Merge pull request NVIDIA#1946 from NvTimLiu/release-tmp

update 0.4.0 with cudf 0.18.1

CHANGELOG.md

@@ -1,5 +1,5 @@
 # Change log
-Generated on 2021-03-02
+Generated on 2021-03-16
 
 ## Release 0.4
 
@@ -104,6 +104,12 @@ Generated on 2021-03-02
 ### PRs
 |||
 |:---|:---|
+|[#1910](https://github.com/NVIDIA/spark-rapids/pull/1910)|Make hash partitioning match CPU|
+|[#1927](https://github.com/NVIDIA/spark-rapids/pull/1927)|Change cuDF dependency to 0.18.1|
+|[#1934](https://github.com/NVIDIA/spark-rapids/pull/1934)|Update documentation to use cudf version 0.18.1|
+|[#1871](https://github.com/NVIDIA/spark-rapids/pull/1871)|Disable coalesce batch spilling to avoid cudf contiguous_split bug|
+|[#1849](https://github.com/NVIDIA/spark-rapids/pull/1849)|Update changelog for 0.4|
+|[#1744](https://github.com/NVIDIA/spark-rapids/pull/1744)|Fix NullPointerException on null partition insert|
 |[#1842](https://github.com/NVIDIA/spark-rapids/pull/1842)|Update to note support for 3.0.2|
 |[#1832](https://github.com/NVIDIA/spark-rapids/pull/1832)|Spark 3.1.1 shim no longer a snapshot shim|
 |[#1831](https://github.com/NVIDIA/spark-rapids/pull/1831)|Spark 3.0.2 shim no longer a snapshot shim|

api_validation/README.md

@@ -17,7 +17,7 @@ It requires cudf, rapids-4-spark and spark jars.
 
 ```
 cd api_validation
-// To run validation script on all version of Spark(3.0.0, 3.0.1 and 3.1.0-SNAPSHOT)
+// To run validation script on all version of Spark(3.0.0, 3.0.1 and 3.1.1)
 sh auditAllVersions.sh
 
 // To run script on particular version we can use profile(spark300, spark301 and spark311)

docs/additional-functionality/cache-serializer.md

@@ -15,7 +15,7 @@ nav_order: 2
   utilize disk space to spill over. To read more about what storage levels are available look
   at `StorageLevel.scala` in Spark.
 
-  Starting in Spark 3.1.0 users can add their own cache serializer, if they desire, by
+  Starting in Spark 3.1.1 users can add their own cache serializer, if they desire, by
   setting the `spark.sql.cache.serializer` configuration. This is a static configuration
   that is set once for the duration of a Spark application which means that you can only set the conf
   before starting a Spark application and cannot be changed for that application's Spark

docs/configs.md

@@ -10,7 +10,7 @@ The following is the list of options that `rapids-plugin-4-spark` supports.
 On startup use: `--conf [conf key]=[conf value]`. For example:
 
 ```
-${SPARK_HOME}/bin/spark --jars 'rapids-4-spark_2.12-0.4.0.jar,cudf-0.18-cuda10-1.jar' \
+${SPARK_HOME}/bin/spark --jars 'rapids-4-spark_2.12-0.4.0.jar,cudf-0.18.1-cuda10-1.jar' \
 --conf spark.plugins=com.nvidia.spark.SQLPlugin \
 --conf spark.rapids.sql.incompatibleOps.enabled=true
 ```

docs/demo/Databricks/generate-init-script.ipynb

@@ -1 +1 @@
-{"cells":[{"cell_type":"code","source":["dbutils.fs.mkdirs(\"dbfs:/databricks/init_scripts/\")\n \ndbutils.fs.put(\"/databricks/init_scripts/init.sh\",\"\"\"\n#!/bin/bash\nsudo wget -O /databricks/jars/rapids-4-spark_2.12-0.4.0.jar https://oss.sonatype.org/content/repositories/staging/com/nvidia/rapids-4-spark_2.12/0.4.0/rapids-4-spark_2.12-0.4.0.jar\nsudo wget -O /databricks/jars/cudf-0.18-cuda10-1.jar https://oss.sonatype.org/content/repositories/staging/ai/rapids/cudf/0.18/cudf-0.18-cuda10-1.jar\"\"\", True)"],"metadata":{},"outputs":[],"execution_count":1},{"cell_type":"code","source":["%sh\ncd ../../dbfs/databricks/init_scripts\npwd\nls -ltr\ncat init.sh"],"metadata":{},"outputs":[],"execution_count":2},{"cell_type":"code","source":[""],"metadata":{},"outputs":[],"execution_count":3}],"metadata":{"name":"generate-init-script","notebookId":2645746662301564},"nbformat":4,"nbformat_minor":0}
