-
Notifications
You must be signed in to change notification settings - Fork 230
/
TimeZonePerfSuite.scala
308 lines (259 loc) · 9.58 KB
/
TimeZonePerfSuite.scala
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
/*
* Copyright (c) 2023-2024, NVIDIA CORPORATION.
*
* Licensed 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 com.nvidia.spark.rapids.timezone
import java.io.File
import java.time.Instant
import com.nvidia.spark.rapids.SparkQueryCompareTestSuite
import com.nvidia.spark.rapids.jni.GpuTimeZoneDB
import org.scalatest.BeforeAndAfterAll
import org.apache.spark.SparkConf
import org.apache.spark.sql._
import org.apache.spark.sql.functions._
import org.apache.spark.sql.timezone._
import org.apache.spark.sql.timezone.TimeZonePerfUtils
import org.apache.spark.sql.types._
/**
* A simple test performance framework for non-UTC features.
* Usage:
*
* argLine="-DTZs=Asia/Shanghai,Japan -DenableTimeZonePerf=true" \
* mvn test -Dbuildver=311 -DwildcardSuites=com.nvidia.spark.rapids.timezone.TimeZonePerfSuite
* Note:
* Generate a Parquet file with 6 columns:
* - c_ts: timestamp column
* - c_long_of_ts: long value which is microseconds
* - c_date: date column
* - c_int_of_date:int value which is days from 1970-01-01
* - c_long_of_ts_seconds: long values of seconds from epoch
* - c_str_for_cast: strings for cast to timestamp, formats are yyyy, yyyy-mm, ...
* - c_str_of_ts: strings with format: yyyy-MM-dd HH:mm:ss
* Each column is high duplicated.
* The generated file is highly compressed since we expect both CPU and GPU can scan quickly.
* When testing operators, we need to add in a max/count aggregator to reduce the result data.
*/
class TimeZonePerfSuite extends SparkQueryCompareTestSuite with BeforeAndAfterAll {
// perf test is disabled by default since it's a long running time in UT.
private val enablePerfTest = java.lang.Boolean.getBoolean("enableTimeZonePerf")
private val timeZoneStrings = System.getProperty("TZs", "Asia/Shanghai")
// rows for perf test
private val numRows: Long = 1024L * 1024L * 10L
private val zones = timeZoneStrings.split(",")
private val path = "/tmp/tmp_TimeZonePerfSuite"
/**
* Create a Parquet file to test
*/
override def beforeAll(): Unit = {
withCpuSparkSession(
spark => createDF(spark).write.mode("overwrite").parquet(path))
}
override def afterAll(): Unit = {
FileUtils.deleteRecursively(new File(path))
}
val year1980 = Instant.parse("1980-01-01T00:00:00Z").getEpochSecond * 1000L * 1000L
val year2000 = Instant.parse("2000-01-01T00:00:00Z").getEpochSecond * 1000L * 1000L
val year2030 = Instant.parse("2030-01-01T00:00:00Z").getEpochSecond * 1000L * 1000L
private val stringsForCast = Array(
"2000",
"2000-01",
"2000-01-02",
"2000-01-02 03:04:05",
"2000-01-02 03:04:05Iran")
private val regularStrings = Array(
"1970-01-01 00:00:00",
"2000-01-02 00:00:00",
"2030-01-02 00:00:00")
/**
* create a data frame with schema:
* - c_ts: timestamp column
* - c_long_of_ts: long value which is microseconds
* - c_date: date column
* - c_int_of_date:int value which is days from 1970-01-01
* - c_str_for_cast: strings for cast to timestamp, formats are yyyy, yyyy-mm, ...
* - c_str_of_ts: strings with format: yyyy-MM-dd HH:mm:ss
*/
def createDF(spark: SparkSession): DataFrame = {
val id = col("id")
val tsArray = Array[Long](year1980, year2000, year2030)
val secondsArray = tsArray.map(e => e / 1000000L)
val dateArray = Array[Int](0, 100, 200)
val columns = Array[Column](
TimeZonePerfUtils.createColumn(id, TimestampType, TsGenFunc(tsArray)).alias("c_ts"),
TimeZonePerfUtils.createColumn(id, LongType, TsGenFunc(tsArray)).alias("c_long_of_ts"),
TimeZonePerfUtils.createColumn(id, DateType, DateGenFunc(dateArray)).alias("c_date"),
TimeZonePerfUtils.createColumn(id, LongType, TsGenFunc(secondsArray))
.alias("c_long_of_ts_seconds"),
TimeZonePerfUtils.createColumn(id, IntegerType, DateGenFunc(dateArray))
.alias("c_int_of_date"),
TimeZonePerfUtils.createColumn(id, StringType, StringGenFunc(stringsForCast))
.alias("c_str_for_cast"),
TimeZonePerfUtils.createColumn(id, StringType, StringGenFunc(regularStrings))
.alias("c_str_of_ts")
)
val range = spark.range(numRows)
range.select(columns: _*)
}
/**
* Run 6 rounds for both Cpu and Gpu,
* but only print the elapsed times for the last 5 rounds.
