use of io.cdap.cdap.api.metadata.MetadataReader in project cdap by caskdata.
the class MapReduceTaskContextProvider method createCacheLoader.
/**
* Creates a {@link CacheLoader} for the task context cache.
*/
private CacheLoader<ContextCacheKey, BasicMapReduceTaskContext> createCacheLoader(final Injector injector) {
DiscoveryServiceClient discoveryServiceClient = injector.getInstance(DiscoveryServiceClient.class);
DatasetFramework datasetFramework = injector.getInstance(DatasetFramework.class);
SecureStore secureStore = injector.getInstance(SecureStore.class);
SecureStoreManager secureStoreManager = injector.getInstance(SecureStoreManager.class);
MessagingService messagingService = injector.getInstance(MessagingService.class);
// Multiple instances of BasicMapReduceTaskContext can share the same program.
AtomicReference<Program> programRef = new AtomicReference<>();
MetadataReader metadataReader = injector.getInstance(MetadataReader.class);
MetadataPublisher metadataPublisher = injector.getInstance(MetadataPublisher.class);
FieldLineageWriter fieldLineageWriter = injector.getInstance(FieldLineageWriter.class);
RemoteClientFactory remoteClientFactory = injector.getInstance(RemoteClientFactory.class);
return new CacheLoader<ContextCacheKey, BasicMapReduceTaskContext>() {
@Override
public BasicMapReduceTaskContext load(ContextCacheKey key) throws Exception {
TaskAttemptID taskAttemptId = key.getTaskAttemptID();
// taskAttemptId could be null if used from a org.apache.hadoop.mapreduce.Partitioner or
// from a org.apache.hadoop.io.RawComparator, in which case we can get the JobId from the conf. Note that the
// JobId isn't in the conf for the OutputCommitter#setupJob method, in which case we use the taskAttemptId
Path txFile = MainOutputCommitter.getTxFile(key.getConfiguration(), taskAttemptId != null ? taskAttemptId.getJobID() : null);
FileSystem fs = txFile.getFileSystem(key.getConfiguration());
Transaction transaction = null;
if (fs.exists(txFile)) {
try (FSDataInputStream txFileInputStream = fs.open(txFile)) {
transaction = new TransactionCodec().decode(ByteStreams.toByteArray(txFileInputStream));
}
}
MapReduceContextConfig contextConfig = new MapReduceContextConfig(key.getConfiguration());
MapReduceClassLoader classLoader = MapReduceClassLoader.getFromConfiguration(key.getConfiguration());
Program program = programRef.get();
if (program == null) {
// Creation of program is relatively cheap, so just create and do compare and set.
programRef.compareAndSet(null, createProgram(contextConfig, classLoader.getProgramClassLoader()));
program = programRef.get();
}
WorkflowProgramInfo workflowInfo = contextConfig.getWorkflowProgramInfo();
DatasetFramework programDatasetFramework = workflowInfo == null ? datasetFramework : NameMappedDatasetFramework.createFromWorkflowProgramInfo(datasetFramework, workflowInfo, program.getApplicationSpecification());
// Setup dataset framework context, if required
if (programDatasetFramework instanceof ProgramContextAware) {
ProgramRunId programRunId = program.getId().run(ProgramRunners.getRunId(contextConfig.getProgramOptions()));
((ProgramContextAware) programDatasetFramework).setContext(new BasicProgramContext(programRunId));
}
MapReduceSpecification spec = program.getApplicationSpecification().getMapReduce().get(program.getName());
MetricsCollectionService metricsCollectionService = null;
MapReduceMetrics.TaskType taskType = null;
String taskId = null;
ProgramOptions options = contextConfig.getProgramOptions();
// from a org.apache.hadoop.io.RawComparator
if (taskAttemptId != null) {
taskId = taskAttemptId.getTaskID().toString();
if (MapReduceMetrics.TaskType.hasType(taskAttemptId.getTaskType())) {
taskType = MapReduceMetrics.TaskType.from(taskAttemptId.getTaskType());
// if this is not for a mapper or a reducer, we don't need the metrics collection service
metricsCollectionService = injector.getInstance(MetricsCollectionService.class);
options = new SimpleProgramOptions(options.getProgramId(), options.getArguments(), new BasicArguments(RuntimeArguments.extractScope("task", taskType.toString().toLowerCase(), contextConfig.getProgramOptions().getUserArguments().asMap())), options.isDebug());
}
}
CConfiguration cConf = injector.getInstance(CConfiguration.class);
TransactionSystemClient txClient = injector.getInstance(TransactionSystemClient.class);
NamespaceQueryAdmin namespaceQueryAdmin = injector.getInstance(NamespaceQueryAdmin.class);
return new BasicMapReduceTaskContext(program, options, cConf, taskType, taskId, spec, workflowInfo, discoveryServiceClient, metricsCollectionService, txClient, transaction, programDatasetFramework, classLoader.getPluginInstantiator(), contextConfig.getLocalizedResources(), secureStore, secureStoreManager, accessEnforcer, authenticationContext, messagingService, mapReduceClassLoader, metadataReader, metadataPublisher, namespaceQueryAdmin, fieldLineageWriter, remoteClientFactory);
}
};
}
Aggregations