RAPIDS Accelerator for Apache Spark Compatibility with Apache Spark

The SQL plugin tries to produce results that are bit for bit identical with Apache Spark. There are a number of cases where there are some differences. In most cases operators that produce different results are off by default, and you can look at the configs for more information on how to enable them. In some cases we felt that enabling the incompatibility by default was worth the performance gain. All of those operators can be disabled through configs if it becomes a problem. Please also look at the current list of bugs which are typically incompatibilities that we have not yet addressed.

Ordering of Output

There are some operators where Spark does not guarantee the order of the output. These are typically things like aggregates and joins that may use a hash to distribute the work load among downstream tasks. In these cases the plugin does not guarantee that it will produce the same output order that Spark does. In cases such as an order by operation where the ordering is explicit the plugin will produce an ordering that is compatible with Spark’s guarantee. It may not be 100% identical if the ordering is ambiguous.

In versions of Spark prior to 3.1.0 -0.0 is always < 0.0 but in 3.1.0 and above this is not true for sorting. For all versions of the plugin -0.0 == 0.0 for sorting.

Spark’s sorting is typically a stable sort. Sort stability cannot be guaranteed in distributed work loads because the order in which upstream data arrives to a task is not guaranteed. Sort stability is only guaranteed in one situation which is reading and sorting data from a file using a single task/partition. The RAPIDS Accelerator does an unstable out of core sort by default. This simply means that the sort algorithm allows for spilling parts of the data if it is larger than can fit in the GPU’s memory, but it does not guarantee ordering of rows when the ordering of the keys is ambiguous. If you do rely on a stable sort in your processing you can request this by setting spark.rapids.sql.stableSort.enabled to true and RAPIDS will try to sort all the data for a given task/partition at once on the GPU. This may change in the future to allow for a spillable stable sort.

Floating Point

For most basic floating-point operations like addition, subtraction, multiplication, and division the plugin will produce a bit for bit identical result as Spark does. For other functions like sin, cos, etc. the output may be different, but within the rounding error inherent in floating-point calculations. The ordering of operations to calculate the value may differ between the underlying JVM implementation used by the CPU and the C++ standard library implementation used by the GPU.

In the case of round and bround the results can be off by more because they can enlarge the difference. This happens in cases where a binary floating-point representation cannot exactly capture a decimal value. For example 1.025 cannot exactly be represented and ends up being closer to 1.02499. The Spark implementation of round converts it first to a decimal value with complex logic to make it 1.025 and then does the rounding. This results in round(1.025, 2) under pure Spark getting a value of 1.03 but under the RAPIDS accelerator it produces 1.02. As a side note Python will produce 1.02, Java does not have the ability to do a round like this built in, but if you do the simple operation of Math.round(1.025 * 100.0)/100.0 you also get 1.02.

For the degrees functions, Spark’s implementation relies on Java JDK’s built-in functions Math.toDegrees. It is angrad * 180.0 / PI in Java 8 while angrad * (180d / PI) in Java 9+. So their results will differ depending on the JDK runtime versions when considering overflow. The RAPIDS Accelerator follows the behavior of Java 9+. Therefore, with JDK 8 or below, the degrees on GPU will not overflow on some very large numbers while the CPU version does.

For aggregations the underlying implementation is doing the aggregations in parallel and due to race conditions within the computation itself the result may not be the same each time the query is run. This is inherent in how the plugin speeds up the calculations and cannot be “fixed.” If a query joins on a floating point value, which is not wise to do anyways, and the value is the result of a floating point aggregation then the join may fail to work properly with the plugin but would have worked with plain Spark. This is behavior is enabled by default but can be disabled with the config spark.rapids.sql.variableFloatAgg.enabled.

`0.0` vs `-0.0`

Floating point allows zero to be encoded as 0.0 and -0.0, but the IEEE standard says that they should be interpreted as the same. Most databases normalize these values to always be 0.0. Spark does this in some cases but not all as is documented here. The underlying implementation of this plugin treats them as the same for essentially all processing. This can result in some differences with Spark for operations, prior to Spark 3.1.0, like sorting, and distinct count. There are still differences with joins, and comparisons even after Spark 3.1.0.

We do not disable operations that produce different results due to -0.0 in the data because it is considered to be a rare occurrence.

`NaN` vs `NaN`

Apache Spark does not have a consistent way to handle NaN comparison. Sometimes, all NaN are considered as one unique value while other times they can be treated as different. The outcome of NaN comparison can differ in various operations and also changed between Spark versions. The RAPIDS Accelerator tries to match its output with Apache Spark except for a few operation(s) listed below:

IN SQL expression: NaN can be treated as different values in Spark 3.1.2 and prior versions, see SPARK-36792 for more details. The RAPIDS Accelerator compares NaN values as equal for this operation which matches the behavior of Apache Spark 3.1.3 and later versions.

Decimal Support

Apache Spark supports decimal values with a precision up to 38. This equates to 128-bits. When processing the data, in most cases, it is temporarily converted to Java’s BigDecimal type which allows for effectively unlimited precision. Overflows will be detected whenever the BigDecimal value is converted back into the Spark decimal type.

