[GitHub] [iceberg] aokolnychyi commented on a diff in pull request #6012: Spark 3.3: Add a procedure to generate table changes

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

aokolnychyi commented on code in PR #6012:
URL: https://github.com/apache/iceberg/pull/6012#discussion_r1018710056


##########
spark/v3.3/spark/src/main/java/org/apache/iceberg/spark/procedures/GenerateChangesProcedure.java:
##########
@@ -0,0 +1,271 @@
+/*
+ * 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.iceberg.spark.procedures;
+
+import java.util.Arrays;
+import java.util.UUID;
+import org.apache.iceberg.ChangelogOperation;
+import org.apache.iceberg.MetadataColumns;
+import org.apache.iceberg.Snapshot;
+import org.apache.iceberg.Table;
+import org.apache.iceberg.spark.SparkReadOptions;
+import org.apache.iceberg.spark.source.SparkChangelogTable;
+import org.apache.iceberg.util.DateTimeUtil;
+import org.apache.iceberg.util.SnapshotUtil;
+import org.apache.spark.sql.Column;
+import org.apache.spark.sql.DataFrameReader;
+import org.apache.spark.sql.Dataset;
+import org.apache.spark.sql.Row;
+import org.apache.spark.sql.catalyst.InternalRow;
+import org.apache.spark.sql.connector.catalog.Identifier;
+import org.apache.spark.sql.connector.catalog.TableCatalog;
+import org.apache.spark.sql.connector.iceberg.catalog.ProcedureParameter;
+import org.apache.spark.sql.expressions.Window;
+import org.apache.spark.sql.functions;
+import org.apache.spark.sql.types.DataTypes;
+import org.apache.spark.sql.types.Metadata;
+import org.apache.spark.sql.types.StructField;
+import org.apache.spark.sql.types.StructType;
+import org.apache.spark.unsafe.types.UTF8String;
+import org.jetbrains.annotations.NotNull;
+
+public class GenerateChangesProcedure extends BaseProcedure {
+
+  public static SparkProcedures.ProcedureBuilder builder() {
+    return new BaseProcedure.Builder<GenerateChangesProcedure>() {
+      @Override
+      protected GenerateChangesProcedure doBuild() {
+        return new GenerateChangesProcedure(tableCatalog());
+      }
+    };
+  }
+
+  private static final ProcedureParameter[] PARAMETERS =
+      new ProcedureParameter[] {
+        ProcedureParameter.required("table", DataTypes.StringType),
+        // the snapshot ids input are ignored when the start/end timestamps are provided
+        ProcedureParameter.optional("start_snapshot_id_exclusive", DataTypes.LongType),
+        ProcedureParameter.optional("end_snapshot_id_inclusive", DataTypes.LongType),
+        ProcedureParameter.optional("table_change_view", DataTypes.StringType),
+        ProcedureParameter.optional("identifier_columns", DataTypes.StringType),
+        ProcedureParameter.optional("start_timestamp", DataTypes.TimestampType),
+        ProcedureParameter.optional("end_timestamp", DataTypes.TimestampType),

Review Comment:
   I am a bit worried about the number of parameters to configure boundaries. What if we replaced all of them with generic `options` and would pass those options along when loading `DataFrame`? Then instead of determining what snapshots match our timestamp range in the procedure, we would do that when scanning the changelog table. That way, users would be able to use timestamp boundaries not only via procedure but also via `DataFrame`. Right now, we only support snapshot ID boundaries.



