beam-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From da...@apache.org
Subject [41/50] [abbrv] incubator-beam git commit: Rename DataflowPipelineRunner to DataflowRunner
Date Mon, 20 Jun 2016 22:16:34 GMT
http://git-wip-us.apache.org/repos/asf/incubator-beam/blob/6d028ac6/runners/google-cloud-dataflow-java/src/main/java/org/apache/beam/runners/dataflow/DataflowRunner.java
----------------------------------------------------------------------
diff --git a/runners/google-cloud-dataflow-java/src/main/java/org/apache/beam/runners/dataflow/DataflowRunner.java b/runners/google-cloud-dataflow-java/src/main/java/org/apache/beam/runners/dataflow/DataflowRunner.java
new file mode 100644
index 0000000..91e34ac
--- /dev/null
+++ b/runners/google-cloud-dataflow-java/src/main/java/org/apache/beam/runners/dataflow/DataflowRunner.java
@@ -0,0 +1,3229 @@
+/*
+ * 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.beam.runners.dataflow;
+
+import static org.apache.beam.sdk.util.StringUtils.approximatePTransformName;
+import static org.apache.beam.sdk.util.StringUtils.approximateSimpleName;
+import static org.apache.beam.sdk.util.WindowedValue.valueInEmptyWindows;
+
+import static com.google.common.base.Preconditions.checkArgument;
+import static com.google.common.base.Preconditions.checkState;
+
+import org.apache.beam.runners.dataflow.DataflowPipelineTranslator.JobSpecification;
+import org.apache.beam.runners.dataflow.DataflowPipelineTranslator.TransformTranslator;
+import org.apache.beam.runners.dataflow.DataflowPipelineTranslator.TranslationContext;
+import org.apache.beam.runners.dataflow.internal.AssignWindows;
+import org.apache.beam.runners.dataflow.internal.DataflowAggregatorTransforms;
+import org.apache.beam.runners.dataflow.internal.IsmFormat;
+import org.apache.beam.runners.dataflow.internal.IsmFormat.IsmRecord;
+import org.apache.beam.runners.dataflow.internal.IsmFormat.IsmRecordCoder;
+import org.apache.beam.runners.dataflow.internal.IsmFormat.MetadataKeyCoder;
+import org.apache.beam.runners.dataflow.internal.ReadTranslator;
+import org.apache.beam.runners.dataflow.options.DataflowPipelineDebugOptions;
+import org.apache.beam.runners.dataflow.options.DataflowPipelineOptions;
+import org.apache.beam.runners.dataflow.options.DataflowPipelineWorkerPoolOptions;
+import org.apache.beam.runners.dataflow.util.DataflowTransport;
+import org.apache.beam.runners.dataflow.util.MonitoringUtil;
+import org.apache.beam.sdk.Pipeline;
+import org.apache.beam.sdk.Pipeline.PipelineVisitor;
+import org.apache.beam.sdk.PipelineResult.State;
+import org.apache.beam.sdk.annotations.Experimental;
+import org.apache.beam.sdk.coders.AvroCoder;
+import org.apache.beam.sdk.coders.BigEndianLongCoder;
+import org.apache.beam.sdk.coders.CannotProvideCoderException;
+import org.apache.beam.sdk.coders.Coder;
+import org.apache.beam.sdk.coders.Coder.NonDeterministicException;
+import org.apache.beam.sdk.coders.CoderException;
+import org.apache.beam.sdk.coders.CoderRegistry;
+import org.apache.beam.sdk.coders.IterableCoder;
+import org.apache.beam.sdk.coders.KvCoder;
+import org.apache.beam.sdk.coders.ListCoder;
+import org.apache.beam.sdk.coders.MapCoder;
+import org.apache.beam.sdk.coders.SerializableCoder;
+import org.apache.beam.sdk.coders.StandardCoder;
+import org.apache.beam.sdk.coders.VarIntCoder;
+import org.apache.beam.sdk.coders.VarLongCoder;
+import org.apache.beam.sdk.io.AvroIO;
+import org.apache.beam.sdk.io.BigQueryIO;
+import org.apache.beam.sdk.io.FileBasedSink;
+import org.apache.beam.sdk.io.PubsubIO;
+import org.apache.beam.sdk.io.PubsubUnboundedSink;
+import org.apache.beam.sdk.io.PubsubUnboundedSource;
+import org.apache.beam.sdk.io.Read;
+import org.apache.beam.sdk.io.ShardNameTemplate;
+import org.apache.beam.sdk.io.TextIO;
+import org.apache.beam.sdk.io.UnboundedSource;
+import org.apache.beam.sdk.io.Write;
+import org.apache.beam.sdk.options.PipelineOptions;
+import org.apache.beam.sdk.options.PipelineOptionsValidator;
+import org.apache.beam.sdk.options.StreamingOptions;
+import org.apache.beam.sdk.runners.AggregatorPipelineExtractor;
+import org.apache.beam.sdk.runners.PipelineRunner;
+import org.apache.beam.sdk.runners.TransformTreeNode;
+import org.apache.beam.sdk.transforms.Aggregator;
+import org.apache.beam.sdk.transforms.Combine;
+import org.apache.beam.sdk.transforms.Combine.CombineFn;
+import org.apache.beam.sdk.transforms.Create;
+import org.apache.beam.sdk.transforms.DoFn;
+import org.apache.beam.sdk.transforms.Flatten;
+import org.apache.beam.sdk.transforms.GroupByKey;
+import org.apache.beam.sdk.transforms.PTransform;
+import org.apache.beam.sdk.transforms.ParDo;
+import org.apache.beam.sdk.transforms.SerializableFunction;
+import org.apache.beam.sdk.transforms.View;
+import org.apache.beam.sdk.transforms.View.CreatePCollectionView;
+import org.apache.beam.sdk.transforms.WithKeys;
+import org.apache.beam.sdk.transforms.windowing.AfterPane;
+import org.apache.beam.sdk.transforms.windowing.BoundedWindow;
+import org.apache.beam.sdk.transforms.windowing.DefaultTrigger;
+import org.apache.beam.sdk.transforms.windowing.GlobalWindow;
+import org.apache.beam.sdk.transforms.windowing.GlobalWindows;
+import org.apache.beam.sdk.transforms.windowing.Window;
+import org.apache.beam.sdk.util.CoderUtils;
+import org.apache.beam.sdk.util.IOChannelUtils;
+import org.apache.beam.sdk.util.InstanceBuilder;
+import org.apache.beam.sdk.util.PCollectionViews;
+import org.apache.beam.sdk.util.PathValidator;
+import org.apache.beam.sdk.util.PropertyNames;
+import org.apache.beam.sdk.util.ReleaseInfo;
+import org.apache.beam.sdk.util.Reshuffle;
+import org.apache.beam.sdk.util.SystemDoFnInternal;
+import org.apache.beam.sdk.util.ValueWithRecordId;
+import org.apache.beam.sdk.util.WindowedValue;
+import org.apache.beam.sdk.util.WindowedValue.FullWindowedValueCoder;
+import org.apache.beam.sdk.util.WindowingStrategy;
+import org.apache.beam.sdk.values.KV;
+import org.apache.beam.sdk.values.PBegin;
+import org.apache.beam.sdk.values.PCollection;
+import org.apache.beam.sdk.values.PCollection.IsBounded;
+import org.apache.beam.sdk.values.PCollectionList;
+import org.apache.beam.sdk.values.PCollectionTuple;
+import org.apache.beam.sdk.values.PCollectionView;
+import org.apache.beam.sdk.values.PDone;
+import org.apache.beam.sdk.values.PInput;
+import org.apache.beam.sdk.values.POutput;
+import org.apache.beam.sdk.values.PValue;
+import org.apache.beam.sdk.values.TupleTag;
+import org.apache.beam.sdk.values.TupleTagList;
+
+import com.google.api.client.googleapis.json.GoogleJsonResponseException;
+import com.google.api.services.clouddebugger.v2.Clouddebugger;
+import com.google.api.services.clouddebugger.v2.model.Debuggee;
+import com.google.api.services.clouddebugger.v2.model.RegisterDebuggeeRequest;
+import com.google.api.services.clouddebugger.v2.model.RegisterDebuggeeResponse;
+import com.google.api.services.dataflow.Dataflow;
+import com.google.api.services.dataflow.model.DataflowPackage;
+import com.google.api.services.dataflow.model.Job;
+import com.google.api.services.dataflow.model.ListJobsResponse;
+import com.google.api.services.dataflow.model.WorkerPool;
+import com.google.common.annotations.VisibleForTesting;
+import com.google.common.base.Function;
+import com.google.common.base.Joiner;
+import com.google.common.base.Optional;
+import com.google.common.base.Preconditions;
+import com.google.common.base.Strings;
+import com.google.common.base.Utf8;
+import com.google.common.collect.ForwardingMap;
+import com.google.common.collect.HashMultimap;
+import com.google.common.collect.ImmutableList;
+import com.google.common.collect.ImmutableMap;
+import com.google.common.collect.Iterables;
+import com.google.common.collect.Maps;
+import com.google.common.collect.Multimap;
+
+import com.fasterxml.jackson.annotation.JsonCreator;
+import com.fasterxml.jackson.annotation.JsonProperty;
+
+import org.joda.time.DateTimeUtils;
+import org.joda.time.DateTimeZone;
+import org.joda.time.Duration;
+import org.joda.time.format.DateTimeFormat;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+import java.io.File;
+import java.io.FileNotFoundException;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.OutputStream;
+import java.io.PrintWriter;
+import java.io.Serializable;
+import java.net.URISyntaxException;
+import java.net.URL;
+import java.net.URLClassLoader;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.Collection;
+import java.util.Collections;
+import java.util.HashMap;
+import java.util.HashSet;
+import java.util.Iterator;
+import java.util.List;
+import java.util.Map;
+import java.util.Random;
+import java.util.Set;
+import java.util.SortedSet;
+import java.util.TreeSet;
+import javax.annotation.Nullable;
+
+/**
+ * A {@link PipelineRunner} that executes the operations in the
+ * pipeline by first translating them to the Dataflow representation
+ * using the {@link DataflowPipelineTranslator} and then submitting
+ * them to a Dataflow service for execution.
+ *
+ * <p><h3>Permissions</h3>
+ * When reading from a Dataflow source or writing to a Dataflow sink using
+ * {@code DataflowRunner}, the Google cloudservices account and the Google compute engine
+ * service account of the GCP project running the Dataflow Job will need access to the corresponding
+ * source/sink.
+ *
+ * <p>Please see <a href="https://cloud.google.com/dataflow/security-and-permissions">Google Cloud
+ * Dataflow Security and Permissions</a> for more details.
+ */
+public class DataflowRunner extends PipelineRunner<DataflowPipelineJob> {
+  private static final Logger LOG = LoggerFactory.getLogger(DataflowRunner.class);
+
+  /** Provided configuration options. */
+  private final DataflowPipelineOptions options;
+
+  /** Client for the Dataflow service. This is used to actually submit jobs. */
+  private final Dataflow dataflowClient;
+
+  /** Translator for this DataflowRunner, based on options. */
+  private final DataflowPipelineTranslator translator;
+
+  /** Custom transforms implementations. */
+  private final Map<Class<?>, Class<?>> overrides;
+
+  /** A set of user defined functions to invoke at different points in execution. */
+  private DataflowRunnerHooks hooks;
+
+  // Environment version information.
+  private static final String ENVIRONMENT_MAJOR_VERSION = "5";
+
+  // Default Docker container images that execute Dataflow worker harness, residing in Google
+  // Container Registry, separately for Batch and Streaming.