The RAPIDS Accelerator does not implement a GPU equivalent of BigDecimal, but it does implement computation on 256-bit values to allow the detection of overflows. The points at which overflows are detected may differ between the CPU and GPU. Spark gives no guarantees that overflows are detected if an intermediate value could overflow the original decimal type during computation but the final value does not (e.g.: a sum of values with many large positive values followed by many large negative values). Spark injects overflow detection at various points during aggregation, and these points can fluctuate depending on cluster shape and number of shuffle partitions.

Unicode

Spark delegates Unicode operations to the underlying JVM. Each version of Java complies with a specific version of the Unicode standard. The SQL plugin does not use the JVM for Unicode support and is compatible with Unicode version 12.1. Because of this there may be corner cases where Spark will produce a different result compared to the plugin.

CSV Reading

Spark allows for stripping leading and trailing white space using various options that are off by default. The plugin will strip leading and trailing space for all values except strings.

There are also discrepancies/issues with specific types that are detailed below.

CSV Strings

Writing strings to a CSV file in general for Spark can be problematic unless you can ensure that your data does not have any line deliminator in it. The GPU accelerated CSV parser handles quoted line deliminators similar to multiLine mode. But there are still a number of issues surrounding it and they should be avoided.

Escaped quote characters '\"' are not supported well as described by this issue.

The GPU accelerated CSV parser does not replace invalid UTF-8 characters with the Unicode replacement character �. Instead it just passes them through as described in this issue.

CSV Dates

Only a limited set of formats are supported when parsing dates.

"yyyy-MM-dd"
"yyyy/MM/dd"
"yyyy-MM"
"yyyy/MM"
"MM-yyyy"
"MM/yyyy"
"MM-dd-yyyy"
"MM/dd/yyyy"
"dd-MM-yyyy"
"dd/MM/yyyy"

CSV Timestamps

The CSV parser does not support time zones. It will ignore any trailing time zone information, despite the format asking for a XXX or [XXX]. The CSV parser does not support the TimestampNTZ type and will fall back to CPU if spark.sql.timestampType is set to TIMESTAMP_NTZ or if an explicit schema is provided that contains the TimestampNTZ type.

The formats supported for timestamps are limited similar to dates. The first part of the format must be a supported date format. The second part must start with a 'T' to separate the time portion followed by one of the following formats:

HH:mm:ss.SSSXXX
HH:mm:ss[.SSS][XXX]
HH:mm:ss[.SSSXXX]
HH:mm
HH:mm:ss
HH:mm[:ss]
HH:mm:ss.SSS
HH:mm:ss[.SSS]

Just like with dates all timestamp formats are actually supported at the same time. The plugin will disable itself if it sees a format it cannot support.

Invalid timestamps in Spark, ones that have the correct format, but the numbers produce invalid dates or times, can result in an exception by default and how they are parsed can be controlled through a config. The RAPIDS Accelerator does not support any of this and will produce an incorrect date. Typically, one that overflowed.

CSV Floating Point

Parsing floating-point values has the same limitations as casting from string to float.

Also parsing of some values will not produce bit for bit identical results to what the CPU does. They are within round-off errors except when they are close enough to overflow to Inf or -Inf which then results in a number being returned when the CPU would have returned null.

CSV ANSI day time interval

This type was added in as a part of Spark 3.3.0, and it’s not supported on Spark versions before 3.3.0. Apache Spark can overflow when reading ANSI day time interval values. The RAPIDS Accelerator does not overflow and as such is not bug for bug compatible with Spark in this case.

Interval string in csv	Spark reads to	The RAPIDS Accelerator reads to	Comments
interval ‘106751992’ day	INTERVAL ‘-106751990’ DAY	NULL	Spark issue
interval ‘2562047789’ hour	INTERVAL ‘-2562047787’ HOUR	NULL	Spark issue

There are two valid textual representations in CSV: the ANSI style and the HIVE style, e.g:

SQL Type	An instance of ANSI style	An instance of HIVE style
INTERVAL DAY	INTERVAL ‘100’ DAY TO SECOND	100
INTERVAL DAY TO HOUR	INTERVAL ‘100 10’ DAY TO HOUR	100 10
INTERVAL DAY TO MINUTE	INTERVAL ‘100 10:30’ DAY TO MINUTE	100 10:30
INTERVAL DAY TO SECOND	INTERVAL ‘100 10:30:40.999999’ DAY TO SECOND	100 10:30:40.999999
INTERVAL HOUR	INTERVAL ‘10’ HOUR	10
INTERVAL HOUR TO MINUTE	INTERVAL ‘10:30’ HOUR TO MINUTE	10:30
INTERVAL HOUR TO SECOND	INTERVAL ‘10:30:40.999999’ HOUR TO SECOND	10:30:40.999999
INTERVAL MINUTE	INTERVAL ‘30’ MINUTE	30
INTERVAL MINUTE TO SECOND	INTERVAL ‘30:40.999999’ MINUTE TO SECOND	30:40.999999
INTERVAL SECOND	INTERVAL ‘40.999999’ SECOND	40.999999

Currently, the RAPIDS Accelerator only supports the ANSI style.

Hive Text File

Hive text files are very similar to CSV, but not exactly the same.

