use of org.apache.drill.exec.record.VectorWrapper in project drill by apache.
the class ExternalSortBatch method innerNext.
@SuppressWarnings("resource")
@Override
public IterOutcome innerNext() {
if (schema != null) {
if (spillCount == 0) {
return (getSelectionVector4().next()) ? IterOutcome.OK : IterOutcome.NONE;
} else {
Stopwatch w = Stopwatch.createStarted();
int count = copier.next(targetRecordCount);
if (count > 0) {
long t = w.elapsed(TimeUnit.MICROSECONDS);
logger.debug("Took {} us to merge {} records", t, count);
container.setRecordCount(count);
return IterOutcome.OK;
} else {
logger.debug("copier returned 0 records");
return IterOutcome.NONE;
}
}
}
int totalCount = 0;
// total number of batches received so far
int totalBatches = 0;
try {
container.clear();
outer: while (true) {
IterOutcome upstream;
if (first) {
upstream = IterOutcome.OK_NEW_SCHEMA;
} else {
upstream = next(incoming);
}
if (upstream == IterOutcome.OK && sorter == null) {
upstream = IterOutcome.OK_NEW_SCHEMA;
}
switch(upstream) {
case NONE:
if (first) {
return upstream;
}
break outer;
case NOT_YET:
throw new UnsupportedOperationException();
case STOP:
return upstream;
case OK_NEW_SCHEMA:
case OK:
VectorContainer convertedBatch;
// only change in the case that the schema truly changes. Artificial schema changes are ignored.
if (upstream == IterOutcome.OK_NEW_SCHEMA && !incoming.getSchema().equals(schema)) {
if (schema != null) {
if (unionTypeEnabled) {
this.schema = SchemaUtil.mergeSchemas(schema, incoming.getSchema());
} else {
throw SchemaChangeException.schemaChanged("Schema changes not supported in External Sort. Please enable Union type", schema, incoming.getSchema());
}
} else {
schema = incoming.getSchema();
}
convertedBatch = SchemaUtil.coerceContainer(incoming, schema, oContext);
for (BatchGroup b : batchGroups) {
b.setSchema(schema);
}
for (BatchGroup b : spilledBatchGroups) {
b.setSchema(schema);
}
this.sorter = createNewSorter(context, convertedBatch);
} else {
convertedBatch = SchemaUtil.coerceContainer(incoming, schema, oContext);
}
if (first) {
first = false;
}
if (convertedBatch.getRecordCount() == 0) {
for (VectorWrapper<?> w : convertedBatch) {
w.clear();
}
break;
}
SelectionVector2 sv2;
if (incoming.getSchema().getSelectionVectorMode() == BatchSchema.SelectionVectorMode.TWO_BYTE) {
sv2 = incoming.getSelectionVector2().clone();
} else {
try {
sv2 = newSV2();
} catch (InterruptedException e) {
return IterOutcome.STOP;
} catch (OutOfMemoryException e) {
throw new OutOfMemoryException(e);
}
}
int count = sv2.getCount();
totalCount += count;
totalBatches++;
sorter.setup(context, sv2, convertedBatch);
sorter.sort(sv2);
RecordBatchData rbd = new RecordBatchData(convertedBatch, oAllocator);
boolean success = false;
try {
rbd.setSv2(sv2);
batchGroups.add(new BatchGroup(rbd.getContainer(), rbd.getSv2(), oContext));
if (peakNumBatches < batchGroups.size()) {
peakNumBatches = batchGroups.size();
stats.setLongStat(Metric.PEAK_BATCHES_IN_MEMORY, peakNumBatches);
}
batchesSinceLastSpill++;
if (// If we haven't spilled so far, do we have enough memory for MSorter if this turns out to be the last incoming batch?
