Examples with BatchAggregator co.cask.cdap.etl.api.batch.BatchAggregator used on opensource projects

Example 1 with BatchAggregator

use of co.cask.cdap.etl.api.batch.BatchAggregator in project cdap by caskdata.

the class ETLMapReduce method initialize.

@Override
@TransactionPolicy(TransactionControl.EXPLICIT)
public void initialize() throws Exception {
    final MapReduceContext context = getContext();
    Map<String, String> properties = context.getSpecification().getProperties();
    if (Boolean.valueOf(properties.get(Constants.STAGE_LOGGING_ENABLED))) {
        LogStageInjector.start();
    }
    PipelineRuntime pipelineRuntime = new PipelineRuntime(context, mrMetrics);
    List<Finisher> finishers = new ArrayList<>();
    final Job job = context.getHadoopJob();
    final Configuration hConf = job.getConfiguration();
    hConf.setBoolean("mapreduce.map.speculative", false);
    hConf.setBoolean("mapreduce.reduce.speculative", false);
    // plugin name -> runtime args for that plugin
    MacroEvaluator evaluator = new DefaultMacroEvaluator(pipelineRuntime.getArguments(), context.getLogicalStartTime(), context, context.getNamespace());
    BatchPhaseSpec phaseSpec = GSON.fromJson(properties.get(Constants.PIPELINEID), BatchPhaseSpec.class);
    Set<String> connectorDatasets = GSON.fromJson(properties.get(Constants.CONNECTOR_DATASETS), CONNECTOR_DATASETS_TYPE);
    for (Map.Entry<String, String> pipelineProperty : phaseSpec.getPipelineProperties().entrySet()) {
        hConf.set(pipelineProperty.getKey(), pipelineProperty.getValue());
    }
    final PipelinePhase phase = phaseSpec.getPhase();
    PipelinePluginInstantiator pluginInstantiator = new PipelinePluginInstantiator(context, mrMetrics, phaseSpec, new MultiConnectorFactory());
    // should never happen if planner is correct
    Set<StageSpec> reducers = phaseSpec.getPhase().getStagesOfType(BatchAggregator.PLUGIN_TYPE, BatchJoiner.PLUGIN_TYPE);
    if (reducers.size() > 1) {
        Iterator<StageSpec> reducerIter = reducers.iterator();
        StringBuilder reducersStr = new StringBuilder(reducerIter.next().getName());
        while (reducerIter.hasNext()) {
            reducersStr.append(",");
            reducersStr.append(reducerIter.next().getName());
        }
        throw new IllegalStateException("Found multiple reducers ( " + reducersStr + " ) in the same pipeline phase. " + "This means there was a bug in planning the pipeline when it was deployed. ");
    }
    job.setMapperClass(ETLMapper.class);
    if (reducers.isEmpty()) {
        job.setNumReduceTasks(0);
    } else {
        job.setReducerClass(ETLReducer.class);
    }
    final Map<String, SinkOutput> sinkOutputs = new HashMap<>();
    final Map<String, String> inputAliasToStage = new HashMap<>();
    // call prepareRun on each stage in order so that any arguments set by a stage will be visible to subsequent stages
    for (final String stageName : phase.getDag().getTopologicalOrder()) {
        final StageSpec stageSpec = phase.getStage(stageName);
        String pluginType = stageSpec.getPluginType();
        boolean isConnectorSource = Constants.Connector.PLUGIN_TYPE.equals(pluginType) && phase.getSources().contains(stageName);
        boolean isConnectorSink = Constants.Connector.PLUGIN_TYPE.equals(pluginType) && phase.getSinks().contains(stageName);
        SubmitterPlugin submitterPlugin = null;
        if (BatchSource.PLUGIN_TYPE.equals(pluginType) || isConnectorSource) {
            BatchConfigurable<BatchSourceContext> batchSource = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<MapReduceBatchContext> contextProvider = new MapReduceBatchContextProvider(context, pipelineRuntime, stageSpec, connectorDatasets);
            submitterPlugin = new SubmitterPlugin<>(stageName, context, batchSource, contextProvider, new SubmitterPlugin.PrepareAction<MapReduceBatchContext>() {

                @Override
                public void act(MapReduceBatchContext sourceContext) {
                    for (String inputAlias : sourceContext.getInputNames()) {
                        inputAliasToStage.put(inputAlias, stageName);
                    }
                }
            });
        } else if (BatchSink.PLUGIN_TYPE.equals(pluginType) || AlertPublisher.PLUGIN_TYPE.equals(pluginType) || isConnectorSink) {
            BatchConfigurable<BatchSinkContext> batchSink = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<MapReduceBatchContext> contextProvider = new MapReduceBatchContextProvider(context, pipelineRuntime, stageSpec, connectorDatasets);
            submitterPlugin = new SubmitterPlugin<>(stageName, context, batchSink, contextProvider, new SubmitterPlugin.PrepareAction<MapReduceBatchContext>() {