+  public static final String BATCH_WORKER_HARNESS_CONTAINER_IMAGE
+      = "dataflow.gcr.io/v1beta3/beam-java-batch:beam-master-20160613";
+  public static final String STREAMING_WORKER_HARNESS_CONTAINER_IMAGE
+      = "dataflow.gcr.io/v1beta3/beam-java-streaming:beam-master-20160613";
+
+  // The limit of CreateJob request size.
+  private static final int CREATE_JOB_REQUEST_LIMIT_BYTES = 10 * 1024 * 1024;
+
+  private final Set<PCollection<?>> pcollectionsRequiringIndexedFormat;
+
+  /**
+   * Project IDs must contain lowercase letters, digits, or dashes.
+   * IDs must start with a letter and may not end with a dash.
+   * This regex isn't exact - this allows for patterns that would be rejected by
+   * the service, but this is sufficient for basic validation of project IDs.
+   */
+  public static final String PROJECT_ID_REGEXP = "[a-z][-a-z0-9:.]+[a-z0-9]";
+
+  /**
+   * Construct a runner from the provided options.
+   *
+   * @param options Properties that configure the runner.
+   * @return The newly created runner.
+   */
+  public static DataflowRunner fromOptions(PipelineOptions options) {
+    // (Re-)register standard IO factories. Clobbers any prior credentials.
+    IOChannelUtils.registerStandardIOFactories(options);
+
+    DataflowPipelineOptions dataflowOptions =
+        PipelineOptionsValidator.validate(DataflowPipelineOptions.class, options);
+    ArrayList<String> missing = new ArrayList<>();
+
+    if (dataflowOptions.getAppName() == null) {
+      missing.add("appName");
+    }
+    if (missing.size() > 0) {
+      throw new IllegalArgumentException(
+          "Missing required values: " + Joiner.on(',').join(missing));
+    }
+
+    PathValidator validator = dataflowOptions.getPathValidator();
+    Preconditions.checkArgument(!(Strings.isNullOrEmpty(dataflowOptions.getTempLocation())
+        && Strings.isNullOrEmpty(dataflowOptions.getStagingLocation())),
+        "Missing required value: at least one of tempLocation or stagingLocation must be set.");
+
+    if (dataflowOptions.getStagingLocation() != null) {
+      validator.validateOutputFilePrefixSupported(dataflowOptions.getStagingLocation());
+    }
+    if (dataflowOptions.getTempLocation() != null) {
+      validator.validateOutputFilePrefixSupported(dataflowOptions.getTempLocation());
+    }
+    if (Strings.isNullOrEmpty(dataflowOptions.getTempLocation())) {
+      dataflowOptions.setTempLocation(dataflowOptions.getStagingLocation());
+    } else if (Strings.isNullOrEmpty(dataflowOptions.getStagingLocation())) {
+      try {
+        dataflowOptions.setStagingLocation(
+            IOChannelUtils.resolve(dataflowOptions.getTempLocation(), "staging"));
+      } catch (IOException e) {
+        throw new IllegalArgumentException("Unable to resolve PipelineOptions.stagingLocation "
+            + "from PipelineOptions.tempLocation. Please set the staging location explicitly.", e);
+      }
+    }
+
+    if (dataflowOptions.getFilesToStage() == null) {
+      dataflowOptions.setFilesToStage(detectClassPathResourcesToStage(
+          DataflowRunner.class.getClassLoader()));
+      LOG.info("PipelineOptions.filesToStage was not specified. "
+          + "Defaulting to files from the classpath: will stage {} files. "
+          + "Enable logging at DEBUG level to see which files will be staged.",
+          dataflowOptions.getFilesToStage().size());
+      LOG.debug("Classpath elements: {}", dataflowOptions.getFilesToStage());
+    }
+
+    // Verify jobName according to service requirements, truncating converting to lowercase if
+    // necessary.
+    String jobName =
+        dataflowOptions
+            .getJobName()
+            .toLowerCase();
+    checkArgument(
+        jobName.matches("[a-z]([-a-z0-9]*[a-z0-9])?"),
+        "JobName invalid; the name must consist of only the characters "
+            + "[-a-z0-9], starting with a letter and ending with a letter "
+            + "or number");
+    if (!jobName.equals(dataflowOptions.getJobName())) {
+      LOG.info(
+          "PipelineOptions.jobName did not match the service requirements. "
+              + "Using {} instead of {}.",
+          jobName,
+          dataflowOptions.getJobName());
+    }
+    dataflowOptions.setJobName(jobName);
+
+    // Verify project
+    String project = dataflowOptions.getProject();
+    if (project.matches("[0-9]*")) {
+      throw new IllegalArgumentException("Project ID '" + project
+          + "' invalid. Please make sure you specified the Project ID, not project number.");
+    } else if (!project.matches(PROJECT_ID_REGEXP)) {
+      throw new IllegalArgumentException("Project ID '" + project
+          + "' invalid. Please make sure you specified the Project ID, not project description.");
+    }
+
+    DataflowPipelineDebugOptions debugOptions =
+        dataflowOptions.as(DataflowPipelineDebugOptions.class);
+    // Verify the number of worker threads is a valid value
+    if (debugOptions.getNumberOfWorkerHarnessThreads() < 0) {
+      throw new IllegalArgumentException("Number of worker harness threads '"
+          + debugOptions.getNumberOfWorkerHarnessThreads()
+          + "' invalid. Please make sure the value is non-negative.");
+    }
+
+    return new DataflowRunner(dataflowOptions);
+  }
+
+  @VisibleForTesting protected DataflowRunner(DataflowPipelineOptions options) {
+    this.options = options;
+    this.dataflowClient = options.getDataflowClient();
+    this.translator = DataflowPipelineTranslator.fromOptions(options);
+    this.pcollectionsRequiringIndexedFormat = new HashSet<>();
+    this.ptransformViewsWithNonDeterministicKeyCoders = new HashSet<>();
+
+    ImmutableMap.Builder<Class<?>, Class<?>> builder = ImmutableMap.<Class<?>, Class<?>>builder();
+    if (options.isStreaming()) {
+      builder.put(Combine.GloballyAsSingletonView.class,
+                  StreamingCombineGloballyAsSingletonView.class);
+      builder.put(Create.Values.class, StreamingCreate.class);
+      builder.put(View.AsMap.class, StreamingViewAsMap.class);
+      builder.put(View.AsMultimap.class, StreamingViewAsMultimap.class);
+      builder.put(View.AsSingleton.class, StreamingViewAsSingleton.class);
+      builder.put(View.AsList.class, StreamingViewAsList.class);
+      builder.put(View.AsIterable.class, StreamingViewAsIterable.class);
+      builder.put(Write.Bound.class, StreamingWrite.class);
+      builder.put(Read.Unbounded.class, StreamingUnboundedRead.class);
+      builder.put(Read.Bounded.class, UnsupportedIO.class);
+      builder.put(AvroIO.Read.Bound.class, UnsupportedIO.class);
+      builder.put(AvroIO.Write.Bound.class, UnsupportedIO.class);
+      builder.put(BigQueryIO.Read.Bound.class, UnsupportedIO.class);
+      builder.put(TextIO.Read.Bound.class, UnsupportedIO.class);
+      builder.put(TextIO.Write.Bound.class, UnsupportedIO.class);
+      builder.put(Window.Bound.class, AssignWindows.class);
+      // In streaming mode must use either the custom Pubsub unbounded source/sink or
+      // defer to Windmill's built-in implementation.
+      builder.put(PubsubIO.Read.Bound.PubsubBoundedReader.class, UnsupportedIO.class);
+      builder.put(PubsubIO.Write.Bound.PubsubBoundedWriter.class, UnsupportedIO.class);
+      if (options.getExperiments() == null
+          || !options.getExperiments().contains("enable_custom_pubsub_source")) {
+        builder.put(PubsubUnboundedSource.class, StreamingPubsubIORead.class);
+      }
+      if (options.getExperiments() == null
+          || !options.getExperiments().contains("enable_custom_pubsub_sink")) {
+        builder.put(PubsubUnboundedSink.class, StreamingPubsubIOWrite.class);
+      }
+    } else {
+      builder.put(Read.Unbounded.class, UnsupportedIO.class);
+      builder.put(Window.Bound.class, AssignWindows.class);
+      builder.put(Write.Bound.class, BatchWrite.class);
+      builder.put(AvroIO.Write.Bound.class, BatchAvroIOWrite.class);
+      builder.put(TextIO.Write.Bound.class, BatchTextIOWrite.class);
+      // In batch mode must use the custom Pubsub bounded source/sink.
+      builder.put(PubsubUnboundedSource.class, UnsupportedIO.class);
+      builder.put(PubsubUnboundedSink.class, UnsupportedIO.class);
+      if (options.getExperiments() == null
+          || !options.getExperiments().contains("disable_ism_side_input")) {
+        builder.put(View.AsMap.class, BatchViewAsMap.class);
+        builder.put(View.AsMultimap.class, BatchViewAsMultimap.class);
+        builder.put(View.AsSingleton.class, BatchViewAsSingleton.class);
+        builder.put(View.AsList.class, BatchViewAsList.class);
+        builder.put(View.AsIterable.class, BatchViewAsIterable.class);
+      }
+    }
+    overrides = builder.build();
+  }
+
+  /**
+   * Applies the given transform to the input. For transforms with customized definitions
+   * for the Dataflow pipeline runner, the application is intercepted and modified here.
+   */
+  @Override
+  public <OutputT extends POutput, InputT extends PInput> OutputT apply(
+      PTransform<InputT, OutputT> transform, InputT input) {
+
+    if (Combine.GroupedValues.class.equals(transform.getClass())
+        || GroupByKey.class.equals(transform.getClass())) {
+
+      // For both Dataflow runners (streaming and batch), GroupByKey and GroupedValues are
+      // primitives. Returning a primitive output instead of the expanded definition
+      // signals to the translator that translation is necessary.
+      @SuppressWarnings("unchecked")
+      PCollection<?> pc = (PCollection<?>) input;
+      @SuppressWarnings("unchecked")
+      OutputT outputT = (OutputT) PCollection.createPrimitiveOutputInternal(
+          pc.getPipeline(),
+          transform instanceof GroupByKey
+              ? ((GroupByKey<?, ?>) transform).updateWindowingStrategy(pc.getWindowingStrategy())
+              : pc.getWindowingStrategy(),
+          pc.isBounded());
+      return outputT;
+    } else if (Window.Bound.class.equals(transform.getClass())) {
+      /*
+       * TODO: make this the generic way overrides are applied (using super.apply() rather than
+       * Pipeline.applyTransform(); this allows the apply method to be replaced without inserting
+       * additional nodes into the graph.
+       */
+      // casting to wildcard
+      @SuppressWarnings("unchecked")
+      OutputT windowed = (OutputT) applyWindow((Window.Bound<?>) transform, (PCollection<?>) input);
+      return windowed;
+    } else if (Flatten.FlattenPCollectionList.class.equals(transform.getClass())
+        && ((PCollectionList<?>) input).size() == 0) {
+      return (OutputT) Pipeline.applyTransform(input, Create.of());
+    } else if (overrides.containsKey(transform.getClass())) {
+      // It is the responsibility of whoever constructs overrides to ensure this is type safe.