(spillCount == 0 && !hasMemoryForInMemorySort(totalCount)) || // If we haven't spilled so far, make sure we don't exceed the maximum number of batches SV4 can address
(spillCount == 0 && totalBatches > Character.MAX_VALUE) || // current memory used is more than 95% of memory usage limit of this operator
(oAllocator.getAllocatedMemory() > .95 * oAllocator.getLimit()) || // since the last spill exceed the defined limit
(batchGroups.size() > SPILL_THRESHOLD && batchesSinceLastSpill >= SPILL_BATCH_GROUP_SIZE)) {
if (firstSpillBatchCount == 0) {
firstSpillBatchCount = batchGroups.size();
}
if (spilledBatchGroups.size() > firstSpillBatchCount / 2) {
logger.info("Merging spills");
final BatchGroup merged = mergeAndSpill(spilledBatchGroups);
if (merged != null) {
spilledBatchGroups.addFirst(merged);
}
}
final BatchGroup merged = mergeAndSpill(batchGroups);
if (merged != null) {
// make sure we don't add null to spilledBatchGroups
spilledBatchGroups.add(merged);
batchesSinceLastSpill = 0;
}
}
success = true;
} finally {
if (!success) {
rbd.clear();
}
}
break;
case OUT_OF_MEMORY:
logger.debug("received OUT_OF_MEMORY, trying to spill");
if (batchesSinceLastSpill > 2) {
final BatchGroup merged = mergeAndSpill(batchGroups);
if (merged != null) {
spilledBatchGroups.add(merged);
batchesSinceLastSpill = 0;
}
} else {
logger.debug("not enough batches to spill, sending OUT_OF_MEMORY downstream");
return IterOutcome.OUT_OF_MEMORY;
}
break;
default:
throw new UnsupportedOperationException();
}
}
if (totalCount == 0) {
return IterOutcome.NONE;
}
if (spillCount == 0) {
if (builder != null) {
builder.clear();
builder.close();
}
builder = new SortRecordBatchBuilder(oAllocator);
for (BatchGroup group : batchGroups) {
RecordBatchData rbd = new RecordBatchData(group.getContainer(), oAllocator);
rbd.setSv2(group.getSv2());
builder.add(rbd);
}
builder.build(context, container);
sv4 = builder.getSv4();
mSorter = createNewMSorter();
mSorter.setup(context, oAllocator, getSelectionVector4(), this.container);
// For testing memory-leak purpose, inject exception after mSorter finishes setup
injector.injectUnchecked(context.getExecutionControls(), INTERRUPTION_AFTER_SETUP);
mSorter.sort(this.container);
// sort may have prematurely exited due to should continue returning false.
if (!context.shouldContinue()) {
return IterOutcome.STOP;
}
// For testing memory-leak purpose, inject exception after mSorter finishes sorting
injector.injectUnchecked(context.getExecutionControls(), INTERRUPTION_AFTER_SORT);
sv4 = mSorter.getSV4();
container.buildSchema(SelectionVectorMode.FOUR_BYTE);
} else {
// some batches were spilled
final BatchGroup merged = mergeAndSpill(batchGroups);
if (merged != null) {
spilledBatchGroups.add(merged);
}
batchGroups.addAll(spilledBatchGroups);
// no need to cleanup spilledBatchGroups, all it's batches are in batchGroups now
spilledBatchGroups = null;
logger.warn("Starting to merge. {} batch groups. Current allocated memory: {}", batchGroups.size(), oAllocator.getAllocatedMemory());
VectorContainer hyperBatch = constructHyperBatch(batchGroups);
createCopier(hyperBatch, batchGroups, container, false);
int estimatedRecordSize = 0;
for (VectorWrapper<?> w : batchGroups.get(0)) {
try {
estimatedRecordSize += TypeHelper.getSize(w.getField().getType());
} catch (UnsupportedOperationException e) {
estimatedRecordSize += 50;
}
}
targetRecordCount = Math.min(MAX_BATCH_SIZE, Math.max(1, COPIER_BATCH_MEM_LIMIT / estimatedRecordSize));
int count = copier.next(targetRecordCount);
container.buildSchema(SelectionVectorMode.NONE);
container.setRecordCount(count);
}
return IterOutcome.OK_NEW_SCHEMA;
} catch (SchemaChangeException ex) {
kill(false);
context.fail(UserException.unsupportedError(ex).message("Sort doesn't currently support sorts with changing schemas").build(logger));
return IterOutcome.STOP;
} catch (ClassTransformationException | IOException ex) {
kill(false);
context.fail(ex);
return IterOutcome.STOP;
} catch (UnsupportedOperationException e) {
throw new RuntimeException(e);
}
}
use of org.apache.drill.exec.record.VectorWrapper in project drill by apache.
the class BatchPrinter method printHyperBatch.
public static void printHyperBatch(VectorAccessible batch, SelectionVector4 sv4) {
List<String> columns = Lists.newArrayList();
List<ValueVector> vectors = Lists.newArrayList();
int numBatches = 0;
for (VectorWrapper vw : batch) {
columns.add(vw.getValueVectors()[0].getField().getPath());
numBatches = vw.getValueVectors().length;
}
int width = columns.size();
for (int j = 0; j < sv4.getCount(); j++) {
if (j % 50 == 0) {
System.out.println(StringUtils.repeat("-", width * 17 + 1));
for (String column : columns) {
System.out.printf("| %-15s", width <= 15 ? column : column.substring(0, 14));
}
System.out.printf("|\n");
System.out.println(StringUtils.repeat("-", width * 17 + 1));
}
for (VectorWrapper vw : batch) {
Object o = vw.getValueVectors()[sv4.get(j) >>> 16].getAccessor().getObject(sv4.get(j) & 65535);
String value;
if (o == null) {
value = "null";
} else if (o instanceof byte[]) {
value = new String((byte[]) o);
} else {
value = o.toString();
}
System.out.printf("| %-15s", value.length() <= 15 ? value : value.substring(0, 14));
}
System.out.printf("|\n");
}
System.out.printf("|\n");
}
use of org.apache.drill.exec.record.VectorWrapper in project drill by apache.
