[GitHub] [iceberg] holdenk commented on a change in pull request #2660: Spark3 structured streaming micro_batch read support

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

holdenk commented on a change in pull request #2660:
URL: https://github.com/apache/iceberg/pull/2660#discussion_r647942939



##########
File path: spark3/src/main/java/org/apache/iceberg/spark/source/SparkMicroBatchStream.java
##########
@@ -0,0 +1,312 @@
+/*
+ * 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.source;
+
+import java.io.BufferedWriter;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.OutputStream;
+import java.io.OutputStreamWriter;
+import java.nio.charset.StandardCharsets;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Optional;
+import org.apache.iceberg.CombinedScanTask;
+import org.apache.iceberg.DataOperations;
+import org.apache.iceberg.FileScanTask;
+import org.apache.iceberg.MicroBatches;
+import org.apache.iceberg.MicroBatches.MicroBatch;
+import org.apache.iceberg.Schema;
+import org.apache.iceberg.SchemaParser;
+import org.apache.iceberg.SerializableTable;
+import org.apache.iceberg.Snapshot;
+import org.apache.iceberg.SnapshotSummary;
+import org.apache.iceberg.Table;
+import org.apache.iceberg.io.CloseableIterable;
+import org.apache.iceberg.relocated.com.google.common.base.Preconditions;
+import org.apache.iceberg.relocated.com.google.common.collect.Iterables;
+import org.apache.iceberg.relocated.com.google.common.collect.Lists;
+import org.apache.iceberg.spark.Spark3Util;
+import org.apache.iceberg.spark.SparkReadOptions;
+import org.apache.iceberg.spark.source.SparkBatchScan.ReadTask;
+import org.apache.iceberg.spark.source.SparkBatchScan.ReaderFactory;
+import org.apache.iceberg.util.PropertyUtil;
+import org.apache.iceberg.util.SnapshotUtil;
+import org.apache.iceberg.util.TableScanUtil;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.broadcast.Broadcast;
+import org.apache.spark.sql.SparkSession;
+import org.apache.spark.sql.connector.read.InputPartition;
+import org.apache.spark.sql.connector.read.PartitionReaderFactory;
+import org.apache.spark.sql.connector.read.streaming.MicroBatchStream;
+import org.apache.spark.sql.connector.read.streaming.Offset;
+import org.apache.spark.sql.execution.streaming.HDFSMetadataLog;
+import org.apache.spark.sql.util.CaseInsensitiveStringMap;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import scala.reflect.ClassTag;
+
+import static org.apache.iceberg.TableProperties.SPLIT_LOOKBACK;
+import static org.apache.iceberg.TableProperties.SPLIT_LOOKBACK_DEFAULT;
+import static org.apache.iceberg.TableProperties.SPLIT_OPEN_FILE_COST;
+import static org.apache.iceberg.TableProperties.SPLIT_OPEN_FILE_COST_DEFAULT;
+import static org.apache.iceberg.TableProperties.SPLIT_SIZE;
+import static org.apache.iceberg.TableProperties.SPLIT_SIZE_DEFAULT;
+
+public class SparkMicroBatchStream implements MicroBatchStream {
+  private static final Logger LOG = LoggerFactory.getLogger(SparkMicroBatchStream.class);
+
+  private final JavaSparkContext sparkContext;
+  private final Table table;
+  private final boolean caseSensitive;
+  private final Schema expectedSchema;
+  private final Long splitSize;
+  private final Integer splitLookback;
+  private final Long splitOpenFileCost;
+  private final boolean localityPreferred;
+  private final InitialOffsetStore initialOffsetStore;
+  private final StreamingOffset initialOffset;
+
+  SparkMicroBatchStream(SparkSession spark,
+                        Table table, boolean caseSensitive, Schema expectedSchema,
+                        CaseInsensitiveStringMap options, String checkpointLocation) {
+    this.sparkContext = JavaSparkContext.fromSparkContext(spark.sparkContext());
+    this.table = table;
+    this.caseSensitive = caseSensitive;
+    this.expectedSchema = expectedSchema;
+    this.localityPreferred = Spark3Util.isLocalityEnabled(table.io(), table.location(), options);
