[ https://issues.apache.org/jira/browse/HDDS-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16956364#comment-16956364 ]
Siddharth Wagle edited comment on HDDS-1574 at 10/21/19 6:28 PM: ----------------------------------------------------------------- {noformat} Do you mean we don't want 2 pipelines to share the exactly same group of datanodes as members? {noformat} Yes, those were my initial thoughts on that matter. With the constraint of raft log per group on separate disks and the fact that we randomly choose a rack-local node should mean this constraint should already be satisfied, however it could be enforced as a part of this jira. Moreover, the same nodes participating in more than one group does not solve _pipeline availability_. The only reason to have such a setup is if Ozone cannot saturate disk bandwidth with one group for the datanodes (assuming traditionally we saturate disk before network). So, I think this constraint is still relevant, we could make enforcement configurable (on by default). We could of course punt this until rest of the multi-raft stuff is implemented. was (Author: swagle): {noformat} Do you mean we don't want 2 pipelines to share the exactly same group of datanodes as members? {noformat} Yes, those were my initial thoughts on that matter. With the constraint of raft log per group on separate disks and the fact that we randomly choose a rack-local node should mean this constraint should already be satisfied, however it could be enforced as a part of this jira. Moreover, the same nodes participating in more than one group does not solve _pipeline availability_. The only reason to have such a setup is if Ozone cannot saturate disk bandwidth with one group (assuming traditionally we saturate disk before network). So, I think this constraint is still relevant, we could of course punt this until rest of the multi-raft stuff is implemented. > Ensure same datanodes are not a part of multiple pipelines > ---------------------------------------------------------- > > Key: HDDS-1574 > URL: https://issues.apache.org/jira/browse/HDDS-1574 > Project: Hadoop Distributed Data Store > Issue Type: Sub-task > Components: SCM > Reporter: Siddharth Wagle > Assignee: Li Cheng > Priority: Major > > Details in design doc. -- This message was sent by Atlassian Jira (v8.3.4#803005) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org