+      @SuppressWarnings("unchecked")
+      Class<PTransform<InputT, OutputT>> transformClass =
+          (Class<PTransform<InputT, OutputT>>) transform.getClass();
+
+      @SuppressWarnings("unchecked")
+      Class<PTransform<InputT, OutputT>> customTransformClass =
+          (Class<PTransform<InputT, OutputT>>) overrides.get(transform.getClass());
+
+      PTransform<InputT, OutputT> customTransform =
+          InstanceBuilder.ofType(customTransformClass)
+          .withArg(DataflowRunner.class, this)
+          .withArg(transformClass, transform)
+          .build();
+
+      return Pipeline.applyTransform(input, customTransform);
+    } else {
+      return super.apply(transform, input);
+    }
+  }
+
+  private <T> PCollection<T> applyWindow(
+      Window.Bound<?> intitialTransform, PCollection<?> initialInput) {
+    // types are matched at compile time
+    @SuppressWarnings("unchecked")
+    Window.Bound<T> transform = (Window.Bound<T>) intitialTransform;
+    @SuppressWarnings("unchecked")
+    PCollection<T> input = (PCollection<T>) initialInput;
+    return super.apply(new AssignWindows<>(transform), input);
+  }
+
+  private String debuggerMessage(String projectId, String uniquifier) {
+    return String.format("To debug your job, visit Google Cloud Debugger at: "
+        + "https://console.developers.google.com/debug?project=%s&dbgee=%s",
+        projectId, uniquifier);
+  }
+
+  private void maybeRegisterDebuggee(DataflowPipelineOptions options, String uniquifier) {
+    if (!options.getEnableCloudDebugger()) {
+      return;
+    }
+
+    if (options.getDebuggee() != null) {
+      throw new RuntimeException("Should not specify the debuggee");
+    }
+
+    Clouddebugger debuggerClient = DataflowTransport.newClouddebuggerClient(options).build();
+    Debuggee debuggee = registerDebuggee(debuggerClient, uniquifier);
+    options.setDebuggee(debuggee);
+
+    System.out.println(debuggerMessage(options.getProject(), debuggee.getUniquifier()));
+  }
+
+  private Debuggee registerDebuggee(Clouddebugger debuggerClient, String uniquifier) {
+    RegisterDebuggeeRequest registerReq = new RegisterDebuggeeRequest();
+    registerReq.setDebuggee(new Debuggee()
+        .setProject(options.getProject())
+        .setUniquifier(uniquifier)
+        .setDescription(uniquifier)
+        .setAgentVersion("google.com/cloud-dataflow-java/v1"));
+
+    try {
+      RegisterDebuggeeResponse registerResponse =
+          debuggerClient.controller().debuggees().register(registerReq).execute();
+      Debuggee debuggee = registerResponse.getDebuggee();
+      if (debuggee.getStatus() != null && debuggee.getStatus().getIsError()) {
+        throw new RuntimeException("Unable to register with the debugger: "
+            + debuggee.getStatus().getDescription().getFormat());
+      }
+
+      return debuggee;
+    } catch (IOException e) {
+      throw new RuntimeException("Unable to register with the debugger: ", e);
+    }
+  }
+
+  @Override
+  public DataflowPipelineJob run(Pipeline pipeline) {
+    logWarningIfPCollectionViewHasNonDeterministicKeyCoder(pipeline);
+
+    LOG.info("Executing pipeline on the Dataflow Service, which will have billing implications "
+        + "related to Google Compute Engine usage and other Google Cloud Services.");
+
+    List<DataflowPackage> packages = options.getStager().stageFiles();
+
+
+    // Set a unique client_request_id in the CreateJob request.
+    // This is used to ensure idempotence of job creation across retried
+    // attempts to create a job. Specifically, if the service returns a job with
+    // a different client_request_id, it means the returned one is a different
+    // job previously created with the same job name, and that the job creation
+    // has been effectively rejected. The SDK should return
+    // Error::Already_Exists to user in that case.
+    int randomNum = new Random().nextInt(9000) + 1000;
+    String requestId = DateTimeFormat.forPattern("YYYYMMddHHmmssmmm").withZone(DateTimeZone.UTC)
+        .print(DateTimeUtils.currentTimeMillis()) + "_" + randomNum;
+
+    // Try to create a debuggee ID. This must happen before the job is translated since it may
+    // update the options.
+    DataflowPipelineOptions dataflowOptions = options.as(DataflowPipelineOptions.class);
+    maybeRegisterDebuggee(dataflowOptions, requestId);
+
+    JobSpecification jobSpecification =
+        translator.translate(pipeline, this, packages);
+    Job newJob = jobSpecification.getJob();
+    newJob.setClientRequestId(requestId);
+
+    String version = ReleaseInfo.getReleaseInfo().getVersion();
+    System.out.println("Dataflow SDK version: " + version);
+
+    newJob.getEnvironment().setUserAgent(ReleaseInfo.getReleaseInfo());
+    // The Dataflow Service may write to the temporary directory directly, so
+    // must be verified.
+    if (!Strings.isNullOrEmpty(options.getTempLocation())) {
+      newJob.getEnvironment().setTempStoragePrefix(
+          dataflowOptions.getPathValidator().verifyPath(options.getTempLocation()));
+    }
+    newJob.getEnvironment().setDataset(options.getTempDatasetId());
+    newJob.getEnvironment().setExperiments(options.getExperiments());
+
+    // Set the Docker container image that executes Dataflow worker harness, residing in Google
+    // Container Registry. Translator is guaranteed to create a worker pool prior to this point.
+    String workerHarnessContainerImage =
+        options.as(DataflowPipelineWorkerPoolOptions.class)
+        .getWorkerHarnessContainerImage();
+    for (WorkerPool workerPool : newJob.getEnvironment().getWorkerPools()) {
+      workerPool.setWorkerHarnessContainerImage(workerHarnessContainerImage);
+    }
+
+    // Requirements about the service.
+    Map<String, Object> environmentVersion = new HashMap<>();
+    environmentVersion.put(PropertyNames.ENVIRONMENT_VERSION_MAJOR_KEY, ENVIRONMENT_MAJOR_VERSION);
+    newJob.getEnvironment().setVersion(environmentVersion);
+    // Default jobType is JAVA_BATCH_AUTOSCALING: A Java job with workers that the job can
+    // autoscale if specified.
+    String jobType = "JAVA_BATCH_AUTOSCALING";
+
+    if (options.isStreaming()) {
+      jobType = "STREAMING";
+    }
+    environmentVersion.put(PropertyNames.ENVIRONMENT_VERSION_JOB_TYPE_KEY, jobType);
+
+    if (hooks != null) {
+      hooks.modifyEnvironmentBeforeSubmission(newJob.getEnvironment());
+    }
+
+    if (!Strings.isNullOrEmpty(options.getDataflowJobFile())) {
+      try (PrintWriter printWriter = new PrintWriter(
+          new File(options.getDataflowJobFile()))) {
+        String workSpecJson = DataflowPipelineTranslator.jobToString(newJob);
+        printWriter.print(workSpecJson);
+        LOG.info("Printed workflow specification to {}", options.getDataflowJobFile());
+      } catch (IllegalStateException ex) {
+        LOG.warn("Cannot translate workflow spec to json for debug.");
+      } catch (FileNotFoundException ex) {
+        LOG.warn("Cannot create workflow spec output file.");
+      }
+    }
+
+    String jobIdToUpdate = null;
+    if (options.isUpdate()) {
+      jobIdToUpdate = getJobIdFromName(options.getJobName());
+      newJob.setTransformNameMapping(options.getTransformNameMapping());
+      newJob.setReplaceJobId(jobIdToUpdate);
+    }
+    Job jobResult;
+    try {
+      jobResult = dataflowClient
+              .projects()
+              .jobs()
+              .create(options.getProject(), newJob)
+              .execute();
+    } catch (GoogleJsonResponseException e) {
+      String errorMessages = "Unexpected errors";
+      if (e.getDetails() != null) {
+        if (Utf8.encodedLength(newJob.toString()) >= CREATE_JOB_REQUEST_LIMIT_BYTES) {
+          errorMessages = "The size of the serialized JSON representation of the pipeline "
+              + "exceeds the allowable limit. "
+              + "For more information, please check the FAQ link below:\n"
+              + "https://cloud.google.com/dataflow/faq";
+        } else {
+          errorMessages = e.getDetails().getMessage();
+        }
+      }
+      throw new RuntimeException("Failed to create a workflow job: " + errorMessages, e);
+    } catch (IOException e) {
+      throw new RuntimeException("Failed to create a workflow job", e);
+    }
+
+    // Obtain all of the extractors from the PTransforms used in the pipeline so the
+    // DataflowPipelineJob has access to them.
+    AggregatorPipelineExtractor aggregatorExtractor = new AggregatorPipelineExtractor(pipeline);
+    Map<Aggregator<?, ?>, Collection<PTransform<?, ?>>> aggregatorSteps =
+        aggregatorExtractor.getAggregatorSteps();
+
+    DataflowAggregatorTransforms aggregatorTransforms =
+        new DataflowAggregatorTransforms(aggregatorSteps, jobSpecification.getStepNames());
+
+    // Use a raw client for post-launch monitoring, as status calls may fail
+    // regularly and need not be retried automatically.
+    DataflowPipelineJob dataflowPipelineJob =
+        new DataflowPipelineJob(options.getProject(), jobResult.getId(),
+            DataflowTransport.newRawDataflowClient(options).build(), aggregatorTransforms);
+
+    // If the service returned client request id, the SDK needs to compare it
+    // with the original id generated in the request, if they are not the same
+    // (i.e., the returned job is not created by this request), throw
+    // DataflowJobAlreadyExistsException or DataflowJobAlreadyUpdatedExcetpion
+    // depending on whether this is a reload or not.
+    if (jobResult.getClientRequestId() != null && !jobResult.getClientRequestId().isEmpty()
+        && !jobResult.getClientRequestId().equals(requestId)) {
+      // If updating a job.
+      if (options.isUpdate()) {
+        throw new DataflowJobAlreadyUpdatedException(dataflowPipelineJob,
+            String.format("The job named %s with id: %s has already been updated into job id: %s "
+                + "and cannot be updated again.",
+                newJob.getName(), jobIdToUpdate, jobResult.getId()));
+      } else {
+        throw new DataflowJobAlreadyExistsException(dataflowPipelineJob,
+            String.format("There is already an active job named %s with id: %s. If you want "
+                + "to submit a second job, try again by setting a different name using --jobName.",
+                newJob.getName(), jobResult.getId()));
+      }
+    }
+
+    LOG.info("To access the Dataflow monitoring console, please navigate to {}",
+        MonitoringUtil.getJobMonitoringPageURL(options.getProject(), jobResult.getId()));
+    System.out.println("Submitted job: " + jobResult.getId());
+
+    LOG.info("To cancel the job using the 'gcloud' tool, run:\n> {}",
+        MonitoringUtil.getGcloudCancelCommand(options, jobResult.getId()));
+
+    return dataflowPipelineJob;
+  }
+
+  /**
+   * Returns the DataflowPipelineTranslator associated with this object.
+   */
+  public DataflowPipelineTranslator getTranslator() {
+    return translator;
+  }
+
+  /**
+   * Sets callbacks to invoke during execution see {@code DataflowRunnerHooks}.
+   */
+  @Experimental
+  public void setHooks(DataflowRunnerHooks hooks) {
+    this.hooks = hooks;
+  }
+
+  /////////////////////////////////////////////////////////////////////////////
+
+  /** Outputs a warning about PCollection views without deterministic key coders. */
+  private void logWarningIfPCollectionViewHasNonDeterministicKeyCoder(Pipeline pipeline) {
+    // We need to wait till this point to determine the names of the transforms since only
+    // at this time do we know the hierarchy of the transforms otherwise we could
+    // have just recorded the full names during apply time.