+{"cells":[{"cell_type":"code","source":["dbutils.fs.mkdirs(\"dbfs:/databricks/init_scripts/\")\n \ndbutils.fs.put(\"/databricks/init_scripts/init.sh\",\"\"\"\n#!/bin/bash\nsudo wget -O /databricks/jars/rapids-4-spark_2.12-0.4.0.jar https://oss.sonatype.org/content/repositories/staging/com/nvidia/rapids-4-spark_2.12/0.4.0/rapids-4-spark_2.12-0.4.0.jar\nsudo wget -O /databricks/jars/cudf-0.18.1-cuda10-1.jar https://oss.sonatype.org/content/repositories/staging/ai/rapids/cudf/0.18.1/cudf-0.18.1-cuda10-1.jar\"\"\", True)"],"metadata":{},"outputs":[],"execution_count":1},{"cell_type":"code","source":["%sh\ncd ../../dbfs/databricks/init_scripts\npwd\nls -ltr\ncat init.sh"],"metadata":{},"outputs":[],"execution_count":2},{"cell_type":"code","source":[""],"metadata":{},"outputs":[],"execution_count":3}],"metadata":{"name":"generate-init-script","notebookId":2645746662301564},"nbformat":4,"nbformat_minor":0}

docs/download.md

@@ -47,9 +47,9 @@ Software Requirements:
 
 ### Download v0.4.0
 * [RAPIDS Spark Package](https://repo1.maven.org/maven2/com/nvidia/rapids-4-spark_2.12/0.4.0/rapids-4-spark_2.12-0.4.0.jar)
-* [cuDF 11.0 Package](https://repo1.maven.org/maven2/ai/rapids/cudf/0.18/cudf-0.18-cuda11.jar)
-* [cuDF 10.2 Package](https://repo1.maven.org/maven2/ai/rapids/cudf/0.18/cudf-0.18-cuda10-2.jar)
-* [cuDF 10.1 Package](https://repo1.maven.org/maven2/ai/rapids/cudf/0.18/cudf-0.18-cuda10-1.jar)
+* [cuDF 11.0 Package](https://repo1.maven.org/maven2/ai/rapids/cudf/0.18.1/cudf-0.18.1-cuda11.jar)
+* [cuDF 10.2 Package](https://repo1.maven.org/maven2/ai/rapids/cudf/0.18.1/cudf-0.18.1-cuda10-2.jar)
+* [cuDF 10.1 Package](https://repo1.maven.org/maven2/ai/rapids/cudf/0.18.1/cudf-0.18.1-cuda10-1.jar)
 
 ## Release v0.3.0
 This release includes additional performance improvements, including

docs/get-started/Dockerfile.cuda

@@ -52,7 +52,7 @@ COPY spark-3.0.1-bin-hadoop3.2/examples /opt/spark/examples
 COPY spark-3.0.1-bin-hadoop3.2/kubernetes/tests /opt/spark/tests
 COPY spark-3.0.1-bin-hadoop3.2/data /opt/spark/data
 
-COPY cudf-0.18-cuda10-1.jar /opt/sparkRapidsPlugin
+COPY cudf-0.18.1-cuda10-1.jar /opt/sparkRapidsPlugin
 COPY rapids-4-spark_2.12-0.4.0.jar /opt/sparkRapidsPlugin
 COPY getGpusResources.sh /opt/sparkRapidsPlugin
 

docs/get-started/getting-started-on-prem.md

@@ -55,15 +55,15 @@ CUDA and will not run on other versions. The jars use a maven classifier to keep
 - CUDA 11.0 => classifier cuda11
 
 For example, here is a sample version of the jars and cudf with CUDA 10.1 support:
-- cudf-0.18-cuda10-1.jar
+- cudf-0.18.1-cuda10-1.jar
 - rapids-4-spark_2.12-0.4.0.jar
 
 
 For simplicity export the location to these jars. This example assumes the sample jars above have
 been placed in the `/opt/sparkRapidsPlugin` directory:
 ```shell 
 export SPARK_RAPIDS_DIR=/opt/sparkRapidsPlugin
-export SPARK_CUDF_JAR=${SPARK_RAPIDS_DIR}/cudf-0.18-cuda10-1.jar
+export SPARK_CUDF_JAR=${SPARK_RAPIDS_DIR}/cudf-0.18.1-cuda10-1.jar
 export SPARK_RAPIDS_PLUGIN_JAR=${SPARK_RAPIDS_DIR}/rapids-4-spark_2.12-0.4.0.jar
 ```
 

docs/supported_ops.md

@@ -9540,19 +9540,19 @@ Accelerator support is described below.
