[
https://issues.apache.org/jira/browse/OMID-142?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16811545#comment-16811545
]
Yonatan Gottesman commented on OMID-142:
----------------------------------------
[~lhofhansl] I have mistaken you.
The TSO holds a cache for the *writeset keys* (rows in phoenix) so it stores
all the rows that where touched during a transaction, and the latest commit
timetstamp for each row.
Once you touch a row that is in the cache the commit timestamp is updated. if
you touch a row that isnt in the cache the row is added and if there is no free
space, the oldest row will be removed and the lowwatermark will be updated with
the committimestamp of the removed row.
The number of rows in the cache is configurable and not hard coded to 1000. In
the configuration file i put in phoenix its:
"conflictMapSize: 100000000"
So only after you update 100M different rows the lowwatermark will bump
> Omid GC fails in Phoenix does not remove all data
> -------------------------------------------------
>
> Key: OMID-142
> URL: https://issues.apache.org/jira/browse/OMID-142
> Project: Apache Omid
> Issue Type: Bug
> Affects Versions: 1.0.1
> Reporter: Lars Hofhansl
> Assignee: Yonatan Gottesman
> Priority: Major
> Attachments: 142-test.txt, omid142.patch
>
>
> When you insert a bunch of data into a table, followed by a DELETE FROM
> <table>, then flush and compact the table in HBase, the data is not removed.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
