[FLINK-3452] Procuring more Travis CI build capacity for Flink

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Assignee: Unassigned
• Reporter: Wes McKinney
• Created: 19/Feb/16 14:18
• Updated: 13/Apr/21 20:31
• Resolved: 30/May/16 10:28

Description

Greetings from Apache Parquet!

In diagnosing Travis CI build queue slowness over the last several weeks we've been looking at overall utilization of the 30 concurrent Travis workers available to all of the ASF's projects on github.com/apache

Finally, Travis CI produced a database dump for me. Here is a total duration in seconds for each of the apache's active projects for the last 180 days:

https://gist.githubusercontent.com/drogus/d01a02aa9e0f7c380bde/raw/05be62eb1dd76085dd6cc9740e496ac2a111bc33/duration-by-project.txt

Indeed, my anecdotal experience has been that we are often waiting on queued Flink builds to free up Travis workers to run parquet-cpp builds (which take about 5 minutes to run).

Given Flink's relative Travis CI utilization and general growth and success of the project (congratulations!), I am wondering if you have explored options to either increase the number of Travis CI workers available to ASF projects like Flink or to procure dedicated build slaves for Flink (similar to what Spark has done). Let me know what you think.

Thank you in advance and best wishes for 2016.