You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "Eugene Koifman (JIRA)" <ji...@apache.org> on 2016/08/02 02:22:20 UTC
[jira] [Comment Edited] (HIVE-14400) Handle concurrent insert with
dynamic partition
[ https://issues.apache.org/jira/browse/HIVE-14400?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15403237#comment-15403237 ]
Eugene Koifman edited comment on HIVE-14400 at 8/2/16 2:22 AM:
---------------------------------------------------------------
Maybe
{noformat}
EnvironmentContext environmentContext = null;
1602 if (hasFollowingStatsTask) {
1603 environmentContext = new EnvironmentContext();
1604 environmentContext.putToProperties(StatsSetupConst.DO_NOT_UPDATE_STATS, StatsSetupConst.TRUE);
1605 }
1606 alterPartition(tbl.getDbName(), tbl.getTableName(), new Partition(tbl, newTPart.getTPartition()), environmentContext);
{noformat}
can be made into a method to be called in both place to make it clear that both are doing exactly the same thing
In the comment, I'd modify the table example not to use transactional table as an example
Otherwise looks fine
was (Author: ekoifman):
Maybe
{noformat}
EnvironmentContext environmentContext = null;
1602 if (hasFollowingStatsTask) {
1603 environmentContext = new EnvironmentContext();
1604 environmentContext.putToProperties(StatsSetupConst.DO_NOT_UPDATE_STATS, StatsSetupConst.TRUE);
1605 }
1606 alterPartition(tbl.getDbName(), tbl.getTableName(), new Partition(tbl, newTPart.getTPartition()), environmentContext);
{noformat}
can be made into a method to be called in both place to make it clear that both are doing exactly the same thing
In the comment, I'd modify the table example not to use transactional table as an example
> Handle concurrent insert with dynamic partition
> -----------------------------------------------
>
> Key: HIVE-14400
> URL: https://issues.apache.org/jira/browse/HIVE-14400
> Project: Hive
> Issue Type: Bug
> Components: Hive
> Affects Versions: 2.2.0
> Reporter: Wei Zheng
> Assignee: Wei Zheng
> Attachments: HIVE-14400.1.patch
>
>
> With multiple users concurrently issuing insert statements on the same partition has a side effect that some queries may not see a partition at the time when they're issued, but will realize the partition is actually there when it is trying to add such partition to the metastore and thus get AlreadyExistsException, because some earlier query just created it (race condition).
> For example, imagine such a table is created:
> {code}
> create table T (name char(50)) partitioned by (ds string) clustered by (name) into 2 buckets stored as orc tblproperties('transactional'='true');
> {code}
> and the following two queries are launched at the same time, from different sessions:
> {code}
> insert into table T partition (ds) values ('Bob', 'today'); -- creates the partition 'today'
> insert into table T partition (ds) values ('Joe', 'today'); -- will fail with AlreadyExistsException
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)