Joe McDonnell created IMPALA-13882:
--------------------------------------
Summary: Iceberg deletes don't work with tuple caching
Key: IMPALA-13882
URL: https://issues.apache.org/jira/browse/IMPALA-13882
Project: IMPALA
Issue Type: Bug
Components: Backend
Affects Versions: Impala 5.0.0
Reporter: Joe McDonnell
Comparing enable_tuple_cache=false to enable_tuple_cache=true shows that tuple
caching produces incorrect results for Iceberg tables with deletes.
Examples:
{noformat}
set enable_tuple_cache=false;
[localhost:21050] functional_parquet> select * from
iceberg_v2_positional_update_all_rows;
+---+---+
| i | s |
+---+---+
| 1 | A |
| 2 | B |
| 3 | C |
+---+---+
Fetched 3 row(s) in 0.11s
set enable_tuple_cache=true;
[localhost:21050] functional_parquet> select * from
iceberg_v2_positional_update_all_rows;
+---+---+
| i | s |
+---+---+
| 1 | A |
| 2 | B |
| 3 | C |
| 1 | a |
| 2 | b |
| 3 | c |
+---+---+
Fetched 6 row(s) in 0.12s{noformat}
One theory is that some logic can depend on the scan node being the immediate
child of the delete node. Tuple caching can insert a TupleCacheNode in between,
so maybe the logic wasn't built to handle that case.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
