Examples with FakeCaptureVectorToRowOutputOperator org.apache.hadoop.hive.ql.exec.vector.util.FakeCaptureVectorToRowOutputOperator used on opensource projects

Example 6 with FakeCaptureVectorToRowOutputOperator

use of org.apache.hadoop.hive.ql.exec.vector.util.FakeCaptureVectorToRowOutputOperator in project hive by apache.

the class TestVectorGroupByOperator method testMemoryPressureFlush.

@Test
public void testMemoryPressureFlush() throws HiveException {
    List<String> mapColumnNames = new ArrayList<String>();
    mapColumnNames.add("Key");
    mapColumnNames.add("Value");
    VectorizationContext ctx = new VectorizationContext("name", mapColumnNames);
    Pair<GroupByDesc, VectorGroupByDesc> pair = buildKeyGroupByDesc(ctx, "max", "Value", TypeInfoFactory.longTypeInfo, "Key", TypeInfoFactory.longTypeInfo);
    GroupByDesc desc = pair.fst;
    VectorGroupByDesc vectorDesc = pair.snd;
    // Set the memory treshold so that we get 100Kb before we need to flush.
    MemoryMXBean memoryMXBean = ManagementFactory.getMemoryMXBean();
    long maxMemory = memoryMXBean.getHeapMemoryUsage().getMax();
    float treshold = 100.0f * 1024.0f / maxMemory;
    desc.setMemoryThreshold(treshold);
    CompilationOpContext cCtx = new CompilationOpContext();
    Operator<? extends OperatorDesc> groupByOp = OperatorFactory.get(cCtx, desc);
    VectorGroupByOperator vgo = (VectorGroupByOperator) Vectorizer.vectorizeGroupByOperator(groupByOp, ctx, vectorDesc);
    FakeCaptureVectorToRowOutputOperator out = FakeCaptureVectorToRowOutputOperator.addCaptureOutputChild(cCtx, vgo);
    vgo.initialize(hconf, null);
    this.outputRowCount = 0;
    out.setOutputInspector(new FakeCaptureVectorToRowOutputOperator.OutputInspector() {

        @Override
        public void inspectRow(Object row, int tag) throws HiveException {
            ++outputRowCount;
        }
    });
    Iterable<Object> it = new Iterable<Object>() {

        @Override
        public Iterator<Object> iterator() {
            return new Iterator<Object>() {

                long value = 0;

                @Override
                public boolean hasNext() {
                    return true;
                }

                @Override
                public Object next() {
                    return ++value;
                }

                @Override
                public void remove() {
                }
            };
        }
    };
    FakeVectorRowBatchFromObjectIterables data = new FakeVectorRowBatchFromObjectIterables(100, new String[] { "long", "long" }, it, it);
    // The 'it' data source will produce data w/o ever ending
    // We want to see that memory pressure kicks in and some
    // entries in the VGBY are flushed.
    long countRowsProduced = 0;
    for (VectorizedRowBatch unit : data) {
        countRowsProduced += 100;
        vgo.process(unit, 0);
        if (0 < outputRowCount) {
            break;
        }
        // Set an upper bound how much we're willing to push before it should flush
        // we've set the memory treshold at 100kb, each key is distinct
        // It should not go beyond 100k/16 (key+data)
        assertTrue(countRowsProduced < 100 * 1024 / 16);
    }
    assertTrue(0 < outputRowCount);
}

Example 7 with FakeCaptureVectorToRowOutputOperator

use of org.apache.hadoop.hive.ql.exec.vector.util.FakeCaptureVectorToRowOutputOperator in project hive by apache.

the class TestVectorGroupByOperator method testKeyTypeAggregate.

private void testKeyTypeAggregate(String aggregateName, FakeVectorRowBatchFromObjectIterables data, Map<Object, Object> expected) throws HiveException {
    List<String> mapColumnNames = new ArrayList<String>();
    mapColumnNames.add("Key");
    mapColumnNames.add("Value");
    VectorizationContext ctx = new VectorizationContext("name", mapColumnNames);
    Set<Object> keys = new HashSet<Object>();
    AggregationDesc agg = buildAggregationDesc(ctx, aggregateName, GenericUDAFEvaluator.Mode.PARTIAL1, "Value", TypeInfoFactory.getPrimitiveTypeInfo(data.getTypes()[1]));
    ArrayList<AggregationDesc> aggs = new ArrayList<AggregationDesc>();
    aggs.add(agg);
    ArrayList<String> outputColumnNames = new ArrayList<String>();
    outputColumnNames.add("_col0");
    outputColumnNames.add("_col1");
    GroupByDesc desc = new GroupByDesc();
    VectorGroupByDesc vectorGroupByDesc = new VectorGroupByDesc();
    desc.setOutputColumnNames(outputColumnNames);
    desc.setAggregators(aggs);
    vectorGroupByDesc.setProcessingMode(ProcessingMode.HASH);
    ExprNodeDesc keyExp = buildColumnDesc(ctx, "Key", TypeInfoFactory.getPrimitiveTypeInfo(data.getTypes()[0]));
    ArrayList<ExprNodeDesc> keysDesc = new ArrayList<ExprNodeDesc>();
    keysDesc.add(keyExp);
    desc.setKeys(keysDesc);
    CompilationOpContext cCtx = new CompilationOpContext();
    Operator<? extends OperatorDesc> groupByOp = OperatorFactory.get(cCtx, desc);
    VectorGroupByOperator vgo = (VectorGroupByOperator) Vectorizer.vectorizeGroupByOperator(groupByOp, ctx, vectorGroupByDesc);
    if (vgo == null) {
        assertTrue(false);
    }
    FakeCaptureVectorToRowOutputOperator out = FakeCaptureVectorToRowOutputOperator.addCaptureOutputChild(cCtx, vgo);
    vgo.initialize(hconf, null);
    out.setOutputInspector(new FakeCaptureVectorToRowOutputOperator.OutputInspector() {

