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

Orc writes don't fully support Booleans with nulls #11763

Merged
merged 10 commits into from
Dec 7, 2024
Merged

Orc writes don't fully support Booleans with nulls #11763

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
022eb61
Initial change, fix one failure in orc write test
kuhushukla Nov 22, 2024
6d0e6f3
Merge branch 'branch-24.12' into issue_11736
kuhushukla Nov 25, 2024
625e2ab
Allow tests for structs sans boolean in orc writes test
kuhushukla Nov 25, 2024
507c75f
Fix test failures and address comments
kuhushukla Dec 2, 2024
eebf787
Fix tests and address review comments
kuhushukla Dec 3, 2024
b26e62c
Fix schema evolution test to avoid booleans for now
kuhushukla Dec 5, 2024
5d582ba
Address comments from reviews
kuhushukla Dec 6, 2024
6fa0db0
Merge remote-tracking branch 'refs/remotes/origin/branch-24.12' into …
kuhushukla Dec 6, 2024
0ced895
Address a missed review comment
kuhushukla Dec 6, 2024
2e2202b
Address missed review comments
kuhushukla Dec 6, 2024
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
37 changes: 30 additions & 7 deletions integration_tests/src/main/python/orc_write_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,10 +26,17 @@
pytestmark = pytest.mark.nightly_resource_consuming_test

jlowe marked this conversation as resolved.
Show resolved Hide resolved
orc_write_basic_gens = [byte_gen, short_gen, int_gen, long_gen, float_gen, double_gen,
string_gen, boolean_gen, DateGen(start=date(1590, 1, 1)),
string_gen, BooleanGen(nullable=False), DateGen(start=date(1590, 1, 1)),
Copy link
Collaborator

Choose a reason for hiding this comment

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

This is removing the test for nullable boolean values. Can we have an explicit test(s) that have a non-nullable struct with nullable values, or many different types, in it? I am fine if this is a follow on issue.

TimestampGen(start=datetime(1970, 1, 1, tzinfo=timezone.utc)) ] + \
decimal_gens
# Use every type except boolean , see https://github.com/NVIDIA/spark-rapids/issues/11762 and
# https://github.com/rapidsai/cudf/issues/6763 .
# Once the first issue is fixed, we can replace this list with
# orc_write_basic_gens
orc_write_basic_gens_for_structs = [byte_gen, short_gen, int_gen, long_gen, float_gen, double_gen,
string_gen, DateGen(start=date(1590, 1, 1)),
TimestampGen(start=datetime(1970, 1, 1, tzinfo=timezone.utc)) ] + \
decimal_gens

all_nulls_string_gen = SetValuesGen(StringType(), [None])
empty_or_null_string_gen = SetValuesGen(StringType(), [None, ""])
all_empty_string_gen = SetValuesGen(StringType(), [""])
Expand All @@ -52,7 +59,8 @@
all_nulls_map_gen,
all_empty_map_gen]

orc_write_basic_struct_gen = StructGen([['child'+str(ind), sub_gen] for ind, sub_gen in enumerate(orc_write_basic_gens)])
orc_write_basic_struct_gen = StructGen(
[['child'+str(ind), sub_gen] for ind, sub_gen in enumerate(orc_write_basic_gens_for_structs)])

orc_write_struct_gens_sample = [orc_write_basic_struct_gen,
StructGen([['child0', byte_gen], ['child1', orc_write_basic_struct_gen]]),
Expand All @@ -64,13 +72,14 @@
ArrayGen(StructGen([['child0', byte_gen], ['child1', string_gen], ['child2', float_gen]]))]

orc_write_basic_map_gens = [simple_string_to_string_map_gen] + [MapGen(f(nullable=False), f()) for f in [
jlowe marked this conversation as resolved.
Show resolved Hide resolved
BooleanGen, ByteGen, ShortGen, IntegerGen, LongGen, FloatGen, DoubleGen,
ByteGen, ShortGen, IntegerGen, LongGen, FloatGen, DoubleGen,
# Using timestamps from 1970 to work around a cudf ORC bug
# https://github.com/NVIDIA/spark-rapids/issues/140.
lambda nullable=True: TimestampGen(start=datetime(1970, 1, 1, tzinfo=timezone.utc), nullable=nullable),
lambda nullable=True: DateGen(start=date(1590, 1, 1), nullable=nullable),
lambda nullable=True: DecimalGen(precision=15, scale=1, nullable=nullable),
lambda nullable=True: DecimalGen(precision=36, scale=5, nullable=nullable)]]
lambda nullable=True: DecimalGen(precision=36, scale=5, nullable=nullable)]] + [MapGen(
f(nullable=False), f(nullable=False)) for f in [BooleanGen]]

