[
https://issues.apache.org/jira/browse/SPARK-38230?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17735519#comment-17735519
]
jeanlyn commented on SPARK-38230:
---------------------------------
We found Hive metastore crash frequently after upgrade Spark from 2.4.7 to
3.3.2. After investigation, I found `InsertIntoHadoopFsRelationCommand` will
pull all partitions when using dynamicPartitionOverwrite, and i find this issue
after solves the problem by using generate paths to get partitions to get
partitions in our environment. So, I have submitted a new pull request, hoping
to help you.
> InsertIntoHadoopFsRelationCommand unnecessarily fetches details of partitions
> in most cases
> -------------------------------------------------------------------------------------------
>
> Key: SPARK-38230
> URL: https://issues.apache.org/jira/browse/SPARK-38230
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 3.0.2, 3.3.0, 3.4.0, 3.5.0
> Reporter: Coal Chan
> Priority: Major
>
> In
> `org.apache.spark.sql.execution.datasources.InsertIntoHadoopFsRelationCommand`,
> `sparkSession.sessionState.catalog.listPartitions` will call method
> `org.apache.hadoop.hive.metastore.listPartitionsPsWithAuth` of hive metastore
> client, this method will produce multiple queries per partition on hive
> metastore db. So when you insert into a table which has too many
> partitions(ie: 10k), it will produce too many queries on hive metastore
> db(ie: n * 10k = 10nk), it puts a lot of strain on the database.
> In fact, it calls method `listPartitions` in order to get locations of
> partitions and get `customPartitionLocations`. But in most cases, we do not
> have custom partitions, we can just get partition names, so we can call
> method listPartitionNames.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