+    if (!ptransformViewsWithNonDeterministicKeyCoders.isEmpty()) {
+      final SortedSet<String> ptransformViewNamesWithNonDeterministicKeyCoders = new TreeSet<>();
+      pipeline.traverseTopologically(new PipelineVisitor() {
+        @Override
+        public void visitValue(PValue value, TransformTreeNode producer) {
+        }
+
+        @Override
+        public void visitPrimitiveTransform(TransformTreeNode node) {
+          if (ptransformViewsWithNonDeterministicKeyCoders.contains(node.getTransform())) {
+            ptransformViewNamesWithNonDeterministicKeyCoders.add(node.getFullName());
+          }
+        }
+
+        @Override
+        public CompositeBehavior enterCompositeTransform(TransformTreeNode node) {
+          if (ptransformViewsWithNonDeterministicKeyCoders.contains(node.getTransform())) {
+            ptransformViewNamesWithNonDeterministicKeyCoders.add(node.getFullName());
+          }
+          return CompositeBehavior.ENTER_TRANSFORM;
+        }
+
+        @Override
+        public void leaveCompositeTransform(TransformTreeNode node) {
+        }
+      });
+
+      LOG.warn("Unable to use indexed implementation for View.AsMap and View.AsMultimap for {} "
+          + "because the key coder is not deterministic. Falling back to singleton implementation "
+          + "which may cause memory and/or performance problems. Future major versions of "
+          + "Dataflow will require deterministic key coders.",
+          ptransformViewNamesWithNonDeterministicKeyCoders);
+    }
+  }
+
+  /**
+   * Returns true if the passed in {@link PCollection} needs to be materialiazed using
+   * an indexed format.
+   */
+  boolean doesPCollectionRequireIndexedFormat(PCollection<?> pcol) {
+    return pcollectionsRequiringIndexedFormat.contains(pcol);
+  }
+
+  /**
+   * Marks the passed in {@link PCollection} as requiring to be materialized using
+   * an indexed format.
+   */
+  private void addPCollectionRequiringIndexedFormat(PCollection<?> pcol) {
+    pcollectionsRequiringIndexedFormat.add(pcol);
+  }
+
+  /** A set of {@link View}s with non-deterministic key coders. */
+  Set<PTransform<?, ?>> ptransformViewsWithNonDeterministicKeyCoders;
+
+  /**
+   * Records that the {@link PTransform} requires a deterministic key coder.
+   */
+  private void recordViewUsesNonDeterministicKeyCoder(PTransform<?, ?> ptransform) {
+    ptransformViewsWithNonDeterministicKeyCoders.add(ptransform);
+  }
+
+  /**
+   * A {@link GroupByKey} transform for the {@link DataflowRunner} which sorts
+   * values using the secondary key {@code K2}.
+   *
+   * <p>The {@link PCollection} created created by this {@link PTransform} will have values in
+   * the empty window. Care must be taken *afterwards* to either re-window
+   * (using {@link Window#into}) or only use {@link PTransform}s that do not depend on the
+   * values being within a window.
+   */
+  static class GroupByKeyAndSortValuesOnly<K1, K2, V>
+      extends PTransform<PCollection<KV<K1, KV<K2, V>>>, PCollection<KV<K1, Iterable<KV<K2, V>>>>> {
+    private GroupByKeyAndSortValuesOnly() {
+    }
+
+    @Override
+    public PCollection<KV<K1, Iterable<KV<K2, V>>>> apply(PCollection<KV<K1, KV<K2, V>>> input) {
+      PCollection<KV<K1, Iterable<KV<K2, V>>>> rval =
+          PCollection.<KV<K1, Iterable<KV<K2, V>>>>createPrimitiveOutputInternal(
+          input.getPipeline(),
+          WindowingStrategy.globalDefault(),
+          IsBounded.BOUNDED);
+
+      @SuppressWarnings({"unchecked", "rawtypes"})
+      KvCoder<K1, KV<K2, V>> inputCoder = (KvCoder) input.getCoder();
+      rval.setCoder(
+          KvCoder.of(inputCoder.getKeyCoder(),
+          IterableCoder.of(inputCoder.getValueCoder())));
+      return rval;
+    }
+  }
+
+  /**
+   * A {@link PTransform} that groups the values by a hash of the window's byte representation
+   * and sorts the values using the windows byte representation.
+   */
+  private static class GroupByWindowHashAsKeyAndWindowAsSortKey<T, W extends BoundedWindow> extends
+      PTransform<PCollection<T>, PCollection<KV<Integer, Iterable<KV<W, WindowedValue<T>>>>>> {
+
+    /**
+     * A {@link DoFn} that for each element outputs a {@code KV} structure suitable for
+     * grouping by the hash of the window's byte representation and sorting the grouped values
+     * using the window's byte representation.
+     */
+    @SystemDoFnInternal
+    private static class UseWindowHashAsKeyAndWindowAsSortKeyDoFn<T, W extends BoundedWindow>
+        extends DoFn<T, KV<Integer, KV<W, WindowedValue<T>>>> implements DoFn.RequiresWindowAccess {
+
+      private final IsmRecordCoder<?> ismCoderForHash;
+      private UseWindowHashAsKeyAndWindowAsSortKeyDoFn(IsmRecordCoder<?> ismCoderForHash) {
+        this.ismCoderForHash = ismCoderForHash;
+      }
+
+      @Override
+      public void processElement(ProcessContext c) throws Exception {
+        @SuppressWarnings("unchecked")
+        W window = (W) c.window();
+        c.output(
+            KV.of(ismCoderForHash.hash(ImmutableList.of(window)),
+                KV.of(window,
+                    WindowedValue.of(
+                        c.element(),
+                        c.timestamp(),
+                        c.window(),
+                        c.pane()))));
+      }
+    }
+
+    private final IsmRecordCoder<?> ismCoderForHash;
+    private GroupByWindowHashAsKeyAndWindowAsSortKey(IsmRecordCoder<?> ismCoderForHash) {
+      this.ismCoderForHash = ismCoderForHash;
+    }
+
+    @Override
+    public PCollection<KV<Integer, Iterable<KV<W, WindowedValue<T>>>>> apply(PCollection<T> input) {
+      @SuppressWarnings("unchecked")
+      Coder<W> windowCoder = (Coder<W>)
+          input.getWindowingStrategy().getWindowFn().windowCoder();
+      PCollection<KV<Integer, KV<W, WindowedValue<T>>>> rval =
+          input.apply(ParDo.of(
+              new UseWindowHashAsKeyAndWindowAsSortKeyDoFn<T, W>(ismCoderForHash)));
+      rval.setCoder(
+          KvCoder.of(
+              VarIntCoder.of(),
+              KvCoder.of(windowCoder,
+                  FullWindowedValueCoder.of(input.getCoder(), windowCoder))));
+      return rval.apply(new GroupByKeyAndSortValuesOnly<Integer, W, WindowedValue<T>>());
+    }
+  }
+
+  /**
+   * Specialized implementation for
+   * {@link org.apache.beam.sdk.transforms.View.AsSingleton View.AsSingleton} for the
+   * Dataflow runner in batch mode.
+   *
+   * <p>Creates a set of files in the {@link IsmFormat} sharded by the hash of the windows
+   * byte representation and with records having:
+   * <ul>
+   *   <li>Key 1: Window</li>
+   *   <li>Value: Windowed value</li>
+   * </ul>
+   */
+  static class BatchViewAsSingleton<T>
+      extends PTransform<PCollection<T>, PCollectionView<T>> {
+
+    /**
+     * A {@link DoFn} that outputs {@link IsmRecord}s. These records are structured as follows:
+     * <ul>
+     *   <li>Key 1: Window
+     *   <li>Value: Windowed value
+     * </ul>
+     */
+    static class IsmRecordForSingularValuePerWindowDoFn<T, W extends BoundedWindow>
+        extends DoFn<KV<Integer, Iterable<KV<W, WindowedValue<T>>>>,
+                     IsmRecord<WindowedValue<T>>> {
+
+      private final Coder<W> windowCoder;
+      IsmRecordForSingularValuePerWindowDoFn(Coder<W> windowCoder) {
+        this.windowCoder = windowCoder;
+      }
+
+      @Override
+      public void processElement(ProcessContext c) throws Exception {
+        Optional<Object> previousWindowStructuralValue = Optional.absent();
+        T previousValue = null;
+
+        Iterator<KV<W, WindowedValue<T>>> iterator = c.element().getValue().iterator();
+        while (iterator.hasNext()) {
+          KV<W, WindowedValue<T>> next = iterator.next();
+          Object currentWindowStructuralValue = windowCoder.structuralValue(next.getKey());
+
+          // Verify that the user isn't trying to have more than one element per window as
+          // a singleton.
+          checkState(!previousWindowStructuralValue.isPresent()
+              || !previousWindowStructuralValue.get().equals(currentWindowStructuralValue),
+              "Multiple values [%s, %s] found for singleton within window [%s].",
+              previousValue,
+              next.getValue().getValue(),
+              next.getKey());
+
+          c.output(
+              IsmRecord.of(
+                  ImmutableList.of(next.getKey()), next.getValue()));
+
+          previousWindowStructuralValue = Optional.of(currentWindowStructuralValue);
+          previousValue = next.getValue().getValue();
+        }
+      }
+    }
+
+    private final DataflowRunner runner;
+    private final View.AsSingleton<T> transform;
+    /**
+     * Builds an instance of this class from the overridden transform.
+     */
+    @SuppressWarnings("unused") // used via reflection in DataflowRunner#apply()
+    public BatchViewAsSingleton(DataflowRunner runner, View.AsSingleton<T> transform) {
+      this.runner = runner;
+      this.transform = transform;
+    }
+
+    @Override
+    public PCollectionView<T> apply(PCollection<T> input) {
+      @SuppressWarnings("unchecked")
+      Coder<BoundedWindow> windowCoder = (Coder<BoundedWindow>)
+          input.getWindowingStrategy().getWindowFn().windowCoder();
+
+      return BatchViewAsSingleton.<T, T, T, BoundedWindow>applyForSingleton(
+          runner,
+          input,
+          new IsmRecordForSingularValuePerWindowDoFn<T, BoundedWindow>(windowCoder),
+          transform.hasDefaultValue(),
+          transform.defaultValue(),
+          input.getCoder());
+    }
+
+    static <T, FinalT, ViewT, W extends BoundedWindow> PCollectionView<ViewT>
+        applyForSingleton(
+            DataflowRunner runner,
+            PCollection<T> input,
+            DoFn<KV<Integer, Iterable<KV<W, WindowedValue<T>>>>,
+                 IsmRecord<WindowedValue<FinalT>>> doFn,
+            boolean hasDefault,
+            FinalT defaultValue,
+            Coder<FinalT> defaultValueCoder) {
+
+      @SuppressWarnings("unchecked")
+      Coder<W> windowCoder = (Coder<W>)
+          input.getWindowingStrategy().getWindowFn().windowCoder();
+
+      @SuppressWarnings({"rawtypes", "unchecked"})
+      PCollectionView<ViewT> view =
+          (PCollectionView<ViewT>) PCollectionViews.<FinalT, W>singletonView(
+              input.getPipeline(),
+              (WindowingStrategy) input.getWindowingStrategy(),
+              hasDefault,
+              defaultValue,
+              defaultValueCoder);
+
+      IsmRecordCoder<WindowedValue<FinalT>> ismCoder =
+          coderForSingleton(windowCoder, defaultValueCoder);
+
+      PCollection<IsmRecord<WindowedValue<FinalT>>> reifiedPerWindowAndSorted = input
+              .apply(new GroupByWindowHashAsKeyAndWindowAsSortKey<T, W>(ismCoder))
+              .apply(ParDo.of(doFn));
+      reifiedPerWindowAndSorted.setCoder(ismCoder);
+
+      runner.addPCollectionRequiringIndexedFormat(reifiedPerWindowAndSorted);
+      return reifiedPerWindowAndSorted.apply(
+          CreatePCollectionView.<IsmRecord<WindowedValue<FinalT>>, ViewT>of(view));
+    }
+
+    @Override
+    protected String getKindString() {
+      return "BatchViewAsSingleton";
+    }
+
+    static <T> IsmRecordCoder<WindowedValue<T>> coderForSingleton(
+        Coder<? extends BoundedWindow> windowCoder, Coder<T> valueCoder) {
+      return IsmRecordCoder.of(
+          1, // We hash using only the window
+          0, // There are no metadata records
+          ImmutableList.<Coder<?>>of(windowCoder),
+          FullWindowedValueCoder.of(valueCoder, windowCoder));
+    }
+  }
+
+  /**
+   * Specialized implementation for
+   * {@link org.apache.beam.sdk.transforms.View.AsIterable View.AsIterable} for the
+   * Dataflow runner in batch mode.
