Skip to content

[SPARK-32130][SQL] Disable the JSON option inferTimestamp by default #28966

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
MaxGekk wants to merge 6 commits into from
Closed

[SPARK-32130][SQL] Disable the JSON option inferTimestamp by default #28966

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
0b1616e
Set inferTimestamp to false by default
MaxGekk Jul 1, 2020
15b34ec
Fix JsonInferSchemaSuite
MaxGekk Jul 1, 2020
277a679
Fix JsonSuite
MaxGekk Jul 1, 2020
e21187e
Re-gen JsonBenchmark-results.txt
MaxGekk Jul 1, 2020
488f8c8
Re-gen JsonBenchmark-jdk11-results.txt
MaxGekk Jul 1, 2020
43fc9aa
Update SQL migration guide
MaxGekk Jul 1, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 5 additions & 1 deletion docs/sql-migration-guide.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,11 @@ license: |
- In Spark 3.1, `from_unixtime`, `unix_timestamp`,`to_unix_timestamp`, `to_timestamp` and `to_date` will fail if the specified datetime pattern is invalid. In Spark 3.0 or earlier, they result `NULL`.

- In Spark 3.1, casting numeric to timestamp will be forbidden by default. It's strongly recommended to use dedicated functions: TIMESTAMP_SECONDS, TIMESTAMP_MILLIS and TIMESTAMP_MICROS. Or you can set `spark.sql.legacy.allowCastNumericToTimestamp` to true to work around it. See more details in SPARK-31710.


## Upgrading from Spark SQL 3.0 to 3.0.1

- In Spark 3.0, JSON datasource and JSON function `schema_of_json` infer TimestampType from string values if they match to the pattern defined by the JSON option `timestampFormat`. Since version 3.0.1, the timestamp type inference is disabled by default. Set the JSON option `inferTimestamp` to `true` to enable such type inference.

## Upgrading from Spark SQL 2.4 to 3.0

### Dataset/DataFrame APIs
Expand Down
2 changes: 1 addition & 1 deletion sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/json/JSONOptions.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,7 +133,7 @@ private[sql] class JSONOptions(
* Enables inferring of TimestampType from strings matched to the timestamp pattern
* defined by the timestampFormat option.
*/
val inferTimestamp: Boolean = parameters.get("inferTimestamp").map(_.toBoolean).getOrElse(true)
val inferTimestamp: Boolean = parameters.get("inferTimestamp").map(_.toBoolean).getOrElse(false)

/** Build a Jackson [[JsonFactory]] using JSON options. */
def buildJsonFactory(): JsonFactory = {
Expand Down
56 changes: 33 additions & 23 deletions sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/json/JsonInferSchemaSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,22 +35,29 @@ class JsonInferSchemaSuite extends SparkFunSuite with SQLHelper {
assert(inferSchema.inferField(parser) === expectedType)
}

def checkTimestampType(pattern: String, json: String): Unit = {
checkType(Map("timestampFormat" -> pattern), json, TimestampType)
def checkTimestampType(pattern: String, json: String, inferTimestamp: Boolean): Unit = {
checkType(
Map("timestampFormat" -> pattern, "inferTimestamp" -> inferTimestamp.toString),
json,
if (inferTimestamp) TimestampType else StringType)
}

test("inferring timestamp type") {
Seq("legacy", "corrected").foreach { legacyParserPolicy =>
withSQLConf(SQLConf.LEGACY_TIME_PARSER_POLICY.key -> legacyParserPolicy) {
checkTimestampType("yyyy", """{"a": "2018"}""")
checkTimestampType("yyyy=MM", """{"a": "2018=12"}""")
checkTimestampType("yyyy MM dd", """{"a": "2018 12 02"}""")
checkTimestampType(
"yyyy-MM-dd'T'HH:mm:ss.SSS",
"""{"a": "2018-12-02T21:04:00.123"}""")
checkTimestampType(
"yyyy-MM-dd'T'HH:mm:ss.SSSSSSXXX",
"""{"a": "2018-12-02T21:04:00.123567+01:00"}""")
Seq(true, false).foreach { inferTimestamp =>
Seq("legacy", "corrected").foreach { legacyParserPolicy =>
withSQLConf(SQLConf.LEGACY_TIME_PARSER_POLICY.key -> legacyParserPolicy) {
checkTimestampType("yyyy", """{"a": "2018"}""", inferTimestamp)
checkTimestampType("yyyy=MM", """{"a": "2018=12"}""", inferTimestamp)
checkTimestampType("yyyy MM dd", """{"a": "2018 12 02"}""", inferTimestamp)
checkTimestampType(
"yyyy-MM-dd'T'HH:mm:ss.SSS",
"""{"a": "2018-12-02T21:04:00.123"}""",
inferTimestamp)
checkTimestampType(
"yyyy-MM-dd'T'HH:mm:ss.SSSSSSXXX",
"""{"a": "2018-12-02T21:04:00.123567+01:00"}""",
inferTimestamp)
}
}
}
}
Expand All @@ -71,16 +78,19 @@ class JsonInferSchemaSuite extends SparkFunSuite with SQLHelper {
}