Hive Text File Floating Point

Parsing floating-point values has the same limitations as casting from string to float.

Hive Text File Decimal

Hive has some limitations in what decimal values it can parse. The GPU kernels that we use to parse decimal values do not have the same limitations. This means that there are times when the CPU version would return a null for an input value, but the GPU version will return a value. This typically happens for numbers with large negative exponents where the GPU will return 0 and Hive will return null. See https://github.com/NVIDIA/spark-rapids/issues/7246

ORC

The ORC format has fairly complete support for both reads and writes. There are only a few known issues. The first is for reading timestamps and dates around the transition between Julian and Gregorian calendars as described here. A similar issue exists for writing dates as described here. Writing timestamps, however only appears to work for dates after the epoch as described here.

The plugin supports reading uncompressed, snappy, zlib and zstd ORC files and writing uncompressed, snappy and zstd ORC files. At this point, the plugin does not have the ability to fall back to the CPU when reading an unsupported compression format, and will error out in that case.

Push Down Aggregates for ORC

Spark-3.3.0+ pushes down certain aggregations (MIN/MAX/COUNT) into ORC when the user-config spark.sql.orc.aggregatePushdown is set to true. By enabling this feature, aggregate query performance will improve as it takes advantage of the statistics information.

Caution

Spark ORC reader/writer assumes that all ORC files must have valid column statistics. This assumption deviates from the ORC-specification which states that statistics are optional. When a Spark-3.3.0+ job reads an ORC file with empty file-statistics, it fails while throwing the following runtime exception:

org.apache.spark.SparkException: Cannot read columns statistics in file: /PATH_TO_ORC_FILE
E    Caused by: java.util.NoSuchElementException
E        at java.util.LinkedList.removeFirst(LinkedList.java:270)
E        at java.util.LinkedList.remove(LinkedList.java:685)
E        at org.apache.spark.sql.execution.datasources.orc.OrcFooterReader.convertStatistics(OrcFooterReader.java:54)
E        at org.apache.spark.sql.execution.datasources.orc.OrcFooterReader.readStatistics(OrcFooterReader.java:45)
E        at org.apache.spark.sql.execution.datasources.orc.OrcUtils$.createAggInternalRowFromFooter(OrcUtils.scala:428)

The Spark community is planning to work on a runtime fallback to read from actual rows when ORC file-statistics are missing (see SPARK-34960 discussion).

Writing ORC Files

There are issues writing ORC files with dates or timestamps that fall within the lost days during the switch from the Julian to Gregorian calendar, i.e.: between October 3rd, 1582 and October 15th,

Dates or timestamps that fall within the range of lost days will not always be written properly by the GPU to the ORC file. The values read by the CPU and the GPU may differ with the CPU often rounding the day up to October 15th, 1582 whereas the GPU does not.

Note that the CPU rounds up dates or timestamps in the lost days range to October 15th, 1582 before writing to the ORC file. If the CPU writes these problematic dates or timestamps to an ORC file, they will be read properly by both the CPU and the GPU.

Reading ORC Files

To take advantage of the aggregate query optimization, where only the ORC metadata is read to satisfy the query, the ORC read falls back to the CPU as it is a metadata-only query.

Parquet

The Parquet format has more configs because there are multiple versions with some compatibility issues between them. Dates and timestamps are where the known issues exist. For reads when spark.sql.legacy.parquet.datetimeRebaseModeInWrite is set to CORRECTED timestamps before the transition between the Julian and Gregorian calendars are wrong, but dates are fine. When spark.sql.legacy.parquet.datetimeRebaseModeInWrite is set to LEGACY, the read may fail for values occurring before the transition between the Julian and Gregorian calendars, i.e.: date <= 1582-10-04.

When writing spark.sql.legacy.parquet.datetimeRebaseModeInWrite is currently ignored as described here.

The plugin supports reading uncompressed, snappy, gzip and zstd Parquet files and writing uncompressed, snappy and zstd Parquet files. At this point, the plugin does not have the ability to fall back to the CPU when reading an unsupported compression format, and will error out in that case.

JSON

The JSON format read is a very experimental feature which is expected to have some issues, so we disable it by default. If you would like to test it, you need to enable spark.rapids.sql.format.json.enabled and spark.rapids.sql.format.json.read.enabled.

Reading input containing invalid JSON format (in any row) will throw runtime exception. An example of valid input is as following:

{"name":"Andy", "age":30}
{"name":"Justin", "age":19}

The following input is invalid and will cause error:

{"name":"Andy", "age":30} ,,,,
{"name":"Justin", "age":19}

{"name":  Justin", "age":19}

Reading input with duplicated json key names is also incompatible with CPU Spark.

JSON supporting types

In the current version, nested types (array, struct, and map types) are not yet supported in regular JSON parsing.

`from_json` function

This particular function supports to output a map or struct type with limited functionalities.

The from_json function is disabled by default because it is experimental and has some known incompatibilities with Spark, and can be enabled by setting spark.rapids.sql.expression.JsonToStructs=true.

Dates are partially supported but there are some known issues:

Only the default dateFormat of yyyy-MM-dd is supported in Spark 3.1.x. The query will fall back to CPU if any other format is specified (#9667)
Strings containing integers with more than four digits will be parsed as null (#9664) whereas Spark versions prior to 3.4 will parse these numbers as number of days since the epoch, and in Spark 3.4 and later, an exception will be thrown.