orc_write_gens_list = [orc_write_basic_gens,
orc_write_struct_gens_sample,
Expand All @@ -79,6 +88,8 @@
pytest.param([date_gen], marks=pytest.mark.xfail(reason='https://github.com/NVIDIA/spark-rapids/issues/139')),
pytest.param([timestamp_gen], marks=pytest.mark.xfail(reason='https://github.com/NVIDIA/spark-rapids/issues/140'))]

nullable_bools_gen = [pytest.param([BooleanGen(nullable=True)],
marks=pytest.mark.allow_non_gpu('ExecutedCommandExec','DataWritingCommandExec'))]
@pytest.mark.parametrize('orc_gens', orc_write_gens_list, ids=idfn)
Copy link
Member

Choose a reason for hiding this comment

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

Suggested change
@pytest.mark.parametrize('orc_gens', orc_write_gens_list, ids=idfn)
@pytest.mark.parametrize('orc_gens', orc_write_gens_list, ids=idfn)

Copy link
Member

Choose a reason for hiding this comment

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

This is a nit that seemed to be missed. Would be nice to have whitespace separating the module variables from the methods.

@pytest.mark.parametrize('orc_impl', ["native", "hive"])
@allow_non_gpu(*non_utc_allow)
Expand All @@ -91,6 +102,18 @@ def test_write_round_trip(spark_tmp_path, orc_gens, orc_impl):
data_path,
conf={'spark.sql.orc.impl': orc_impl, 'spark.rapids.sql.format.orc.write.enabled': True})

@pytest.mark.parametrize('orc_gens', nullable_bools_gen, ids=idfn)
@pytest.mark.parametrize('orc_impl', ["native", "hive"])
@allow_non_gpu('ExecutedCommandExec', 'DataWritingCommandExec', 'WriteFilesExec')
def test_write_round_trip_null_bool(spark_tmp_path, orc_gens, orc_impl):
gen_list = [('_c' + str(i), gen) for i, gen in enumerate(orc_gens)]
data_path = spark_tmp_path + '/ORC_DATA'
assert_gpu_and_cpu_writes_are_equal_collect(
lambda spark, path: gen_df(spark, gen_list).coalesce(1).write.orc(path),
lambda spark, path: spark.read.orc(path),
data_path,
conf={'spark.sql.orc.impl': orc_impl, 'spark.rapids.sql.format.orc.write.enabled': True})

@pytest.mark.parametrize('orc_gen', orc_write_odd_empty_strings_gens_sample, ids=idfn)
@pytest.mark.parametrize('orc_impl', ["native", "hive"])
def test_write_round_trip_corner(spark_tmp_path, orc_gen, orc_impl):
Expand All @@ -103,7 +126,7 @@ def test_write_round_trip_corner(spark_tmp_path, orc_gen, orc_impl):
conf={'spark.sql.orc.impl': orc_impl, 'spark.rapids.sql.format.orc.write.enabled': True})

orc_part_write_gens = [
byte_gen, short_gen, int_gen, long_gen, boolean_gen,
byte_gen, short_gen, int_gen, long_gen, BooleanGen(nullable=False),
# Some file systems have issues with UTF8 strings so to help the test pass even there
StringGen('(\\w| ){0,50}'),
# Once https://github.com/NVIDIA/spark-rapids/issues/139 is fixed replace this with
Expand Down Expand Up @@ -345,7 +368,7 @@ def test_orc_write_column_name_with_dots(spark_tmp_path):
("f.g", int_gen),
("h", string_gen)])),
("i.j", long_gen)])),
("k", boolean_gen)]
("k", BooleanGen(nullable=False))]
assert_gpu_and_cpu_writes_are_equal_collect(
lambda spark, path: gen_df(spark, gens).coalesce(1).write.orc(path),
lambda spark, path: spark.read.orc(path),
Expand Down
10 changes: 10 additions & 0 deletions sql-plugin/src/main/scala/com/nvidia/spark/rapids/RapidsConf.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1243,6 +1243,14 @@ val GPU_COREDUMP_PIPE_PATTERN = conf("spark.rapids.gpu.coreDump.pipePattern")
.booleanConf
.createWithDefault(true)

val ENABLE_ORC_NULLABLE_BOOL = conf("spark.rapids.sql.format.orc.write.boolType.enabled")
Copy link
Collaborator