+   *
+   * <p>Creates a set of {@code Ism} files sharded by the hash of the windows byte representation
+   * and with records having:
+   * <ul>
+   *   <li>Key 1: Window</li>
+   *   <li>Key 2: Index offset within window</li>
+   *   <li>Value: Windowed value</li>
+   * </ul>
+   */
+  static class BatchViewAsIterable<T>
+      extends PTransform<PCollection<T>, PCollectionView<Iterable<T>>> {
+
+    private final DataflowRunner runner;
+    /**
+     * Builds an instance of this class from the overridden transform.
+     */
+    @SuppressWarnings("unused") // used via reflection in DataflowRunner#apply()
+    public BatchViewAsIterable(DataflowRunner runner, View.AsIterable<T> transform) {
+      this.runner = runner;
+    }
+
+    @Override
+    public PCollectionView<Iterable<T>> apply(PCollection<T> input) {
+      PCollectionView<Iterable<T>> view = PCollectionViews.iterableView(
+          input.getPipeline(), input.getWindowingStrategy(), input.getCoder());
+      return BatchViewAsList.applyForIterableLike(runner, input, view);
+    }
+  }
+
+  /**
+   * Specialized implementation for
+   * {@link org.apache.beam.sdk.transforms.View.AsList View.AsList} for the
+   * Dataflow runner in batch mode.
+   *
+   * <p>Creates a set of {@code Ism} files sharded by the hash of the window's byte representation
+   * and with records having:
+   * <ul>
+   *   <li>Key 1: Window</li>
+   *   <li>Key 2: Index offset within window</li>
+   *   <li>Value: Windowed value</li>
+   * </ul>
+   */
+  static class BatchViewAsList<T>
+      extends PTransform<PCollection<T>, PCollectionView<List<T>>> {
+    /**
+     * A {@link DoFn} which creates {@link IsmRecord}s assuming that each element is within the
+     * global window. Each {@link IsmRecord} has
+     * <ul>
+     *   <li>Key 1: Global window</li>
+     *   <li>Key 2: Index offset within window</li>
+     *   <li>Value: Windowed value</li>
+     * </ul>
+     */
+    @SystemDoFnInternal
+    static class ToIsmRecordForGlobalWindowDoFn<T>
+        extends DoFn<T, IsmRecord<WindowedValue<T>>> {
+
+      long indexInBundle;
+      @Override
+      public void startBundle(Context c) throws Exception {
+        indexInBundle = 0;
+      }
+
+      @Override
+      public void processElement(ProcessContext c) throws Exception {
+        c.output(IsmRecord.of(
+            ImmutableList.of(GlobalWindow.INSTANCE, indexInBundle),
+            WindowedValue.of(
+                c.element(),
+                c.timestamp(),
+                GlobalWindow.INSTANCE,
+                c.pane())));
+        indexInBundle += 1;
+      }
+    }
+
+    /**
+     * A {@link DoFn} which creates {@link IsmRecord}s comparing successive elements windows
+     * to locate the window boundaries. The {@link IsmRecord} has:
+     * <ul>
+     *   <li>Key 1: Window</li>
+     *   <li>Key 2: Index offset within window</li>
+     *   <li>Value: Windowed value</li>
+     * </ul>
+     */
+    @SystemDoFnInternal
+    static class ToIsmRecordForNonGlobalWindowDoFn<T, W extends BoundedWindow>
+        extends DoFn<KV<Integer, Iterable<KV<W, WindowedValue<T>>>>,
+                     IsmRecord<WindowedValue<T>>> {
+
+      private final Coder<W> windowCoder;
+      ToIsmRecordForNonGlobalWindowDoFn(Coder<W> windowCoder) {
+        this.windowCoder = windowCoder;
+      }
+
+      @Override
+      public void processElement(ProcessContext c) throws Exception {
+        long elementsInWindow = 0;
+        Optional<Object> previousWindowStructuralValue = Optional.absent();
+        for (KV<W, WindowedValue<T>> value : c.element().getValue()) {
+          Object currentWindowStructuralValue = windowCoder.structuralValue(value.getKey());
+          // Compare to see if this is a new window so we can reset the index counter i
+          if (previousWindowStructuralValue.isPresent()
+              && !previousWindowStructuralValue.get().equals(currentWindowStructuralValue)) {
+            // Reset i since we have a new window.
+            elementsInWindow = 0;
+          }
+          c.output(IsmRecord.of(
+              ImmutableList.of(value.getKey(), elementsInWindow),
+              value.getValue()));
+          previousWindowStructuralValue = Optional.of(currentWindowStructuralValue);
+          elementsInWindow += 1;
+        }
+      }
+    }
+
+    private final DataflowRunner runner;
+    /**
+     * Builds an instance of this class from the overridden transform.
+     */
+    @SuppressWarnings("unused") // used via reflection in DataflowRunner#apply()
+    public BatchViewAsList(DataflowRunner runner, View.AsList<T> transform) {
+      this.runner = runner;
+    }
+
+    @Override
+    public PCollectionView<List<T>> apply(PCollection<T> input) {
+      PCollectionView<List<T>> view = PCollectionViews.listView(
+          input.getPipeline(), input.getWindowingStrategy(), input.getCoder());
+      return applyForIterableLike(runner, input, view);
+    }
+
+    static <T, W extends BoundedWindow, ViewT> PCollectionView<ViewT> applyForIterableLike(
+        DataflowRunner runner,
+        PCollection<T> input,
+        PCollectionView<ViewT> view) {
+
+      @SuppressWarnings("unchecked")
+      Coder<W> windowCoder = (Coder<W>)
+          input.getWindowingStrategy().getWindowFn().windowCoder();
+
+      IsmRecordCoder<WindowedValue<T>> ismCoder = coderForListLike(windowCoder, input.getCoder());
+
+      // If we are working in the global window, we do not need to do a GBK using the window
+      // as the key since all the elements of the input PCollection are already such.
+      // We just reify the windowed value while converting them to IsmRecords and generating
+      // an index based upon where we are within the bundle. Each bundle
+      // maps to one file exactly.
+      if (input.getWindowingStrategy().getWindowFn() instanceof GlobalWindows) {
+        PCollection<IsmRecord<WindowedValue<T>>> reifiedPerWindowAndSorted =
+            input.apply(ParDo.of(new ToIsmRecordForGlobalWindowDoFn<T>()));
+        reifiedPerWindowAndSorted.setCoder(ismCoder);
+
+        runner.addPCollectionRequiringIndexedFormat(reifiedPerWindowAndSorted);
+        return reifiedPerWindowAndSorted.apply(
+            CreatePCollectionView.<IsmRecord<WindowedValue<T>>, ViewT>of(view));
+      }
+
+      PCollection<IsmRecord<WindowedValue<T>>> reifiedPerWindowAndSorted = input
+              .apply(new GroupByWindowHashAsKeyAndWindowAsSortKey<T, W>(ismCoder))
+              .apply(ParDo.of(new ToIsmRecordForNonGlobalWindowDoFn<T, W>(windowCoder)));
+      reifiedPerWindowAndSorted.setCoder(ismCoder);
+
+      runner.addPCollectionRequiringIndexedFormat(reifiedPerWindowAndSorted);
+      return reifiedPerWindowAndSorted.apply(
+          CreatePCollectionView.<IsmRecord<WindowedValue<T>>, ViewT>of(view));
+    }
+
+    @Override
+    protected String getKindString() {
+      return "BatchViewAsList";
+    }
+
+    static <T> IsmRecordCoder<WindowedValue<T>> coderForListLike(
+        Coder<? extends BoundedWindow> windowCoder, Coder<T> valueCoder) {
+      // TODO: swap to use a variable length long coder which has values which compare
+      // the same as their byte representation compare lexicographically within the key coder
+      return IsmRecordCoder.of(
+          1, // We hash using only the window
+          0, // There are no metadata records
+          ImmutableList.of(windowCoder, BigEndianLongCoder.of()),
+          FullWindowedValueCoder.of(valueCoder, windowCoder));
+    }
+  }
+
+  /**
+   * Specialized implementation for
+   * {@link org.apache.beam.sdk.transforms.View.AsMap View.AsMap} for the
+   * Dataflow runner in batch mode.
+   *
+   * <p>Creates a set of {@code Ism} files sharded by the hash of the key's byte
+   * representation. Each record is structured as follows:
+   * <ul>
+   *   <li>Key 1: User key K</li>
+   *   <li>Key 2: Window</li>
+   *   <li>Key 3: 0L (constant)</li>
+   *   <li>Value: Windowed value</li>
+   * </ul>
+   *
+   * <p>Alongside the data records, there are the following metadata records:
+   * <ul>
+   *   <li>Key 1: Metadata Key</li>
+   *   <li>Key 2: Window</li>
+   *   <li>Key 3: Index [0, size of map]</li>
+   *   <li>Value: variable length long byte representation of size of map if index is 0,
+   *              otherwise the byte representation of a key</li>
+   * </ul>
+   * The {@code [META, Window, 0]} record stores the number of unique keys per window, while
+   * {@code [META, Window, i]}  for {@code i} in {@code [1, size of map]} stores a the users key.
+   * This allows for one to access the size of the map by looking at {@code [META, Window, 0]}
+   * and iterate over all the keys by accessing {@code [META, Window, i]} for {@code i} in
+   * {@code [1, size of map]}.
+   *
+   * <p>Note that in the case of a non-deterministic key coder, we fallback to using
+   * {@link org.apache.beam.sdk.transforms.View.AsSingleton View.AsSingleton} printing
+   * a warning to users to specify a deterministic key coder.
+   */
+  static class BatchViewAsMap<K, V>
+      extends PTransform<PCollection<KV<K, V>>, PCollectionView<Map<K, V>>> {
+
+    /**
+     * A {@link DoFn} which groups elements by window boundaries. For each group,
+     * the group of elements is transformed into a {@link TransformedMap}.
+     * The transformed {@code Map<K, V>} is backed by a {@code Map<K, WindowedValue<V>>}
+     * and contains a function {@code WindowedValue<V> -> V}.
