[
https://issues.apache.org/jira/browse/HIVE-20434?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Shelukhin updated HIVE-20434:
------------------------------------
Description:
{noformat}
set hive.stats.dbclass=fs;
set hive.stats.fetch.column.stats=true;
set datanucleus.cache.collections=false;
set hive.merge.mapfiles=false;
set hive.merge.mapredfiles=false;
set hive.mapred.mode=nonstrict;
set hive.stats.autogather=true;
set hive.stats.column.autogather=true;
set hive.compute.query.using.stats=true;
set hive.explain.user=false;
set hive.fetch.task.conversion=none;
set hive.query.results.cache.enabled=false;
create table stats_part(key int,value string) partitioned by (p int)
tblproperties ("transactional"="false");
insert into table stats_part partition(p=101) values (1, "foo");
explain select count(key) from stats_part; -- <== stats are accurate
ALTER TABLE stats_part CHANGE COLUMN key key2 int;
explain select count(key2) from stats_part; -- <== stats are now inaccurate
analyze table stats_part partition(p) compute statistics for columns;
explain select count(key2) from stats_part; -- <== stats are now accurate again
alter table stats_part add partition(p=105);
explain select count(key2) from stats_part; -- <== stats are now inaccurate
analyze table stats_part partition(p) compute statistics for columns;
explain select count(key2) from stats_part; -- <== stats are still inaccurate!
drop table stats_part;
{noformat}
Seems like it's impossible to have correct stats on a table with an empty
partition.
was:
{noformat}
set hive.stats.dbclass=fs;
set hive.stats.fetch.column.stats=true;
set datanucleus.cache.collections=false;
set hive.merge.mapfiles=false;
set hive.merge.mapredfiles=false;
set hive.mapred.mode=nonstrict;
set hive.stats.autogather=true;
set hive.stats.column.autogather=true;
set hive.compute.query.using.stats=true;
set hive.explain.user=false;
set hive.fetch.task.conversion=none;
set hive.query.results.cache.enabled=false;
create table stats_part(key int,value string) partitioned by (p int)
tblproperties ("transactional"="false");
insert into table stats_part partition(p=101) values (1, "foo");
explain select count(key) from stats_part; -- <== stats are correct
ALTER TABLE stats_part CHANGE COLUMN key key2 int;
explain select count(key2) from stats_part; -- <== stats are now incorrect
analyze table stats_part partition(p) compute statistics for columns;
explain select count(key2) from stats_part; -- <== stats are now correct again
alter table stats_part add partition(p=105);
explain select count(key2) from stats_part; -- <== stats are now incorrect
analyze table stats_part partition(p) compute statistics for columns;
explain select count(key2) from stats_part; -- <== stats are still incorrect!
drop table stats_part;
{noformat}
Seems like it's impossible to have correct stats on a table with an empty
partition.
> analyze on an empty partition doesn't mark stats as accurate/produce stats
> --------------------------------------------------------------------------
>
> Key: HIVE-20434
> URL: https://issues.apache.org/jira/browse/HIVE-20434
> Project: Hive
> Issue Type: Bug
> Reporter: Sergey Shelukhin
> Assignee: Ashutosh Chauhan
> Priority: Major
>
> {noformat}
> set hive.stats.dbclass=fs;
> set hive.stats.fetch.column.stats=true;
> set datanucleus.cache.collections=false;
> set hive.merge.mapfiles=false;
> set hive.merge.mapredfiles=false;
> set hive.mapred.mode=nonstrict;
> set hive.stats.autogather=true;
> set hive.stats.column.autogather=true;
> set hive.compute.query.using.stats=true;
> set hive.explain.user=false;
> set hive.fetch.task.conversion=none;
> set hive.query.results.cache.enabled=false;
> create table stats_part(key int,value string) partitioned by (p int)
> tblproperties ("transactional"="false");
> insert into table stats_part partition(p=101) values (1, "foo");
> explain select count(key) from stats_part; -- <== stats are accurate
> ALTER TABLE stats_part CHANGE COLUMN key key2 int;
> explain select count(key2) from stats_part; -- <== stats are now inaccurate
> analyze table stats_part partition(p) compute statistics for columns;
> explain select count(key2) from stats_part; -- <== stats are now accurate
> again
> alter table stats_part add partition(p=105);
> explain select count(key2) from stats_part; -- <== stats are now inaccurate
> analyze table stats_part partition(p) compute statistics for columns;
> explain select count(key2) from stats_part; -- <== stats are still inaccurate!
> drop table stats_part;
> {noformat}
> Seems like it's impossible to have correct stats on a table with an empty
> partition.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