+    this.splitSize = Optional.ofNullable(Spark3Util.propertyAsLong(options, SparkReadOptions.SPLIT_SIZE, null))
+        .orElseGet(() -> PropertyUtil.propertyAsLong(table.properties(), SPLIT_SIZE, SPLIT_SIZE_DEFAULT));
+    this.splitLookback = Optional.ofNullable(Spark3Util.propertyAsInt(options, SparkReadOptions.LOOKBACK, null))
+        .orElseGet(() -> PropertyUtil.propertyAsInt(table.properties(), SPLIT_LOOKBACK, SPLIT_LOOKBACK_DEFAULT));
+    this.splitOpenFileCost = Optional.ofNullable(
+        Spark3Util.propertyAsLong(options, SparkReadOptions.FILE_OPEN_COST, null))
+        .orElseGet(() -> PropertyUtil.propertyAsLong(table.properties(), SPLIT_OPEN_FILE_COST,
+            SPLIT_OPEN_FILE_COST_DEFAULT));
+    this.initialOffsetStore = InitialOffsetStore.getInstance(spark, checkpointLocation);
+    this.initialOffset = getOrWriteInitialOffset(table, initialOffsetStore);
+  }
+
+  @Override
+  public Offset latestOffset() {
+    Snapshot latestSnapshot = table.currentSnapshot();
+    if (latestSnapshot == null) {
+      return StreamingOffset.START_OFFSET;
+    }
+
+    String addedFilesValue = latestSnapshot.summary().get(SnapshotSummary.ADDED_FILES_PROP);
+    long addedFiles = addedFilesValue == null ?
+        Iterables.size(latestSnapshot.addedFiles()) :
+        Long.parseLong(addedFilesValue);
+
+    boolean scanAllFiles = !StreamingOffset.START_OFFSET.equals(initialOffset) &&
+        latestSnapshot.snapshotId() == initialOffset.snapshotId();
+    return new StreamingOffset(
+        latestSnapshot.snapshotId(),
+        addedFiles,
+        scanAllFiles);
+  }
+
+  @Override
+  public InputPartition[] planInputPartitions(Offset start, Offset end) {
+    if (end.equals(StreamingOffset.START_OFFSET)) {
+      return new InputPartition[0];
+    }
+
+    // broadcast the table metadata as input partitions will be sent to executors
+    Broadcast<Table> tableBroadcast = sparkContext.broadcast(SerializableTable.copyOf(table));
+    String expectedSchemaString = SchemaParser.toJson(expectedSchema);
+
+    Preconditions.checkState(
+        end instanceof StreamingOffset,
+        "The end offset passed to planInputPartitions() is not an instance of StreamingOffset.");
+
+    Preconditions.checkState(
+        start instanceof StreamingOffset,
+        "The start offset passed to planInputPartitions() is not an instance of StreamingOffset.");
+
+    StreamingOffset endOffset = (StreamingOffset) end;
+    StreamingOffset startOffset = (StreamingOffset) start;
+
+    List<FileScanTask> fileScanTasks = getFileScanTasks(startOffset, endOffset);
+
+    CloseableIterable<FileScanTask> splitTasks = TableScanUtil.splitFiles(
+        CloseableIterable.withNoopClose(fileScanTasks),
+        splitSize);
+    List<CombinedScanTask> combinedScanTasks = Lists.newArrayList(
+        TableScanUtil.planTasks(splitTasks, splitSize, splitLookback, splitOpenFileCost));
+    InputPartition[] readTasks = new InputPartition[combinedScanTasks.size()];
+
+    for (int i = 0; i < combinedScanTasks.size(); i++) {
+      readTasks[i] = new ReadTask(
+          combinedScanTasks.get(i), tableBroadcast, expectedSchemaString,
+          caseSensitive, localityPreferred);
+    }
+
+    return readTasks;
+  }
+
+  @Override
+  public PartitionReaderFactory createReaderFactory() {
+    int batchSizeValueToDisableColumnarReads = 0;
+    return new ReaderFactory(batchSizeValueToDisableColumnarReads);
+  }
+
+  @Override
+  public Offset initialOffset() {
+    return initialOffset;
+  }
+
+  @Override
+  public Offset deserializeOffset(String json) {
+    return StreamingOffset.fromJson(json);
+  }
+
+  @Override
+  public void commit(Offset end) {
+  }
+
+  @Override
+  public void stop() {
+  }
+
+  private static StreamingOffset getOrWriteInitialOffset(Table table, InitialOffsetStore initialOffsetStore) {

Review comment:
       I think we might not need this if my assumptions above are correct. But I could be missing something.

