[
https://issues.apache.org/jira/browse/PHOENIX-2591?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
James Taylor updated PHOENIX-2591:
----------------------------------
Attachment: PHOENIX-2591.patch
Please review, [~tdsilva]. We can use the write pointer and not have to do the
funky commit stuff as long as we use tx.getTransactionId() which doesn't change
(it's the initial write pointer).
> Minimize transaction commit/rollback for DDL
> --------------------------------------------
>
> Key: PHOENIX-2591
> URL: https://issues.apache.org/jira/browse/PHOENIX-2591
> Project: Phoenix
> Issue Type: Bug
> Reporter: James Taylor
> Assignee: Thomas D'Silva
> Fix For: 4.7.0
>
> Attachments: PHOENIX-2591.patch
>
>
> Seems that the number of times we commit/rollback transactions during DDL
> operations could be improved. See TransactionUtil.getTableTimestamp() for
> example. There'd also be another couple when MutationState.commitWriteFence()
> is called when a CREATE INDEX is performed too.
> I realize we're doing this to get the transaction read pointer to "catch up"
> to the current time, as we use the read pointer as our "current time" for
> transactional tables. However, what would the impact be if we used the
> transaction write pointer instead?
> At a minimum, we need to document what we're doing before we forget.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
