[
https://issues.apache.org/jira/browse/PHOENIX-6798?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
chenglei updated PHOENIX-6798:
------------------------------
Description:
For HBase, reversed scan is slower than normal forward scan.
Consider following table:
{code}
create table test_table
(group_id integer not null,
keyword varchar not null,
cost integer,
CONSTRAINT TEST_PK PRIMARY KEY (group_id,keyword))
{code}
For the following sql :
{code}
select keyword,sum(cost) from test group by keyword order by keyword desc
{code}
The compiled {{AggregatePlan}} would set the scan for HBase as reversed scan,
but that is very strange, because for this sql, {{GroupBy.isOrderPreserving}}
is false, we have no need to set the scan as reversed scan because we have to
hash-aggregate the scanned results from HBase in {{Coprocessor}} before sending
them to client.
was:
For HBase, reversed scan is slower than normal forward scan.
Consider following table:
{code}
create table test_table
(group_id integer not null,
keyword varchar not null,
cost integer,
CONSTRAINT TEST_PK PRIMARY KEY (group_id,keyword))
{code}
For the following sql :
{code}
select keyword,sum(cost) from test group by keyword order by keyword desc
{code}
The compiled {{AggregatePlan}} would set the scan for HBase as reversed scan,
but that is very strange, because for this sql, {{GroupBy.isOrderPreserving}}
is false, we have no need to set the scan as reversed scan because we have to
hash-aggregate the scanned results from HBase.
> Eliminate unnecessary reversed scan
> -----------------------------------
>
> Key: PHOENIX-6798
> URL: https://issues.apache.org/jira/browse/PHOENIX-6798
> Project: Phoenix
> Issue Type: Bug
> Components: core
> Affects Versions: 5.1.2
> Reporter: chenglei
> Assignee: chenglei
> Priority: Major
>
> For HBase, reversed scan is slower than normal forward scan.
> Consider following table:
> {code}
> create table test_table
> (group_id integer not null,
> keyword varchar not null,
> cost integer,
> CONSTRAINT TEST_PK PRIMARY KEY (group_id,keyword))
> {code}
> For the following sql :
> {code}
> select keyword,sum(cost) from test group by keyword order by keyword desc
> {code}
> The compiled {{AggregatePlan}} would set the scan for HBase as reversed scan,
> but that is very strange, because for this sql, {{GroupBy.isOrderPreserving}}
> is false, we have no need to set the scan as reversed scan because we have to
> hash-aggregate the scanned results from HBase in {{Coprocessor}} before
> sending them to client.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