        private int rowIndex;

        private String aggregateName;

        private Map<Object, Object> expected;

        private Set<Object> keys;

        @Override
        public void inspectRow(Object row, int tag) throws HiveException {
            assertTrue(row instanceof Object[]);
            Object[] fields = (Object[]) row;
            assertEquals(2, fields.length);
            Object key = fields[0];
            Object keyValue = null;
            if (null == key) {
                keyValue = null;
            } else if (key instanceof ByteWritable) {
                ByteWritable bwKey = (ByteWritable) key;
                keyValue = bwKey.get();
            } else if (key instanceof ShortWritable) {
                ShortWritable swKey = (ShortWritable) key;
                keyValue = swKey.get();
            } else if (key instanceof IntWritable) {
                IntWritable iwKey = (IntWritable) key;
                keyValue = iwKey.get();
            } else if (key instanceof LongWritable) {
                LongWritable lwKey = (LongWritable) key;
                keyValue = lwKey.get();
            } else if (key instanceof TimestampWritable) {
                TimestampWritable twKey = (TimestampWritable) key;
                keyValue = twKey.getTimestamp();
            } else if (key instanceof DoubleWritable) {
                DoubleWritable dwKey = (DoubleWritable) key;
                keyValue = dwKey.get();
            } else if (key instanceof FloatWritable) {
                FloatWritable fwKey = (FloatWritable) key;
                keyValue = fwKey.get();
            } else if (key instanceof BooleanWritable) {
                BooleanWritable bwKey = (BooleanWritable) key;
                keyValue = bwKey.get();
            } else if (key instanceof HiveDecimalWritable) {
                HiveDecimalWritable hdwKey = (HiveDecimalWritable) key;
                keyValue = hdwKey.getHiveDecimal();
            } else {
                Assert.fail(String.format("Not implemented key output type %s: %s", key.getClass().getName(), key));
            }
            String keyValueAsString = String.format("%s", keyValue);
            assertTrue(expected.containsKey(keyValue));
            Object expectedValue = expected.get(keyValue);
            Object value = fields[1];
            Validator validator = getValidator(aggregateName);
            validator.validate(keyValueAsString, expectedValue, new Object[] { value });
            keys.add(keyValue);
        }

        private FakeCaptureVectorToRowOutputOperator.OutputInspector init(String aggregateName, Map<Object, Object> expected, Set<Object> keys) {
            this.aggregateName = aggregateName;
            this.expected = expected;
            this.keys = keys;
            return this;
        }
    }.init(aggregateName, expected, keys));
    for (VectorizedRowBatch unit : data) {
        vgo.process(unit, 0);
    }
    vgo.close(false);
    List<Object> outBatchList = out.getCapturedRows();
    assertNotNull(outBatchList);
    assertEquals(expected.size(), outBatchList.size());
    assertEquals(expected.size(), keys.size());
}

Example 8 with FakeCaptureVectorToRowOutputOperator

use of org.apache.hadoop.hive.ql.exec.vector.util.FakeCaptureVectorToRowOutputOperator in project hive by apache.

the class TestVectorGroupByOperator method testAggregateLongKeyIterable.

public void testAggregateLongKeyIterable(String aggregateName, Iterable<VectorizedRowBatch> data, HashMap<Object, Object> expected) throws HiveException {
    List<String> mapColumnNames = new ArrayList<String>();
    mapColumnNames.add("Key");
    mapColumnNames.add("Value");
    VectorizationContext ctx = new VectorizationContext("name", mapColumnNames);
    Set<Object> keys = new HashSet<Object>();
    Pair<GroupByDesc, VectorGroupByDesc> pair = buildKeyGroupByDesc(ctx, aggregateName, "Value", TypeInfoFactory.longTypeInfo, "Key", TypeInfoFactory.longTypeInfo);
    GroupByDesc desc = pair.fst;
    VectorGroupByDesc vectorDesc = pair.snd;
    CompilationOpContext cCtx = new CompilationOpContext();
    Operator<? extends OperatorDesc> groupByOp = OperatorFactory.get(cCtx, desc);
    VectorGroupByOperator vgo = (VectorGroupByOperator) Vectorizer.vectorizeGroupByOperator(groupByOp, ctx, vectorDesc);
    FakeCaptureVectorToRowOutputOperator out = FakeCaptureVectorToRowOutputOperator.addCaptureOutputChild(cCtx, vgo);
    vgo.initialize(hconf, null);
    out.setOutputInspector(new FakeCaptureVectorToRowOutputOperator.OutputInspector() {

        private String aggregateName;

        private HashMap<Object, Object> expected;

        private Set<Object> keys;

