[
https://issues.apache.org/jira/browse/HDDS-3698?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17129495#comment-17129495
]
Yiqun Lin commented on HDDS-3698:
---------------------------------
Hi [~avijayan], a very good design for this. In this design, we talk about many
upgrade challenges things in different components. But I am interested in
detailed about what will happened when we upgrade failed and have to do
downgrade behavior.
Will we ensure the 100% consistency compared with trigger downgraded behavior
before? I see we introduce the new table TransactionInfoTable and let new
transactions write into this table. So from my understanding for this, this is
a temporary table to store the transaction during upgrade procedure (not
finalized). Since new feature operation is now allowed before finalized, so
this table won't store incompatible type transaction. Then in downgrading, OM
won't throw error when replying transaction table, right?
To protect our data before cluster be finalized, another thing I am thinking
that we would be better not directly delete file data. We can move deleted
container into a trash directory in Datanode. if downgraded happened, we will
restore trash data back, if finalized, then delete them. This is the way that
HDFS currently used in its upgrade, this also makes sense to Ozone upgrade.
> Ozone Non-Rolling upgrades.
> ---------------------------
>
> Key: HDDS-3698
> URL: https://issues.apache.org/jira/browse/HDDS-3698
> Project: Hadoop Distributed Data Store
> Issue Type: New Feature
> Reporter: Aravindan Vijayan
> Assignee: Aravindan Vijayan
> Priority: Major
> Attachments: Ozone Non-Rolling Upgrades.pdf
>
>
> Support for Non-rolling upgrades in Ozone.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org
