use of org.apache.flink.table.connector.source.DynamicTableSource in project flink by apache.
the class FactoryUtilTest method testAllOptions.
@Test
public void testAllOptions() {
final Map<String, String> options = createAllOptions();
final DynamicTableSource actualSource = createTableSource(SCHEMA, options);
final DynamicTableSource expectedSource = new DynamicTableSourceMock("MyTarget", null, new DecodingFormatMock(",", false), new DecodingFormatMock("|", true));
assertThat(actualSource).isEqualTo(expectedSource);
final DynamicTableSink actualSink = createTableSink(SCHEMA, options);
final DynamicTableSink expectedSink = new DynamicTableSinkMock("MyTarget", 1000L, new EncodingFormatMock(","), new EncodingFormatMock("|"));
assertThat(actualSink).isEqualTo(expectedSink);
}
use of org.apache.flink.table.connector.source.DynamicTableSource in project flink by apache.
the class FactoryUtilTest method testManagedConnector.
@Test
public void testManagedConnector() {
final Map<String, String> options = createAllOptions();
options.remove("connector");
final DynamicTableSource actualSource = createTableSource(SCHEMA, options);
assertThat(actualSource).isExactlyInstanceOf(TestManagedTableSource.class);
}
use of org.apache.flink.table.connector.source.DynamicTableSource in project flink by apache.
the class DataGenTableSourceFactoryTest method testSequenceCheckpointRestore.
@Test
public void testSequenceCheckpointRestore() throws Exception {
DescriptorProperties descriptor = new DescriptorProperties();
descriptor.putString(FactoryUtil.CONNECTOR.key(), "datagen");
descriptor.putString(DataGenConnectorOptionsUtil.FIELDS + ".f0." + DataGenConnectorOptionsUtil.KIND, DataGenConnectorOptionsUtil.SEQUENCE);
descriptor.putLong(DataGenConnectorOptionsUtil.FIELDS + ".f0." + DataGenConnectorOptionsUtil.START, 0);
descriptor.putLong(DataGenConnectorOptionsUtil.FIELDS + ".f0." + DataGenConnectorOptionsUtil.END, 100);
DynamicTableSource dynamicTableSource = createTableSource(ResolvedSchema.of(Column.physical("f0", DataTypes.BIGINT())), descriptor.asMap());
DataGenTableSource dataGenTableSource = (DataGenTableSource) dynamicTableSource;
DataGeneratorSource<RowData> source = dataGenTableSource.createSource();
final int initElement = 0;
final int maxElement = 100;
final Set<RowData> expectedOutput = new HashSet<>();
for (long i = initElement; i <= maxElement; i++) {
expectedOutput.add(GenericRowData.of(i));
}
DataGeneratorSourceTest.innerTestDataGenCheckpointRestore(() -> {
try {
return InstantiationUtil.clone(source);
} catch (IOException | ClassNotFoundException e) {
throw new RuntimeException(e);
}
}, expectedOutput);
}
use of org.apache.flink.table.connector.source.DynamicTableSource in project flink by apache.
the class PushPartitionIntoTableSourceScanRule method onMatch.
