Joe McDonnell created IMPALA-13899:
--------------------------------------
Summary: Tuple caching does not preserve warnings from the
original run
Key: IMPALA-13899
URL: https://issues.apache.org/jira/browse/IMPALA-13899
Project: IMPALA
Issue Type: Task
Components: Backend
Affects Versions: Impala 5.0.0
Reporter: Joe McDonnell
When running queries with abort_on_error=0, the query can report warnings about
issues reading certain files. For example, test cases from
data_errors/test_data_errors.py can rely on the query reporting errors like
this:
{noformat}
use functional_avro_snap;
select * from bad_avro_snap_strings;
WARNINGS: Problem parsing file
hdfs://localhost:20500/test-warehouse/bad_avro_snap_strings_avro_snap/invalid_union.avro
at 174 (1 of 4 similar)
File
'hdfs://localhost:20500/test-warehouse/bad_avro_snap_strings_avro_snap/truncated_string.avro'
is corrupt: truncated data block at offset 155
File
'hdfs://localhost:20500/test-warehouse/bad_avro_snap_strings_avro_snap/invalid_union.avro'
is corrupt: invalid union value 4 at offset 174 (1 of 2 similar)
File
'hdfs://localhost:20500/test-warehouse/bad_avro_snap_strings_avro_snap/negative_string_len.avro'
is corrupt: invalid length -7 at offset 164{noformat}
These warnings are not stored along with the tuple cache results, so they won't
be displayed if there is a tuple cache hit (unless tuple cache verification is
on). This can cause some tests to fail when tuple caching is on (particularly
data_errors/test_data_errors.py).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
