace as valid._
https://commons.apache.org/proper/commons-codec/apidocs/org/apache/commons/codec/binary/Base64.html#isArrayByteBase64-byte:A-
The
[qtest|https://github.com/apache/hive/blob/f98e136bdd5642e3de10d2fd1a4c14d1d6762113/ql/src/test/queries/clientpositive/compute_stats_binary.q]
for this feature uses full sentences (which includes spaces)
[here|https://github.com/apache/hive/blob/f98e136bdd5642e3de10d2fd1a4c14d1d6762113/data/files/binary.txt]
and therefore it thinks this data is Base-64 and returns an incorrect
estimation for size.
This should really not auto-detect Base64 data and instead it should be enabled
with a table property.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
