[ 
https://issues.apache.org/jira/browse/HDDS-17725?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

ASF GitHub Bot updated HDDS-17725:
----------------------------------
    Labels: pull-request-available  (was: )

> Enforce same-storage-type moves in Disk Balancer regardless of the configured 
> choosing policy
> ---------------------------------------------------------------------------------------------
>
>                 Key: HDDS-17725
>                 URL: https://issues.apache.org/jira/browse/HDDS-17725
>             Project: Apache Ozone
>          Issue Type: Sub-task
>          Components: Ozone Datanode
>            Reporter: Devesh Kumar Singh
>            Assignee: Devesh Kumar Singh
>            Priority: Major
>              Labels: pull-request-available
>
> Disk Balancer allows a configurable ContainerChoosingPolicy to select a 
> container and its source/destination volumes. The default policy already 
> selects volumes of the same storage type, but DiskBalancerService does not 
> independently enforce this rule before copying data.
> A custom policy could therefore return a cross-type pair, causing ordinary 
> disk balancing to change a replica’s storage placement.
> The configuration is:
> *hdds.datanode.disk.balancer.container.choosing.policy*
> For example, consider these volumes on one datanode:
> | Volume | Storage type | Utilization |
> |  SSD-1  |         SSD       |      80%     |
> |  SSD-2  |        SSD        |     40%      |
> |  DISK-1  |        DISK       |     10%       |
> A custom policy choosing the fullest and emptiest volumes could select SSD-1 
> → DISK-1. Disk Balancer should reject that move. An eligible SSD-1 → SSD-2 
> move should remain allowed.
> Add a service/task-level check that rejects moves between different source 
> and destination volume storage types before copying data. Ensure rejection 
> releases any destination space reserved during selection and leaves the 
> source container unchanged.
> *Acceptance criteria:*
> - Enforce same-type movement regardless of the configured choosing policy.
> - Reject cross-type candidates before data copy and clean up 
> reservations/accounting.
> - Preserve existing same-type movement, container eligibility and failure 
> handling.
> - Add regression tests using an injected custom policy that returns a 
> cross-type candidate.
> - Keep cross-tier policy migration/promotion separate from ordinary Disk 
> Balancer behavior.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

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

Reply via email to