                @Override
                public void act(MapReduceBatchContext sinkContext) {
                    sinkOutputs.put(stageName, new SinkOutput(sinkContext.getOutputNames()));
                }
            });
        } else if (Transform.PLUGIN_TYPE.equals(pluginType)) {
            Transform<?, ?> transform = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<MapReduceBatchContext> contextProvider = new MapReduceBatchContextProvider(context, pipelineRuntime, stageSpec, connectorDatasets);
            submitterPlugin = new SubmitterPlugin<>(stageName, context, transform, contextProvider);
        } else if (BatchAggregator.PLUGIN_TYPE.equals(pluginType)) {
            final BatchAggregator<?, ?, ?> aggregator = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<DefaultAggregatorContext> contextProvider = new AggregatorContextProvider(pipelineRuntime, stageSpec, context.getAdmin());
            submitterPlugin = new SubmitterPlugin<>(stageName, context, aggregator, contextProvider, new SubmitterPlugin.PrepareAction<DefaultAggregatorContext>() {

                @Override
                public void act(DefaultAggregatorContext aggregatorContext) {
                    if (aggregatorContext.getNumPartitions() != null) {
                        job.setNumReduceTasks(aggregatorContext.getNumPartitions());
                    }
                    Class<?> outputKeyClass = aggregatorContext.getGroupKeyClass();
                    Class<?> outputValClass = aggregatorContext.getGroupValueClass();
                    if (outputKeyClass == null) {
                        outputKeyClass = TypeChecker.getGroupKeyClass(aggregator);
                    }
                    if (outputValClass == null) {
                        outputValClass = TypeChecker.getGroupValueClass(aggregator);
                    }
                    hConf.set(MAP_KEY_CLASS, outputKeyClass.getName());
                    hConf.set(MAP_VAL_CLASS, outputValClass.getName());
                    job.setMapOutputKeyClass(getOutputKeyClass(stageName, outputKeyClass));
                    job.setMapOutputValueClass(getOutputValClass(stageName, outputValClass));
                }
            });
        } else if (BatchJoiner.PLUGIN_TYPE.equals(pluginType)) {
            final BatchJoiner<?, ?, ?> batchJoiner = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<DefaultJoinerContext> contextProvider = new JoinerContextProvider(pipelineRuntime, stageSpec, context.getAdmin());
            submitterPlugin = new SubmitterPlugin<>(stageName, context, batchJoiner, contextProvider, new SubmitterPlugin.PrepareAction<DefaultJoinerContext>() {

                @Override
                public void act(DefaultJoinerContext joinerContext) {
                    if (joinerContext.getNumPartitions() != null) {
                        job.setNumReduceTasks(joinerContext.getNumPartitions());
                    }
                    Class<?> outputKeyClass = joinerContext.getJoinKeyClass();
                    Class<?> inputRecordClass = joinerContext.getJoinInputRecordClass();
                    if (outputKeyClass == null) {
                        outputKeyClass = TypeChecker.getJoinKeyClass(batchJoiner);
                    }
                    if (inputRecordClass == null) {
                        inputRecordClass = TypeChecker.getJoinInputRecordClass(batchJoiner);
                    }
                    hConf.set(MAP_KEY_CLASS, outputKeyClass.getName());
                    hConf.set(MAP_VAL_CLASS, inputRecordClass.getName());
                    job.setMapOutputKeyClass(getOutputKeyClass(stageName, outputKeyClass));
                    getOutputValClass(stageName, inputRecordClass);
                    // for joiner plugin map output is tagged with stageName
                    job.setMapOutputValueClass(TaggedWritable.class);
                }
            });
        }
        if (submitterPlugin != null) {
            submitterPlugin.prepareRun();
            finishers.add(submitterPlugin);
        }
    }
    hConf.set(SINK_OUTPUTS_KEY, GSON.toJson(sinkOutputs));
    hConf.set(INPUT_ALIAS_KEY, GSON.toJson(inputAliasToStage));
    finisher = new CompositeFinisher(finishers);
    job.setMapperClass(ETLMapper.class);
    WorkflowToken token = context.getWorkflowToken();
    if (token != null) {
        for (Map.Entry<String, String> entry : pipelineRuntime.getArguments().getAddedArguments().entrySet()) {
            token.put(entry.getKey(), entry.getValue());
        }
    }
    // token is null when just the mapreduce job is run but not the entire workflow
    // we still want things to work in that case.
    hConf.set(RUNTIME_ARGS_KEY, GSON.toJson(pipelineRuntime.getArguments().asMap()));
}

Example 2 with BatchAggregator

use of co.cask.cdap.etl.api.batch.BatchAggregator in project cdap by caskdata.

the class ETLSpark method initialize.

