[
https://issues.apache.org/jira/browse/HIVE-16383?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15957497#comment-15957497
]
Thejas M Nair commented on HIVE-16383:
--------------------------------------
Looks like we end up having 2 separate connection pools, one in ObjectStore and
one in TxnHandler.
Regarding the above formula, the core_count and effective_spindle_count should
be the numbers for host where the RDBMS is deployed, is that right ?
Also, the recommendation from HikariCp doesn't seem to take into consideration
number of application instances. We would need to account for number of
metastore hosts and HS2 with embedded metastore instances to arrive at a proper
value. For now, I think lets leave this to the user. The default of 10 seems
like an OK value.
[~ekoifman] [~alangates]
Any thoughts on using the configured connection pool size for DN (used by
ObjectStore) for the pool in TxnHandler as well, instead of hard coded value of
10? (Thats what this patch does). Do we expect the same level of concurrent
requests as what we would see via ObjectStore ?
> Switch to HikariCP as default connection pooling
> ------------------------------------------------
>
> Key: HIVE-16383
> URL: https://issues.apache.org/jira/browse/HIVE-16383
> Project: Hive
> Issue Type: Improvement
> Components: Metastore
> Affects Versions: 3.0.0
> Reporter: Prasanth Jayachandran
> Assignee: Prasanth Jayachandran
> Attachments: HIVE-16383.1.patch
>
>
> Since 3.0 is planned to move to JDK8, we can now switch to HikariCP as
> default connection pooling for DN because of its improved performance over
> others.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
