HDDS-9090. NullPointerException thrown by ReplicationManagerMetrics because of the order of initialization

[aswinshakil]

What changes were proposed in this pull request?

ECUnderReplicationHandler is initialized before initializing ReplicationManagerMetrics, So whenever any ReplicationManagerMetrics is used, it throws NPE. The behavior is same for all these other classes and will throw NPEs,
ECOverReplicationHandler, ECMisReplicationHandler, RatisUnderReplicationHandler, RatisMisReplicationHandler

  ECUnderReplicationHandler(final PlacementPolicy containerPlacement,
      final ConfigurationSource conf, ReplicationManager replicationManager) {
    this.containerPlacement = containerPlacement;
    this.currentContainerSize = (long) conf
        .getStorageSize(ScmConfigKeys.OZONE_SCM_CONTAINER_SIZE,
            ScmConfigKeys.OZONE_SCM_CONTAINER_SIZE_DEFAULT, StorageUnit.BYTES);
    this.replicationManager = replicationManager;
    this.metrics = replicationManager.getMetrics(); --> still null because replication manager metrics is not initialized yet.
 }

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-9090

How was this patch tested?

Manually tested by adding a line and verified it threw NPE.

  ECUnderReplicationHandler(final PlacementPolicy containerPlacement,
      final ConfigurationSource conf, ReplicationManager replicationManager) {
    this.containerPlacement = containerPlacement;
    this.currentContainerSize = (long) conf
        .getStorageSize(ScmConfigKeys.OZONE_SCM_CONTAINER_SIZE,
            ScmConfigKeys.OZONE_SCM_CONTAINER_SIZE_DEFAULT, StorageUnit.BYTES);
    this.replicationManager = replicationManager;
    this.metrics = replicationManager.getMetrics();
    this.metrics.getEcPartialReconstructionNoneOverloadedTotal(); ---> Added line manually just to check NPE
}
java.lang.NullPointerException
	at org.apache.hadoop.hdds.scm.container.replication.ECUnderReplicationHandler.<init>(ECUnderReplicationHandler.java:78)
	at org.apache.hadoop.hdds.scm.container.replication.ReplicationManager.<init>(ReplicationManager.java:247)
	at org.apache.hadoop.hdds.scm.container.replication.TestReplicationManager$1.<init>(TestReplicationManager.java:177)
	at 

Even after metrics is initialized, the metrics inside other classes are null

[adoroszlai]

Thanks @aswinshakil for the fix. Please note that ReplicationManager can be stopped/restarted by admin from CLI.
stop() unregisters the metrics source, and start() used to (re)create it. But now it isn't re-created, so it won't work after restart. One way to fix this is to add a method in ReplicationManagerMetrics to register it with the metrics system (without creating a new instance), and call this new method from start().

[aswinshakil]

Thanks for pointing out @adoroszlai . I wasn't aware of that, feel free to assign the JIRA to me if created, or I can create one.

[adoroszlai]

Thanks @aswinshakil, created HDDS-9235.