Joe McDonnell created IMPALA-15556:
--------------------------------------
Summary: Reduce variation between runs for
bin/single_node_perf_run.py and make data more accessible
Key: IMPALA-15556
URL: https://issues.apache.org/jira/browse/IMPALA-15556
Project: IMPALA
Issue Type: Task
Components: Infrastructure
Affects Versions: Impala 5.0.0
Reporter: Joe McDonnell
There are a series of improvements for bin/single_node_perf_run.py that should
help reduce variation and make the data easier to analyze:
* Currently, bin/single_node_perf_run.py runs queries round-robin on each
impalad. Using an exclusive coordinator would avoid duplicating the metadata
and could reduce variation slightly. It's also more similar to typical Impala
deployments
* Exporting the raw data as a CSV would make it easier to pull into a
spreadsheet and analyze
* Clearing the OS buffer cache and compacting memory could reduce variation
and avoid a bias towards the second run (because the data is in the cache)
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