##########
spark/v3.3/spark/src/main/java/org/apache/iceberg/spark/procedures/GenerateChangesProcedure.java:
##########
@@ -0,0 +1,271 @@
+/*
+ * 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.iceberg.spark.procedures;
+
+import java.util.Arrays;
+import java.util.UUID;
+import org.apache.iceberg.ChangelogOperation;
+import org.apache.iceberg.MetadataColumns;
+import org.apache.iceberg.Snapshot;
+import org.apache.iceberg.Table;
+import org.apache.iceberg.spark.SparkReadOptions;
+import org.apache.iceberg.spark.source.SparkChangelogTable;
+import org.apache.iceberg.util.DateTimeUtil;
+import org.apache.iceberg.util.SnapshotUtil;
+import org.apache.spark.sql.Column;
+import org.apache.spark.sql.DataFrameReader;
+import org.apache.spark.sql.Dataset;
+import org.apache.spark.sql.Row;
+import org.apache.spark.sql.catalyst.InternalRow;
+import org.apache.spark.sql.connector.catalog.Identifier;
+import org.apache.spark.sql.connector.catalog.TableCatalog;
+import org.apache.spark.sql.connector.iceberg.catalog.ProcedureParameter;
+import org.apache.spark.sql.expressions.Window;
+import org.apache.spark.sql.functions;
+import org.apache.spark.sql.types.DataTypes;
+import org.apache.spark.sql.types.Metadata;
+import org.apache.spark.sql.types.StructField;
+import org.apache.spark.sql.types.StructType;
+import org.apache.spark.unsafe.types.UTF8String;
+import org.jetbrains.annotations.NotNull;
+
+public class GenerateChangesProcedure extends BaseProcedure {
+
+  public static SparkProcedures.ProcedureBuilder builder() {
+    return new BaseProcedure.Builder<GenerateChangesProcedure>() {
+      @Override
+      protected GenerateChangesProcedure doBuild() {
+        return new GenerateChangesProcedure(tableCatalog());
+      }
+    };
+  }
+
+  private static final ProcedureParameter[] PARAMETERS =
+      new ProcedureParameter[] {
+        ProcedureParameter.required("table", DataTypes.StringType),
+        // the snapshot ids input are ignored when the start/end timestamps are provided
+        ProcedureParameter.optional("start_snapshot_id_exclusive", DataTypes.LongType),
+        ProcedureParameter.optional("end_snapshot_id_inclusive", DataTypes.LongType),
+        ProcedureParameter.optional("table_change_view", DataTypes.StringType),
+        ProcedureParameter.optional("identifier_columns", DataTypes.StringType),

Review Comment:
   Each snapshot schema may have a list of identifier fields. Can we use those if set and make this list a fallback if real identifier fields are not known?