@Override
public void onMatch(RelOptRuleCall call) {
Filter filter = call.rel(0);
LogicalTableScan scan = call.rel(1);
TableSourceTable tableSourceTable = scan.getTable().unwrap(TableSourceTable.class);
RelDataType inputFieldTypes = filter.getInput().getRowType();
List<String> inputFieldNames = inputFieldTypes.getFieldNames();
List<String> partitionFieldNames = tableSourceTable.contextResolvedTable().<ResolvedCatalogTable>getResolvedTable().getPartitionKeys();
// extract partition predicates
RelBuilder relBuilder = call.builder();
RexBuilder rexBuilder = relBuilder.getRexBuilder();
Tuple2<Seq<RexNode>, Seq<RexNode>> allPredicates = RexNodeExtractor.extractPartitionPredicateList(filter.getCondition(), FlinkRelOptUtil.getMaxCnfNodeCount(scan), inputFieldNames.toArray(new String[0]), rexBuilder, partitionFieldNames.toArray(new String[0]));
RexNode partitionPredicate = RexUtil.composeConjunction(rexBuilder, JavaConversions.seqAsJavaList(allPredicates._1));
if (partitionPredicate.isAlwaysTrue()) {
return;
}
// build pruner
LogicalType[] partitionFieldTypes = partitionFieldNames.stream().map(name -> {
int index = inputFieldNames.indexOf(name);
if (index < 0) {
throw new TableException(String.format("Partitioned key '%s' isn't found in input columns. " + "Validator should have checked that.", name));
}
return inputFieldTypes.getFieldList().get(index).getType();
}).map(FlinkTypeFactory::toLogicalType).toArray(LogicalType[]::new);
RexNode finalPartitionPredicate = adjustPartitionPredicate(inputFieldNames, partitionFieldNames, partitionPredicate);
FlinkContext context = ShortcutUtils.unwrapContext(scan);
Function<List<Map<String, String>>, List<Map<String, String>>> defaultPruner = partitions -> PartitionPruner.prunePartitions(context.getTableConfig(), partitionFieldNames.toArray(new String[0]), partitionFieldTypes, partitions, finalPartitionPredicate);
// prune partitions
List<Map<String, String>> remainingPartitions = readPartitionsAndPrune(rexBuilder, context, tableSourceTable, defaultPruner, allPredicates._1(), inputFieldNames);
// apply push down
DynamicTableSource dynamicTableSource = tableSourceTable.tableSource().copy();
PartitionPushDownSpec partitionPushDownSpec = new PartitionPushDownSpec(remainingPartitions);
partitionPushDownSpec.apply(dynamicTableSource, SourceAbilityContext.from(scan));
// build new statistic
TableStats newTableStat = null;
if (tableSourceTable.contextResolvedTable().isPermanent()) {
ObjectIdentifier identifier = tableSourceTable.contextResolvedTable().getIdentifier();
ObjectPath tablePath = identifier.toObjectPath();
Catalog catalog = tableSourceTable.contextResolvedTable().getCatalog().get();
for (Map<String, String> partition : remainingPartitions) {
Optional<TableStats> partitionStats = getPartitionStats(catalog, tablePath, partition);
if (!partitionStats.isPresent()) {
// clear all information before
newTableStat = null;
break;
} else {
newTableStat = newTableStat == null ? partitionStats.get() : newTableStat.merge(partitionStats.get());
}
}
}
FlinkStatistic newStatistic = FlinkStatistic.builder().statistic(tableSourceTable.getStatistic()).tableStats(newTableStat).build();
TableSourceTable newTableSourceTable = tableSourceTable.copy(dynamicTableSource, newStatistic, new SourceAbilitySpec[] { partitionPushDownSpec });
LogicalTableScan newScan = LogicalTableScan.create(scan.getCluster(), newTableSourceTable, scan.getHints());
// transform to new node
RexNode nonPartitionPredicate = RexUtil.composeConjunction(rexBuilder, JavaConversions.seqAsJavaList(allPredicates._2()));
if (nonPartitionPredicate.isAlwaysTrue()) {
call.transformTo(newScan);
} else {
Filter newFilter = filter.copy(filter.getTraitSet(), newScan, nonPartitionPredicate);
call.transformTo(newFilter);
}
}
use of org.apache.flink.table.connector.source.DynamicTableSource in project flink by apache.
the class PushPartitionIntoTableSourceScanRule method matches.
@Override
public boolean matches(RelOptRuleCall call) {
Filter filter = call.rel(0);
if (filter.getCondition() == null) {
return false;
}
TableSourceTable tableSourceTable = call.rel(1).getTable().unwrap(TableSourceTable.class);
if (tableSourceTable == null) {
return false;
}
DynamicTableSource dynamicTableSource = tableSourceTable.tableSource();
if (!(dynamicTableSource instanceof SupportsPartitionPushDown)) {
return false;
}
CatalogTable catalogTable = tableSourceTable.contextResolvedTable().getTable();
if (!catalogTable.isPartitioned() || catalogTable.getPartitionKeys().isEmpty()) {
return false;
}
return Arrays.stream(tableSourceTable.abilitySpecs()).noneMatch(spec -> spec instanceof PartitionPushDownSpec);
}
Aggregations