You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@mesos.apache.org by "Chun-Hung Hsiao (JIRA)" <ji...@apache.org> on 2019/01/28 20:10:00 UTC

[jira] [Created] (MESOS-9540) Support `DESTROY_DISK` on `RAW` disks.

Chun-Hung Hsiao created MESOS-9540:
--------------------------------------

             Summary: Support `DESTROY_DISK` on `RAW` disks.
                 Key: MESOS-9540
                 URL: https://issues.apache.org/jira/browse/MESOS-9540
             Project: Mesos
          Issue Type: Improvement
          Components: storage
            Reporter: Chun-Hung Hsiao


Currently the experimental {{DESTROY_DISK}} operation only applies to {{BLOCK}} or {{MOUNT}} disk resources. We should consider supporting {{DESTROY_DISK}} on {{RAW}} disk resources with source IDs as well. This could be handy in e.g., the following scenario:

1. The framework issued a {{CREATE_DISK}}.
2. The SLRP received {{CREATE_DISK}} and translated to a {{CreateVolume}} CSI call.
3. While the {{CreateVolume}} is ongoing, the agent was restarted with a new agent ID, causing the SLRP to lose its bookkeeping and start with a new RP ID as well, and hence the {{CREATE_DISK}} operation was lost.
4. The {{CreateVolume}} call succeeded and the new SLRP picked it up as a pre-existing CSI volume.

In the above case, the framework should be able to choose to either "import" the CSI volume through {{CREATE_DISK}}, or directly reclaim the space through {{DESTROY_DISK}}. Currently the framework needs to always import the CSI volume before reclaiming the space.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)