##########
File path: spark3/src/main/java/org/apache/iceberg/spark/source/SparkMicroBatchStream.java
##########
@@ -0,0 +1,312 @@
+/*
+ * 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.source;
+
+import java.io.BufferedWriter;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.OutputStream;
+import java.io.OutputStreamWriter;
+import java.nio.charset.StandardCharsets;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Optional;
+import org.apache.iceberg.CombinedScanTask;
+import org.apache.iceberg.DataOperations;
+import org.apache.iceberg.FileScanTask;
+import org.apache.iceberg.MicroBatches;
+import org.apache.iceberg.MicroBatches.MicroBatch;
+import org.apache.iceberg.Schema;
+import org.apache.iceberg.SchemaParser;
+import org.apache.iceberg.SerializableTable;
+import org.apache.iceberg.Snapshot;
+import org.apache.iceberg.SnapshotSummary;
+import org.apache.iceberg.Table;
+import org.apache.iceberg.io.CloseableIterable;
+import org.apache.iceberg.relocated.com.google.common.base.Preconditions;
+import org.apache.iceberg.relocated.com.google.common.collect.Iterables;
+import org.apache.iceberg.relocated.com.google.common.collect.Lists;
+import org.apache.iceberg.spark.Spark3Util;
+import org.apache.iceberg.spark.SparkReadOptions;
+import org.apache.iceberg.spark.source.SparkBatchScan.ReadTask;
+import org.apache.iceberg.spark.source.SparkBatchScan.ReaderFactory;
+import org.apache.iceberg.util.PropertyUtil;
+import org.apache.iceberg.util.SnapshotUtil;
+import org.apache.iceberg.util.TableScanUtil;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.broadcast.Broadcast;
+import org.apache.spark.sql.SparkSession;
+import org.apache.spark.sql.connector.read.InputPartition;
+import org.apache.spark.sql.connector.read.PartitionReaderFactory;
+import org.apache.spark.sql.connector.read.streaming.MicroBatchStream;
+import org.apache.spark.sql.connector.read.streaming.Offset;
+import org.apache.spark.sql.execution.streaming.HDFSMetadataLog;
+import org.apache.spark.sql.util.CaseInsensitiveStringMap;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import scala.reflect.ClassTag;
+
+import static org.apache.iceberg.TableProperties.SPLIT_LOOKBACK;
+import static org.apache.iceberg.TableProperties.SPLIT_LOOKBACK_DEFAULT;
+import static org.apache.iceberg.TableProperties.SPLIT_OPEN_FILE_COST;
+import static org.apache.iceberg.TableProperties.SPLIT_OPEN_FILE_COST_DEFAULT;
+import static org.apache.iceberg.TableProperties.SPLIT_SIZE;
+import static org.apache.iceberg.TableProperties.SPLIT_SIZE_DEFAULT;
+
+public class SparkMicroBatchStream implements MicroBatchStream {
+  private static final Logger LOG = LoggerFactory.getLogger(SparkMicroBatchStream.class);
+
+  private final JavaSparkContext sparkContext;
+  private final Table table;
+  private final boolean caseSensitive;
+  private final Schema expectedSchema;
+  private final Long splitSize;
+  private final Integer splitLookback;
+  private final Long splitOpenFileCost;
+  private final boolean localityPreferred;
+  private final InitialOffsetStore initialOffsetStore;
+  private final StreamingOffset initialOffset;
+
+  SparkMicroBatchStream(SparkSession spark,
+                        Table table, boolean caseSensitive, Schema expectedSchema,
+                        CaseInsensitiveStringMap options, String checkpointLocation) {
+    this.sparkContext = JavaSparkContext.fromSparkContext(spark.sparkContext());
+    this.table = table;
+    this.caseSensitive = caseSensitive;
+    this.expectedSchema = expectedSchema;
+    this.localityPreferred = Spark3Util.isLocalityEnabled(table.io(), table.location(), options);
+    this.splitSize = Optional.ofNullable(Spark3Util.propertyAsLong(options, SparkReadOptions.SPLIT_SIZE, null))
+        .orElseGet(() -> PropertyUtil.propertyAsLong(table.properties(), SPLIT_SIZE, SPLIT_SIZE_DEFAULT));
+    this.splitLookback = Optional.ofNullable(Spark3Util.propertyAsInt(options, SparkReadOptions.LOOKBACK, null))
+        .orElseGet(() -> PropertyUtil.propertyAsInt(table.properties(), SPLIT_LOOKBACK, SPLIT_LOOKBACK_DEFAULT));
+    this.splitOpenFileCost = Optional.ofNullable(
+        Spark3Util.propertyAsLong(options, SparkReadOptions.FILE_OPEN_COST, null))
+        .orElseGet(() -> PropertyUtil.propertyAsLong(table.properties(), SPLIT_OPEN_FILE_COST,
+            SPLIT_OPEN_FILE_COST_DEFAULT));
+    this.initialOffsetStore = InitialOffsetStore.getInstance(spark, checkpointLocation);
+    this.initialOffset = getOrWriteInitialOffset(table, initialOffsetStore);
+  }
+
+  @Override
+  public Offset latestOffset() {
+    Snapshot latestSnapshot = table.currentSnapshot();
+    if (latestSnapshot == null) {
+      return StreamingOffset.START_OFFSET;
+    }
+
+    String addedFilesValue = latestSnapshot.summary().get(SnapshotSummary.ADDED_FILES_PROP);
+    long addedFiles = addedFilesValue == null ?
+        Iterables.size(latestSnapshot.addedFiles()) :
+        Long.parseLong(addedFilesValue);
+
+    boolean scanAllFiles = !StreamingOffset.START_OFFSET.equals(initialOffset) &&

