[
https://issues.apache.org/jira/browse/OMID-146?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16824589#comment-16824589
]
Lars Hofhansl edited comment on OMID-146 at 4/23/19 11:18 PM:
--------------------------------------------------------------
Looked at the PR. Can't claim I fully understand it, but looks right.
I did notice some printlns forgotten in there.
Also, will the commit cache be a problem now? As we run the CompactScanner,
once we cache a commit ts as absent, we won't check it again, right?
was (Author: lhofhansl):
Looked at the PR. Can't claim I fully understand it, but looks right.
I did notice some printlns in there.
> Persist transaction commit before updating low watermark
> --------------------------------------------------------
>
> Key: OMID-146
> URL: https://issues.apache.org/jira/browse/OMID-146
> Project: Apache Omid
> Issue Type: Bug
> Reporter: Ohad Shacham
> Assignee: Yonatan Gottesman
> Priority: Minor
> Fix For: 1.0.1
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> When a transaction t updates the conflict table at the TSO, it mights remove
> entries of transactions that committed after t started, or even entries of t.
> Therefore, in order to make sure that the garbage collector does not removes
> entries of t from the data tables, thinking accidentally that t was not
> committed yet. We should first persist t's commit in the commit table and
> only then update the low watermark. This also applies to the low latency
> which persists the commit by the clients.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
