[ 
https://issues.apache.org/jira/browse/HDDS-5526?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Sammi Chen resolved HDDS-5526.
------------------------------
    Fix Version/s: 1.4.0
       Resolution: Fixed

> ContainerBalancer#checkConditionsForBalancing pre-emptively checks iteration 
> limits.
> ------------------------------------------------------------------------------------
>
>                 Key: HDDS-5526
>                 URL: https://issues.apache.org/jira/browse/HDDS-5526
>             Project: Apache Ozone
>          Issue Type: Sub-task
>          Components: SCM
>            Reporter: Siddhant Sangwan
>            Assignee: Siddhant Sangwan
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 1.4.0
>
>
> ContainerBalancer#checkConditionsForBalancing is used to check whether the 
> max size to move or max datanodes to involve limits have been reached.
> Currently, ContainerBalancer#doIteration uses this method before finding a 
> ContainerMoveSelection for a source datanode. Since the check is performed 
> before selecting the next container to move, the configuration 
> OZONE_SCM_CONTAINER_SIZE (default is 5GB) is used to predetermine if:
> sizeMovedPerIteration + configured container size >= 
> maxSizeToMovePerIteration.
> This check will cause the current iteration to stop in cases like: 
> maxSizeToMovePerIteration = 100GB, sizeMovedPerIteration = 96GB, and 
> OZONE_SCM_CONTAINER_SIZE = 5GB.
> On one hand the current implementation can avoid extra work by not letting 
> balancer continue when we are very close to the limit. On the other hand, 
> because of this implementation, we will never reach the limit. This Jira 
> tracks if it's better to check conditions using container size in 
> ContainerBalancer#checkConditionsForBalancing after having found a 
> ContainerMoveSelection in ContainerBalancer#doIteration.
> Another solution is to allow one container move before checking the limit. 
> The downside is that if the limits are really low (size lesser than 5GB, 
> number of datanodes lesser than 2), we will fail to respect them.
> Another way is to check for container size _while_ picking a 
> ContainerMoveSelection.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

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

Reply via email to