chihsuan commented on PR #10648:
URL: https://github.com/apache/ozone/pull/10648#issuecomment-4991686156

   > I think this is good. Though wonder if we want to put this change behind a 
flag.
   
   I agree that a flag would make this safer for existing EC deployments. I'll 
update the PR.
   
   > One corner case that's not going too well after this is when a new empty 
datanode is added to the rack, in which case the node will receive most traffic 
until it becomes even with the rest in the rack.
   
   That is a valid concern. The same power-of-two rationale is documented in 
`SCMContainerPlacementCapacity`:
   
   
https://github.com/apache/ozone/blob/3fdf0015721d85053fd2c4dc34010155ee4e52c9/hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/container/placement/algorithms/SCMContainerPlacementCapacity.java#L54-L57
   
   Unlike a greedy least-utilized-node policy, a newly added empty datanode is 
considered only when it is sampled.
   
   In `RackScatter`, with N eligible datanodes in a rack, the new node is 
included in the two distinct candidates in 2/N of selections. The two-node rack 
is the degenerate case: both nodes are always compared, so the empty node is 
selected until utilization converges.
   
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to