[
https://issues.apache.org/jira/browse/HDDS-8824?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17822381#comment-17822381
]
Sumit Agrawal commented on HDDS-8824:
-------------------------------------
for quota repair tool, it may not be required offline and CLI option to trigger
quota repair is possible with below approach,
At OM when qouta count repair request comes,
# It can create a temp snapshot of DB of OM
# Can do repair count (as existing logic in HDDS-8742 can be helpful) with
this new db handler
# After having consolidated count, can update differential count for every
bucket.
Advantage:
* No need bring OM down till quote count is repaired (to avoid parallel update
of count)
* Existing implementation of HDDS-8742 can use this mechanism to reduce
blocking of OM for upgrade from old version
Related to snapshot feature, quota needs capture files in snapshot exclusively
present as part of that. Earlier this is not captured/handled as part of
HDDS-8742 as snapshot feature was not present during that point of time.
As part of this impact,
* Need mark quota consumed for those keys/files which are present in snapshot
but not present in active db (snap diff)
* This Needs avoid deletion of snapshot parallel till this operation is
running OR also need have snapshot copy. This need further discuss for handling.
I think when files/keys/directory deleted from snapshot, it handles reducing
quota. This can be verified as normal impact of snapshot to quota.
> Quota count repair tool
> -----------------------
>
> Key: HDDS-8824
> URL: https://issues.apache.org/jira/browse/HDDS-8824
> Project: Apache Ozone
> Issue Type: Sub-task
> Reporter: Sumit Agrawal
> Assignee: Dave Teng
> Priority: Minor
>
> quota count can go wrong based on bug in the system. All the known bug is
> fixed with parent task HDDS-7479 related to quota calculation whenever
> file/keys/directories are added/removed.
> And upgrade is provided with HDDS-8742.
> CLI based tool provide offline capability for recalculating quota,
> * For keys/files with 1 billion takes approx 30 minutes, so this can be run
> in offline mode
> * can be triggered at bucket level
>
> Few impact to be takenup:
> * snapshot impact needs to be handled as quota is shared between snapshot db
> and active db.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
