Uploaded image for project: 'Apache Ozone'
  1. Apache Ozone
  2. HDDS-7349

Flaky integration test have memory leak for RatisDropwizardExports

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Resolved
    • Minor
    • Resolution: Fixed
    • None
    • 1.4.0
    • None

    Description

      Memory leak observed in flaky integration test causing OutOfMemory multiple times, and causing the cancel of the case.

      1. registeration of RatisDropwizardExports is done multiple time, even its registered.
      – This is called by Ratis CopyOnWrite when another metrics get registered.
      – Since the track of above object is lost, no way to cleanup later

      ------------------stack showing registeration-----------
        at io.prometheus.client.CollectorRegistry.register(Lio/prometheus/client/Collector;)V (CollectorRegistry.java:60)
        at org.apache.hadoop.hdds.server.http.RatisDropwizardExports.registerDropwizard(Lorg/apache/ratis/metrics/RatisMetricRegistry;Ljava/util/Map;)V (RatisDropwizardExports.java:75)
        at org.apache.hadoop.hdds.server.http.RatisDropwizardExports.lambda$registerRatisMetricReporters$0(Ljava/util/Map;Lorg/apache/ratis/metrics/RatisMetricRegistry;)V (RatisDropwizardExports.java:53)
        at org.apache.hadoop.hdds.server.http.RatisDropwizardExports$$Lambda$384.accept(Ljava/lang/Object;)V (Unknown Source)
        at org.apache.ratis.metrics.impl.MetricRegistriesImpl.lambda$null$0(Lorg/apache/ratis/metrics/RatisMetricRegistry;Ljava/util/function/Consumer;)V (MetricRegistriesImpl.java:72)
        at org.apache.ratis.metrics.impl.MetricRegistriesImpl$$Lambda$618.accept(Ljava/lang/Object;)V (Unknown Source)
        at java.util.concurrent.CopyOnWriteArrayList.forEach(Ljava/util/function/Consumer;)V (CopyOnWriteArrayList.java:895)
        at org.apache.ratis.metrics.impl.MetricRegistriesImpl.lambda$create$1(Lorg/apache/ratis/metrics/MetricRegistryInfo;)Lorg/apache/ratis/metrics/RatisMetricRegistry; (MetricRegistriesImpl.java:72)
      ------------------------------------
      

      2. after stop of Datanode, still ratis registeration going on causing to get accumulated.

      This is observed that Datanode is stopped, but still Datanode having "ratisMetricsMap" having new metrics registered.

      Attachments

        1. MetricRegistries.patch
          4 kB
          Tsz-wo Sze
        2. 3858_review2.patch
          6 kB
          Tsz-wo Sze
        3. 3858_review.patch
          8 kB
          Tsz-wo Sze

        Issue Links

          Activity

            People

              sumitagrawl Sumit Agrawal
              sumitagrawl Sumit Agrawal
              Votes:
              0 Vote for this issue
              Watchers:
              2 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: