HDDS-5536. Add metrics for ReplicationSupervisor

[ChenSammi]

https://issues.apache.org/jira/browse/HDDS-5536

[jojochuang]

we create and register one MetricsSystem per DN, which is not a problem for typical deployments.
But i imagine it can potentially be problematic for unit tests because there are multiple DN objects in unit test heap and only one of them will successfully register.

[jojochuang]

Other than that the code looks good to me.

[ChenSammi]

Yea, that will be a "metrics already exists" warning in unit test, not only for this metrics, but also for other DN metries.
What's a general way to handle this? Maybe we can append DN's hostname+port to each metrics source name?
I'm not sure how HDFS handle this.

[jojochuang]

HDFS DataNodeMetrics.java
dnName is host name + port. So yeah, it would be a good idea to try to make them unique.
public static DataNodeMetrics create(Configuration conf, String dnName) {
String sessionId = conf.get(DFSConfigKeys.DFS_METRICS_SESSION_ID_KEY);
MetricsSystem ms = DefaultMetricsSystem.instance();
JvmMetrics jm = JvmMetrics.create("DataNode", sessionId, ms);
String name = "DataNodeActivity-"+ (dnName.isEmpty()
? "UndefinedDataNodeName"+ ThreadLocalRandom.current().nextInt()
: dnName.replace(':', '-'));

// Percentile measurement is off by default, by watching no intervals
int[] intervals = 
    conf.getInts(DFSConfigKeys.DFS_METRICS_PERCENTILES_INTERVALS_KEY);

return ms.register(name, null, new DataNodeMetrics(name, sessionId,
    intervals, jm));

}

[jojochuang]

https://github.com/apache/hadoop/blob/95454d821c6e4c08d49df195d7c1e0ea1aacafb9/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/metrics/DataNodeMetrics.java#L259

[ChenSammi]

I see. @jojochuang , can we do it in a followup JIRA? since other DN metrics also need be improved.

[adoroszlai]

@jojochuang can we commit this and file a follow-up task?

[ChenSammi]

Fired HDDS-5838 to track the comments.

[adoroszlai]

Thanks @ChenSammi for the patch.

[ChenSammi]

Thanks @jojochuang and @adoroszlai for the code review.