[FLINK-4087] JMXReporter can't handle port conflicts

• Type: Improvement
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 1.1.0
• Fix Version/s: 1.1.0
• Component/s: Runtime / Metrics
• Assignee: Chesnay Schepler
• Reporter: Chesnay Schepler
• Created: 17/Jun/16 13:46
• Updated: 27/Jun/16 11:18
• Resolved: 27/Jun/16 11:18

Description

The JMXReporter is currently configured to use a single port that is set as a JVM argument.

This approach has a few disadvantages:

If multiple TaskManagers run on the same machine only 1 can expose metrics. This issue is compounded by the upcoming JobManager metrics, which would then prevent TM metrics from being exposed in local setups.

Currently, we prevent other TM's from exposing metrics by checking the the start-daemon-sh whether a TM is already running, and if so clear the arguments. This isn't a particular safe way to do it, and this script is not used when deploying on yarn, leading to TM failures since the JVM can't allocate the JMX port.

We should find a way to specifiy port-ranges for JMX and log the final port used.

Issue Links

• links to: GitHub Pull Request #2127
• links to: GitHub Pull Request #2145