[HADOOP-17804] Prometheus metrics only include the last set of labels

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.3.1
• Fix Version/s: 3.4.0, 3.3.2
• Component/s: common
• Labels: pull-request-available
• Hadoop Flags: Reviewed
• Assignee: Adam Binford
• Reporter: Adam Binford
• Created: 15/Jul/21 16:02
• Updated: 09/Sep/21 08:13
• Resolved: 09/Sep/21 08:12

Description

A prometheus endpoint was added in https://issues.apache.org/jira/browse/HADOOP-16398, but the logic that puts them into a map based on the "key" incorrectly hides any metrics with the same key but different labels. The relevant code is here: https://github.com/apache/hadoop/blob/trunk/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/sink/PrometheusMetricsSink.java#L55

The labels/tags need to be taken into account, as different tags mean different metrics. For example, I came across this while trying to scrape metrics for all the queues in our scheduler. Only the last queue is included because all the metrics have the same "key" but a different "queue" label/tag.

Issue Links

• is related to: HADOOP-17893 Improve PrometheusSink for Namenode TopMetrics
• links to: GitHub Pull Request #3369