the class BatchPrinter method printBatch.
public static void printBatch(VectorAccessible batch, SelectionVector2 sv2) {
List<String> columns = Lists.newArrayList();
List<ValueVector> vectors = Lists.newArrayList();
for (VectorWrapper vw : batch) {
columns.add(vw.getValueVector().getField().getPath());
vectors.add(vw.getValueVector());
}
int width = columns.size();
int rows = vectors.get(0).getMetadata().getValueCount();
for (int i = 0; i < rows; i++) {
if (i % 50 == 0) {
System.out.println(StringUtils.repeat("-", width * 17 + 1));
for (String column : columns) {
System.out.printf("| %-15s", width <= 15 ? column : column.substring(0, 14));
}
System.out.printf("|\n");
System.out.println(StringUtils.repeat("-", width * 17 + 1));
}
int row = sv2.getIndex(i);
for (ValueVector vv : vectors) {
Object o = vv.getAccessor().getObject(row);
String value;
if (o == null) {
value = "null";
} else if (o instanceof byte[]) {
value = new String((byte[]) o);
} else {
value = o.toString();
}
System.out.printf("| %-15s", value.length() <= 15 ? value : value.substring(0, 14));
}
System.out.printf("|\n");
}
}
use of org.apache.drill.exec.record.VectorWrapper in project drill by apache.
the class BatchPrinter method printBatch.
public static void printBatch(VectorAccessible batch) {
List<String> columns = Lists.newArrayList();
List<ValueVector> vectors = Lists.newArrayList();
for (VectorWrapper vw : batch) {
columns.add(vw.getValueVector().getField().getPath());
vectors.add(vw.getValueVector());
}
int width = columns.size();
int rows = vectors.get(0).getMetadata().getValueCount();
for (int row = 0; row < rows; row++) {
if (row % 50 == 0) {
System.out.println(StringUtils.repeat("-", width * 17 + 1));
for (String column : columns) {
System.out.printf("| %-15s", width <= 15 ? column : column.substring(0, 14));
}
System.out.printf("|\n");
System.out.println(StringUtils.repeat("-", width * 17 + 1));
}
for (ValueVector vv : vectors) {
Object o = vv.getAccessor().getObject(row);
String value;
if (o == null) {
value = "null";
} else if (o instanceof byte[]) {
value = new String((byte[]) o);
} else {
value = o.toString();
}
System.out.printf("| %-15s", value.length() <= 15 ? value : value.substring(0, 14));
}
System.out.printf("|\n");
}
}
use of org.apache.drill.exec.record.VectorWrapper in project drill by apache.
the class TestHashJoin method simpleEqualityJoin.
@Test
public void simpleEqualityJoin() throws Throwable {
// Function checks hash join with single equality condition
try (RemoteServiceSet serviceSet = RemoteServiceSet.getLocalServiceSet();
Drillbit bit = new Drillbit(CONFIG, serviceSet);
DrillClient client = new DrillClient(CONFIG, serviceSet.getCoordinator())) {
// run query.
bit.run();
client.connect();
List<QueryDataBatch> results = client.runQuery(org.apache.drill.exec.proto.UserBitShared.QueryType.PHYSICAL, Files.toString(FileUtils.getResourceAsFile("/join/hash_join.json"), Charsets.UTF_8).replace("#{TEST_FILE_1}", FileUtils.getResourceAsFile("/build_side_input.json").toURI().toString()).replace("#{TEST_FILE_2}", FileUtils.getResourceAsFile("/probe_side_input.json").toURI().toString()));
RecordBatchLoader batchLoader = new RecordBatchLoader(bit.getContext().getAllocator());
QueryDataBatch batch = results.get(1);
assertTrue(batchLoader.load(batch.getHeader().getDef(), batch.getData()));
Iterator<VectorWrapper<?>> itr = batchLoader.iterator();
// Just test the join key
long[] colA = { 1, 1, 2, 2, 1, 1 };
// Check the output of decimal9
ValueVector.Accessor intAccessor1 = itr.next().getValueVector().getAccessor();
for (int i = 0; i < intAccessor1.getValueCount(); i++) {
assertEquals(intAccessor1.getObject(i), colA[i]);
}
assertEquals(6, intAccessor1.getValueCount());
batchLoader.clear();
for (QueryDataBatch result : results) {
result.release();
}
}
}
Aggregations