[
https://issues.apache.org/jira/browse/HIVE-23052?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17071138#comment-17071138
]
Marton Bod commented on HIVE-23052:
-----------------------------------
Benchmark test results:
|*Mode*|*dbType*|*partCount*|*tblCount*|*Cnt*|*Score*|*Error*|*Units*|
|non-patched|MYSQL|30|4|20|257.759|2.283|ms/op|
|patched|MYSQL|30|4|20|247.602|3.178|ms/op|
|non-patched|MYSQL|300|4|20|582.5|18.627|ms/op|
|patched|MYSQL|300|4|20|516.164|17.316|ms/op|
|non-patched|MYSQL|600|4|20|947.159|28.219|ms/op|
|patched|MYSQL|600|4|20|770.398|43.109|ms/op|
|non-patched|MYSQL|1200|4|20|1645.015|41.583|ms/op|
|patched|MYSQL|1200|4|20|1370.759|104.764|ms/op|
|non-patched|ORACLE|30|4|20|491.062|18.127|ms/op|
|patched|ORACLE|30|4|20|300.348|5.314|ms/op|
|non-patched|ORACLE|300|4|20|3433.233|105.651|ms/op|
|patched|ORACLE|300|4|20|804.947|383.411|ms/op|
|non-patched|ORACLE|600|4|20|4793.748|250.258|ms/op|
|patched|ORACLE|600|4|20|1293.439|581.037|ms/op|
|non-patched|ORACLE|1200|4|20|8950.897|460.234|ms/op|
|patched|ORACLE|1200|4|20|2358.546|947.063|ms/op|
|non-patched|POSTGRES|30|4|20|638.023|34.243|ms/op|
|patched|POSTGRES|30|4|20|355.213|11.115|ms/op|
|non-patched|POSTGRES|300|4|20|4017.29|290.869|ms/op|
|patched|POSTGRES|300|4|20|1130.358|76.175|ms/op|
|non-patched|POSTGRES|600|4|20|7447.98|545.179|ms/op|
|patched|POSTGRES|600|4|20|2267.942|436.401|ms/op|
|non-patched|POSTGRES|1200|4|20|18506.656|1645.762|ms/op|
|patched|POSTGRES|1200|4|20|4543.454|897.753|ms/op|
> Optimize lock enqueueing in TxnHandler
> --------------------------------------
>
> Key: HIVE-23052
> URL: https://issues.apache.org/jira/browse/HIVE-23052
> Project: Hive
> Issue Type: Improvement
> Reporter: Marton Bod
> Assignee: Marton Bod
> Priority: Major
> Attachments: HIVE-23052.1.patch, HIVE-23052.2.patch,
> HIVE-23052.3.patch, HIVE-23052.4.patch, HIVE-23052.5.patch,
> HIVE-23052.6.patch, HIVE-23052.7.patch, HIVE-23052.8.patch,
> HIVE-23052.8.patch, HIVE-23052.8.patch, HIVE-23052.9.patch
>
>
> * Reduce scope of next_lock_id select-for-update by moving the txn_component
> inserts before the S4U + inserting the hive_locks entries before the S4U
> (first with a temp ID, which will be replaced later in a single update). This
> helps decrease the overall time that the next_lock_id table is locked,
> thereby increasing concurrency
> * Insert txn_components in a batch instead of one-by-one (also in
> TxnHandler::addDynamicPartition)
> * Increment next_lock_id and update hive_locks table in a single batch
> statement
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