test("skip decimal type inferring") {
Seq("legacy", "corrected").foreach { legacyParserPolicy =>
withSQLConf(SQLConf.LEGACY_TIME_PARSER_POLICY.key -> legacyParserPolicy) {
checkType(
options = Map(
"prefersDecimal" -> "false",
"timestampFormat" -> "yyyyMMdd.HHmmssSSS"
),
json = """{"a": "20181202.210400123"}""",
dt = TimestampType
)
Seq(true, false).foreach { inferTimestamp =>
Seq("legacy", "corrected").foreach { legacyParserPolicy =>
withSQLConf(SQLConf.LEGACY_TIME_PARSER_POLICY.key -> legacyParserPolicy) {
checkType(
options = Map(
"prefersDecimal" -> "false",
"timestampFormat" -> "yyyyMMdd.HHmmssSSS",
"inferTimestamp" -> inferTimestamp.toString
),
json = """{"a": "20181202.210400123"}""",
dt = if (inferTimestamp) TimestampType else StringType
)
}
}
}
}
Expand Down
86 changes: 43 additions & 43 deletions sql/core/benchmarks/JsonBenchmark-jdk11-results.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,106 +7,106 @@ OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-106
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
JSON schema inferring: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
No encoding 68879 68993 116 1.5 688.8 1.0X
UTF-8 is set 115270 115602 455 0.9 1152.7 0.6X
No encoding 69219 69342 116 1.4 692.2 1.0X
UTF-8 is set 143950 143986 55 0.7 1439.5 0.5X

Preparing data for benchmarking ...
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
count a short column: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
No encoding 47452 47538 113 2.1 474.5 1.0X
UTF-8 is set 77330 77354 30 1.3 773.3 0.6X
No encoding 57828 57913 136 1.7 578.3 1.0X
UTF-8 is set 83649 83711 60 1.2 836.5 0.7X

Preparing data for benchmarking ...
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
count a wide column: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
No encoding 60470 60900 534 0.2 6047.0 1.0X
UTF-8 is set 104733 104931 189 0.1 10473.3 0.6X
No encoding 64560 65193 1023 0.2 6456.0 1.0X
UTF-8 is set 102925 103174 216 0.1 10292.5 0.6X

Preparing data for benchmarking ...
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
select wide row: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
No encoding 130302 131072 976 0.0 260604.6 1.0X
UTF-8 is set 150860 151284 377 0.0 301720.1 0.9X
No encoding 131002 132316 1160 0.0 262003.1 1.0X
UTF-8 is set 152128 152371 332 0.0 304256.5 0.9X

Preparing data for benchmarking ...
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Select a subset of 10 columns: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
Select 10 columns 18619 18684 99 0.5 1861.9 1.0X
Select 1 column 24227 24270 38 0.4 2422.7 0.8X
Select 10 columns 19376 19514 160 0.5 1937.6 1.0X
Select 1 column 24089 24156 58 0.4 2408.9 0.8X

Preparing data for benchmarking ...
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
creation of JSON parser per line: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
Short column without encoding 7947 7971 21 1.3 794.7 1.0X
Short column with UTF-8 12700 12753 58 0.8 1270.0 0.6X
Wide column without encoding 92632 92955 463 0.1 9263.2 0.1X
Wide column with UTF-8 147013 147170 188 0.1 14701.3 0.1X
Short column without encoding 8131 8219 103 1.2 813.1 1.0X
Short column with UTF-8 13464 13508 44 0.7 1346.4 0.6X
Wide column without encoding 108012 108598 914 0.1 10801.2 0.1X
Wide column with UTF-8 150988 151369 412 0.1 15098.8 0.1X

Preparing data for benchmarking ...
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
JSON functions: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
Text read 713 734 19 14.0 71.3 1.0X
from_json 22019 22429 456 0.5 2201.9 0.0X
json_tuple 27987 28047 74 0.4 2798.7 0.0X
get_json_object 21468 21870 350 0.5 2146.8 0.0X
Text read 753 765 18 13.3 75.3 1.0X
from_json 23182 23446 230 0.4 2318.2 0.0X
json_tuple 31129 31304 181 0.3 3112.9 0.0X
get_json_object 22821 23073 225 0.4 2282.1 0.0X

