[ https://issues.apache.org/jira/browse/HBASE-18775?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16176929#comment-16176929 ]
Zach York commented on HBASE-18775: ----------------------------------- [~ashish singhi] I'm not convinced that this should be implemented as a coprocessor. The current read-only regions aren't implemented as a coprocessor. I'd like to hear your pros/cons for each approach, perhaps there is something I'm missing. As for the ZK option, it seems like there is a desire (especially for large clusters) to use ZK as little as possible (although maybe this is only for assignment). Is there an inherent scaling issue with using ZK? Perhaps [~tedyu] [~stack] might know better (than me) about the limitations of using ZK for large clusters. I do like the more dynamic aspect of this (although in our initial approach this setting was final unless the server was restarted with new configs). It could be useful to turn a read-replica into a primary cluster. > Add a Global Read-Only property to turn off all writes for the cluster > ---------------------------------------------------------------------- > > Key: HBASE-18775 > URL: https://issues.apache.org/jira/browse/HBASE-18775 > Project: HBase > Issue Type: Sub-task > Components: master, regionserver > Affects Versions: HBASE-18477 > Reporter: Zach York > Assignee: Zach York > Attachments: HBASE-18775.HBASE-18477.001.patch, > HBASE-18775.HBASE-18477.002.patch, HBASE-18775.HBASE-18477.003.patch, > HBASE-18775.HBASE-18477.004.patch > > > As part of HBASE-18477, we need a way to turn off all modification for a > cluster. This patch extends the read only mode used by replication to disable > all data and metadata operations. -- This message was sent by Atlassian JIRA (v6.4.14#64029)