Re: [PR] HDDS-9645. Recon doesn't exclude out-of-service nodes when checking for healthy containers [ozone]

["xBis7 (via GitHub)" <gi...@apache.org>]

xBis7 commented on PR #5651:
URL: https://github.com/apache/ozone/pull/5651#issuecomment-1839300428

   @sodonnel After your comments regarding maintenance, I started looking more into it. We need to filter decommissioned nodes but not the ones in maintenance.
   
   Let's say that we have 3 replicas
   
   IN_SERVICE
   IN_SERVICE
   IN_SERVICE
   
   and then we put 1 in maintenance
   
   IN_SERVICE
   IN_SERVICE
   IN_MAINTENANCE
   
   we have 2 healthy replicas and no copy is created.
   
   If we put one more in maintenance, then we will have 1 healthy replica and therefore 1 copy is made to meet the replication requirements.
   
   IN_SERVICE
   IN_SERVICE
   IN_MAINTENANCE
   IN_MAINTENANCE
   
   For SCM there is no issue, for Recon the container is over-replicated as it has 4/3 replicas.
   
   The container will be over-replicated when both nodes are out of maintenance but until then, the SCM correctly ignores the excessive replicas.
   
   If we filter the nodes in maintenance, Recon will report under-replication even with 2 healthy replicas, which is worse because that means potential data loss.
   
   I'll temporarily convert this PR into a draft while trying to figure out a way to be consistent with SCM regarding maintenance nodes.
   


-- 
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: issues-unsubscribe@ozone.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@ozone.apache.org
For additional commands, e-mail: issues-help@ozone.apache.org