[GitHub] [flink] qiuxiafei commented on a change in pull request #9373: [FLINK-13596][ml] Add two utils for Table transformations

[GitBox <gi...@apache.org>]

qiuxiafei commented on a change in pull request #9373: [FLINK-13596][ml] Add two utils for Table transformations
URL: https://github.com/apache/flink/pull/9373#discussion_r344544335
 
 

 ##########
 File path: flink-ml-parent/flink-ml-lib/src/main/java/org/apache/flink/ml/common/utils/DataSetConversionUtil.java
 ##########
 @@ -0,0 +1,171 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+package org.apache.flink.ml.common.utils;
+
+import org.apache.flink.api.common.functions.MapFunction;
+import org.apache.flink.api.common.typeinfo.TypeInformation;
+import org.apache.flink.api.java.DataSet;
+import org.apache.flink.api.java.operators.SingleInputUdfOperator;
+import org.apache.flink.api.java.operators.TwoInputUdfOperator;
+import org.apache.flink.api.java.typeutils.RowTypeInfo;
+import org.apache.flink.ml.common.MLEnvironment;
+import org.apache.flink.ml.common.MLEnvironmentFactory;
+import org.apache.flink.table.api.Table;
+import org.apache.flink.table.api.TableSchema;
+import org.apache.flink.types.Row;
+
+/**
+ * Provide functions of conversions between DataSet and Table.
+ */
+public class DataSetConversionUtil {
+	/**
+	 * Convert the given Table to {@link DataSet}<{@link Row}>.
+	 *
+	 * @param sessionId the sessionId of {@link MLEnvironmentFactory}
+	 * @param table the Table to convert.
+	 * @return the converted DataSet.
+	 */
+	public static DataSet <Row> fromTable(Long sessionId, Table table) {
+		return MLEnvironmentFactory
+			.get(sessionId)
+			.getBatchTableEnvironment()
+			.toDataSet(table, Row.class);
+	}
+
+	/**
+	 * Convert the given DataSet into a Table with specified TableSchema.
+	 *
+	 * @param sessionId the sessionId of {@link MLEnvironmentFactory}
+	 * @param data   the DataSet to convert.
+	 * @param schema the specified TableSchema.
+	 * @return the converted Table.
+	 */
+	public static Table toTable(Long sessionId, DataSet <Row> data, TableSchema schema) {
+		return toTable(sessionId, data, schema.getFieldNames(), schema.getFieldTypes());
+	}
+
+	/**
+	 * Convert the given DataSet into a Table with specified colNames and colTypes.
+	 *
+	 * @param sessionId sessionId the sessionId of {@link MLEnvironmentFactory}.
+	 * @param data     the DataSet to convert.
+	 * @param colNames the specified colNames.
+	 * @param colTypes the specified colTypes. This variable is used only when the
+	 *                 DataSet is produced by a function and Flink cannot determine
+	 *                 automatically what the produced type is.
+	 * @return the converted Table.
+	 */
+	public static Table toTable(Long sessionId, DataSet <Row> data, String[] colNames, TypeInformation <?>[] colTypes) {
+		return toTable(MLEnvironmentFactory.get(sessionId), data, colNames, colTypes);
+	}
+
+	/**
+	 * Convert the given DataSet into a Table with specified colNames.
+	 *
+	 * @param sessionId sessionId the sessionId of {@link MLEnvironmentFactory}.
+	 * @param data     the DataSet to convert.
+	 * @param colNames the specified colNames.
+	 * @return the converted Table.
+	 */
+	public static Table toTable(Long sessionId, DataSet <Row> data, String[] colNames) {
+		return toTable(MLEnvironmentFactory.get(sessionId), data, colNames);
+	}
+
+	/**
+	 * Convert the given DataSet into a Table with specified colNames and colTypes.
+	 *
+	 * @param session the MLEnvironment using to convert DataSet to Table.
+	 * @param data     the DataSet to convert.
+	 * @param colNames the specified colNames.
+	 * @param colTypes the specified colTypes. This variable is used only when the
+	 *                 DataSet is produced by a function and Flink cannot determine
+	 *                 automatically what the produced type is.
+	 * @return the converted Table.
+	 */
+	public static Table toTable(MLEnvironment session, DataSet <Row> data, String[] colNames, TypeInformation <?>[] colTypes) {
+		try {
+			// Try to add row type information for the dataset to be converted.
+			// In most case, this keeps us from the rolling back logic in the catch block,
+			// which adds an unnecessary map function just in order to add row type information.
+			if (data instanceof SingleInputUdfOperator) {
+				((SingleInputUdfOperator) data).returns(new RowTypeInfo(colTypes, colNames));
+			} else if (data instanceof TwoInputUdfOperator) {
+				((TwoInputUdfOperator) data).returns(new RowTypeInfo(colTypes, colNames));
+			}
+			return toTable(session, data, colNames);
+		} catch (Exception ex) {
 
 Review comment:
   The actual exception thrown here is this `ValidationException`:
   https://github.com/apache/flink/blob/ef4055ab989db431e29e0071573619d267960000/flink-table/flink-table-api-java/src/main/java/org/apache/flink/table/typeutils/FieldInfoUtils.java#L184-L189
   That's because sometimes the type information of Row bassed dataset is not set.
   
   As you mentioned, our initial solution is doing some pre-check, saying checking if the `TypeInformation` is `GenericTypeInfo<Row>`. To check the type, we need to `getType()`. As a side effect, `typeUsed` is set to `true`:
   
   https://github.com/apache/flink/blob/937356f786b5edac28a7ff44fc31410761c6d7f0/flink-java/src/main/java/org/apache/flink/api/java/DataSet.java#L171-L182
   
    But what if the input dataset really has no specified type information? We have to set it's TypeInformation via the `.return()` methods, which calls the following code, which says "no modification after getType called":
   https://github.com/apache/flink/blob/937356f786b5edac28a7ff44fc31410761c6d7f0/flink-java/src/main/java/org/apache/flink/api/java/DataSet.java#L155-L161
   
   That's the deadlock we faced. So we just try the best to complete the type information (for `SingleInputUdfOperator` and `TwoInputUdfOperator` here) to avoid falling into the exception branch and let the `getDataSetWithExplicitTypeDefine` method work as a final guard. 
   
   BTW, catching `ValidationException` instead of a generic `Exception` is a good idea I think.
   

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services