Timestamps are partially supported but there are some known issues:

Only the default timestampFormat of yyyy-MM-dd'T'HH:mm:ss[.SSS][XXX] is supported. The query will fall back to CPU if any other format is specified (#9273)
Strings containing integers with more than four digits will be parsed as null (#9664) whereas Spark versions prior to 3.4 will parse these numbers as number of days since the epoch, and in Spark 3.4 and later, an exception will be thrown.
Strings containing special date constant values such as now and today will parse as null (#9724), which differs from the behavior in Spark 3.1.x

When reading numeric values, the GPU implementation always supports leading zeros regardless of the setting for the JSON option allowNumericLeadingZeros (#9588).

For struct output type, the function only supports struct of struct, array, string, integral, floating-point, and decimal types. The output is incompatible if duplicated json key names are present in the input strings. For schemas that include IntegerType, if arbitrarily large numbers are specified in the JSON strings, the GPU implementation will cast the numbers to IntegerType, whereas CPU Spark will return null.

In particular, the output map is not resulted from a regular JSON parsing but instead it will just contain plain text of key-value pairs extracted directly from the input JSON string. Due to such limitations, the input JSON map type schema must be MAP<STRING,STRING> and nothing else. Furthermore, there is no validation, no error tolerance, no data conversion as well as string formatting is performed. This may lead to some minor differences in the output if compared to the result of Spark CPU’s from_json, such as:

Floating point numbers in the input JSON string such as 1.2000 will not be reformatted to 1.2. Instead, the output will be the same as the input.
If the input JSON is given as multiple rows, any row containing invalid JSON format will be parsed as an empty struct instead of a null value (#9592).

When a JSON attribute contains mixed types (different types in different rows), such as a mix of dictionaries and lists, Spark will return a string representation of the JSON, but when running on GPU, the default behavior is to throw an exception. There is an experimental setting spark.rapids.sql.json.read.mixedTypesAsString.enabled that can be set to true to support reading mixed types as string, but there are known issues where it could also read structs as string in some cases. There can also be minor formatting differences. Spark will return a parsed and formatted representation, but the GPU implementation returns the unparsed JSON string.

`to_json` function

The to_json function is disabled by default because it is experimental and has some known incompatibilities with Spark, and can be enabled by setting spark.rapids.sql.expression.StructsToJson=true.

Known issues are:

There can be rounding differences when formatting floating-point numbers as strings. For example, Spark may produce -4.1243574E26 but the GPU may produce -4.124357351E26.
Not all JSON options are respected

JSON Floating Point

Parsing floating-point values has the same limitations as casting from string to float.

Prior to Spark 3.3.0, reading JSON strings such as "+Infinity" when specifying that the data type is FloatType or DoubleType caused these values to be parsed even when allowNonNumericNumbers is set to false. Also, Spark versions prior to 3.3.0 only supported the "Infinity" and "-Infinity" representations of infinity and did not support "+INF", "-INF", or "+Infinity", which Spark considers valid when unquoted. The GPU JSON reader is consistent with the behavior in Spark 3.3.0 and later.

Another limitation of the GPU JSON reader is that it will parse strings containing non-string boolean or numeric values where Spark will treat them as invalid inputs and will just return null.

JSON Timestamps

The JSON parser does not support the TimestampNTZ type and will fall back to CPU if spark.sql.timestampType is set to TIMESTAMP_NTZ or if an explicit schema is provided that contains the TimestampNTZ type.

There is currently no support for reading numeric values as timestamps and null values are returned instead (#4940). A workaround would be to read as longs and then cast to timestamp.

JSON Schema discovery

Spark SQL can automatically infer the schema of a JSON dataset if schema is not provided explicitly. The CPU handles schema discovery and there is no GPU acceleration of this. By default Spark will read/parse the entire dataset to determine the schema. This means that some options/errors which are ignored by the GPU may still result in an exception if used with schema discovery.

JSON options

Spark supports passing options to the JSON parser when reading a dataset. In most cases if the RAPIDS Accelerator sees one of these options that it does not support it will fall back to the CPU. In some cases we do not. The following options are documented below.

allowNumericLeadingZeros - Allows leading zeros in numbers (e.g. 00012). By default this is set to false. When it is false Spark throws an exception if it encounters this type of number. The RAPIDS Accelerator strips off leading zeros from all numbers and this config has no impact on it.
allowUnquotedControlChars - Allows JSON Strings to contain unquoted control characters (ASCII characters with value less than 32, including tab and line feed characters) or not. By default this is set to false. If the schema is provided while reading JSON file, then this flag has no impact on the RAPIDS Accelerator as it always allows unquoted control characters but Spark reads these entries incorrectly as null. However, if the schema is not provided and when the option is false, then RAPIDS Accelerator’s behavior is same as Spark where an exception is thrown as discussed in JSON Schema discovery section.
allowNonNumericNumbers - Allows NaN and Infinity values to be parsed (note that these are not valid numeric values in the JSON specification). Spark versions prior to 3.3.0 have inconsistent behavior and will parse some variants of NaN and Infinity even when this option is disabled (SPARK-38060). The RAPIDS Accelerator behavior is consistent with Spark version 3.3.0 and later.

get_json_object

The GetJsonObject operator takes a JSON formatted string and a JSON path string as input. The code base for this is currently separate from GPU parsing of JSON for files and FromJsonObject. Because of this the results can be different from each other. Because of several incompatibilities and bugs in the GPU version of GetJsonObject it will be on the CPU by default. If you are aware of the current limitations with the GPU version, you might see a significant performance speedup if you enable it by setting spark.rapids.sql.expression.GetJsonObject to true.

