[
https://issues.apache.org/jira/browse/PHOENIX-2307?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Samarth Jain updated PHOENIX-2307:
----------------------------------
Description: On region splits, there possibly could be cases where Phoenix
would be able to set the min/max timestamps itself. One such scenario being
when the part of row key prior to the timestamp hasn't changed in the region
that is to be split. Or when the leading part of the row key is timestamp, with
or without salting. In such cases, it would be relatively easy to figure out
what the min and max timestamps should be on the newly created child regions.
To build this we would need HBase apis to set the min/max timestamps on the
HFiles which could be then called in coprocessor hooks for splits. (was: As
part of the work done in PHOENIX-914, we noticed that because of region splits,
the time range information cannot be used to filter out the child regions. When
splitting a region HBase (understandably) doesn't read every row of the region
being split and so it doesn't know the min and max timestamps to add as
metadata on the newly created store files.
There possibly could be cases where Phoenix would be able to set the min/ma x
timestamps itself. One such scenario being when the part of row key prior to
the timestamp hasn't changed in the region that is to be split. We would need
HBase APIs to set the min/max timestamp which could be then called in the
coprocessor hook for splits.)
> Build means to set min/max timestamp on store files on region splits.
> ---------------------------------------------------------------------
>
> Key: PHOENIX-2307
> URL: https://issues.apache.org/jira/browse/PHOENIX-2307
> Project: Phoenix
> Issue Type: Improvement
> Reporter: Samarth Jain
>
> On region splits, there possibly could be cases where Phoenix would be able
> to set the min/max timestamps itself. One such scenario being when the part
> of row key prior to the timestamp hasn't changed in the region that is to be
> split. Or when the leading part of the row key is timestamp, with or without
> salting. In such cases, it would be relatively easy to figure out what the
> min and max timestamps should be on the newly created child regions. To build
> this we would need HBase apis to set the min/max timestamps on the HFiles
> which could be then called in coprocessor hooks for splits.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