Choose a reason for hiding this comment

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

Can we just fall back for all booleans instead of only nullable ones? Spark already marks almost everything as nullable, so there is very little value in trying to distinguish between the two. But then I see things like #11762 where it scares me that CUDF might end up writing something out that they think is valid, but in practice is not.

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

ack. yes.

.doc("When set to false disables nullable boolean columns for ORC writes." +
"Set to true if your data does not have null booleans and want tp experiment" +
"See https://github.com/NVIDIA/spark-rapids/issues/11736.")
.internal()
.booleanConf
.createWithDefault(false)

val ENABLE_EXPAND_PREPROJECT = conf("spark.rapids.sql.expandPreproject.enabled")
.doc("When set to false disables the pre-projection for GPU Expand. " +
"Pre-projection leverages the tiered projection to evaluate expressions that " +
Expand Down Expand Up @@ -2964,6 +2972,8 @@ class RapidsConf(conf: Map[String, String]) extends Logging {

lazy val maxNumOrcFilesParallel: Int = get(ORC_MULTITHREAD_READ_MAX_NUM_FILES_PARALLEL)

lazy val isOrcBoolNullTypeEnabled: Boolean = get(ENABLE_ORC_NULLABLE_BOOL)

lazy val isCsvEnabled: Boolean = get(ENABLE_CSV)

lazy val isCsvReadEnabled: Boolean = get(ENABLE_CSV_READ)
Expand Down
31 changes: 31 additions & 0 deletions sql-plugin/src/main/scala/org/apache/spark/sql/rapids/GpuOrcFileFormat.scala
View file
Open in desktop
jlowe marked this conversation as resolved.
Show resolved Hide resolved
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,27 @@ object GpuOrcFileFormat extends Logging {
cls == classOf[OrcFileFormat] || cls.getCanonicalName.equals(HIVE_IMPL_CLASS)
}

private def checkForBoolNulls(dataType: DataType): Boolean = {
dataType match {
case ArrayType(elementType, t) => elementType == BooleanType && t
case StructType(fields) =>
fields.exists { f =>
hasBoolNulls(f.dataType, f.nullable)
}
case MapType(_, valueType, t) => hasBoolNulls(valueType, t)
}
}

private def hasBoolNulls(d: DataType, nulls: Boolean) = {
if (nulls && d == BooleanType) {
true
} else if (DataTypeUtils.isNestedType(d)) {
checkForBoolNulls(d)
} else {
false
}
}

def tagGpuSupport(meta: RapidsMeta[_, _, _],
spark: SparkSession,
options: Map[String, String],
Expand Down Expand Up @@ -83,6 +104,12 @@ object GpuOrcFileFormat extends Logging {
// [[org.apache.spark.sql.execution.datasources.DaysWritable]] object
// which is a subclass of [[org.apache.hadoop.hive.serde2.io.DateWritable]].
val types = schema.map(_.dataType).toSet
val res = schema.exists {
case field if field.dataType == BooleanType && field.nullable => true
case field if DataTypeUtils.isNestedType(field.dataType) => checkForBoolNulls(field.dataType)
case _ => false
}

if (types.exists(GpuOverrides.isOrContainsDateOrTimestamp(_))) {
if (!GpuOverrides.isUTCTimezone()) {
meta.willNotWorkOnGpu("Only UTC timezone is supported for ORC. " +
Expand All @@ -91,6 +118,10 @@ object GpuOrcFileFormat extends Logging {
}
}

if (res && !meta.conf.isOrcBoolNullTypeEnabled) {
meta.willNotWorkOnGpu("Nullable Booleans can not work in certain cases with ORC writer." +
"See https://github.com/rapidsai/cudf/issues/6763")
}
FileFormatChecks.tag(meta, schema, OrcFormatType, WriteFileOp)

val sqlConf = spark.sessionState.conf
Expand Down