The following is a list of known differences.

No input validation. If the input string is not valid JSON Apache Spark returns a null result, but ours will still try to find a match.
Escapes are not properly processed for Strings. When returning a result for a quoted string Apache Spark will remove the quotes and replace any escape sequences with the proper characters. The escape sequence processing does not happen on the GPU.
Invalid JSON paths could throw exceptions If a JSON path is not valid Apache Spark returns a null result, but ours may throw an exception and fail the query.
Non-string output is not normalized When returning a result for things other than strings, a number of things are normalized by Apache Spark, but are not normalized by the GPU, like removing unnecessary white space, parsing and then serializing floating point numbers, turning single quotes to double quotes, and removing unneeded escapes for single quotes.

The following is a list of bugs in either the GPU version or arguably in Apache Spark itself.

https://github.com/NVIDIA/spark-rapids/issues/10219 non-matching quotes in quoted strings
https://github.com/NVIDIA/spark-rapids/issues/10213 array index notation works without root
https://github.com/NVIDIA/spark-rapids/issues/10214 unquoted array index notation is not supported
https://github.com/NVIDIA/spark-rapids/issues/10215 leading spaces can be stripped from named keys.
https://github.com/NVIDIA/spark-rapids/issues/10216 It appears that Spark is flattening some output, which is different from other implementations including the GPU version.
https://github.com/NVIDIA/spark-rapids/issues/10217 a JSON path execution bug
https://issues.apache.org/jira/browse/SPARK-46761 Apache Spark does not allow the ? character in a quoted JSON path string.

Avro

The Avro format read is a very experimental feature which is expected to have some issues, so we disable it by default. If you would like to test it, you need to enable spark.rapids.sql.format.avro.enabled and spark.rapids.sql.format.avro.read.enabled.

Currently, the GPU accelerated Avro reader doesn’t support reading the Avro version 1.2 files.

Supported types

The boolean, byte, short, int, long, float, double, string are supported in current version.

Regular Expressions

The following Apache Spark regular expression functions and expressions are supported on the GPU:

RLIKE
regexp
regexp_extract
regexp_extract_all
regexp_like
regexp_replace
string_split
str_to_map

Regular expression evaluation on the GPU is enabled by default when the UTF-8 character set is used by the current locale. Execution will fall back to the CPU for regular expressions that are not yet supported on the GPU, and in environments where the locale does not use UTF-8. However, there are some edge cases that will still execute on the GPU and produce different results to the CPU. To disable regular expressions on the GPU, set spark.rapids.sql.regexp.enabled=false.

These are the known edge cases where running on the GPU will produce different results to the CPU:

Regular expressions that contain an end of line anchor ‘$’ or end of string anchor ‘\Z’ immediately next to a newline or a repetition that produces zero or more results (#5610)`
Word and non-word boundaries, \b and \B
Line anchor $ will incorrectly match any of the unicode characters \u0085, \u2028, or \u2029 followed by another line-terminator, such as \n. For example, the pattern TEST$ will match TEST\u0085\n on the GPU but not on the CPU (#7585).

The following regular expression patterns are not yet supported on the GPU and will fall back to the CPU.

Line anchors ^ and $ are not supported in some contexts, such as when combined with a choice (^|a or $|a).
String anchor \Z is not supported by regexp_replace, and in some rare contexts.
String anchor \z is not supported
Patterns containing an end of line or string anchor immediately next to a newline or repetition that produces zero or more results
Line anchor $ and string anchors \Z are not supported in patterns containing \W or \D
Line and string anchors are not supported by string_split and str_to_map
Lazy quantifiers within a choice block such as (2|\u2029??)+
Possessive quantifiers, such as a*+
Character classes that use union, intersection, or subtraction semantics, such as [a-d[m-p]], [a-z&&[def]], or [a-z&&[^bc]]
Empty groups: ()

Work is ongoing to increase the range of regular expressions that can run on the GPU.

URL Parsing

parse_url QUERY with a column key could produce different results on CPU and GPU. In Spark, the key in parse_url could act like a regex, but GPU will match the key exactly. If key is literal, GPU will check if key contains regex special characters and fallback to CPU if it does, but if key is column, it will not be able to fallback. For example, parse_url("http://foo/bar?abc=BAD&a.c=GOOD", QUERY, "a.c") will return “BAD” on CPU, but “GOOD” on GPU. See the Spark issue: https://issues.apache.org/jira/browse/SPARK-44500

Timestamps

Spark stores timestamps internally relative to the JVM time zone. Converting an arbitrary timestamp between time zones is not currently supported on the GPU. Therefore operations involving timestamps will only be GPU-accelerated if the time zone used by the JVM is UTC.

