Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
-
ghx-label-2
Description
Our tests (e.g., https://jenkins.impala.io/job/ubuntu-16.04-from-scratch/buildTimeTrend) tend to take about 4 hours. This can be improved.
I'm opening this JIRA track those changes. I'm currently looking at:
- Parallelizing multiple data-load steps: TPC-DS, TPC-H, and Functional take ~65 minutes when serialized. They take 35 minutes if running in parallel.
- Parallelizing compute stats: this takes ~10 minutes; probably can be faster.
The trickier thing is parallelizing fe tests, ee tests, and custom cluster tests. The approach I'm taking is to create a docker container with everything in it (including data load), and then running tests in parallel. This is a bit messier, but I think it has some legs when it comes to using machines with many cores.