Skip to content
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

Jump to bottom

support interval.microseconds for range window TimeStampType #2525

Merged
merged 3 commits into from
Jun 1, 2021
Merged

support interval.microseconds for range window TimeStampType #2525

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
431ffb8
support interval.microseconds for TimeStampType
wbo4958 May 27, 2021
8a5deec
add overflow checking
wbo4958 May 27, 2021
2e4b5b4
add more clear message when falling back to CPU
wbo4958 May 28, 2021
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
111 changes: 61 additions & 50 deletions integration_tests/src/main/python/window_function_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,16 +31,26 @@
('b', IntegerGen()),
('c', IntegerGen())]

_grpkey_longs_with_timestamps = [
_grpkey_longs_with_dates = [
('a', RepeatSeqGen(LongGen(), length=2048)),
('b', DateGen(nullable=False, start=date(year=2020, month=1, day=1), end=date(year=2020, month=12, day=31))),
('c', IntegerGen())]

_grpkey_longs_with_nullable_timestamps = [
_grpkey_longs_with_nullable_dates = [
('a', RepeatSeqGen(LongGen(nullable=False), length=20)),
('b', DateGen(nullable=(True, 5.0), start=date(year=2020, month=1, day=1), end=date(year=2020, month=12, day=31))),
('c', IntegerGen())]

_grpkey_longs_with_timestamps = [
('a', RepeatSeqGen(LongGen(), length=2048)),
('b', TimestampGen(nullable=False)),
('c', IntegerGen())]

_grpkey_longs_with_nullable_timestamps = [
('a', RepeatSeqGen(LongGen(nullable=False), length=20)),
('b', TimestampGen(nullable=(True, 5.0))),
('c', IntegerGen())]

_grpkey_longs_with_decimals = [
('a', RepeatSeqGen(LongGen(nullable=False), length=20)),
('b', DecimalGen(precision=18, scale=3, nullable=False)),
Expand All @@ -59,19 +69,32 @@

_grpkey_byte_with_nulls = [
('a', RepeatSeqGen(int_gen, length=20)),
('b', ByteGen(nullable=True, min_val=-98, max_val=98, special_cases=[]))]
# restrict the values generated by min_val/max_val not to be overflow when calculating
('b', ByteGen(nullable=True, min_val=-98, max_val=98, special_cases=[])),
('c', IntegerGen())]

_grpkey_short_with_nulls = [
('a', RepeatSeqGen(int_gen, length=20)),
('b', ShortGen(nullable=True, min_val=-32700, max_val=32700, special_cases=[]))]
# restrict the values generated by min_val/max_val not to be overflow when calculating
('b', ShortGen(nullable=True, min_val=-32700, max_val=32700, special_cases=[])),
('c', IntegerGen())]

_grpkey_int_with_nulls = [
('a', RepeatSeqGen(int_gen, length=20)),
('b', IntegerGen(nullable=True, min_val=-2147483000, max_val=2147483000, special_cases=[]))]
# restrict the values generated by min_val/max_val not to be overflow when calculating
('b', IntegerGen(nullable=True, min_val=-2147483000, max_val=2147483000, special_cases=[])),
('c', IntegerGen())]

_grpkey_long_with_nulls = [
('a', RepeatSeqGen(int_gen, length=20)),
('b', LongGen(nullable=True, min_val=-9223372036854775000, max_val=9223372036854775000, special_cases=[]))]
# restrict the values generated by min_val/max_val not to be overflow when calculating
('b', LongGen(nullable=True, min_val=-9223372036854775000, max_val=9223372036854775000, special_cases=[])),
('c', IntegerGen())]

_grpkey_date_with_nulls = [
('a', RepeatSeqGen(int_gen, length=20)),
('b', DateGen(nullable=(True, 5.0), start=date(year=2020, month=1, day=1), end=date(year=2020, month=12, day=31))),
('c', IntegerGen())]