Preparing data for benchmarking ...
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Dataset of json strings: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
Text read 2887 2910 24 17.3 57.7 1.0X
schema inferring 31793 31843 43 1.6 635.9 0.1X
parsing 36791 37104 294 1.4 735.8 0.1X
Text read 3078 3101 26 16.2 61.6 1.0X
schema inferring 30225 30434 333 1.7 604.5 0.1X
parsing 32237 32308 63 1.6 644.7 0.1X

Preparing data for benchmarking ...
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Json files in the per-line mode: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
Text read 10570 10611 45 4.7 211.4 1.0X
Schema inferring 48729 48763 41 1.0 974.6 0.2X
Parsing without charset 35490 35648 141 1.4 709.8 0.3X
Parsing with UTF-8 63853 63994 163 0.8 1277.1 0.2X
Text read 10835 10900 86 4.6 216.7 1.0X
Schema inferring 37720 37805 110 1.3 754.4 0.3X
Parsing without charset 35464 35538 100 1.4 709.3 0.3X
Parsing with UTF-8 67311 67738 381 0.7 1346.2 0.2X

OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Write dates and timestamps: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
Create a dataset of timestamps 2187 2190 5 4.6 218.7 1.0X
to_json(timestamp) 16262 16503 323 0.6 1626.2 0.1X
write timestamps to files 11679 11692 12 0.9 1167.9 0.2X
Create a dataset of dates 2297 2310 12 4.4 229.7 1.0X
to_json(date) 10904 10956 46 0.9 1090.4 0.2X
write dates to files 6610 6645 35 1.5 661.0 0.3X
Create a dataset of timestamps 2208 2222 14 4.5 220.8 1.0X
to_json(timestamp) 14299 14570 285 0.7 1429.9 0.2X
write timestamps to files 12955 12969 13 0.8 1295.5 0.2X
Create a dataset of dates 2297 2323 30 4.4 229.7 1.0X
to_json(date) 8509 8561 74 1.2 850.9 0.3X
write dates to files 6786 6827 45 1.5 678.6 0.3X

OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Read dates and timestamps: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
read timestamp text from files 2524 2530 9 4.0 252.4 1.0X
read timestamps from files 41002 41052 59 0.2 4100.2 0.1X
infer timestamps from files 84621 84939 526 0.1 8462.1 0.0X
read date text from files 2292 2302 9 4.4 229.2 1.1X
read date from files 16954 16976 21 0.6 1695.4 0.1X
timestamp strings 3067 3077 13 3.3 306.7 0.8X
parse timestamps from Dataset[String] 48690 48971 243 0.2 4869.0 0.1X
infer timestamps from Dataset[String] 97463 97786 338 0.1 9746.3 0.0X
date strings 3952 3956 3 2.5 395.2 0.6X
parse dates from Dataset[String] 24210 24241 30 0.4 2421.0 0.1X
from_json(timestamp) 71710 72242 629 0.1 7171.0 0.0X
from_json(date) 42465 42481 13 0.2 4246.5 0.1X
read timestamp text from files 2598 2613 18 3.8 259.8 1.0X
read timestamps from files 42007 42028 19 0.2 4200.7 0.1X
infer timestamps from files 18102 18120 28 0.6 1810.2 0.1X
Copy link
Member

@dongjoon-hyun dongjoon-hyun Jul 1, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry, but this test case seems to be supposed to run infer timestamp always. Why does this become faster?

- infer timestamps from files                       84621          84939         526          0.1        8462.1       0.0X
+ infer timestamps from files                       18102          18120          28          0.6        1810.2       0.1X

Copy link
Member

@dongjoon-hyun dongjoon-hyun Jul 1, 2020 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It seems that we need to update JsonBenchmark together.

readBench.addCase("infer timestamps from files", numIters) { _ =>
    spark.read.json(timestampDir).noop()
}

Copy link
Member

@dongjoon-hyun dongjoon-hyun Jul 1, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Never mind. Let's update the benchmark suite later as a follow-up.

read date text from files 2355 2360 5 4.2 235.5 1.1X
read date from files 17420 17458 33 0.6 1742.0 0.1X
timestamp strings 3099 3101 3 3.2 309.9 0.8X
parse timestamps from Dataset[String] 48188 48215 25 0.2 4818.8 0.1X
infer timestamps from Dataset[String] 22929 22988 102 0.4 2292.9 0.1X
date strings 4090 4103 11 2.4 409.0 0.6X
parse dates from Dataset[String] 24952 25068 139 0.4 2495.2 0.1X
from_json(timestamp) 66038 66352 413 0.2 6603.8 0.0X
from_json(date) 43755 43782 27 0.2 4375.5 0.1X


Loading