*/
def runAndRecordTime(
testName: String,
func: (SparkSession, String) => DataFrame,
conf: SparkConf = new SparkConf()): Any = {
if (!enablePerfTest) {
// by default skip perf test
return None
}
println(s"test,type,zone,used MS")
for (zoneStr <- zones) {
// run 6 rounds, but ignore the first round.
val elapses = (1 to 6).map { i =>
// run on Cpu
val startOnCpu = System.nanoTime()
withCpuSparkSession(
spark => func(spark, zoneStr).collect(),
// set session time zone
conf.set("spark.sql.session.timeZone", zoneStr))
val endOnCpu = System.nanoTime()
val elapseOnCpuMS = (endOnCpu - startOnCpu) / 1000000L
if (i != 1) {
println(s"$testName,Cpu,$zoneStr,$elapseOnCpuMS")
}
// run on Gpu
val startOnGpu = System.nanoTime()
withGpuSparkSession(
spark => func(spark, zoneStr).collect(),
// set session time zone
conf.set("spark.sql.session.timeZone", zoneStr))
val endOnGpu = System.nanoTime()
val elapseOnGpuMS = (endOnGpu - startOnGpu) / 1000000L
if (i != 1) {
println(s"$testName,Gpu,$zoneStr,$elapseOnGpuMS")
(elapseOnCpuMS, elapseOnGpuMS)
} else {
(0L, 0L) // skip the first round
}
}
val meanCpu = elapses.map(_._1).sum / 5.0
val meanGpu = elapses.map(_._2).sum / 5.0
val speedup = meanCpu.toDouble / meanGpu.toDouble
println(f"$testName, $zoneStr: mean cpu time: $meanCpu%.2f ms, " +
f"mean gpu time: $meanGpu%.2f ms, speedup: $speedup%.2f x")
}
}
test("test from_utc_timestamp") {
assume(enablePerfTest)
// cache time zone DB in advance
GpuTimeZoneDB.cacheDatabase()
Thread.sleep(5L)
def perfTest(spark: SparkSession, zone: String): DataFrame = {
spark.read.parquet(path).select(functions.max( // use max to reduce the result data
functions.from_utc_timestamp(functions.col("c_ts"), zone)
))
}
runAndRecordTime("from_utc_timestamp", perfTest)
}
test("test to_utc_timestamp") {
assume(enablePerfTest)
// cache time zone DB in advance
GpuTimeZoneDB.cacheDatabase()
Thread.sleep(5L)
def perfTest(spark: SparkSession, zone: String): DataFrame = {
spark.read.parquet(path).select(functions.count(
functions.to_utc_timestamp(functions.col("c_ts"), zone)
))
}
runAndRecordTime("to_utc_timestamp", perfTest)
}
test("test hour") {
assume(enablePerfTest)
// cache time zone DB in advance
GpuTimeZoneDB.cacheDatabase()
Thread.sleep(5L)
def perfTest(spark: SparkSession, zone: String): DataFrame = {
spark.read.parquet(path).select(functions.count(
functions.hour(functions.col("c_ts"))
))
}
runAndRecordTime("hour",
perfTest)
}
test("test minute") {
assume(enablePerfTest)
// cache time zone DB in advance
GpuTimeZoneDB.cacheDatabase()
Thread.sleep(5L)
def perfTest(spark: SparkSession, zone: String): DataFrame = {
spark.read.parquet(path).select(functions.count(
functions.minute(functions.col("c_ts"))
))
}
runAndRecordTime("minute",
perfTest)
}
test("test second") {
assume(enablePerfTest)
// cache time zone DB in advance
GpuTimeZoneDB.cacheDatabase()
Thread.sleep(5L)
def perfTest(spark: SparkSession, zone: String): DataFrame = {
spark.read.parquet(path).select(functions.count(
functions.second(functions.col("c_ts"))
))
}
runAndRecordTime("second",
perfTest)
}
test("test unix_timestamp") {
assume(enablePerfTest)
// cache time zone DB in advance
GpuTimeZoneDB.cacheDatabase()
Thread.sleep(5L)
def perfTest(spark: SparkSession, zone: String): DataFrame = {
spark.read.parquet(path).select(functions.count(
functions.unix_timestamp(functions.col("c_str_of_ts"))
))
}
runAndRecordTime("unix_timestamp",
perfTest)
}
test("test from_unixtime") {
assume(enablePerfTest)
// cache time zone DB in advance
GpuTimeZoneDB.cacheDatabase()
Thread.sleep(5L)
def perfTest(spark: SparkSession, zone: String): DataFrame = {
spark.read.parquet(path).select(functions.count(
functions.from_unixtime(functions.col("c_long_of_ts_seconds"))
))
}
runAndRecordTime("from_unixtime",
perfTest)
}
test("test date_format") {
assume(enablePerfTest)
// cache time zone DB in advance
GpuTimeZoneDB.cacheDatabase()
Thread.sleep(5L)
def perfTest(spark: SparkSession, zone: String): DataFrame = {
spark.read.parquet(path).select(functions.count(
functions.date_format(functions.col("c_ts"), "yyyy-MM-dd HH:mm:ss")
))
}
runAndRecordTime("date_format",
perfTest)
}
}