Windowing

Window Functions

Because of ordering differences between the CPU and the GPU window functions especially row based window functions like row_number, lead, and lag can produce different results if the ordering includes both -0.0 and 0.0, or if the ordering is ambiguous. Spark can produce different results from one run to another if the ordering is ambiguous on a window function too.

Range Window

When the order-by column of a range based window is numeric type like byte/short/int/long and the range boundary calculated for a value has overflow, CPU and GPU will get different results.

For example, consider the following dataset:

+------+---------+
| id   | dollars |
+------+---------+
|    1 |    NULL |
|    1 |      13 |
|    1 |      14 |
|    1 |      15 |
|    1 |      15 |
|    1 |      17 |
|    1 |      18 |
|    1 |      52 |
|    1 |      53 |
|    1 |      61 |
|    1 |      65 |
|    1 |      72 |
|    1 |      73 |
|    1 |      75 |
|    1 |      78 |
|    1 |      84 |
|    1 |      85 |
|    1 |      86 |
|    1 |      92 |
|    1 |      98 |
+------+---------+

After executing the SQL statement:

SELECT
 COUNT(dollars) over
    (PARTITION BY id
    ORDER BY CAST (dollars AS Byte) ASC
    RANGE BETWEEN 127 PRECEDING AND 127 FOLLOWING)
FROM table

The results will differ between the CPU and GPU due to overflow handling.

CPU: WrappedArray([0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0], [0])
GPU: WrappedArray([0], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19], [19])

To enable byte-range windowing on the GPU, set spark.rapids.sql.window.range.byte.enabled to true.

We also provide configurations for other integral range types:

The reason why we default the configurations to false for byte/short and to true for int/long is that we think the most real-world queries are based on int or long.

Parsing strings as dates or timestamps

When converting strings to dates or timestamps using functions like to_date and unix_timestamp, the specified format string will fall into one of three categories:

Supported on GPU and 100% compatible with Spark
Supported on GPU but may produce different results to Spark
Unsupported on GPU

The formats which are supported on GPU vary depending on the setting for timeParserPolicy.

CORRECTED and EXCEPTION timeParserPolicy

With timeParserPolicy set to CORRECTED or EXCEPTION (the default), the following formats are supported on the GPU without requiring any additional settings.

yyyy-MM-dd
yyyy/MM/dd
yyyy-MM
yyyy/MM
dd/MM/yyyy
yyyy-MM-dd HH:mm:ss
MM-dd
MM/dd
dd-MM
dd/MM
MM/yyyy
MM-yyyy
MM/dd/yyyy
MM-dd-yyyy
MMyyyy

Valid Spark date/time formats that do not appear in the list above may also be supported but have not been extensively tested and may produce different results compared to the CPU. Known issues include:

Valid dates and timestamps followed by trailing characters (including whitespace) may be parsed to non-null values on GPU where Spark would treat the data as invalid and return null

To attempt to use other formats on the GPU, set spark.rapids.sql.incompatibleDateFormats.enabled to true.

Formats that contain any of the following characters are unsupported and will fall back to CPU:

'k', 'K','z', 'V', 'c', 'F', 'W', 'Q', 'q', 'G', 'A', 'n', 'N',
'O', 'X', 'p', '\'', '[', ']', '#', '{', '}', 'Z', 'w', 'e', 'E', 'x', 'Z', 'Y'

Formats that contain any of the following words are unsupported and will fall back to CPU:

"u", "uu", "uuu", "uuuu", "uuuuu", "uuuuuu", "uuuuuuu", "uuuuuuuu", "uuuuuuuuu", "uuuuuuuuuu",
"y", "yy", yyy", "yyyyy", "yyyyyy", "yyyyyyy", "yyyyyyyy", "yyyyyyyyy", "yyyyyyyyyy",
"D", "DD", "DDD", "s", "m", "H", "h", "M", "MMM", "MMMM", "MMMMM", "L", "LLL", "LLLL", "LLLLL",
"d", "S", "SS", "SSS", "SSSS", "SSSSS", "SSSSSSSSS", "SSSSSSS", "SSSSSSSS"

LEGACY timeParserPolicy

With timeParserPolicy set to LEGACY and spark.rapids.sql.incompatibleDateFormats.enabled set to true, and spark.sql.ansi.enabled set to false, the following formats are supported but not guaranteed to produce the same results as the CPU:

dd-MM-yyyy
dd/MM/yyyy
yyyy/MM/dd
yyyy-MM-dd
yyyy/MM/dd HH:mm:ss
yyyy-MM-dd HH:mm:ss

LEGACY timeParserPolicy support has the following limitations when running on the GPU:

Only 4 digit years are supported
The proleptic Gregorian calendar is used instead of the hybrid Julian+Gregorian calendar that Spark uses in legacy mode

Formatting dates and timestamps as strings

When formatting dates and timestamps as strings using functions such as from_unixtime, only a subset of valid format strings are supported on the GPU.