 <td>S</td>
 <td>S</td>
 <td>S</td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
 <td>S</td>
+<td>S*</td>
 <td>S</td>
-<td> </td>
-<td> </td>
-<td>S</td>
-<td> </td>
-<td>S</td>
-<td> </td>
-<td> </td>
-<td> </td>
-<td> </td>
-<td> </td>
-<td> </td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
 </tr>
 <tr>
 <td>result</td>
@@ -9585,17 +9585,17 @@ Accelerator support is described below.
 <td><b>NS</b></td>
 <td><b>NS</b></td>
 <td><b>NS</b></td>
-<td> </td>
-<td> </td>
 <td><b>NS</b></td>
-<td> </td>
 <td><b>NS</b></td>
-<td> </td>
-<td> </td>
-<td> </td>
-<td> </td>
-<td> </td>
-<td> </td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
+<td><b>NS</b></td>
 </tr>
 <tr>
 <td>result</td>

integration_tests/README.md

@@ -131,7 +131,7 @@ If you just want to verify the SQL replacement is working you will need to add t
 example assumes CUDA 10.1 is being used.
 
 ```
-$SPARK_HOME/bin/spark-submit --jars "rapids-4-spark_2.12-0.4.0.jar,rapids-4-spark-udf-examples_2.12-0.4.0.jar,cudf-0.18-cuda10-1.jar" ./runtests.py
+$SPARK_HOME/bin/spark-submit --jars "rapids-4-spark_2.12-0.4.0.jar,rapids-4-spark-udf-examples_2.12-0.4.0.jar,cudf-0.18.1-cuda10-1.jar" ./runtests.py
 ```
 
 You don't have to enable the plugin for this to work, the test framework will do that for you.
