[
https://issues.apache.org/jira/browse/SPARK-35494?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Raphael Luta updated SPARK-35494:
---------------------------------
Description:
In Spark SQL 2.4, when converting a datetime string column to timestamp with
cast or to_timestamp, we have noticed a major performance issue when the source
string contains timezone information (for example 2021-05-24T00:00:00+02:00)
This simple benchmark illustrates the difference:
{noformat}
OpenJDK 64-Bit Server VM 1.8.0_292-b10 on Mac OS X 10.16
Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz
Timestamp Conversion: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
----------------------------------------------------------------------
no tz 1368 / 1379 3,1 326,1 1,0X
with UTC tz 5940 / 5947 0,7 1416,2 0,2X
with hours tz 5940 / 5962 0,7 1416,2 0,2X
{noformat}
With the provided patch, the benchmark results are:
{noformat}
OpenJDK 64-Bit Server VM 1.8.0_292-b10 on Mac OS X 10.16
Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz
Timestamp Conversion: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------
no tz 1383 / 1392 3,0 329,7 1,0X
with UTC tz 1550 / 1587 2,7 369,6 0,9X
with hours tz 1570 / 1574 2,7 374,2 0,9X
{noformat}
This does not occur on the 3.x branches
was:
In Spark SQL 2.4, when converting a datetime string column to timestamp with
cast or to_timestamp, we have noticed a major performance issue when the source
string contains timezone information (for example 2021-05-24T00:00:00+02:00)
This simple benchmark illustrates the difference:
{noformat}
OpenJDK 64-Bit Server VM 1.8.0_292-b10 on Mac OS X 10.16
Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz
Timestamp Conversion: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
----------------------------------------------------------------------
no tz 1368 / 1379 3,1 326,1 1,0X
with UTC tz 5940 / 5947 0,7 1416,2 0,2X
with hours tz 5940 / 5962 0,7 1416,2 0,2X
{noformat}
With the provided patch, the benchmark results are:
{noformat}
OpenJDK 64-Bit Server VM 1.8.0_292-b10 on Mac OS X 10.16
Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz
Timestamp Conversion: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------
no tz 1383 / 1392 3,0 329,7 1,0X
with UTC tz 1550 / 1587 2,7 369,6 0,9X
with hours tz 1570 / 1574 2,7 374,2 0,9X
{noformat}
This does not occur on the 3.x branches
> Timestamp casting performance issue when invoked with timezone
> --------------------------------------------------------------
>
> Key: SPARK-35494
> URL: https://issues.apache.org/jira/browse/SPARK-35494
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.4.7, 2.4.8
> Reporter: Raphael Luta
> Priority: Major
>
> In Spark SQL 2.4, when converting a datetime string column to timestamp with
> cast or to_timestamp, we have noticed a major performance issue when the
> source string contains timezone information (for example
> 2021-05-24T00:00:00+02:00)
> This simple benchmark illustrates the difference:
> {noformat}
> OpenJDK 64-Bit Server VM 1.8.0_292-b10 on Mac OS X 10.16
> Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz
> Timestamp Conversion: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
> ----------------------------------------------------------------------
> no tz 1368 / 1379 3,1 326,1 1,0X
> with UTC tz 5940 / 5947 0,7 1416,2 0,2X
> with hours tz 5940 / 5962 0,7 1416,2 0,2X
> {noformat}
> With the provided patch, the benchmark results are:
> {noformat}
> OpenJDK 64-Bit Server VM 1.8.0_292-b10 on Mac OS X 10.16
> Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz
> Timestamp Conversion: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
> ------------------------------------------------------------------------------------------------
> no tz 1383 / 1392 3,0 329,7 1,0X
> with UTC tz 1550 / 1587 2,7 369,6 0,9X
> with hours tz 1570 / 1574 2,7 374,2 0,9X
> {noformat}
> This does not occur on the 3.x branches
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
