You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@impala.apache.org by "Joe McDonnell (Jira)" <ji...@apache.org> on 2025/06/03 22:00:00 UTC
[jira] [Created] (IMPALA-14120) Test tuple cache with one node having tuple caching disabled
Joe McDonnell created IMPALA-14120:
--------------------------------------
Summary: Test tuple cache with one node having tuple caching disabled
Key: IMPALA-14120
URL: https://issues.apache.org/jira/browse/IMPALA-14120
Project: IMPALA
Issue Type: Test
Components: Test
Affects Versions: Impala 5.0.0
Reporter: Joe McDonnell
The build side of the hash join is always feeding from an exchange. Since we want to cache past a hash join, there can be a case where some executors have cache hits above the hash join and others do not. The ones that do not have cache hits need to read from the exchange on the build side. The ones that have cache hits won't read from the exchange.
We should have tests for scenarios like this to make sure that the exchange makes progress. One way to do that is to have one executor not have a tuple cache at all, so it always needs the exchange.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)