[FLINK-9665] PrometheusReporter does not properly unregister metrics

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.4.2, 1.5.0, 1.6.0
• Fix Version/s: 1.4.3, 1.5.1, 1.6.0
• Component/s: Runtime / Metrics
• Labels: pull-request-available
• Assignee: Jelmer Kuperus
• Reporter: Chesnay Schepler
• Created: 26/Jun/18 16:46
• Updated: 17/May/21 04:08
• Resolved: 03/Jul/18 08:13

Description

The PrometheusReporter groups metrics with the same logical scope in a single Collector which are periodically polled by Prometheus.

New metrics are added to an existing collector, and a reference count is maintained so we can eventually cleanup the Collector itself.

For removed metrics we decrease the reference count, do not however remove the metrics that were added. As a result the collector will continue to expose metrics, as long as at least 1 metric exists with the same logical scope.

If the collector is a io.prometheus.client.Gauge we can use the #remove() method. For histograms we will have to modify our HistogramSummaryProxy class to allow removing individual histograms.

Issue Links

• links to: GitHub Pull Request #6211