GitHub user tejasapatil opened a pull request:
https://github.com/apache/spark/pull/17184
[SPARK-19843] [SQL] UTF8String => (int / long) conversion expensive for
invalid inputs
## What changes were proposed in this pull request?
Jira : https://issues.apache.org/jira/browse/SPARK-19843
Added cheap checks to `UTF8String` to check if it can be converted to int /
long. The logic is motivated from [LazyUtils.isNumberMaybe in
Hive](https://github.com/apache/hive/blob/ff67cdda1c538dc65087878eeba3e165cf3230f4/serde/src/java/org/apache/hadoop/hive/serde2/lazy/LazyUtils.java#L90).
## How was this patch tested?
- Added new unit tests
- Ran a prod job which had conversion from string -> int and verified the
outputs
## Performance
Attached `UTF8StringBenchmark` with the PR. Here is the summary:
```
conversion to int: Best/Avg Time(ms) Rate(M/s)
Per Row(ns) Relative
------------------------------------------------------------------------------------------------
without check over valid integers 500 / 503 33.6
29.8 1.0X
with check over valid integers 1069 / 1072 15.7
63.7 0.5X
without check over in-valid integers 34515 / 36096 0.5
2057.3 0.0X
with check over in-valid integers 718 / 895 23.4
42.8 0.7X
```
Note that this PR is a tradeoff. If the data does not have any invalid
inputs, then performance of conversion is hurt by 2x. Based on the numbers in
the benchmark, if more than 2.5% of input cannot be converted to `int`, then it
will add perf. gains. For my workloads, I could see on average 37% savings
(best case 2.08x performance gain).
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/tejasapatil/spark SPARK-19843_is_numeric_maybe
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/17184.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #17184
----
commit adcb405bc889345c662ddda1742d3ea95191570d
Author: Tejas Patil <[email protected]>
Date: 2017-03-07T00:56:32Z
[SPARK-19843] [SQL] UTF8String => (int / long) conversion expensive for
invalid inputs
----
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