_grpkey_byte_with_nulls_with_overflow = [
('a', IntegerGen()),
Expand Down Expand Up @@ -158,41 +181,42 @@ def test_window_aggs_for_ranges_numeric_long_overflow(data_gen):
_grpkey_byte_with_nulls,
_grpkey_short_with_nulls,
_grpkey_int_with_nulls,
_grpkey_long_with_nulls
_grpkey_long_with_nulls,
_grpkey_date_with_nulls,
], ids=idfn)
def test_window_aggs_for_range_numeric(data_gen, batch_size):
def test_window_aggs_for_range_numeric_date(data_gen, batch_size):
conf = {'spark.rapids.sql.batchSizeBytes': batch_size,
'spark.rapids.sql.window.range.byte.enabled': True,
'spark.rapids.sql.window.range.short.enabled': True}
assert_gpu_and_cpu_are_equal_sql(
lambda spark: gen_df(spark, data_gen, length=2048),
"window_agg_table",
'window_agg_table',
'select '
' sum(b) over '
' sum(c) over '
' (partition by a order by b asc '
' range between 1 preceding and 3 following) as sum_c_asc, '
' avg(b) over '
' avg(c) over '
' (partition by a order by b asc '
' range between 1 preceding and 3 following) as avg_b_asc, '
' max(b) over '
' max(c) over '
' (partition by a order by b asc '
' range between 1 preceding and 3 following) as max_b_desc, '
' min(b) over '
' min(c) over '
' (partition by a order by b asc '
' range between 1 preceding and 3 following) as min_b_asc, '
' count(1) over '
' (partition by a order by b asc '
' range between CURRENT ROW and UNBOUNDED following) as count_1_asc, '
' count(b) over '
' count(c) over '
' (partition by a order by b asc '
' range between CURRENT ROW and UNBOUNDED following) as count_b_asc, '
' avg(b) over '
' avg(c) over '
' (partition by a order by b asc '
' range between UNBOUNDED preceding and CURRENT ROW) as avg_b_unbounded, '
' sum(b) over '
' sum(c) over '
' (partition by a order by b asc '
' range between UNBOUNDED preceding and CURRENT ROW) as sum_b_unbounded, '
' max(b) over '
' max(c) over '
' (partition by a order by b asc '
' range between UNBOUNDED preceding and UNBOUNDED following) as max_b_unbounded '
'from window_agg_table ',
Expand All @@ -202,8 +226,8 @@ def test_window_aggs_for_range_numeric(data_gen, batch_size):
@pytest.mark.parametrize('batch_size', ['1000', '1g'], ids=idfn) # set the batch size so we can test multiple stream batches
@pytest.mark.parametrize('data_gen', [_grpkey_longs_with_no_nulls,
_grpkey_longs_with_nulls,
_grpkey_longs_with_timestamps,
_grpkey_longs_with_nullable_timestamps,
_grpkey_longs_with_dates,
_grpkey_longs_with_nullable_dates,
_grpkey_longs_with_decimals,
_grpkey_longs_with_nullable_decimals,
_grpkey_decimals_with_nulls], ids=idfn)
Expand Down Expand Up @@ -348,62 +372,49 @@ def do_it(spark):


