[
https://issues.apache.org/jira/browse/PHOENIX-1550?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14362752#comment-14362752
]
Sun Fulin edited comment on PHOENIX-1550 at 3/16/15 6:15 AM:
-------------------------------------------------------------
[~jamestaylor] - The priority lies that snapshot scan can improve aggregation
work performance greatly as provided through HBASE-8369. We had integrated
spark with hbase and had tested a lot for hbase scan VS snapshot, which had got
results for that snapshot upgrade performance a lot in use cases of some
hourly, daily, weekly and monthly aggregation work. We would prefer more like
[~maghamraviki...@gmail.com] designed for mapreduce support, e.g. issuing more
like {{PhoenixSnapshotInputFormat}}. Thus we believe can achieve our goal of
resolving some kind of phoenix full table scan performance issues.
Sorry for some misunderstanding about my comments.
was (Author: sunfl):
[~jamestaylor] - The priority lies that snapshot scan can improve aggregation
work performance greatly as provided through HBASE-8369. We had integrated
spark with hbase and had tested a lot for hbase scan VS snapshot, which had got
results for that snapshot upgrade performance a lot in use cases of some
hourly, daily, weekly and monthly aggregation work. We would prefer more like
[~maghamraviki...@gmail.com] designed for mapreduce support, e.g. issuing more
like PhoenixSnapshotInputFormat. Thus we believe can achieve our goal of
resolving some kind of phoenix full table scan performance issues.
Sorry for some misunderstanding about my comments.
> Freeze and thaw Phoenix tables
> ------------------------------
>
> Key: PHOENIX-1550
> URL: https://issues.apache.org/jira/browse/PHOENIX-1550
> Project: Phoenix
> Issue Type: New Feature
> Reporter: Andrew Purtell
>
> HBase snapshots provide a way to freeze the state of a table and then later
> restore that state from the snapshot or clone it into a new table. This can
> be very convenient for a wide range of use cases. However, we can't apply
> this technique directly to Phoenix tables because Phoenix generates and uses
> more state in more locations than a basic HBase table - a Phoenix table may
> have one or more secondary indexes, global or local, schema in the system
> catalog, sequence state, statistics.
> This issue is about considering a freeze and thaw feature for Phoenix tables,
> perhaps based on HBase snapshots. A freeze operation would create a
> consistent point in time snapshot of a Phoenix table and all related index
> data and metadata. A thaw operation would restore frozen state either by
> overwriting the table or by cloning the state into a new table.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
