[ https://issues.apache.org/jira/browse/KAFKA-5199?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16387294#comment-16387294 ]
Ewen Cheslack-Postava commented on KAFKA-5199: ---------------------------------------------- [~cmccabe] Do we still need this? I spent time quite awhile ago cleaning up the tests to not rely on fixed ports as this, as well as things like fixed /tmp locations for data, are well known causes of transient failures. This is also why ducktape avoids sharing resources as aggressively (and expensively!) as it does. Docker has definitely become more popular, but I don't think it is safe to assume every developer will have it available. While we don't emphasize testing for them, Windows developers seem the least likely to have it. Even developers on OS X may not, and anyone on a BSD that we'd otherwise expect to work would probably be unable to run tests. Is baking a Linux-specific tool into the build something we really want to do? > Run Kafka JUnit build inside Docker > ----------------------------------- > > Key: KAFKA-5199 > URL: https://issues.apache.org/jira/browse/KAFKA-5199 > Project: Kafka > Issue Type: Bug > Reporter: Colin P. McCabe > Assignee: Colin P. McCabe > Priority: Major > > We should run the Kafka JUnit build inside Docker. This would avoid "port in > use" conflicts when running multiple jobs on the same node with a fixed port. > It would also help lock down the specific Linux environment and make > failures more reproducible. -- This message was sent by Atlassian JIRA (v7.6.3#76005)