@Override
@TransactionPolicy(TransactionControl.EXPLICIT)
public void initialize() throws Exception {
    final SparkClientContext context = getContext();
    cleanupFiles = new ArrayList<>();
    List<Finisher> finishers = new ArrayList<>();
    SparkConf sparkConf = new SparkConf();
    sparkConf.set("spark.driver.extraJavaOptions", "-XX:MaxPermSize=256m");
    sparkConf.set("spark.executor.extraJavaOptions", "-XX:MaxPermSize=256m");
    sparkConf.set("spark.speculation", "false");
    context.setSparkConf(sparkConf);
    Map<String, String> properties = context.getSpecification().getProperties();
    BatchPhaseSpec phaseSpec = GSON.fromJson(properties.get(Constants.PIPELINEID), BatchPhaseSpec.class);
    for (Map.Entry<String, String> pipelineProperty : phaseSpec.getPipelineProperties().entrySet()) {
        sparkConf.set(pipelineProperty.getKey(), pipelineProperty.getValue());
    }
    MacroEvaluator evaluator = new DefaultMacroEvaluator(new BasicArguments(context), context.getLogicalStartTime(), context, context.getNamespace());
    final SparkBatchSourceFactory sourceFactory = new SparkBatchSourceFactory();
    final SparkBatchSinkFactory sinkFactory = new SparkBatchSinkFactory();
    final Map<String, Integer> stagePartitions = new HashMap<>();
    PluginContext pluginContext = new SparkPipelinePluginContext(context, context.getMetrics(), phaseSpec.isStageLoggingEnabled(), phaseSpec.isProcessTimingEnabled());
    PipelinePluginInstantiator pluginInstantiator = new PipelinePluginInstantiator(pluginContext, context.getMetrics(), phaseSpec, new SingleConnectorFactory());
    final PipelineRuntime pipelineRuntime = new PipelineRuntime(context);
    final Admin admin = context.getAdmin();
    PipelinePhase phase = phaseSpec.getPhase();
    // go through in topological order so that arguments set by one stage are seen by stages after it
    for (final String stageName : phase.getDag().getTopologicalOrder()) {
        final StageSpec stageSpec = phase.getStage(stageName);
        String pluginType = stageSpec.getPluginType();
        boolean isConnectorSource = Constants.Connector.PLUGIN_TYPE.equals(pluginType) && phase.getSources().contains(stageName);
        boolean isConnectorSink = Constants.Connector.PLUGIN_TYPE.equals(pluginType) && phase.getSinks().contains(stageName);
        SubmitterPlugin submitterPlugin = null;
        if (BatchSource.PLUGIN_TYPE.equals(pluginType) || isConnectorSource) {
            BatchConfigurable<BatchSourceContext> batchSource = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<BatchSourceContext> contextProvider = new ContextProvider<BatchSourceContext>() {

                @Override
                public BatchSourceContext getContext(DatasetContext datasetContext) {
                    return new SparkBatchSourceContext(sourceFactory, context, pipelineRuntime, datasetContext, stageSpec);
                }
            };
            submitterPlugin = new SubmitterPlugin(stageName, context, batchSource, contextProvider);
        } else if (Transform.PLUGIN_TYPE.equals(pluginType)) {
            Transform transform = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<StageSubmitterContext> contextProvider = new ContextProvider<StageSubmitterContext>() {

                @Override
                public StageSubmitterContext getContext(DatasetContext datasetContext) {
                    return new SparkBatchSourceContext(sourceFactory, context, pipelineRuntime, datasetContext, stageSpec);
                }
            };
            submitterPlugin = new SubmitterPlugin(stageName, context, transform, contextProvider);
        } else if (BatchSink.PLUGIN_TYPE.equals(pluginType) || isConnectorSink) {
            BatchConfigurable<BatchSinkContext> batchSink = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<BatchSinkContext> contextProvider = new ContextProvider<BatchSinkContext>() {