# Test for RANGE queries, with timestamp order-by expressions.
# Non-timestamp order-by columns are currently unsupported for RANGE queries.
# See https://github.com/NVIDIA/spark-rapids/issues/216
@ignore_order
@pytest.mark.parametrize('data_gen', [_grpkey_longs_with_timestamps,
pytest.param(_grpkey_longs_with_nullable_timestamps)],
ids=idfn)
def test_window_aggs_for_ranges(data_gen):
def test_window_aggs_for_ranges_timestamps(data_gen):
assert_gpu_and_cpu_are_equal_sql(
lambda spark: gen_df(spark, data_gen, length=2048),
"window_agg_table",
'select '
' sum(c) over '
' (partition by a order by cast(b as timestamp) asc '
' range between interval 1 day preceding and interval 1 day following) as sum_c_asc, '
' (partition by a order by b asc '
' range between interval 1 DAY 5 HOUR 3 MINUTE 2 SECOND 1 MILLISECOND 5 MICROSECOND preceding '
' and interval 1 DAY 5 HOUR 3 MINUTE 2 SECOND 1 MILLISECOND 5 MICROSECOND following) as sum_c_asc, '
' avg(c) over '
' (partition by a order by cast(b as timestamp) asc '
' range between interval 1 day preceding and interval 1 day following) as avg_c_asc, '
' (partition by a order by b asc '
' range between interval 1 DAY 5 HOUR 3 MINUTE 2 SECOND 1 MILLISECOND 5 MICROSECOND preceding '
' and interval 1 DAY 5 HOUR 3 MINUTE 2 SECOND 1 MILLISECOND 5 MICROSECOND following) as avg_c_asc, '
' max(c) over '
' (partition by a order by cast(b as timestamp) desc '
' range between interval 2 days preceding and interval 1 days following) as max_c_desc, '
' (partition by a order by b desc '
' range between interval 2 DAY 5 HOUR 3 MINUTE 2 SECOND 1 MILLISECOND 5 MICROSECOND preceding '
' and interval 1 DAY 5 HOUR 3 MINUTE 2 SECOND 1 MILLISECOND 5 MICROSECOND following) as max_c_desc, '
' min(c) over '
' (partition by a order by cast(b as timestamp) asc '
' range between interval 2 days preceding and current row) as min_c_asc, '
' (partition by a order by b asc '
' range between interval 2 DAY 5 HOUR 3 MINUTE 2 SECOND 1 MILLISECOND 5 MICROSECOND preceding '
' and current row) as min_c_asc, '
' count(1) over '
' (partition by a order by cast(b as timestamp) asc '
' (partition by a order by b asc '
' range between CURRENT ROW and UNBOUNDED following) as count_1_asc, '
' count(c) over '
' (partition by a order by cast(b as timestamp) asc '
' (partition by a order by b asc '
' range between CURRENT ROW and UNBOUNDED following) as count_c_asc, '
' avg(c) over '
' (partition by a order by cast(b as timestamp) asc '
' (partition by a order by b asc '
' range between UNBOUNDED preceding and CURRENT ROW) as avg_c_unbounded, '
' sum(c) over '
' (partition by a order by cast(b as timestamp) asc '
' (partition by a order by b asc '
' range between UNBOUNDED preceding and CURRENT ROW) as sum_c_unbounded, '
' max(c) over '
' (partition by a order by cast(b as timestamp) asc '
' (partition by a order by b asc '
' range between UNBOUNDED preceding and UNBOUNDED following) as max_c_unbounded '
'from window_agg_table',
conf = {'spark.rapids.sql.castFloatToDecimal.enabled': True})

@pytest.mark.xfail(reason="[UNSUPPORTED] Ranges over non-timestamp columns "
"(https://github.com/NVIDIA/spark-rapids/issues/216)")
@ignore_order
@pytest.mark.parametrize('data_gen', [_grpkey_longs_with_timestamps], ids=idfn)
def test_window_aggs_for_ranges_of_dates(data_gen):
assert_gpu_and_cpu_are_equal_sql(
lambda spark: gen_df(spark, data_gen, length=2048),
"window_agg_table",
'select '
' sum(c) over '
' (partition by a order by b asc '
' range between 1 preceding and 1 following) as sum_c_asc '
'from window_agg_table'
)

_gen_data_for_collect = [
('a', RepeatSeqGen(LongGen(), length=20)),
('b', IntegerGen()),
Expand Down
46 changes: 27 additions & 19 deletions sql-plugin/src/main/scala/com/nvidia/spark/rapids/GpuWindowExpression.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,8 @@

package com.nvidia.spark.rapids

import scala.concurrent.duration.{DAYS, Duration}
import java.util.concurrent.TimeUnit

