Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
2.0.0-alpha
-
None
-
None
-
None
Description
Currently a datanode reports the transceiver count to namenode. Namenode aggregates this as TotalLoad metrics and is used for monitoring cluster activity. It is also used for making block placement decision, to qualify if a datanode is a good target.
Currently transceiver count is = 1 (for XceiverServer) + 1 * (number of readers) + 2 * (Number of writers in pipeline) + 1 * (number of datanode replications) + 1 * (number of recover blocks)
Should the number of transceiver just reflect number of readers + writers, instead of reporting it as is currently done? We should perhaps report readers and writers as separate count (I will file a separate jira for this).