@@ -192,7 +192,7 @@ To run cudf_udf tests, need following configuration changes:
 As an example, here is the `spark-submit` command with the cudf_udf parameter on CUDA 10.1:
 
 ```
-$SPARK_HOME/bin/spark-submit --jars "rapids-4-spark_2.12-0.4.0.jar,rapids-4-spark-udf-examples_2.12-0.4.0.jar,cudf-0.18-cuda10-1.jar,rapids-4-spark-tests_2.12-0.4.0.jar" --conf spark.rapids.memory.gpu.allocFraction=0.3 --conf spark.rapids.python.memory.gpu.allocFraction=0.3 --conf spark.rapids.python.concurrentPythonWorkers=2 --py-files "rapids-4-spark_2.12-0.4.0.jar" --conf spark.executorEnv.PYTHONPATH="rapids-4-spark_2.12-0.4.0.jar" ./runtests.py --cudf_udf
+$SPARK_HOME/bin/spark-submit --jars "rapids-4-spark_2.12-0.4.0.jar,rapids-4-spark-udf-examples_2.12-0.4.0.jar,cudf-0.18.1-cuda10-1.jar,rapids-4-spark-tests_2.12-0.4.0.jar" --conf spark.rapids.memory.gpu.allocFraction=0.3 --conf spark.rapids.python.memory.gpu.allocFraction=0.3 --conf spark.rapids.python.concurrentPythonWorkers=2 --py-files "rapids-4-spark_2.12-0.4.0.jar" --conf spark.executorEnv.PYTHONPATH="rapids-4-spark_2.12-0.4.0.jar" ./runtests.py --cudf_udf
 ```
 
 ## Writing tests

integration_tests/src/main/python/data_gen.py

@@ -276,6 +276,21 @@ def start(self, rand):
         self._start(rand, self._loop_values)
         self._vals = [self._child.gen() for _ in range(0, self._length)]
 
+class SetValuesGen(DataGen):
+    """A set of values that are randomly selected"""
+    def __init__(self, data_type, data):
+        super().__init__(data_type, nullable=False)
+        self.nullable = any(x is None for x in data)
+        self._vals = data
+
+    def __repr__(self):
+        return super().__repr__() + '(' + str(self._child) + ')'
+
+    def start(self, rand):
+        data = self._vals
+        length = len(data)
+        self._start(rand, lambda : data[rand.randrange(0, length)])
+
 FLOAT_MIN = -3.4028235E38
 FLOAT_MAX = 3.4028235E38
 NEG_FLOAT_NAN_MIN_VALUE = struct.unpack('f', struct.pack('I', 0xffffffff))[0]

integration_tests/src/main/python/join_test.py

@@ -201,3 +201,40 @@ def do_join(spark):
                 return testurls.join(resolved, "Url", "inner")
     assert_gpu_and_cpu_are_equal_collect(do_join, conf={'spark.sql.autoBroadcastJoinThreshold': '-1'})
 
+@pytest.mark.parametrize('join_type', ['Left', 'Right', 'Inner', 'LeftSemi', 'LeftAnti'], ids=idfn)
+@pytest.mark.parametrize('cache_side', ['cache_left', 'cache_right'], ids=idfn)
+@pytest.mark.parametrize('cpu_side', ['cache', 'not_cache'], ids=idfn)
+@ignore_order
+def test_half_cache_join(join_type, cache_side, cpu_side):
+    left_gen = [('a', SetValuesGen(LongType(), range(500))), ('b', IntegerGen())]
+    right_gen = [('r_a', SetValuesGen(LongType(), range(500))), ('c', LongGen())]
+    def do_join(spark):
+        # Try to force the shuffle to be split between CPU and GPU for the join
+        # so don't let the shuffle be on the GPU/CPU depending on how the test is configured
+        # when we repartition and cache the data
+        spark.conf.set('spark.rapids.sql.exec.ShuffleExchangeExec', cpu_side != 'cache')
+        left = gen_df(spark, left_gen, length=500)
+        right = gen_df(spark, right_gen, length=500)
+
+        if (cache_side == 'cache_left'):
+            # Try to force the shuffle to be split between CPU and GPU for the join
+            # by default if the operation after the shuffle is not on the GPU then 
+            # don't do a GPU shuffle, so do something simple after the repartition
+            # to make sure that the GPU shuffle is used.
+            left = left.repartition('a').selectExpr('b + 1 as b', 'a').cache()
+            left.count() # populate the cache
+        else:
+            #cache_right
+            # Try to force the shuffle to be split between CPU and GPU for the join
+            # by default if the operation after the shuffle is not on the GPU then 
+            # don't do a GPU shuffle, so do something simple after the repartition
+            # to make sure that the GPU shuffle is used.
+            right = right.repartition('r_a').selectExpr('c + 1 as c', 'r_a').cache()
+            right.count() # populate the cache
+        # Now turn it back so the other half of the shuffle will be on the oposite side
+        spark.conf.set('spark.rapids.sql.exec.ShuffleExchangeExec', cpu_side == 'cache')
+        return left.join(right, left.a == right.r_a, join_type)
+
+    # Even though Spark does not know the size of an RDD input so it will not do a broadcast join unless
+    # we tell it to, this is just to be safe
+    assert_gpu_and_cpu_are_equal_collect(do_join, {'spark.sql.autoBroadcastJoinThreshold': '1'})

integration_tests/src/main/python/repart_test.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2020, NVIDIA CORPORATION.
