spark-instrumented-optimizer/common/unsafe
Kent Yao 2dd6807e42 [SPARK-28023][SQL] Add trim logic in UTF8String's toInt/toLong to make it consistent with other string-numeric casting
### What changes were proposed in this pull request?

Modify `UTF8String.toInt/toLong` to support trim spaces for both sides before converting it to byte/short/int/long.

With this kind of "cheap" trim can help improve performance for casting string to integrals. The idea is from https://github.com/apache/spark/pull/24872#issuecomment-556917834

### Why are the changes needed?

make the behavior consistent.

### Does this PR introduce any user-facing change?
yes, cast string to an integral type, and binary comparison between string and integrals will trim spaces first. their behavior will be consistent with float and double.
### How was this patch tested?
1. add ut.
2. benchmark tests
 the benchmark is modified based on https://github.com/apache/spark/pull/24872#issuecomment-503827016

```scala
/*
 * Licensed to the Apache Software Foundation (ASF) under one or more
 * contributor license agreements.  See the NOTICE file distributed with
 * this work for additional information regarding copyright ownership.
 * The ASF licenses this file to You under the Apache License, Version 2.0
 * (the "License"); you may not use this file except in compliance with
 * the License.  You may obtain a copy of the License at
 *
 *    http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

package org.apache.spark.sql.execution.benchmark

import org.apache.spark.benchmark.Benchmark

/**
 * Benchmark trim the string when casting string type to Boolean/Numeric types.
 * To run this benchmark:
 * {{{
 *   1. without sbt:
 *      bin/spark-submit --class <this class> --jars <spark core test jar> <spark sql test jar>
 *   2. build/sbt "sql/test:runMain <this class>"
 *   3. generate result: SPARK_GENERATE_BENCHMARK_FILES=1 build/sbt "sql/test:runMain <this class>"
 *      Results will be written to "benchmarks/CastBenchmark-results.txt".
 * }}}
 */
object CastBenchmark extends SqlBasedBenchmark {
This conversation was marked as resolved by yaooqinn

  override def runBenchmarkSuite(mainArgs: Array[String]): Unit = {
    val title = "Cast String to Integral"
    runBenchmark(title) {
      withTempPath { dir =>
        val N = 500L << 14
        val df = spark.range(N)
        val types = Seq("int", "long")
        (1 to 5).by(2).foreach { i =>
          df.selectExpr(s"concat(id, '${" " * i}') as str")
            .write.mode("overwrite").parquet(dir + i.toString)
        }

        val benchmark = new Benchmark(title, N, minNumIters = 5, output = output)
        Seq(true, false).foreach { trim =>
          types.foreach { t =>
            val str = if (trim) "trim(str)" else "str"
            val expr = s"cast($str as $t) as c_$t"
            (1 to 5).by(2).foreach { i =>
              benchmark.addCase(expr + s" - with $i spaces") { _ =>
                spark.read.parquet(dir + i.toString).selectExpr(expr).collect()
              }
            }
          }
        }
        benchmark.run()
      }
    }
  }
}
```
#### benchmark result.
normal trim v.s. trim in toInt/toLong
```java
================================================================================================
Cast String to Integral
================================================================================================

Java HotSpot(TM) 64-Bit Server VM 1.8.0_231-b11 on Mac OS X 10.15.1
Intel(R) Core(TM) i5-5287U CPU  2.90GHz
Cast String to Integral:                  Best Time(ms)   Avg Time(ms)   Stdev(ms)    Rate(M/s)   Per Row(ns)   Relative
------------------------------------------------------------------------------------------------------------------------
cast(trim(str) as int) as c_int - with 1 spaces          10220          12994        1337          0.8        1247.5       1.0X
cast(trim(str) as int) as c_int - with 3 spaces           4763           8356         357          1.7         581.4       2.1X
cast(trim(str) as int) as c_int - with 5 spaces           4791           8042         NaN          1.7         584.9       2.1X
cast(trim(str) as long) as c_long - with 1 spaces           4014           6755         NaN          2.0         490.0       2.5X
cast(trim(str) as long) as c_long - with 3 spaces           4737           6938         NaN          1.7         578.2       2.2X
cast(trim(str) as long) as c_long - with 5 spaces           4478           6919        1404          1.8         546.6       2.3X
cast(str as int) as c_int - with 1 spaces           4443           6222         NaN          1.8         542.3       2.3X
cast(str as int) as c_int - with 3 spaces           3659           3842         170          2.2         446.7       2.8X
cast(str as int) as c_int - with 5 spaces           4372           7996         NaN          1.9         533.7       2.3X
cast(str as long) as c_long - with 1 spaces           3866           5838         NaN          2.1         471.9       2.6X
cast(str as long) as c_long - with 3 spaces           3793           5449         NaN          2.2         463.0       2.7X
cast(str as long) as c_long - with 5 spaces           4947           5961        1198          1.7         603.9       2.1X
```

Closes #26622 from yaooqinn/cheapstringtrim.

Authored-by: Kent Yao <yaooqinn@hotmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2019-11-22 19:32:27 +08:00
..
src [SPARK-28023][SQL] Add trim logic in UTF8String's toInt/toLong to make it consistent with other string-numeric casting 2019-11-22 19:32:27 +08:00
pom.xml Revert "Prepare Spark release v3.0.0-preview-rc2" 2019-10-30 17:45:44 -07:00