+     *
+     * <p>Outputs {@link IsmRecord}s having:
+     * <ul>
+     *   <li>Key 1: Window</li>
+     *   <li>Value: Transformed map containing a transform that removes the encapsulation
+     *              of the window around each value,
+     *              {@code Map<K, WindowedValue<V>> -> Map<K, V>}.</li>
+     * </ul>
+     */
+    static class ToMapDoFn<K, V, W extends BoundedWindow>
+        extends DoFn<KV<Integer, Iterable<KV<W, WindowedValue<KV<K, V>>>>>,
+                     IsmRecord<WindowedValue<TransformedMap<K,
+                                             WindowedValue<V>,
+                                             V>>>> {
+
+      private final Coder<W> windowCoder;
+      ToMapDoFn(Coder<W> windowCoder) {
+        this.windowCoder = windowCoder;
+      }
+
+      @Override
+      public void processElement(ProcessContext c)
+          throws Exception {
+        Optional<Object> previousWindowStructuralValue = Optional.absent();
+        Optional<W> previousWindow = Optional.absent();
+        Map<K, WindowedValue<V>> map = new HashMap<>();
+        for (KV<W, WindowedValue<KV<K, V>>> kv : c.element().getValue()) {
+          Object currentWindowStructuralValue = windowCoder.structuralValue(kv.getKey());
+          if (previousWindowStructuralValue.isPresent()
+              && !previousWindowStructuralValue.get().equals(currentWindowStructuralValue)) {
+            // Construct the transformed map containing all the elements since we
+            // are at a window boundary.
+            c.output(IsmRecord.of(
+                ImmutableList.of(previousWindow.get()),
+                valueInEmptyWindows(new TransformedMap<>(WindowedValueToValue.<V>of(), map))));
+            map = new HashMap<>();
+          }
+
+          // Verify that the user isn't trying to insert the same key multiple times.
+          checkState(!map.containsKey(kv.getValue().getValue().getKey()),
+              "Multiple values [%s, %s] found for single key [%s] within window [%s].",
+              map.get(kv.getValue().getValue().getKey()),
+              kv.getValue().getValue().getValue(),
+              kv.getKey());
+          map.put(kv.getValue().getValue().getKey(),
+                  kv.getValue().withValue(kv.getValue().getValue().getValue()));
+          previousWindowStructuralValue = Optional.of(currentWindowStructuralValue);
+          previousWindow = Optional.of(kv.getKey());
+        }
+
+        // The last value for this hash is guaranteed to be at a window boundary
+        // so we output a transformed map containing all the elements since the last
+        // window boundary.
+        c.output(IsmRecord.of(
+            ImmutableList.of(previousWindow.get()),
+            valueInEmptyWindows(new TransformedMap<>(WindowedValueToValue.<V>of(), map))));
+      }
+    }
+
+    private final DataflowRunner runner;
+    /**
+     * Builds an instance of this class from the overridden transform.
+     */
+    @SuppressWarnings("unused") // used via reflection in DataflowRunner#apply()
+    public BatchViewAsMap(DataflowRunner runner, View.AsMap<K, V> transform) {
+      this.runner = runner;
+    }
+
+    @Override
+    public PCollectionView<Map<K, V>> apply(PCollection<KV<K, V>> input) {
+      return this.<BoundedWindow>applyInternal(input);
+    }
+
+    private <W extends BoundedWindow> PCollectionView<Map<K, V>>
+        applyInternal(PCollection<KV<K, V>> input) {
+
+      @SuppressWarnings({"rawtypes", "unchecked"})
+      KvCoder<K, V> inputCoder = (KvCoder) input.getCoder();
+      try {
+        PCollectionView<Map<K, V>> view = PCollectionViews.mapView(
+            input.getPipeline(), input.getWindowingStrategy(), inputCoder);
+        return BatchViewAsMultimap.applyForMapLike(runner, input, view, true /* unique keys */);
+      } catch (NonDeterministicException e) {
+        runner.recordViewUsesNonDeterministicKeyCoder(this);
+
+        // Since the key coder is not deterministic, we convert the map into a singleton
+        // and return a singleton view equivalent.
+        return applyForSingletonFallback(input);
+      }
+    }
+
+    @Override
+    protected String getKindString() {
+      return "BatchViewAsMap";
+    }
+
+    /** Transforms the input {@link PCollection} into a singleton {@link Map} per window. */
+    private <W extends BoundedWindow> PCollectionView<Map<K, V>>
+        applyForSingletonFallback(PCollection<KV<K, V>> input) {
+      @SuppressWarnings("unchecked")
+      Coder<W> windowCoder = (Coder<W>)
+          input.getWindowingStrategy().getWindowFn().windowCoder();
+
+      @SuppressWarnings({"rawtypes", "unchecked"})
+      KvCoder<K, V> inputCoder = (KvCoder) input.getCoder();
+
+      @SuppressWarnings({"unchecked", "rawtypes"})
+      Coder<Function<WindowedValue<V>, V>> transformCoder =
+          (Coder) SerializableCoder.of(WindowedValueToValue.class);
+
+      Coder<TransformedMap<K, WindowedValue<V>, V>> finalValueCoder =
+          TransformedMapCoder.of(
+          transformCoder,
+          MapCoder.of(
+              inputCoder.getKeyCoder(),
+              FullWindowedValueCoder.of(inputCoder.getValueCoder(), windowCoder)));
+
+      TransformedMap<K, WindowedValue<V>, V> defaultValue = new TransformedMap<>(
+          WindowedValueToValue.<V>of(),
+          ImmutableMap.<K, WindowedValue<V>>of());
+
+      return BatchViewAsSingleton.<KV<K, V>,
+                                   TransformedMap<K, WindowedValue<V>, V>,
+                                   Map<K, V>,
+                                   W> applyForSingleton(
+          runner,
+          input,
+          new ToMapDoFn<K, V, W>(windowCoder),
+          true,
+          defaultValue,
+          finalValueCoder);
+    }
+  }
+
+  /**
+   * Specialized implementation for
+   * {@link org.apache.beam.sdk.transforms.View.AsMultimap View.AsMultimap} for the
+   * Dataflow runner in batch mode.
+   *
+   * <p>Creates a set of {@code Ism} files sharded by the hash of the key's byte
+   * representation. Each record is structured as follows:
+   * <ul>
+   *   <li>Key 1: User key K</li>
+   *   <li>Key 2: Window</li>
+   *   <li>Key 3: Index offset for a given key and window.</li>
+   *   <li>Value: Windowed value</li>
+   * </ul>
+   *
+   * <p>Alongside the data records, there are the following metadata records:
+   * <ul>
+   *   <li>Key 1: Metadata Key</li>
+   *   <li>Key 2: Window</li>
+   *   <li>Key 3: Index [0, size of map]</li>
+   *   <li>Value: variable length long byte representation of size of map if index is 0,
+   *              otherwise the byte representation of a key</li>
+   * </ul>
+   * The {@code [META, Window, 0]} record stores the number of unique keys per window, while
+   * {@code [META, Window, i]}  for {@code i} in {@code [1, size of map]} stores a the users key.
+   * This allows for one to access the size of the map by looking at {@code [META, Window, 0]}
+   * and iterate over all the keys by accessing {@code [META, Window, i]} for {@code i} in
+   * {@code [1, size of map]}.
+   *
+   * <p>Note that in the case of a non-deterministic key coder, we fallback to using
+   * {@link org.apache.beam.sdk.transforms.View.AsSingleton View.AsSingleton} printing
+   * a warning to users to specify a deterministic key coder.
+   */
+  static class BatchViewAsMultimap<K, V>
+      extends PTransform<PCollection<KV<K, V>>, PCollectionView<Map<K, Iterable<V>>>> {
+    /**
+     * A {@link PTransform} that groups elements by the hash of window's byte representation
+     * if the input {@link PCollection} is not within the global window. Otherwise by the hash
+     * of the window and key's byte representation. This {@link PTransform} also sorts
+     * the values by the combination of the window and key's byte representations.
+     */
+    private static class GroupByKeyHashAndSortByKeyAndWindow<K, V, W extends BoundedWindow>
+        extends PTransform<PCollection<KV<K, V>>,
+                           PCollection<KV<Integer, Iterable<KV<KV<K, W>, WindowedValue<V>>>>>> {
+
+      @SystemDoFnInternal
+      private static class GroupByKeyHashAndSortByKeyAndWindowDoFn<K, V, W>
+          extends DoFn<KV<K, V>, KV<Integer, KV<KV<K, W>, WindowedValue<V>>>>
+          implements DoFn.RequiresWindowAccess {
+
+        private final IsmRecordCoder<?> coder;
+        private GroupByKeyHashAndSortByKeyAndWindowDoFn(IsmRecordCoder<?> coder) {
+          this.coder = coder;
+        }
+
+        @Override
+        public void processElement(ProcessContext c) throws Exception {
+          @SuppressWarnings("unchecked")
+          W window = (W) c.window();
+
+          c.output(
+              KV.of(coder.hash(ImmutableList.of(c.element().getKey())),
+                  KV.of(KV.of(c.element().getKey(), window),
+                      WindowedValue.of(
+                          c.element().getValue(),
+                          c.timestamp(),
+                          (BoundedWindow) window,
+                          c.pane()))));
+        }
+      }
+
+      private final IsmRecordCoder<?> coder;
+      public GroupByKeyHashAndSortByKeyAndWindow(IsmRecordCoder<?> coder) {
+        this.coder = coder;
+      }
+
+      @Override
+      public PCollection<KV<Integer, Iterable<KV<KV<K, W>, WindowedValue<V>>>>>
+          apply(PCollection<KV<K, V>> input) {
+
+        @SuppressWarnings("unchecked")
+        Coder<W> windowCoder = (Coder<W>)
+            input.getWindowingStrategy().getWindowFn().windowCoder();
+        @SuppressWarnings("unchecked")
+        KvCoder<K, V> inputCoder = (KvCoder<K, V>) input.getCoder();
+
+        PCollection<KV<Integer, KV<KV<K, W>, WindowedValue<V>>>> keyedByHash;
+        keyedByHash = input.apply(
+            ParDo.of(new GroupByKeyHashAndSortByKeyAndWindowDoFn<K, V, W>(coder)));
+        keyedByHash.setCoder(
+            KvCoder.of(
+                VarIntCoder.of(),
+                KvCoder.of(KvCoder.of(inputCoder.getKeyCoder(), windowCoder),
+                    FullWindowedValueCoder.of(inputCoder.getValueCoder(), windowCoder))));
+
+        return keyedByHash.apply(
+            new GroupByKeyAndSortValuesOnly<Integer, KV<K, W>, WindowedValue<V>>());
+      }
+    }
+
+    /**
+     * A {@link DoFn} which creates {@link IsmRecord}s comparing successive elements windows
+     * and keys to locate window and key boundaries. The main output {@link IsmRecord}s have:
+     * <ul>
+     *   <li>Key 1: Window</li>
+     *   <li>Key 2: User key K</li>
+     *   <li>Key 3: Index offset for a given key and window.</li>
+     *   <li>Value: Windowed value</li>
+     * </ul>
+     *
+     * <p>Additionally, we output all the unique keys per window seen to {@code outputForEntrySet}
+     * and the unique key count per window to {@code outputForSize}.
+     *
+     * <p>Finally, if this DoFn has been requested to perform unique key checking, it will
+     * throw an {@link IllegalStateException} if more than one key per window is found.
