Have found the Reserved Space For Replicas is not released on some cases in a Cx Prod cluster. There are few fixes like
HDFS-9530 and HDFS-8072 but still the issue is not completely fixed. Have tried to debug the root cause but this will take lot of time as it is Cx Prod Cluster.
But we have an easier way to fix the issue completely by releasing any remaining reserved space of the Replica from the Volume. DataXceiver#writeBlock finally will call BlockReceiver#close which will check if the ReplicaInfo has any remaining reserved space, if so release from the Volume.