+# Copyright (c) 2020-2021, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -17,6 +17,7 @@
 from asserts import assert_gpu_and_cpu_are_equal_collect
 from data_gen import *
 from marks import ignore_order
+import pyspark.sql.functions as f
 
 @pytest.mark.parametrize('data_gen', all_gen, ids=idfn)
 def test_union(data_gen):
@@ -45,3 +46,39 @@ def test_repartion_df(num_parts, length):
     assert_gpu_and_cpu_are_equal_collect(
             lambda spark : gen_df(spark, gen_list, length=length).repartition(num_parts),
             conf = allow_negative_scale_of_decimal_conf)
+
+@ignore_order(local=True) # To avoid extra data shuffle by 'sort on Spark' for this repartition test.
+@pytest.mark.parametrize('num_parts', [1, 2, 10, 17, 19, 32], ids=idfn)
+@pytest.mark.parametrize('gen', [
+    ([('a', boolean_gen)], ['a']), 
+    ([('a', byte_gen)], ['a']), 
+    ([('a', short_gen)], ['a']),
+    ([('a', int_gen)], ['a']),
+    ([('a', long_gen)], ['a']),
+    pytest.param(([('a', float_gen)], ['a']), marks=pytest.mark.xfail(reason='https://github.com/NVIDIA/spark-rapids/issues/1914')),
+    pytest.param(([('a', double_gen)], ['a']), marks=pytest.mark.xfail(reason='https://github.com/NVIDIA/spark-rapids/issues/1914')),
+    ([('a', decimal_gen_default)], ['a']),
+    ([('a', decimal_gen_neg_scale)], ['a']),
+    ([('a', decimal_gen_scale_precision)], ['a']),
+    ([('a', decimal_gen_same_scale_precision)], ['a']),
+    ([('a', decimal_gen_64bit)], ['a']),
+    ([('a', string_gen)], ['a']),
+    ([('a', null_gen)], ['a']),
+    ([('a', byte_gen)], [f.col('a') - 5]), 
+    ([('a', long_gen)], [f.col('a') + 15]), 
+    ([('a', byte_gen), ('b', boolean_gen)], ['a', 'b']),
+    ([('a', short_gen), ('b', string_gen)], ['a', 'b']),
+    ([('a', int_gen), ('b', byte_gen)], ['a', 'b']),
+    ([('a', long_gen), ('b', null_gen)], ['a', 'b']),
+    ([('a', byte_gen), ('b', boolean_gen), ('c', short_gen)], ['a', 'b', 'c']),
+    ([('a', short_gen), ('b', string_gen), ('c', int_gen)], ['a', 'b', 'c']),
+    ([('a', decimal_gen_default), ('b', decimal_gen_64bit), ('c', decimal_gen_scale_precision)], ['a', 'b', 'c']),
+    ], ids=idfn)
+def test_hash_repartition_exact(gen, num_parts):
+    data_gen = gen[0]
+    part_on = gen[1]
+    assert_gpu_and_cpu_are_equal_collect(
+            lambda spark : gen_df(spark, data_gen)\
+                    .repartition(num_parts, *part_on)\
+                    .selectExpr('spark_partition_id() as id', '*', 'hash(*)', 'pmod(hash(*),{})'.format(num_parts)),
+            conf = allow_negative_scale_of_decimal_conf)

integration_tests/src/test/scala/com/nvidia/spark/rapids/JoinsSuite.scala

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020, NVIDIA CORPORATION.