Review comment:
       So it looks like this is propegated to MicroBatches.java where if scanAllFiles is set to true it looks at all the manifests. I feel like this logic might be wrong, at the very least it's being a bit complicated for me to reason about (maybe I need more coffee) but maybe we could have a comment explaining why were doing this check?
   

##########
File path: spark3/src/main/java/org/apache/iceberg/spark/source/SparkMicroBatchStream.java
##########
@@ -0,0 +1,312 @@
+/*
+ * 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.source;
+
+import java.io.BufferedWriter;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.OutputStream;
+import java.io.OutputStreamWriter;
+import java.nio.charset.StandardCharsets;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Optional;
+import org.apache.iceberg.CombinedScanTask;
+import org.apache.iceberg.DataOperations;
+import org.apache.iceberg.FileScanTask;
+import org.apache.iceberg.MicroBatches;
+import org.apache.iceberg.MicroBatches.MicroBatch;
+import org.apache.iceberg.Schema;
+import org.apache.iceberg.SchemaParser;
+import org.apache.iceberg.SerializableTable;
+import org.apache.iceberg.Snapshot;
+import org.apache.iceberg.SnapshotSummary;
+import org.apache.iceberg.Table;
+import org.apache.iceberg.io.CloseableIterable;
+import org.apache.iceberg.relocated.com.google.common.base.Preconditions;
+import org.apache.iceberg.relocated.com.google.common.collect.Iterables;
+import org.apache.iceberg.relocated.com.google.common.collect.Lists;
+import org.apache.iceberg.spark.Spark3Util;
+import org.apache.iceberg.spark.SparkReadOptions;
+import org.apache.iceberg.spark.source.SparkBatchScan.ReadTask;
+import org.apache.iceberg.spark.source.SparkBatchScan.ReaderFactory;
+import org.apache.iceberg.util.PropertyUtil;
+import org.apache.iceberg.util.SnapshotUtil;
+import org.apache.iceberg.util.TableScanUtil;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.broadcast.Broadcast;
+import org.apache.spark.sql.SparkSession;
+import org.apache.spark.sql.connector.read.InputPartition;
+import org.apache.spark.sql.connector.read.PartitionReaderFactory;
+import org.apache.spark.sql.connector.read.streaming.MicroBatchStream;
+import org.apache.spark.sql.connector.read.streaming.Offset;
+import org.apache.spark.sql.execution.streaming.HDFSMetadataLog;
+import org.apache.spark.sql.util.CaseInsensitiveStringMap;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import scala.reflect.ClassTag;
+
+import static org.apache.iceberg.TableProperties.SPLIT_LOOKBACK;
+import static org.apache.iceberg.TableProperties.SPLIT_LOOKBACK_DEFAULT;
+import static org.apache.iceberg.TableProperties.SPLIT_OPEN_FILE_COST;
+import static org.apache.iceberg.TableProperties.SPLIT_OPEN_FILE_COST_DEFAULT;
+import static org.apache.iceberg.TableProperties.SPLIT_SIZE;
