[
https://issues.apache.org/jira/browse/SPARK-22010?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16166136#comment-16166136
]
Maciej Bryński edited comment on SPARK-22010 at 9/14/17 12:05 PM:
------------------------------------------------------------------
The reason of this Jira is this profiling (attachment).
!profile_fact_dok.jpg|thumbnail!
As you can see about 80% of pyspark time is spent in Spark internals.
was (Author: maver1ck):
The reason of this Jira is this profiling (attachment).
As you can see about 80% of pyspark time is spent in Spark internals.
> Slow fromInternal conversion for TimestampType
> ----------------------------------------------
>
> Key: SPARK-22010
> URL: https://issues.apache.org/jira/browse/SPARK-22010
> Project: Spark
> Issue Type: Bug
> Components: PySpark
> Affects Versions: 2.2.0
> Reporter: Maciej Bryński
> Attachments: profile_fact_dok.png
>
>
> To convert timestamp type to python we are using
> `datetime.datetime.fromtimestamp(ts // 1000000).replace(microsecond=ts %
> 1000000)`
> code.
> {code}
> In [34]: %%timeit
> ...:
> datetime.datetime.fromtimestamp(1505383647).replace(microsecond=12344)
> ...:
> 4.2 µs ± 558 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)
> {code}
> It's slow, because:
> # we're trying to get TZ on every conversion
> # we're using replace method
> Proposed solution: custom datetime conversion and move calculation of TZ to
> module
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
