Package org.apache.druid.segment.realtime.appenderator

Class UnifiedIndexerAppenderatorsManager

java.lang.Object

org.apache.druid.segment.realtime.appenderator.UnifiedIndexerAppenderatorsManager

All Implemented Interfaces:

AppenderatorsManager

public class UnifiedIndexerAppenderatorsManager
extends Object
implements AppenderatorsManager

Manages Appenderator instances for the CliIndexer task execution service, which runs all tasks in a single process. This class keeps a map of UnifiedIndexerAppenderatorsManager.DatasourceBundle objects, keyed by datasource name. Each bundle contains: - A per-datasource SinkQuerySegmentWalker (with an associated per-datasource timeline) - A map that associates a taskId with a list of Appenderators created for that task Access to the datasource bundle map and the task->appenderator maps is synchronized. The methods on this class can be called concurrently from multiple task threads. If there are no remaining appenderators for a given datasource, the corresponding bundle will be removed from the bundle map. Appenderators created by this class will use the shared per-datasource SinkQuerySegmentWalkers. The per-datasource SinkQuerySegmentWalkers share a common queryExecutorService. Each task that requests an Appenderator from this AppenderatorsManager will receive a heap memory limit equal to WorkerConfig.globalIngestionHeapLimitBytes evenly divided by WorkerConfig.capacity. This assumes that each task will only ingest to one Appenderator simultaneously. The Appenderators created by this class share an executor pool for IndexMerger persist and merge operations, with concurrent operations limited to `druid.worker.capacity` divided 2. This limit is imposed to reduce overall memory usage.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
class	UnifiedIndexerAppenderatorsManager.DatasourceBundle
static class	UnifiedIndexerAppenderatorsManager.LimitedPoolIndexMerger	This wrapper around IndexMerger limits concurrent calls to the merge/persist methods used by StreamAppenderator with a shared executor service.

Constructor Summary

Constructors

Constructor

Description

UnifiedIndexerAppenderatorsManager(QueryProcessingPool queryProcessingPool, WorkerConfig workerConfig, Cache cache, CacheConfig cacheConfig, CachePopulatorStats cachePopulatorStats, PolicyEnforcer policyEnforcer, com.fasterxml.jackson.databind.ObjectMapper objectMapper, ServiceEmitter serviceEmitter, com.google.inject.Provider<QueryRunnerFactoryConglomerate> queryRunnerFactoryConglomerateProvider)

Method Summary

Modifier and Type	Method	Description
Appenderator	createBatchAppenderatorForTask(String taskId, DataSchema schema, AppenderatorConfig config, TaskDirectory taskDirectory, SegmentGenerationMetrics metrics, DataSegmentPusher dataSegmentPusher, com.fasterxml.jackson.databind.ObjectMapper objectMapper, IndexIO indexIO, IndexMerger indexMerger, RowIngestionMeters rowIngestionMeters, ParseExceptionHandler parseExceptionHandler, CentralizedDatasourceSchemaConfig centralizedDatasourceSchemaConfig)	Creates a BatchAppenderator suitable for batch ingestion with no ability to process queries against the processed data.
Appenderator	createRealtimeAppenderatorForTask(SegmentLoaderConfig segmentLoaderConfig, String taskId, DataSchema schema, AppenderatorConfig config, TaskDirectory taskDirectory, SegmentGenerationMetrics metrics, DataSegmentPusher dataSegmentPusher, com.fasterxml.jackson.databind.ObjectMapper objectMapper, IndexIO indexIO, IndexMerger indexMerger, QueryRunnerFactoryConglomerate conglomerate, DataSegmentAnnouncer segmentAnnouncer, ServiceEmitter emitter, QueryProcessingPool queryProcessingPool, JoinableFactory joinableFactory, Cache cache, CacheConfig cacheConfig, CachePopulatorStats cachePopulatorStats, PolicyEnforcer policyEnforcer, RowIngestionMeters rowIngestionMeters, ParseExceptionHandler parseExceptionHandler, CentralizedDatasourceSchemaConfig centralizedDatasourceSchemaConfig)	Creates an StreamAppenderator suited for realtime ingestion.
Map<String,UnifiedIndexerAppenderatorsManager.DatasourceBundle>	getDatasourceBundles()
<T> QueryRunner<T>	getQueryRunnerForIntervals(Query<T> query, Iterable<org.joda.time.Interval> intervals)	Returns a query runner for the given intervals over the Appenderators managed by this AppenderatorsManager.
<T> QueryRunner<T>	getQueryRunnerForSegments(Query<T> query, Iterable<SegmentDescriptor> specs)	Returns a query runner for the given segment specs over the Appenderators managed by this AppenderatorsManager.
WorkerConfig	getWorkerConfig()
void	removeAppenderatorsForTask(String taskId, String dataSource)	Removes any internal Appenderator-tracking state associated with the provided taskId.
boolean	shouldTaskMakeNodeAnnouncements()	As AppenderatorsManager implementions are service dependent (i.e., Peons and Indexers have different impls), this method allows Tasks to know whether they should announce themselves as nodes and segment servers to the rest of the cluster.
void	shutdown()	Shut down the AppenderatorsManager.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

