[
https://issues.apache.org/jira/browse/HIVE-23721?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17149997#comment-17149997
]
zhangbutao commented on HIVE-23721:
-----------------------------------
We set hive.in.test=true; MetaStoreDirectSql.ensureDbInit will check metadate
before each sql request.
[https://github.com/apache/hive/blob/4942a7c0b4be3a5b0c889a89b903e9a70c57d494/standalone-metastore/metastore-server/src/main/java/org/apache/hadoop/hive/metastore/MetaStoreDirectSql.java#L212]
However, the two queries do not use the index correctly:
[https://github.com/apache/hive/blob/4942a7c0b4be3a5b0c889a89b903e9a70c57d494/standalone-metastore/metastore-server/src/main/java/org/apache/hadoop/hive/metastore/MetaStoreDirectSql.java#L280]
[https://github.com/apache/hive/blob/4942a7c0b4be3a5b0c889a89b903e9a70c57d494/standalone-metastore/metastore-server/src/main/java/org/apache/hadoop/hive/metastore/MetaStoreDirectSql.java#L281]
This is because table TAB_COL_STATS and PART_COL_STATS have combined index:
[https://github.com/apache/hive/blob/4942a7c0b4be3a5b0c889a89b903e9a70c57d494/standalone-metastore/metastore-server/src/main/sql/mysql/hive-schema-4.0.0.mysql.sql#L742]
[https://github.com/apache/hive/blob/4942a7c0b4be3a5b0c889a89b903e9a70c57d494/standalone-metastore/metastore-server/src/main/sql/mysql/hive-schema-4.0.0.mysql.sql#L774]
According to the leftmost matching principle of the combined index, we should
use "catName == ''" instead of "dbName == ''"，because “catName” is the first
index column.
> MetaStoreDirectSql.ensureDbInit() need to optimize QuerySQL
> -----------------------------------------------------------
>
> Key: HIVE-23721
> URL: https://issues.apache.org/jira/browse/HIVE-23721
> Project: Hive
> Issue Type: Bug
> Affects Versions: 3.1.2
> Environment: Hadoop 3.1（1700+ nodes）
> YARN 3.1 （with timelineserver enabled，https enabled)
> Hive 3.1 (15 HS2 instance)
> 60000+ YARN Applications every day
> Reporter: YulongZ
> Assignee: zhangbutao
> Priority: Critical
> Fix For: 4.0.0
>
> Attachments: HIVE-23721.01.patch
>
>
> From Hive3.0，catalog added to hivemeta，many schema of metastore added column
> “catName”，and index for table added column “catName”。
> In MetaStoreDirectSql.ensureDbInit() ，two queries below
> “
> initQueries.add(pm.newQuery(MTableColumnStatistics.class, "dbName ==
> ''"));
> initQueries.add(pm.newQuery(MPartitionColumnStatistics.class, "dbName
> == ''"));
> ”
> should use "catName == ''" instead of "dbName == ''"，because “catName” is the
> first index column。
> When data of metastore become large，for example， table of
> MPartitionColumnStatistics have millions of lines。The
> “newQuery(MPartitionColumnStatistics.class, "dbName == ''")” for metastore
> executed very slowly，and the query “show tables“ for hiveserver2 executed
> very slowly too。
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
