[ 
https://issues.apache.org/jira/browse/HDDS-5905?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17528542#comment-17528542
 ] 

UENISHI Kota commented on HDDS-5905:
------------------------------------

The document implies that enabling Ratis is just one-way config change, but I'm 
not sure. Is disabling Ratis (turning from true to false) is officially 
supported? If I understand correctly, 1.1.0 does not have OM HA (Ratis).

 

Example scenario with 1.1 would be like (I know it's very complex):
 # Upgrade the cluster to 1.3 (Ratis enabled)
 # Downgrade to 1.2, and then after downgrade to 1.1 (Ratis disabled, 
transaction index reset)
 # Upgrade to 1.2 with Ratis disabled
 # Upgrade to 1.3 with Ratis enabled

 

Either is fine to me, but I just want to know the right assumption - if we 
don't support turning Ratis off, then I'll update my design document and my 
pull request (currently, only for HDDS-6483). Please let me know, [~Sammi] 

> Race condition of deletion service and active object deletion
> -------------------------------------------------------------
>
>                 Key: HDDS-5905
>                 URL: https://issues.apache.org/jira/browse/HDDS-5905
>             Project: Apache Ozone
>          Issue Type: Bug
>          Components: OM
>    Affects Versions: 0.5.0, 1.0.0, 1.1.0, 1.2.0
>            Reporter: UENISHI Kota
>            Assignee: UENISHI Kota
>            Priority: Major
>
> Race condition of deletion service - the deletion service does delete blocks 
> and later delete the entry in delete table without any locking. After the 
> deletion service fetches the keys and before deleting them from the table, a 
> user's concurrent  deletion of active key (and addition to the delete table) 
> will be lost without deleting blocks.
> From Bharat's [Slack 
> comment|https://the-asf.slack.com/archives/C5RK7PWA1/p1635135579007300?thread_ts=1635134167.007000&cid=C5RK7PWA1]:
> There seems to issue When same key is created/deleted we might miss deleting 
> some blocks.  Scenario below.
> 1. lets say key deleted we add to delete table
> 2. the BG picks up and completed sending to SCM,
> 3. After SCM ack, it deletes from delete table using purgekey
> 4. Now there is a new key addeed/deleted between 2 and 3 we add to delete key 
> table, we will delete newly added entries which are not deleted by purge key.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)

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

Reply via email to