+import static org.apache.iceberg.TableProperties.SPLIT_SIZE_DEFAULT;
+
+public class SparkMicroBatchStream implements MicroBatchStream {
+  private static final Logger LOG = LoggerFactory.getLogger(SparkMicroBatchStream.class);
+
+  private final JavaSparkContext sparkContext;
+  private final Table table;
+  private final boolean caseSensitive;
+  private final Schema expectedSchema;
+  private final Long splitSize;
+  private final Integer splitLookback;
+  private final Long splitOpenFileCost;
+  private final boolean localityPreferred;
+  private final InitialOffsetStore initialOffsetStore;
+  private final StreamingOffset initialOffset;
+
+  SparkMicroBatchStream(SparkSession spark,
+                        Table table, boolean caseSensitive, Schema expectedSchema,
+                        CaseInsensitiveStringMap options, String checkpointLocation) {
+    this.sparkContext = JavaSparkContext.fromSparkContext(spark.sparkContext());
+    this.table = table;
+    this.caseSensitive = caseSensitive;
+    this.expectedSchema = expectedSchema;
+    this.localityPreferred = Spark3Util.isLocalityEnabled(table.io(), table.location(), options);
+    this.splitSize = Optional.ofNullable(Spark3Util.propertyAsLong(options, SparkReadOptions.SPLIT_SIZE, null))
+        .orElseGet(() -> PropertyUtil.propertyAsLong(table.properties(), SPLIT_SIZE, SPLIT_SIZE_DEFAULT));
+    this.splitLookback = Optional.ofNullable(Spark3Util.propertyAsInt(options, SparkReadOptions.LOOKBACK, null))
+        .orElseGet(() -> PropertyUtil.propertyAsInt(table.properties(), SPLIT_LOOKBACK, SPLIT_LOOKBACK_DEFAULT));
+    this.splitOpenFileCost = Optional.ofNullable(
+        Spark3Util.propertyAsLong(options, SparkReadOptions.FILE_OPEN_COST, null))
+        .orElseGet(() -> PropertyUtil.propertyAsLong(table.properties(), SPLIT_OPEN_FILE_COST,
+            SPLIT_OPEN_FILE_COST_DEFAULT));
+    this.initialOffsetStore = InitialOffsetStore.getInstance(spark, checkpointLocation);
+    this.initialOffset = getOrWriteInitialOffset(table, initialOffsetStore);

Review comment:
       Summarizing an offline conversation around this:
   
   I was thinking that we could maybe skip `getOrWriteInitialOffset` and instead always compute the initialOffset since this is the one place where we use it. Inside of Spark the initialOffset will be used if we're starting from "scratch" (e.g. no records processed).
   
   I don't think caching initialOffset does anything for us, the only thing which it could do is change the behaviour in `scanAllFiles` but we would probably want to set scanAllFiles to true in the situation where the old initial offset no longer exists and we have only one snapShot anyways.




-- 
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



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