##########
spark/v3.3/spark/src/main/java/org/apache/iceberg/spark/procedures/GenerateChangesProcedure.java:
##########
@@ -0,0 +1,210 @@
+/*
+ * 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.iceberg.spark.procedures;
+
+import java.util.Arrays;
+import org.apache.iceberg.ChangelogOperation;
+import org.apache.iceberg.MetadataColumns;
+import org.apache.iceberg.spark.SparkReadOptions;
+import org.apache.iceberg.spark.source.SparkChangelogTable;
+import org.apache.spark.sql.Column;
+import org.apache.spark.sql.DataFrameReader;
+import org.apache.spark.sql.Dataset;
+import org.apache.spark.sql.Row;
+import org.apache.spark.sql.catalyst.InternalRow;
+import org.apache.spark.sql.connector.catalog.TableCatalog;
+import org.apache.spark.sql.connector.iceberg.catalog.ProcedureParameter;
+import org.apache.spark.sql.expressions.Window;
+import org.apache.spark.sql.functions;
+import org.apache.spark.sql.types.DataTypes;
+import org.apache.spark.sql.types.Metadata;
+import org.apache.spark.sql.types.StructField;
+import org.apache.spark.sql.types.StructType;
+import org.apache.spark.unsafe.types.UTF8String;
+import org.jetbrains.annotations.NotNull;
+
+public class GenerateChangesProcedure extends BaseProcedure {
+
+  public static SparkProcedures.ProcedureBuilder builder() {
+    return new BaseProcedure.Builder<GenerateChangesProcedure>() {
+      @Override
+      protected GenerateChangesProcedure doBuild() {
+        return new GenerateChangesProcedure(tableCatalog());
+      }
+    };
+  }
+
+  private static final ProcedureParameter[] PARAMETERS =
+      new ProcedureParameter[] {
+        ProcedureParameter.required("table", DataTypes.StringType),
+        ProcedureParameter.optional("start_snapshot_id_exclusive", DataTypes.LongType),
+        ProcedureParameter.optional("end_snapshot_id_inclusive", DataTypes.LongType),
+        ProcedureParameter.optional("table_change_view", DataTypes.StringType),
+        ProcedureParameter.optional("identifier_columns", DataTypes.StringType),
+      };
+
+  private static final StructType OUTPUT_TYPE =
+      new StructType(
+          new StructField[] {
+            new StructField("orphan_file_location", DataTypes.StringType, false, Metadata.empty())
+          });
+
+  private GenerateChangesProcedure(TableCatalog tableCatalog) {
+    super(tableCatalog);
+  }
+
+  @Override
+  public ProcedureParameter[] parameters() {
+    return PARAMETERS;
+  }
+
+  @Override
+  public StructType outputType() {
+    return OUTPUT_TYPE;
+  }
+
+  @Override
+  public InternalRow[] call(InternalRow args) {
+    String tableName = args.getString(0);
+
+    // Read data from the table.changes
+    Dataset<Row> df = changelogRecords(tableName, args);
+
+    // Compute the pre-image and post-images if the identifier columns are provided.
+    if (!args.isNullAt(4)) {
+      String[] identifierColumns = args.getString(4).split(",");
+      if (identifierColumns == null || identifierColumns.length > 0) {
+        df = withUpdate(df, identifierColumns);
+      }
+    }
+
+    String viewName = viewName(args, tableName);
+
+    // Create a view for users to query
+    df.createOrReplaceTempView(viewName);
+
+    return toOutputRows(viewName);
+  }
+
+  private Dataset<Row> changelogRecords(String tableName, InternalRow args) {
+    DataFrameReader reader = spark().read();
+
+    // we don't have to validate the snapshot ids here because the reader will do it for us.
+    if (!args.isNullAt(1)) {
+      long startSnapshotId = args.getLong(1);
+      reader = reader.option(SparkReadOptions.START_SNAPSHOT_ID, startSnapshotId);
+    }
+
+    if (!args.isNullAt(2)) {
+      long endSnapshotId = args.getLong(2);
+      reader = reader.option(SparkReadOptions.END_SNAPSHOT_ID, endSnapshotId);
+    }
+
+    return reader.table(tableName + "." + SparkChangelogTable.TABLE_NAME);
+  }
+
+  @NotNull
+  private static String viewName(InternalRow args, String tableName) {
+    String viewName = args.isNullAt(3) ? null : args.getString(3);
+    if (viewName == null) {
+      String shortTableName =
+          tableName.contains(".") ? tableName.substring(tableName.lastIndexOf(".") + 1) : tableName;
+      viewName = shortTableName + "_changes";
+    }
+    return viewName;
+  }
+
+  private Dataset<Row> withUpdate(Dataset<Row> df, String[] identifiers) {
+    Column[] partitionSpec = getPartitionSpec(df, identifiers);
+
+    Dataset<Row> dfWithUpdate =
+        df.withColumn("count", functions.count("*").over(Window.partitionBy(partitionSpec)))
+            .withColumn(
+                "rank",
+                functions
+                    .rank()
+                    .over(
+                        Window.partitionBy(partitionSpec)
+                            .orderBy(MetadataColumns.CHANGE_TYPE.name())));
+
+    Dataset<Row> preImageDf =
+        dfWithUpdate
+            .filter("rank = 1")
+            .filter("count = 2")
+            .drop("rank", "count")
+            .withColumn(
+                MetadataColumns.CHANGE_TYPE.name(),
+                functions.lit(ChangelogOperation.UPDATE_PREIMAGE.name()));
+
+    Dataset<Row> postImageDf =
+        dfWithUpdate
+            .filter("rank = 2")
+            .filter("count = 2")
+            .drop("rank", "count")
+            .withColumn(
+                MetadataColumns.CHANGE_TYPE.name(),
+                functions.lit(ChangelogOperation.UPDATE_POSTIMAGE.name()));
+
+    // remove the carry-over rows
+    Dataset<Row> dfWithoutCarryOver = removeCarryOvers(preImageDf.union(postImageDf));

Review Comment:
   Is there another algorithm we can consider that would make it cheaper? Will something like this work?
   
   ```
   - Load DELETEs and INSERTs as a DF
   - Repartition the DF by primary key, _change_ordinal and locally sort by primary key, _change_ordinal, _operation_type
   - Call mapPartitions with a closure that would look at the previous, current and next rows
     - If the previous, current, next row keys are different, output the current row as-is
     - If the next row key is same, the current row must be DELETE and the next row must be INSERT (if not -> exception)
         - If other columns beyond the key are same, it is a copied over row
             - Output null if unchanged rows should be ignored
             - Output the current row as-is if all rows should be produced 
         - If other columns beyond key are different, it is an update
             - Output the current row as pre-update
     - If the previous row key is same as the current one, the current row must be INSERT and the previous row must be DELETE
         - If other columns beyond the key are same, it is a copied over row
             - Output null if unchanged rows should be ignored
             - Output the current row as-is if all rows should be produced 
         - If other columns beyond key are different, it is an update
             - Output the current row as post-update
   ```
   
   That would require reading the changes only once, doing a single hash-based shuffle to co-locate rows for the same key and change ordinal, keeping  at most 3 rows in memory at a time. Seems fairly cheap?



-- 
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.

To unsubscribe, e-mail: issues-unsubscribe@iceberg.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@iceberg.apache.org
For additional commands, e-mail: issues-help@iceberg.apache.org