I'm looking into available JMX metrics for namenode. What can be used as a representative metric for plotting namenode load and traffic from other nodes?
I see some of the metrics related to CPU currently that can be used for load.
Any inputs are appreciated! Thanks!
You could use the following:
1. NumOpenConnections - provides the number of open connections to Namenode on port 8020 from other nodes.
Object Name: Hadoop:service=NameNode,name=RpcActivityForPort8020
2. BlockReportAvgTime could be used to roughly gauge how quickly the name node is getting block report. A peak could indicate namenode is loaded.
3. RpcQueueTimeAvgTime - An abnormal value could indicate some load on the namenode. This is present for both namenode and datanode