import scala.language.{existentials, implicitConversions}

import ai.rapids.cudf.{Aggregation, AggregationOnColumn, ColumnVector, DType, RollingAggregation, Scalar, WindowOptions}
Expand Down Expand Up @@ -143,10 +144,10 @@ class GpuWindowExpressionMeta(
Seq(spec.lower, spec.upper).foreach {
case l @ Literal(_, ByteType | ShortType | IntegerType | LongType) =>
checkRangeBoundaryConfig(l.dataType)
case Literal(value, CalendarIntervalType) =>
val ci = value.asInstanceOf[CalendarInterval]
if (ci.months != 0 || ci.microseconds != 0) {
willNotWorkOnGpu("only days are supported for window range intervals")
case Literal(ci: CalendarInterval, CalendarIntervalType) =>
// interval is only working for TimeStampType
if (ci.months != 0) {
willNotWorkOnGpu("interval months isn't supported")
}
case UnboundedFollowing | UnboundedPreceding | CurrentRow =>
case anythings =>
Expand Down Expand Up @@ -416,15 +417,15 @@ object GpuWindowExpression extends Arm {
* @return Scalar holding boundary value
*/
def createRangeWindowBoundary(orderByType: DType, value: Long): Scalar = {
implicit def daysToDuration(days: Long) = Duration(value, DAYS)
orderByType match {
case DType.INT8 => Scalar.fromByte(value.toByte)
case DType.INT16 => Scalar.fromShort(value.toShort)
case DType.INT32 => Scalar.fromInt(value.toInt)
case DType.INT64 => Scalar.fromLong(value)
case DType.TIMESTAMP_DAYS => Scalar.durationFromLong(DType.DURATION_DAYS, value.toDays)
// Interval is not working for DateType
case DType.TIMESTAMP_DAYS => Scalar.durationFromLong(DType.DURATION_DAYS, value)
case DType.TIMESTAMP_MICROSECONDS =>
Scalar.durationFromLong(DType.DURATION_MICROSECONDS, value.toMicros)
Scalar.durationFromLong(DType.DURATION_MICROSECONDS, value)
case _ => throw new RuntimeException(s"Not supported order by type, Found $orderByType")
}
}
Expand All @@ -443,10 +444,10 @@ object GpuWindowExpression extends Arm {
val isUnBounded = special.isUnBounded
(isUnBounded, if (isUnBounded) None else orderByType.map(
createRangeWindowBoundary(_, special.value)))
case GpuLiteral(value, CalendarIntervalType) =>
// TimeStampDays -> DurationDays
var x = value.asInstanceOf[CalendarInterval].days
if (x == Int.MinValue) x = Int.MaxValue
case GpuLiteral(ci: CalendarInterval, CalendarIntervalType) =>
// Get the total microseconds for TIMESTAMP_MICROSECONDS
var x = ci.days * TimeUnit.DAYS.toMicros(1) + ci.microseconds
if (x == Long.MinValue) x = Long.MaxValue
(false, orderByType.map(createRangeWindowBoundary(_, Math.abs(x))))
case GpuLiteral(value, ByteType) =>
var x = value.asInstanceOf[Byte]
Expand Down Expand Up @@ -600,16 +601,23 @@ class GpuSpecifiedWindowFrameMeta(
case Literal(value, ShortType) => value.asInstanceOf[Short].toLong
case Literal(value, IntegerType) => value.asInstanceOf[Int].toLong
case Literal(value, LongType) => value.asInstanceOf[Long]
case Literal(value, CalendarIntervalType) =>
val interval = value.asInstanceOf[CalendarInterval]
if (interval.microseconds != 0 || interval.months != 0) { // DAYS == 0 is permitted.
willNotWorkOnGpu(s"Bounds for Range-based window frames must be specified" +
s" only in DAYS. Found $interval")
case Literal(ci: CalendarInterval, CalendarIntervalType) =>
if (ci.months != 0) {
willNotWorkOnGpu("interval months isn't supported")
}
// return the total microseconds
try {
Math.addExact(
Math.multiplyExact(ci.days.toLong, TimeUnit.DAYS.toMicros(1)),
ci.microseconds)
} catch {
case e: ArithmeticException =>
willNotWorkOnGpu("interval total microseconds has overflow")
Copy link
Collaborator

Choose a reason for hiding this comment

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

I would like to see a better message. It is not clear how the user can fix this or what exactly has happened.

Perhaps something more like

willNotWorkOnGpu(s"windows over timestamps are converted to microseconds and $ci is too large to fit")

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

Thx @revans2. Done

if (isLower) -1 else 1 // not check again
}
interval.days.toLong
case _ =>
willNotWorkOnGpu(s"Bounds for Range-based window frames must be specified in Integral" +
s" type (Boolean exclusive) or DAYS. Found ${bounds.dataType}")
s" type (Boolean exclusive) or CalendarInterval. Found ${bounds.dataType}")
if (isLower) -1 else 1 // not check again
}

Expand Down