Examples with SchemaRelationProvider org.apache.spark.sql.sources.SchemaRelationProvider used on opensource projects

Example 1 with SchemaRelationProvider

use of org.apache.spark.sql.sources.SchemaRelationProvider in project OpenLineage by OpenLineage.

the class SaveIntoDataSourceCommandVisitor method apply.

@Override
public List<OpenLineage.OutputDataset> apply(SparkListenerEvent event) {
    BaseRelation relation;
    SaveIntoDataSourceCommand command = (SaveIntoDataSourceCommand) context.getQueryExecution().get().optimizedPlan();
    // as other impls of CreatableRelationProvider may not be able to be handled in the generic way.
    if (KafkaRelationVisitor.isKafkaSource(command.dataSource())) {
        return KafkaRelationVisitor.createKafkaDatasets(outputDataset(), command.dataSource(), command.options(), command.mode(), command.schema());
    }
    if (command.dataSource().getClass().getName().contains("DeltaDataSource")) {
        if (command.options().contains("path")) {
            URI uri = URI.create(command.options().get("path").get());
            return Collections.singletonList(outputDataset().getDataset(PathUtils.fromURI(uri, "file"), command.schema()));
        }
    }
    SQLContext sqlContext = context.getSparkSession().get().sqlContext();
    try {
        if (command.dataSource() instanceof RelationProvider) {
            RelationProvider p = (RelationProvider) command.dataSource();
            relation = p.createRelation(sqlContext, command.options());
        } else {
            SchemaRelationProvider p = (SchemaRelationProvider) command.dataSource();
            relation = p.createRelation(sqlContext, command.options(), command.schema());
        }
    } catch (Exception ex) {
        // Bad detection of errors in scala
        if (ex instanceof SQLException) {
            // This can happen on SparkListenerSQLExecutionStart for example for sqlite, when database
            // does not exist yet - it will be created as command execution
            // Still, we can just ignore it on start, because it will work on end
            // see SparkReadWriteIntegTest.testReadFromFileWriteToJdbc
            log.warn("Can't create relation: ", ex);
            return Collections.emptyList();
        }
        throw ex;
    }
    LogicalRelation logicalRelation = new LogicalRelation(relation, relation.schema().toAttributes(), Option.empty(), command.isStreaming());
    return delegate(context.getOutputDatasetQueryPlanVisitors(), context.getOutputDatasetBuilders(), event).applyOrElse(logicalRelation, ScalaConversionUtils.toScalaFn((lp) -> Collections.<OutputDataset>emptyList())).stream().map(ds -> {
        Builder<String, OpenLineage.DatasetFacet> facetsMap = ImmutableMap.<String, OpenLineage.DatasetFacet>builder();
        if (ds.getFacets().getAdditionalProperties() != null) {
            facetsMap.putAll(ds.getFacets().getAdditionalProperties());
        }
        ds.getFacets().getAdditionalProperties().putAll(facetsMap.build());
        if (SaveMode.Overwrite == command.mode()) {
            // rebuild whole dataset with a LifecycleStateChange facet added
            OpenLineage.DatasetFacets facets = context.getOpenLineage().newDatasetFacets(ds.getFacets().getDocumentation(), ds.getFacets().getDataSource(), ds.getFacets().getVersion(), ds.getFacets().getSchema(), context.getOpenLineage().newLifecycleStateChangeDatasetFacet(OpenLineage.LifecycleStateChangeDatasetFacet.LifecycleStateChange.OVERWRITE, null));
            OpenLineage.OutputDataset newDs = context.getOpenLineage().newOutputDataset(ds.getNamespace(), ds.getName(), facets, ds.getOutputFacets());
            return newDs;
        }
        return ds;
    }).collect(Collectors.toList());
}