public SqoopOutputFormatLoadExecutorSpark(JobContext jobctx) {
context = jobctx;
loaderName = context.getConfiguration().get(MRJobConstants.JOB_ETL_LOADER);
writer = new SqoopRecordWriter();
// jackh: This must be conditional - Extract schema using credentials in case of MR and simply
// extract from the
// credentials object in case of Spark (due to known issue with Hadoop/Spark that the
// credentials are never added
// for serialization)
// matcher = MatcherFactory.getMatcher(
// MRConfigurationUtils.getConnectorSchema(Direction.FROM, context.getConfiguration()),
// MRConfigurationUtils.getConnectorSchema(Direction.TO, context.getConfiguration()));
matcher =
MatcherFactory.getMatcher(
MRConfigurationUtils.getConnectorSchemaUnsafe(
Direction.FROM, context.getConfiguration()),
MRConfigurationUtils.getConnectorSchemaUnsafe(
Direction.TO, context.getConfiguration()));
toDataFormat =
(IntermediateDataFormat<?>)
ClassUtils.instantiate(
context.getConfiguration().get(MRJobConstants.TO_INTERMEDIATE_DATA_FORMAT));
// Using the TO schema since the SqoopDataWriter in the SqoopMapper encapsulates the
// toDataFormat
toDataFormat.setSchema(matcher.getToSchema());
}
@SuppressWarnings({"unchecked", "rawtypes"})
@Override
public void run(Context context) throws IOException, InterruptedException {
Configuration conf = context.getConfiguration();
String extractorName = conf.get(MRJobConstants.JOB_ETL_EXTRACTOR);
Extractor extractor = (Extractor) ClassUtils.instantiate(extractorName);
Schema fromSchema = MRConfigurationUtils.getConnectorSchema(Direction.FROM, conf);
Schema toSchema = MRConfigurationUtils.getConnectorSchema(Direction.TO, conf);
matcher = MatcherFactory.getMatcher(fromSchema, toSchema);
String fromIDFClass = conf.get(MRJobConstants.FROM_INTERMEDIATE_DATA_FORMAT);
fromIDF = (IntermediateDataFormat<Object>) ClassUtils.instantiate(fromIDFClass);
fromIDF.setSchema(matcher.getFromSchema());
String toIDFClass = conf.get(MRJobConstants.TO_INTERMEDIATE_DATA_FORMAT);
toIDF = (IntermediateDataFormat<Object>) ClassUtils.instantiate(toIDFClass);
toIDF.setSchema(matcher.getToSchema());
// Objects that should be passed to the Executor execution
PrefixContext subContext =
new PrefixContext(conf, MRJobConstants.PREFIX_CONNECTOR_FROM_CONTEXT);
Object fromConfig = MRConfigurationUtils.getConnectorLinkConfig(Direction.FROM, conf);
Object fromJob = MRConfigurationUtils.getConnectorJobConfig(Direction.FROM, conf);
SqoopSplit split = context.getCurrentKey();
ExtractorContext extractorContext =
new ExtractorContext(
subContext, new SqoopDataWriter(context, fromIDF, toIDF, matcher), fromSchema);
try {
LOG.info("Running extractor class " + extractorName);
extractor.extract(extractorContext, fromConfig, fromJob, split.getPartition());
LOG.info("Extractor has finished");
context.getCounter(SqoopCounters.ROWS_READ).increment(extractor.getRowsRead());
} catch (Exception e) {
throw new SqoopException(MRExecutionError.MAPRED_EXEC_0017, e);
} finally {
LOG.info("Stopping progress service");
}
}
@SuppressWarnings({"rawtypes", "unchecked"})
@Override
public void run() {
LOG.info("SqoopOutputFormatLoadExecutor consumer thread is starting");
try {
DataReader reader = new SqoopOutputFormatDataReader();
Configuration conf = context.getConfiguration();
Loader loader = (Loader) ClassUtils.instantiate(loaderName);
// Objects that should be passed to the Loader
PrefixContext subContext =
new PrefixContext(conf, MRJobConstants.PREFIX_CONNECTOR_TO_CONTEXT);
Object connectorLinkConfig =
MRConfigurationUtils.getConnectorLinkConfigUnsafe(Direction.TO, conf);
Object connectorToJobConfig =
MRConfigurationUtils.getConnectorJobConfigUnsafe(Direction.TO, conf);
// Using the TO schema since the SqoopDataWriter in the SqoopMapper
// encapsulates the toDataFormat
// Create loader context
LoaderContext loaderContext = new LoaderContext(subContext, reader, matcher.getToSchema());
LOG.info("Running loader class " + loaderName);
loader.load(loaderContext, connectorLinkConfig, connectorToJobConfig);
LOG.info("Loader has finished");
((TaskAttemptContext) jobctx)
.getCounter(SqoopCounters.ROWS_WRITTEN)
.increment(loader.getRowsWritten());
} catch (Throwable t) {
readerFinished = true;
LOG.error("Error while loading data out of MR job.", t);
// Release so that the writer can tell Sqoop something went
// wrong.
free.release();
throw new SqoopException(SparkExecutionError.SPARK_EXEC_0018, t);
}
// if no exception happens yet and reader finished before writer,
// something went wrong
if (!writerFinished) {
// throw exception if data are not all consumed
readerFinished = true;
LOG.error("Reader terminated, but writer is still running!");
// Release so that the writer can tell Sqoop something went
// wrong.
free.release();
throw new SqoopException(SparkExecutionError.SPARK_EXEC_0019);
}
// inform writer that reader is finished
readerFinished = true;
}