+     */
+    static class ToIsmRecordForMapLikeDoFn<K, V, W extends BoundedWindow>
+        extends DoFn<KV<Integer, Iterable<KV<KV<K, W>, WindowedValue<V>>>>,
+                     IsmRecord<WindowedValue<V>>> {
+
+      private final TupleTag<KV<Integer, KV<W, Long>>> outputForSize;
+      private final TupleTag<KV<Integer, KV<W, K>>> outputForEntrySet;
+      private final Coder<W> windowCoder;
+      private final Coder<K> keyCoder;
+      private final IsmRecordCoder<WindowedValue<V>> ismCoder;
+      private final boolean uniqueKeysExpected;
+      ToIsmRecordForMapLikeDoFn(
+          TupleTag<KV<Integer, KV<W, Long>>> outputForSize,
+          TupleTag<KV<Integer, KV<W, K>>> outputForEntrySet,
+          Coder<W> windowCoder,
+          Coder<K> keyCoder,
+          IsmRecordCoder<WindowedValue<V>> ismCoder,
+          boolean uniqueKeysExpected) {
+        this.outputForSize = outputForSize;
+        this.outputForEntrySet = outputForEntrySet;
+        this.windowCoder = windowCoder;
+        this.keyCoder = keyCoder;
+        this.ismCoder = ismCoder;
+        this.uniqueKeysExpected = uniqueKeysExpected;
+      }
+
+      @Override
+      public void processElement(ProcessContext c) throws Exception {
+        long currentKeyIndex = 0;
+        // We use one based indexing while counting
+        long currentUniqueKeyCounter = 1;
+        Iterator<KV<KV<K, W>, WindowedValue<V>>> iterator = c.element().getValue().iterator();
+
+        KV<KV<K, W>, WindowedValue<V>> currentValue = iterator.next();
+        Object currentKeyStructuralValue =
+            keyCoder.structuralValue(currentValue.getKey().getKey());
+        Object currentWindowStructuralValue =
+            windowCoder.structuralValue(currentValue.getKey().getValue());
+
+        while (iterator.hasNext()) {
+          KV<KV<K, W>, WindowedValue<V>> nextValue = iterator.next();
+          Object nextKeyStructuralValue =
+              keyCoder.structuralValue(nextValue.getKey().getKey());
+          Object nextWindowStructuralValue =
+              windowCoder.structuralValue(nextValue.getKey().getValue());
+
+          outputDataRecord(c, currentValue, currentKeyIndex);
+
+          final long nextKeyIndex;
+          final long nextUniqueKeyCounter;
+
+          // Check to see if its a new window
+          if (!currentWindowStructuralValue.equals(nextWindowStructuralValue)) {
+            // The next value is a new window, so we output for size the number of unique keys
+            // seen and the last key of the window. We also reset the next key index the unique
+            // key counter.
+            outputMetadataRecordForSize(c, currentValue, currentUniqueKeyCounter);
+            outputMetadataRecordForEntrySet(c, currentValue);
+
+            nextKeyIndex = 0;
+            nextUniqueKeyCounter = 1;
+          } else if (!currentKeyStructuralValue.equals(nextKeyStructuralValue)){
+            // It is a new key within the same window so output the key for the entry set,
+            // reset the key index and increase the count of unique keys seen within this window.
+            outputMetadataRecordForEntrySet(c, currentValue);
+
+            nextKeyIndex = 0;
+            nextUniqueKeyCounter = currentUniqueKeyCounter + 1;
+          } else if (!uniqueKeysExpected) {
+            // It is not a new key so we don't have to output the number of elements in this
+            // window or increase the unique key counter. All we do is increase the key index.
+
+            nextKeyIndex = currentKeyIndex + 1;
+            nextUniqueKeyCounter = currentUniqueKeyCounter;
+          } else {
+            throw new IllegalStateException(String.format(
+                "Unique keys are expected but found key %s with values %s and %s in window %s.",
+                currentValue.getKey().getKey(),
+                currentValue.getValue().getValue(),
+                nextValue.getValue().getValue(),
+                currentValue.getKey().getValue()));
+          }
+
+          currentValue = nextValue;
+          currentWindowStructuralValue = nextWindowStructuralValue;
+          currentKeyStructuralValue = nextKeyStructuralValue;
+          currentKeyIndex = nextKeyIndex;
+          currentUniqueKeyCounter = nextUniqueKeyCounter;
+        }
+
+        outputDataRecord(c, currentValue, currentKeyIndex);
+        outputMetadataRecordForSize(c, currentValue, currentUniqueKeyCounter);
+        // The last value for this hash is guaranteed to be at a window boundary
+        // so we output a record with the number of unique keys seen.
+        outputMetadataRecordForEntrySet(c, currentValue);
+      }
+
+      /** This outputs the data record. */
+      private void outputDataRecord(
+          ProcessContext c, KV<KV<K, W>, WindowedValue<V>> value, long keyIndex) {
+        IsmRecord<WindowedValue<V>> ismRecord = IsmRecord.of(
+            ImmutableList.of(
+                value.getKey().getKey(),
+                value.getKey().getValue(),
+                keyIndex),
+            value.getValue());
+        c.output(ismRecord);
+      }
+
+      /**
+       * This outputs records which will be used to compute the number of keys for a given window.
+       */
+      private void outputMetadataRecordForSize(
+          ProcessContext c, KV<KV<K, W>, WindowedValue<V>> value, long uniqueKeyCount) {
+        c.sideOutput(outputForSize,
+            KV.of(ismCoder.hash(ImmutableList.of(IsmFormat.getMetadataKey(),
+                                                 value.getKey().getValue())),
+                KV.of(value.getKey().getValue(), uniqueKeyCount)));
+      }
+
+      /** This outputs records which will be used to construct the entry set. */
+      private void outputMetadataRecordForEntrySet(
+          ProcessContext c, KV<KV<K, W>, WindowedValue<V>> value) {
+        c.sideOutput(outputForEntrySet,
+            KV.of(ismCoder.hash(ImmutableList.of(IsmFormat.getMetadataKey(),
+                                                 value.getKey().getValue())),
+                KV.of(value.getKey().getValue(), value.getKey().getKey())));
+      }
+    }
+
+    /**
+     * A {@link DoFn} which outputs a metadata {@link IsmRecord} per window of:
+       * <ul>
+       *   <li>Key 1: META key</li>
+       *   <li>Key 2: window</li>
+       *   <li>Key 3: 0L (constant)</li>
+       *   <li>Value: sum of values for window</li>
+       * </ul>
+       *
+       * <p>This {@link DoFn} is meant to be used to compute the number of unique keys
+       * per window for map and multimap side inputs.
+       */
+    static class ToIsmMetadataRecordForSizeDoFn<K, V, W extends BoundedWindow>
+        extends DoFn<KV<Integer, Iterable<KV<W, Long>>>, IsmRecord<WindowedValue<V>>> {
+      private final Coder<W> windowCoder;
+      ToIsmMetadataRecordForSizeDoFn(Coder<W> windowCoder) {
+        this.windowCoder = windowCoder;
+      }
+
+      @Override
+      public void processElement(ProcessContext c) throws Exception {
+        Iterator<KV<W, Long>> iterator = c.element().getValue().iterator();
+        KV<W, Long> currentValue = iterator.next();
+        Object currentWindowStructuralValue = windowCoder.structuralValue(currentValue.getKey());
+        long size = 0;
+        while (iterator.hasNext()) {
+          KV<W, Long> nextValue = iterator.next();
+          Object nextWindowStructuralValue = windowCoder.structuralValue(nextValue.getKey());
+
+          size += currentValue.getValue();
+          if (!currentWindowStructuralValue.equals(nextWindowStructuralValue)) {
+            c.output(IsmRecord.<WindowedValue<V>>meta(
+                ImmutableList.of(IsmFormat.getMetadataKey(), currentValue.getKey(), 0L),
+                CoderUtils.encodeToByteArray(VarLongCoder.of(), size)));
+            size = 0;
+          }
+
+          currentValue = nextValue;
+          currentWindowStructuralValue = nextWindowStructuralValue;
+        }
+
+        size += currentValue.getValue();
+        // Output the final value since it is guaranteed to be on a window boundary.
+        c.output(IsmRecord.<WindowedValue<V>>meta(
+            ImmutableList.of(IsmFormat.getMetadataKey(), currentValue.getKey(), 0L),
+            CoderUtils.encodeToByteArray(VarLongCoder.of(), size)));
+      }
+    }
+
+    /**
+     * A {@link DoFn} which outputs a metadata {@link IsmRecord} per window and key pair of:
+       * <ul>
+       *   <li>Key 1: META key</li>
+       *   <li>Key 2: window</li>
+       *   <li>Key 3: index offset (1-based index)</li>
+       *   <li>Value: key</li>
+       * </ul>
+       *
+       * <p>This {@link DoFn} is meant to be used to output index to key records
+       * per window for map and multimap side inputs.
+       */
+    static class ToIsmMetadataRecordForKeyDoFn<K, V, W extends BoundedWindow>
+        extends DoFn<KV<Integer, Iterable<KV<W, K>>>, IsmRecord<WindowedValue<V>>> {
+
+      private final Coder<K> keyCoder;
+      private final Coder<W> windowCoder;
+      ToIsmMetadataRecordForKeyDoFn(Coder<K> keyCoder, Coder<W> windowCoder) {
+        this.keyCoder = keyCoder;
+        this.windowCoder = windowCoder;
+      }
+
+      @Override
+      public void processElement(ProcessContext c) throws Exception {
+        Iterator<KV<W, K>> iterator = c.element().getValue().iterator();
+        KV<W, K> currentValue = iterator.next();
+        Object currentWindowStructuralValue = windowCoder.structuralValue(currentValue.getKey());
+        long elementsInWindow = 1;
+        while (iterator.hasNext()) {
+          KV<W, K> nextValue = iterator.next();
+          Object nextWindowStructuralValue = windowCoder.structuralValue(nextValue.getKey());
+
+          c.output(IsmRecord.<WindowedValue<V>>meta(
+              ImmutableList.of(IsmFormat.getMetadataKey(), currentValue.getKey(), elementsInWindow),
+              CoderUtils.encodeToByteArray(keyCoder, currentValue.getValue())));
+          elementsInWindow += 1;
+
+          if (!currentWindowStructuralValue.equals(nextWindowStructuralValue)) {
+            elementsInWindow = 1;
+          }
+
+          currentValue = nextValue;
+          currentWindowStructuralValue = nextWindowStructuralValue;
+        }
+
+        // Output the final value since it is guaranteed to be on a window boundary.
+        c.output(IsmRecord.<WindowedValue<V>>meta(
+            ImmutableList.of(IsmFormat.getMetadataKey(), currentValue.getKey(), elementsInWindow),
+            CoderUtils.encodeToByteArray(keyCoder, currentValue.getValue())));
+      }
+    }
+
+    /**
+     * A {@link DoFn} which partitions sets of elements by window boundaries. Within each
+     * partition, the set of elements is transformed into a {@link TransformedMap}.
+     * The transformed {@code Map<K, Iterable<V>>} is backed by a
+     * {@code Map<K, Iterable<WindowedValue<V>>>} and contains a function
+     * {@code Iterable<WindowedValue<V>> -> Iterable<V>}.
+     *
+     * <p>Outputs {@link IsmRecord}s having:
+     * <ul>
+     *   <li>Key 1: Window</li>
+     *   <li>Value: Transformed map containing a transform that removes the encapsulation
+     *              of the window around each value,
+     *              {@code Map<K, Iterable<WindowedValue<V>>> -> Map<K, Iterable<V>>}.</li>
+     * </ul>
+     */
+    static class ToMultimapDoFn<K, V, W extends BoundedWindow>
+        extends DoFn<KV<Integer, Iterable<KV<W, WindowedValue<KV<K, V>>>>>,
+                     IsmRecord<WindowedValue<TransformedMap<K,
+                                                            Iterable<WindowedValue<V>>,
+                                                            Iterable<V>>>>> {
+
+      private final Coder<W> windowCoder;
+      ToMultimapDoFn(Coder<W> windowCoder) {
+        this.windowCoder = windowCoder;
+      }
+
+      @Override
+      public void processElement(ProcessContext c)
+          throws Exception {
+        Optional<Object> previousWindowStructuralValue = Optional.absent();
+        Optional<W> previousWindow = Optional.absent();
+        Multimap<K, WindowedValue<V>> multimap = HashMultimap.create();
+        for (KV<W, WindowedValue<KV<K, V>>> kv : c.element().getValue()) {
+          Object currentWindowStructuralValue = windowCoder.structuralValue(kv.getKey());
+          if (previousWindowStructuralValue.isPresent()
+              && !previousWindowStructuralValue.get().equals(currentWindowStructuralValue)) {
+            // Construct the transformed map containing all the elements since we
+            // are at a window boundary.
