[ https://issues.apache.org/jira/browse/HDFS-10391?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16179157#comment-16179157 ]
Daryn Sharp commented on HDFS-10391: ------------------------------------ bq. In what situation do you see the extra RPC endpoint as disadvantageous? I used to often consider switching to a service port for the DNs because many handlers were tied up with IBRs. I reconsidered after I added the async IBR processing which queues and batches IBRs. This massively reduced write lock contention. Reasonable throughput is maintained even under heavy replication loads that would have previously crippled the NN. So let me ask: What scenario leads you to believe a service port should be mandatory? > Always enable NameNode service RPC port > --------------------------------------- > > Key: HDFS-10391 > URL: https://issues.apache.org/jira/browse/HDFS-10391 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode, namenode > Reporter: Arpit Agarwal > Assignee: Gergely Novák > Labels: Incompatible > Attachments: HDFS-10391.001.patch, HDFS-10391.002.patch, > HDFS-10391.003.patch, HDFS-10391.004.patch, HDFS-10391.005.patch, > HDFS-10391.006.patch, HDFS-10391.007.patch, HDFS-10391.008.patch, > HDFS-10391.009.patch, HDFS-10391.010.patch, HDFS-10391.v5-v6-delta.patch > > > The NameNode should always be setup with a service RPC port so that it does > not have to be explicitly enabled by an administrator. -- This message was sent by Atlassian JIRA (v6.4.14#64029) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org