UnifiedIndexerAppenderatorsManager

@Inject
public UnifiedIndexerAppenderatorsManager(QueryProcessingPool queryProcessingPool,
 WorkerConfig workerConfig,
 Cache cache,
 CacheConfig cacheConfig,
 CachePopulatorStats cachePopulatorStats,
 PolicyEnforcer policyEnforcer,
 com.fasterxml.jackson.databind.ObjectMapper objectMapper,
 ServiceEmitter serviceEmitter,
 com.google.inject.Provider<QueryRunnerFactoryConglomerate> queryRunnerFactoryConglomerateProvider)

Method Details

createRealtimeAppenderatorForTask

public Appenderator createRealtimeAppenderatorForTask(SegmentLoaderConfig segmentLoaderConfig,
 String taskId,
 DataSchema schema,
 AppenderatorConfig config,
 TaskDirectory taskDirectory,
 SegmentGenerationMetrics metrics,
 DataSegmentPusher dataSegmentPusher,
 com.fasterxml.jackson.databind.ObjectMapper objectMapper,
 IndexIO indexIO,
 IndexMerger indexMerger,
 QueryRunnerFactoryConglomerate conglomerate,
 DataSegmentAnnouncer segmentAnnouncer,
 ServiceEmitter emitter,
 QueryProcessingPool queryProcessingPool,
 JoinableFactory joinableFactory,
 Cache cache,
 CacheConfig cacheConfig,
 CachePopulatorStats cachePopulatorStats,
 PolicyEnforcer policyEnforcer,
 RowIngestionMeters rowIngestionMeters,
 ParseExceptionHandler parseExceptionHandler,
 CentralizedDatasourceSchemaConfig centralizedDatasourceSchemaConfig)

Description copied from interface: AppenderatorsManager

Creates an StreamAppenderator suited for realtime ingestion. Note that this method's parameters include objects used for query processing. Intermediary segments are persisted to disk and memory mapped to be available for query processing.

Specified by:

createRealtimeAppenderatorForTask in interface AppenderatorsManager

createBatchAppenderatorForTask

public Appenderator createBatchAppenderatorForTask(String taskId,
 DataSchema schema,
 AppenderatorConfig config,
 TaskDirectory taskDirectory,
 SegmentGenerationMetrics metrics,
 DataSegmentPusher dataSegmentPusher,
 com.fasterxml.jackson.databind.ObjectMapper objectMapper,
 IndexIO indexIO,
 IndexMerger indexMerger,
 RowIngestionMeters rowIngestionMeters,
 ParseExceptionHandler parseExceptionHandler,
 CentralizedDatasourceSchemaConfig centralizedDatasourceSchemaConfig)

Description copied from interface: AppenderatorsManager

Creates a BatchAppenderator suitable for batch ingestion with no ability to process queries against the processed data. Intermediary segments are persisted to temporary disk and then merged into the final set of segments at publishing time.

Specified by:

createBatchAppenderatorForTask in interface AppenderatorsManager

removeAppenderatorsForTask

public void removeAppenderatorsForTask(String taskId,
 String dataSource)

Description copied from interface: AppenderatorsManager

Removes any internal Appenderator-tracking state associated with the provided taskId. This method should be called when a task is finished using its Appenderators that were previously created by createRealtimeAppenderatorForTask or createOfflineAppenderatorForTask. The method can be called by the entity managing Tasks when the Tasks finish, such as ThreadingTaskRunner.

Specified by:

removeAppenderatorsForTask in interface AppenderatorsManager

getQueryRunnerForIntervals

public <T> QueryRunner<T> getQueryRunnerForIntervals(Query<T> query,
 Iterable<org.joda.time.Interval> intervals)

Description copied from interface: AppenderatorsManager

Returns a query runner for the given intervals over the Appenderators managed by this AppenderatorsManager.

Specified by:

getQueryRunnerForIntervals in interface AppenderatorsManager

getQueryRunnerForSegments

public <T> QueryRunner<T> getQueryRunnerForSegments(Query<T> query,
 Iterable<SegmentDescriptor> specs)

Description copied from interface: AppenderatorsManager

Returns a query runner for the given segment specs over the Appenderators managed by this AppenderatorsManager.

Specified by:

getQueryRunnerForSegments in interface AppenderatorsManager

shouldTaskMakeNodeAnnouncements

public boolean shouldTaskMakeNodeAnnouncements()

Description copied from interface: AppenderatorsManager

As AppenderatorsManager implementions are service dependent (i.e., Peons and Indexers have different impls), this method allows Tasks to know whether they should announce themselves as nodes and segment servers to the rest of the cluster. Only Tasks running in Peons (i.e., as separate processes) should make their own individual node announcements.

Specified by:

shouldTaskMakeNodeAnnouncements in interface AppenderatorsManager

shutdown

public void shutdown()

Description copied from interface: AppenderatorsManager

Shut down the AppenderatorsManager.

Specified by:

shutdown in interface AppenderatorsManager

getDatasourceBundles

public Map<String,UnifiedIndexerAppenderatorsManager.DatasourceBundle> getDatasourceBundles()

getWorkerConfig

public WorkerConfig getWorkerConfig()