                @Override
                public BatchSinkContext getContext(DatasetContext datasetContext) {
                    return new SparkBatchSinkContext(sinkFactory, context, pipelineRuntime, datasetContext, stageSpec);
                }
            };
            submitterPlugin = new SubmitterPlugin(stageName, context, batchSink, contextProvider);
        } else if (SparkSink.PLUGIN_TYPE.equals(pluginType)) {
            BatchConfigurable<SparkPluginContext> sparkSink = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<SparkPluginContext> contextProvider = new ContextProvider<SparkPluginContext>() {

                @Override
                public SparkPluginContext getContext(DatasetContext datasetContext) {
                    return new BasicSparkPluginContext(context, pipelineRuntime, stageSpec, datasetContext, admin);
                }
            };
            submitterPlugin = new SubmitterPlugin(stageName, context, sparkSink, contextProvider);
        } else if (BatchAggregator.PLUGIN_TYPE.equals(pluginType)) {
            BatchAggregator aggregator = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<DefaultAggregatorContext> contextProvider = new AggregatorContextProvider(pipelineRuntime, stageSpec, admin);
            submitterPlugin = new SubmitterPlugin(stageName, context, aggregator, contextProvider);
        } else if (BatchJoiner.PLUGIN_TYPE.equals(pluginType)) {
            BatchJoiner joiner = pluginInstantiator.newPluginInstance(stageName, evaluator);
            ContextProvider<DefaultJoinerContext> contextProvider = new JoinerContextProvider(pipelineRuntime, stageSpec, admin);
            submitterPlugin = new SubmitterPlugin<>(stageName, context, joiner, contextProvider, new SubmitterPlugin.PrepareAction<DefaultJoinerContext>() {

                @Override
                public void act(DefaultJoinerContext sparkJoinerContext) {
                    stagePartitions.put(stageName, sparkJoinerContext.getNumPartitions());
                }
            });
        }
        if (submitterPlugin != null) {
            submitterPlugin.prepareRun();
            finishers.add(submitterPlugin);
        }
    }
    File configFile = File.createTempFile("HydratorSpark", ".config");
    cleanupFiles.add(configFile);
    try (Writer writer = Files.newBufferedWriter(configFile.toPath(), StandardCharsets.UTF_8)) {
        SparkBatchSourceSinkFactoryInfo sourceSinkInfo = new SparkBatchSourceSinkFactoryInfo(sourceFactory, sinkFactory, stagePartitions);
        writer.write(GSON.toJson(sourceSinkInfo));
    }
    finisher = new CompositeFinisher(finishers);
    context.localize("HydratorSpark.config", configFile.toURI());
    WorkflowToken token = context.getWorkflowToken();
    if (token != null) {
        for (Map.Entry<String, String> entry : pipelineRuntime.getArguments().getAddedArguments().entrySet()) {
            token.put(entry.getKey(), entry.getValue());
        }
    }
}

Example 3 with BatchAggregator

use of co.cask.cdap.etl.api.batch.BatchAggregator in project cdap by caskdata.

the class PipelinePluginContext method wrapPlugin.

private Object wrapPlugin(String pluginId, Object plugin) {
    Caller caller = getCaller(pluginId);
    StageMetrics stageMetrics = new DefaultStageMetrics(metrics, pluginId);
    OperationTimer operationTimer = processTimingEnabled ? new MetricsOperationTimer(stageMetrics) : NoOpOperationTimer.INSTANCE;
    if (plugin instanceof Action) {
        return new WrappedAction((Action) plugin, caller);
    } else if (plugin instanceof BatchSource) {
        return new WrappedBatchSource<>((BatchSource) plugin, caller, operationTimer);
    } else if (plugin instanceof BatchSink) {
        return new WrappedBatchSink<>((BatchSink) plugin, caller, operationTimer);
    } else if (plugin instanceof ErrorTransform) {
        return new WrappedErrorTransform<>((ErrorTransform) plugin, caller, operationTimer);
    } else if (plugin instanceof Transform) {
        return new WrappedTransform<>((Transform) plugin, caller, operationTimer);
    } else if (plugin instanceof BatchAggregator) {
        return new WrappedBatchAggregator<>((BatchAggregator) plugin, caller, operationTimer);
    } else if (plugin instanceof BatchJoiner) {
        return new WrappedBatchJoiner<>((BatchJoiner) plugin, caller, operationTimer);
    } else if (plugin instanceof PostAction) {
        return new WrappedPostAction((PostAction) plugin, caller);
    } else if (plugin instanceof SplitterTransform) {
        return new WrappedSplitterTransform<>((SplitterTransform) plugin, caller, operationTimer);
    }
    return wrapUnknownPlugin(pluginId, plugin, caller);
}