Formats that contain any of the following characters are unsupported and will fall back to CPU:

'k', 'K','z', 'V', 'c', 'F', 'W', 'Q', 'q', 'G', 'A', 'n', 'N',
'O', 'X', 'p', '\'', '[', ']', '#', '{', '}', 'Z', 'w', 'e', 'E', 'x', 'Z', 'Y'

Formats that contain any of the following words are unsupported and will fall back to CPU:

"u", "uu", "uuu", "uuuu", "uuuuu", "uuuuuu", "uuuuuuu", "uuuuuuuu", "uuuuuuuuu", "uuuuuuuuuu",
"y", yyy", "yyyyy", "yyyyyy", "yyyyyyy", "yyyyyyyy", "yyyyyyyyy", "yyyyyyyyyy",
"D", "DD", "DDD", "s", "m", "H", "h", "M", "MMM", "MMMM", "MMMMM", "L", "LLL", "LLLL", "LLLLL",
"d", "S", "SS", "SSSS", "SSSSS", "SSSSSSSSS", "SSSSSSS", "SSSSSSSS"

Note that this list differs very slightly from the list given in the previous section for parsing strings to dates because the two-digit year format "yy" is supported when formatting dates as strings but not when parsing strings to dates.

Casting between types

In general, performing cast and ansi_cast operations on the GPU is compatible with the same operations on the CPU. However, there are some exceptions. For this reason, certain casts are disabled on the GPU by default and require configuration options to be specified to enable them.

Float to Decimal

The GPU will use a different strategy from Java’s BigDecimal to handle/store decimal values, which leads to restrictions:

Float values cannot be larger than 1e18 or smaller than -1e18 after conversion.
The results produced by GPU slightly differ from the default results of Spark.

This configuration is enabled by default. To disable this operation on the GPU set spark.rapids.sql.castFloatToDecimal.enabled to false

Float to Integral Types

With both cast and ansi_cast, Spark uses the expression Math.floor(x) <= MAX && Math.ceil(x) >= MIN to determine whether a floating-point value can be converted to an integral type. Prior to Spark 3.1.0 the MIN and MAX values were floating-point values such as Int.MaxValue.toFloat but starting with 3.1.0 these are now integral types such as Int.MaxValue so this has slightly affected the valid range of values and now differs slightly from the behavior on GPU in some cases.

This configuration is enabled by default. To disable this operation on the GPU set spark.rapids.sql.castFloatToIntegralTypes.enabled to false.

Float to String

The Rapids Accelerator for Apache Spark uses uses a method based on ryu when converting floating point data type to string. As a result the computed string can differ from the output of Spark in some cases: sometimes the output is shorter (which is arguably more accurate) and sometimes the output may differ in the precise digits output.

This configuration is enabled by default. To disable this operation on the GPU set spark.rapids.sql.castFloatToString.enabled to false.

The format_number function also uses ryu as the solution when formatting floating-point data types to strings, so results may differ from Spark in the same way. To disable this on the GPU, set spark.rapids.sql.formatNumberFloat.enabled to false.

String to Float

Casting from string to double on the GPU returns incorrect results when the string represents any number in the following ranges. In both cases the GPU returns Double.MaxValue. The default behavior in Apache Spark is to return +Infinity and -Infinity, respectively.

1.7976931348623158E308 <= x < 1.7976931348623159E308
-1.7976931348623159E308 < x <= -1.7976931348623158E308

Casting from string to double on the GPU could also sometimes return incorrect results if the string contains high precision values. Apache Spark rounds the values to the nearest double, while the GPU truncates the values directly.

Also, the GPU does not support casting from strings containing hex values to floating-point types.

This configuration is enabled by default. To disable this operation on the GPU set spark.rapids.sql.castStringToFloat.enabled to false.

String to Date

The following formats/patterns are supported on the GPU. Timezone of UTC is assumed.

Format or Pattern	Supported on GPU?
`"yyyy"`	Yes
`"yyyy-[M]M"`	Yes
`"yyyy-[M]M "`	Yes
`"yyyy-[M]M-[d]d"`	Yes
`"yyyy-[M]M-[d]d "`	Yes
`"yyyy-[M]M-[d]d *"`	Yes
`"yyyy-[M]M-[d]d T*"`	Yes
`"epoch"`	Yes
`"now"`	Yes
`"today"`	Yes
`"tomorrow"`	Yes
`"yesterday"`	Yes

String to Timestamp

To allow casts from string to timestamp on the GPU, enable the configuration property spark.rapids.sql.castStringToTimestamp.enabled.

Casting from string to timestamp currently has the following limitations.

Format or Pattern	Supported on GPU?
`"yyyy"`	Yes
`"yyyy-[M]M"`	Yes
`"yyyy-[M]M "`	Yes
`"yyyy-[M]M-[d]d"`	Yes
`"yyyy-[M]M-[d]d "`	Yes
`"yyyy-[M]M-[d]dT[h]h:[m]m:[s]s[zone_id]"`	Partial [1]
`"yyyy-[M]M-[d]d [h]h:[m]m:[s]s[zone_id]"`	Partial [1]
`"yyyy-[M]M-[d]dT[h]h:[m]m:[s]s.[ms][ms][ms][us][us][us][zone_id]"`	Partial [1]
`"yyyy-[M]M-[d]d [h]h:[m]m:[s]s.[ms][ms][ms][us][us][us][zone_id]"`	Partial [1]
`"[h]h:[m]m:[s]s.[ms][ms][ms][us][us][us][zone_id]"`	Partial [1]
`"T[h]h:[m]m:[s]s.[ms][ms][ms][us][us][us][zone_id]"`	Partial [1]
`"epoch"`	Yes
`"now"`	Yes
`"today"`	Yes
`"tomorrow"`	Yes
`"yesterday"`	Yes

[1] Leap seconds are not supported. If a zone_id is provided then only timezone ‘Z’ (UTC) is supported. Casting unsupported formats will result in null values.