+ * Copyright (c) 2019-2021, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -101,65 +101,4 @@ class JoinsSuite extends SparkQueryCompareTestSuite {
     mixedDfWithNulls, mixedDfWithNulls, sortBeforeRepart = true) {
     (A, B) => A.join(B, A("longs") === B("longs"), "LeftAnti")
   }
-
-  test("fixUpJoinConsistencyIfNeeded AQE on") {
-    // this test is only valid in Spark 3.0.1 and later due to AQE supporting the plugin
-    val isValidTestForSparkVersion = ShimLoader.getSparkShims.getSparkShimVersion match {
-      case SparkShimVersion(3, 0, 0) => false
-      case DatabricksShimVersion(3, 0, 0) => false
-      case _ => true
-    }
-    assume(isValidTestForSparkVersion)
-    testFixUpJoinConsistencyIfNeeded(true)
-  }
-
-  test("fixUpJoinConsistencyIfNeeded AQE off") {
-    testFixUpJoinConsistencyIfNeeded(false)
-  }
-
-  private def testFixUpJoinConsistencyIfNeeded(aqe: Boolean) {
-
-    val conf = shuffledJoinConf.clone()
-        .set("spark.sql.adaptive.enabled", String.valueOf(aqe))
-        .set("spark.rapids.sql.test.allowedNonGpu",
-          "BroadcastHashJoinExec,SortMergeJoinExec,SortExec,Upper")
-        .set("spark.rapids.sql.incompatibleOps.enabled", "false") // force UPPER onto CPU
-
-    withGpuSparkSession(spark => {
-      import spark.implicits._
-
-      def createStringDF(name: String, upper: Boolean = false): DataFrame = {
-        val countryNames = (0 until 1000).map(i => s"country_$i")
-        if (upper) {
-          countryNames.map(_.toUpperCase).toDF(name)
-        } else {
-          countryNames.toDF(name)
-        }
-      }
-
-      val left = createStringDF("c1")
-          .join(createStringDF("c2"), col("c1") === col("c2"))
-
-      val right = createStringDF("c3")
-          .join(createStringDF("c4"), col("c3") === col("c4"))
-
-      val join = left.join(right, upper(col("c1")) === col("c4"))
-
-      // call collect so that we get the final executed plan when AQE is on
-      join.collect()
-
-      val shuffleExec = TestUtils
-          .findOperator(join.queryExecution.executedPlan, _.isInstanceOf[ShuffleExchangeExec])
-          .get
-
-      val gpuSupportedTag = TreeNodeTag[Set[String]]("rapids.gpu.supported")
-      val reasons = shuffleExec.getTagValue(gpuSupportedTag).getOrElse(Set.empty)
-      assert(reasons.contains(
-          "other exchanges that feed the same join are on the CPU, and GPU " +
-          "hashing is not consistent with the CPU version"))
-
-    }, conf)
-
-  }
-
 }

integration_tests/src/test/scala/com/nvidia/spark/rapids/UnaryOperatorsSuite.scala

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020, NVIDIA CORPORATION.
+ * Copyright (c) 2019-2021, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -37,6 +37,6 @@ class UnaryOperatorsSuite extends SparkQueryCompareTestSuite {
   }
 
   testSparkResultsAreEqual("Test murmur3", mixedDfWithNulls) {
-    frame => frame.selectExpr("hash(longs, doubles, 1, null, 'stock string', ints, strings)")
+    frame => frame.selectExpr("hash(longs, 1, null, 'stock string', ints, strings)")
   }
 }

jenkins/version-def.sh

@@ -26,7 +26,7 @@ for VAR in $OVERWRITE_PARAMS;do
 done
 IFS=$PRE_IFS
 
-CUDF_VER=${CUDF_VER:-"0.18"}
+CUDF_VER=${CUDF_VER:-"0.18.1"}
 CUDA_CLASSIFIER=${CUDA_CLASSIFIER:-"cuda10-1"}
 PROJECT_VER=${PROJECT_VER:-"0.4.0"}
 PROJECT_TEST_VER=${PROJECT_TEST_VER:-"0.4.0"}

pom.xml

@@ -147,7 +147,7 @@
         <maven.compiler.target>1.8</maven.compiler.target>
         <spark.version>${spark300.version}</spark.version>
         <cuda.version>cuda10-1</cuda.version>
-        <cudf.version>0.18</cudf.version>
+        <cudf.version>0.18.1</cudf.version>
         <scala.binary.version>2.12</scala.binary.version>
         <scala.version>2.12.8</scala.version>
         <orc.version>1.5.8</orc.version>