+            @SuppressWarnings({"unchecked", "rawtypes"})
+            Map<K, Iterable<WindowedValue<V>>> resultMap = (Map) multimap.asMap();
+            c.output(IsmRecord.<WindowedValue<TransformedMap<K,
+                                                             Iterable<WindowedValue<V>>,
+                                                             Iterable<V>>>>of(
+                ImmutableList.of(previousWindow.get()),
+                valueInEmptyWindows(
+                    new TransformedMap<>(
+                        IterableWithWindowedValuesToIterable.<V>of(), resultMap))));
+            multimap = HashMultimap.create();
+          }
+
+          multimap.put(kv.getValue().getValue().getKey(),
+                       kv.getValue().withValue(kv.getValue().getValue().getValue()));
+          previousWindowStructuralValue = Optional.of(currentWindowStructuralValue);
+          previousWindow = Optional.of(kv.getKey());
+        }
+
+        // The last value for this hash is guaranteed to be at a window boundary
+        // so we output a transformed map containing all the elements since the last
+        // window boundary.
+        @SuppressWarnings({"unchecked", "rawtypes"})
+        Map<K, Iterable<WindowedValue<V>>> resultMap = (Map) multimap.asMap();
+        c.output(IsmRecord.<WindowedValue<TransformedMap<K,
+                                                         Iterable<WindowedValue<V>>,
+                                                         Iterable<V>>>>of(
+            ImmutableList.of(previousWindow.get()),
+            valueInEmptyWindows(
+                new TransformedMap<>(IterableWithWindowedValuesToIterable.<V>of(), resultMap))));
+      }
+    }
+
+    private final DataflowRunner runner;
+    /**
+     * Builds an instance of this class from the overridden transform.
+     */
+    @SuppressWarnings("unused") // used via reflection in DataflowRunner#apply()
+    public BatchViewAsMultimap(DataflowRunner runner, View.AsMultimap<K, V> transform) {
+      this.runner = runner;
+    }
+
+    @Override
+    public PCollectionView<Map<K, Iterable<V>>> apply(PCollection<KV<K, V>> input) {
+      return this.<BoundedWindow>applyInternal(input);
+    }
+
+    private <W extends BoundedWindow> PCollectionView<Map<K, Iterable<V>>>
+        applyInternal(PCollection<KV<K, V>> input) {
+      @SuppressWarnings({"rawtypes", "unchecked"})
+      KvCoder<K, V> inputCoder = (KvCoder) input.getCoder();
+      try {
+        PCollectionView<Map<K, Iterable<V>>> view = PCollectionViews.multimapView(
+            input.getPipeline(), input.getWindowingStrategy(), inputCoder);
+
+        return applyForMapLike(runner, input, view, false /* unique keys not expected */);
+      } catch (NonDeterministicException e) {
+        runner.recordViewUsesNonDeterministicKeyCoder(this);
+
+        // Since the key coder is not deterministic, we convert the map into a singleton
+        // and return a singleton view equivalent.
+        return applyForSingletonFallback(input);
+      }
+    }
+
+    /** Transforms the input {@link PCollection} into a singleton {@link Map} per window. */
+    private <W extends BoundedWindow> PCollectionView<Map<K, Iterable<V>>>
+        applyForSingletonFallback(PCollection<KV<K, V>> input) {
+      @SuppressWarnings("unchecked")
+      Coder<W> windowCoder = (Coder<W>)
+          input.getWindowingStrategy().getWindowFn().windowCoder();
+
+      @SuppressWarnings({"rawtypes", "unchecked"})
+      KvCoder<K, V> inputCoder = (KvCoder) input.getCoder();
+
+      @SuppressWarnings({"unchecked", "rawtypes"})
+      Coder<Function<Iterable<WindowedValue<V>>, Iterable<V>>> transformCoder =
+          (Coder) SerializableCoder.of(IterableWithWindowedValuesToIterable.class);
+
+      Coder<TransformedMap<K, Iterable<WindowedValue<V>>, Iterable<V>>> finalValueCoder =
+          TransformedMapCoder.of(
+          transformCoder,
+          MapCoder.of(
+              inputCoder.getKeyCoder(),
+              IterableCoder.of(
+                  FullWindowedValueCoder.of(inputCoder.getValueCoder(), windowCoder))));
+
+      TransformedMap<K, Iterable<WindowedValue<V>>, Iterable<V>> defaultValue =
+          new TransformedMap<>(
+              IterableWithWindowedValuesToIterable.<V>of(),
+              ImmutableMap.<K, Iterable<WindowedValue<V>>>of());
+
+      return BatchViewAsSingleton.<KV<K, V>,
+                                   TransformedMap<K, Iterable<WindowedValue<V>>, Iterable<V>>,
+                                   Map<K, Iterable<V>>,
+                                   W> applyForSingleton(
+          runner,
+          input,
+          new ToMultimapDoFn<K, V, W>(windowCoder),
+          true,
+          defaultValue,
+          finalValueCoder);
+    }
+
+    private static <K, V, W extends BoundedWindow, ViewT> PCollectionView<ViewT> applyForMapLike(
+        DataflowRunner runner,
+        PCollection<KV<K, V>> input,
+        PCollectionView<ViewT> view,
+        boolean uniqueKeysExpected) throws NonDeterministicException {
+
+      @SuppressWarnings("unchecked")
+      Coder<W> windowCoder = (Coder<W>)
+          input.getWindowingStrategy().getWindowFn().windowCoder();
+
+      @SuppressWarnings({"rawtypes", "unchecked"})
+      KvCoder<K, V> inputCoder = (KvCoder) input.getCoder();
+
+      // If our key coder is deterministic, we can use the key portion of each KV
+      // part of a composite key containing the window , key and index.
+      inputCoder.getKeyCoder().verifyDeterministic();
+
+      IsmRecordCoder<WindowedValue<V>> ismCoder =
+          coderForMapLike(windowCoder, inputCoder.getKeyCoder(), inputCoder.getValueCoder());
+
+      // Create the various output tags representing the main output containing the data stream
+      // and the side outputs containing the metadata about the size and entry set.
+      TupleTag<IsmRecord<WindowedValue<V>>> mainOutputTag = new TupleTag<>();
+      TupleTag<KV<Integer, KV<W, Long>>> outputForSizeTag = new TupleTag<>();
+      TupleTag<KV<Integer, KV<W, K>>> outputForEntrySetTag = new TupleTag<>();
+
+      // Process all the elements grouped by key hash, and sorted by key and then window
+      // outputting to all the outputs defined above.
+      PCollectionTuple outputTuple = input
+           .apply("GBKaSVForData", new GroupByKeyHashAndSortByKeyAndWindow<K, V, W>(ismCoder))
+           .apply(ParDo.of(new ToIsmRecordForMapLikeDoFn<K, V, W>(
+                   outputForSizeTag, outputForEntrySetTag,
+                   windowCoder, inputCoder.getKeyCoder(), ismCoder, uniqueKeysExpected))
+                       .withOutputTags(mainOutputTag,
+                                       TupleTagList.of(
+                                           ImmutableList.<TupleTag<?>>of(outputForSizeTag,
+                                                                         outputForEntrySetTag))));
+
+      // Set the coder on the main data output.
+      PCollection<IsmRecord<WindowedValue<V>>> perHashWithReifiedWindows =
+          outputTuple.get(mainOutputTag);
+      perHashWithReifiedWindows.setCoder(ismCoder);
+
+      // Set the coder on the metadata output for size and process the entries
+      // producing a [META, Window, 0L] record per window storing the number of unique keys
+      // for each window.
+      PCollection<KV<Integer, KV<W, Long>>> outputForSize = outputTuple.get(outputForSizeTag);
+      outputForSize.setCoder(
+          KvCoder.of(VarIntCoder.of(),
+                     KvCoder.of(windowCoder, VarLongCoder.of())));
+      PCollection<IsmRecord<WindowedValue<V>>> windowMapSizeMetadata = outputForSize
+          .apply("GBKaSVForSize", new GroupByKeyAndSortValuesOnly<Integer, W, Long>())
+          .apply(ParDo.of(new ToIsmMetadataRecordForSizeDoFn<K, V, W>(windowCoder)));
+      windowMapSizeMetadata.setCoder(ismCoder);
+
+      // Set the coder on the metadata output destined to build the entry set and process the
+      // entries producing a [META, Window, Index] record per window key pair storing the key.
+      PCollection<KV<Integer, KV<W, K>>> outputForEntrySet =
+          outputTuple.get(outputForEntrySetTag);
+      outputForEntrySet.setCoder(
+          KvCoder.of(VarIntCoder.of(),
+                     KvCoder.of(windowCoder, inputCoder.getKeyCoder())));
+      PCollection<IsmRecord<WindowedValue<V>>> windowMapKeysMetadata = outputForEntrySet
+          .apply("GBKaSVForKeys", new GroupByKeyAndSortValuesOnly<Integer, W, K>())
+          .apply(ParDo.of(
+              new ToIsmMetadataRecordForKeyDoFn<K, V, W>(inputCoder.getKeyCoder(), windowCoder)));
+      windowMapKeysMetadata.setCoder(ismCoder);
+
+      // Set that all these outputs should be materialized using an indexed format.
+      runner.addPCollectionRequiringIndexedFormat(perHashWithReifiedWindows);
+      runner.addPCollectionRequiringIndexedFormat(windowMapSizeMetadata);
+      runner.addPCollectionRequiringIndexedFormat(windowMapKeysMetadata);
+
+      PCollectionList<IsmRecord<WindowedValue<V>>> outputs =
+          PCollectionList.of(ImmutableList.of(
+              perHashWithReifiedWindows, windowMapSizeMetadata, windowMapKeysMetadata));
+
+      return Pipeline.applyTransform(outputs,
+                                     Flatten.<IsmRecord<WindowedValue<V>>>pCollections())
+          .apply(CreatePCollectionView.<IsmRecord<WindowedValue<V>>,
+                                        ViewT>of(view));
+    }
+
+    @Override
+    protected String getKindString() {
+      return "BatchViewAsMultimap";
+    }
+
+    static <V> IsmRecordCoder<WindowedValue<V>> coderForMapLike(
+        Coder<? extends BoundedWindow> windowCoder, Coder<?> keyCoder, Coder<V> valueCoder) {
+      // TODO: swap to use a variable length long coder which has values which compare
+      // the same as their byte representation compare lexicographically within the key coder
+      return IsmRecordCoder.of(
+          1, // We use only the key for hashing when producing value records
+          2, // Since the key is not present, we add the window to the hash when
+             // producing metadata records
+          ImmutableList.of(
+              MetadataKeyCoder.of(keyCoder),
+              windowCoder,
+              BigEndianLongCoder.of()),
+          FullWindowedValueCoder.of(valueCoder, windowCoder));
+    }
+  }
+
+  /**
+   * A {@code Map<K, V2>} backed by a {@code Map<K, V1>} and a function that transforms
+   * {@code V1 -> V2}.
+   */
+  static class TransformedMap<K, V1, V2>
+      extends ForwardingMap<K, V2> {
+    private final Function<V1, V2> transform;
+    private final Map<K, V1> originalMap;
+    private final Map<K, V2> transformedMap;
+
+    private TransformedMap(Function<V1, V2> transform, Map<K, V1> originalMap) {
+      this.transform = transform;
+      this.originalMap = Collections.unmodifiable

<TRUNCATED>


Mime
View raw message