Spark is very lenient when casting from string to timestamp because all date and time components are optional, meaning that input values such as T, T2, :, ::, 1:, :1, and ::1 are considered valid timestamps. The GPU will treat these values as invalid and cast them to null values.

Constant Folding

ConstantFolding is an operator optimization rule in Catalyst that replaces expressions that can be statically evaluated with their equivalent literal values. The RAPIDS Accelerator relies on constant folding and parts of the query will not be accelerated if org.apache.spark.sql.catalyst.optimizer.ConstantFolding is excluded as a rule.

long/double to Timestamp

Spark 330+ has an issue when casting a big enough long/double as timestamp, refer to https://issues.apache.org/jira/browse/SPARK-39209. Spark 330+ throws errors while the RAPIDS Accelerator can handle correctly when casting a big enough long/double as timestamp.

JSON string handling

The 0.5 release introduces the get_json_object operation. The JSON specification only allows double quotes around strings in JSON data, whereas Spark allows single quotes around strings in JSON data. The RAPIDS Spark get_json_object operation on the GPU will return None in PySpark or Null in Scala when trying to match a string surrounded by single quotes. This behavior will be updated in a future release to more closely match Spark.

If the JSON has a single quote ' in the path, the GPU query may fail with ai.rapids.cudf.CudfException. More examples are in issue-12483.

Approximate Percentile

The GPU implementation of approximate_percentile uses t-Digests which have high accuracy, particularly near the tails of a distribution. The results are not bit-for-bit identical with the Apache Spark implementation of approximate_percentile. This feature is enabled by default and can be disabled by setting spark.rapids.sql.expression.ApproximatePercentile=false.

Conditionals and operations with side effects (ANSI mode)

In Apache Spark condition operations like if, coalesce, and case/when lazily evaluate their parameters on a row by row basis. On the GPU it is generally more efficient to evaluate the parameters regardless of the condition and then select which result to return based on the condition. This is fine so long as there are no side effects caused by evaluating a parameter. For most expressions in Spark this is true, but in ANSI mode many expressions can throw exceptions, like for the Add expression if an overflow happens. This is also true of UDFs, because by their nature they are user defined and can have side effects like throwing exceptions.

Currently, the RAPIDS Accelerator assumes that there are no side effects. This can result it situations, specifically in ANSI mode, where the RAPIDS Accelerator will always throw an exception, but Spark on the CPU will not. For example:

spark.conf.set("spark.sql.ansi.enabled", "true")

Seq(0L, Long.MaxValue).toDF("val")
    .repartition(1) // The repartition makes Spark not optimize selectExpr away
    .selectExpr("IF(val > 1000, null, val + 1) as ret")
    .show()

If the above example is run on the CPU you will get a result like.

+----+
| ret|
+----+
|   1|
|null|
+----+

But if it is run on the GPU an overflow exception is thrown. As was explained before this is because the RAPIDS Accelerator will evaluate both val + 1 and null regardless of the result of the condition. In some cases you can work around this. The above example could be re-written so the if happens before the Add operation.

Seq(0L, Long.MaxValue).toDF("val")
    .repartition(1) // The repartition makes Spark not optimize selectExpr away
    .selectExpr("IF(val > 1000, null, val) + 1 as ret")
    .show()

But this is not something that can be done generically and requires inner knowledge about what can trigger a side effect.

RAPIDS Accelerator for Apache Spark Compatibility with Apache Spark

Ordering of Output

Floating Point

0.0 vs -0.0

NaN vs NaN

Decimal Support

Unicode

CSV Reading

CSV Strings

CSV Dates

CSV Timestamps

CSV Floating Point

CSV ANSI day time interval

Hive Text File

Hive Text File Floating Point

Hive Text File Decimal

ORC

Push Down Aggregates for ORC

Parquet

JSON

JSON supporting types

from_json function

to_json function

JSON Floating Point

JSON Timestamps

JSON Schema discovery

JSON options

get_json_object

Avro

Supported types

Regular Expressions

URL Parsing

Timestamps

Windowing

Window Functions

Range Window

Parsing strings as dates or timestamps

CORRECTED and EXCEPTION timeParserPolicy

LEGACY timeParserPolicy

Formatting dates and timestamps as strings

Casting between types

Float to Decimal

Float to Integral Types

Float to String

String to Float

String to Date

String to Timestamp

Constant Folding

long/double to Timestamp

JSON string handling

Approximate Percentile

Conditionals and operations with side effects (ANSI mode)

`0.0` vs `-0.0`

`NaN` vs `NaN`

`from_json` function

`to_json` function