        @Override
        public void inspectRow(Object row, int tag) throws HiveException {
            assertTrue(row instanceof Object[]);
            Object[] fields = (Object[]) row;
            assertEquals(2, fields.length);
            Object key = fields[0];
            Long keyValue = null;
            if (null != key) {
                assertTrue(key instanceof LongWritable);
                LongWritable lwKey = (LongWritable) key;
                keyValue = lwKey.get();
            }
            assertTrue(expected.containsKey(keyValue));
            String keyAsString = String.format("%s", key);
            Object expectedValue = expected.get(keyValue);
            Object value = fields[1];
            Validator validator = getValidator(aggregateName);
            validator.validate(keyAsString, expectedValue, new Object[] { value });
            keys.add(keyValue);
        }

        private FakeCaptureVectorToRowOutputOperator.OutputInspector init(String aggregateName, HashMap<Object, Object> expected, Set<Object> keys) {
            this.aggregateName = aggregateName;
            this.expected = expected;
            this.keys = keys;
            return this;
        }
    }.init(aggregateName, expected, keys));
    for (VectorizedRowBatch unit : data) {
        vgo.process(unit, 0);
    }
    vgo.close(false);
    List<Object> outBatchList = out.getCapturedRows();
    assertNotNull(outBatchList);
    assertEquals(expected.size(), outBatchList.size());
    assertEquals(expected.size(), keys.size());
}

Example 9 with FakeCaptureVectorToRowOutputOperator

use of org.apache.hadoop.hive.ql.exec.vector.util.FakeCaptureVectorToRowOutputOperator in project hive by apache.

the class TestVectorGroupByOperator method testMultiKey.

private void testMultiKey(String aggregateName, FakeVectorRowBatchFromObjectIterables data, HashMap<Object, Object> expected) throws HiveException {
    Map<String, Integer> mapColumnNames = new HashMap<String, Integer>();
    ArrayList<String> outputColumnNames = new ArrayList<String>();
    ArrayList<ExprNodeDesc> keysDesc = new ArrayList<ExprNodeDesc>();
    Set<Object> keys = new HashSet<Object>();
    // The types array tells us the number of columns in the data
    final String[] columnTypes = data.getTypes();
    // Columns 0..N-1 are keys. Column N is the aggregate value input
    int i = 0;
    for (; i < columnTypes.length - 1; ++i) {
        String columnName = String.format("_col%d", i);
        mapColumnNames.put(columnName, i);
        outputColumnNames.add(columnName);
    }
    mapColumnNames.put("value", i);
    outputColumnNames.add("value");
    VectorizationContext ctx = new VectorizationContext("name", outputColumnNames);
    ArrayList<AggregationDesc> aggs = new ArrayList(1);
    aggs.add(buildAggregationDesc(ctx, aggregateName, GenericUDAFEvaluator.Mode.PARTIAL1, "value", TypeInfoFactory.getPrimitiveTypeInfo(columnTypes[i])));
    for (i = 0; i < columnTypes.length - 1; ++i) {
        String columnName = String.format("_col%d", i);
        keysDesc.add(buildColumnDesc(ctx, columnName, TypeInfoFactory.getPrimitiveTypeInfo(columnTypes[i])));
    }
    GroupByDesc desc = new GroupByDesc();
    VectorGroupByDesc vectorGroupByDesc = new VectorGroupByDesc();
    desc.setOutputColumnNames(outputColumnNames);
    desc.setAggregators(aggs);
    desc.setKeys(keysDesc);
    vectorGroupByDesc.setProcessingMode(ProcessingMode.HASH);
    CompilationOpContext cCtx = new CompilationOpContext();
    Operator<? extends OperatorDesc> groupByOp = OperatorFactory.get(cCtx, desc);
    VectorGroupByOperator vgo = (VectorGroupByOperator) Vectorizer.vectorizeGroupByOperator(groupByOp, ctx, vectorGroupByDesc);
    FakeCaptureVectorToRowOutputOperator out = FakeCaptureVectorToRowOutputOperator.addCaptureOutputChild(cCtx, vgo);
    vgo.initialize(hconf, null);
    out.setOutputInspector(new FakeCaptureVectorToRowOutputOperator.OutputInspector() {

        private int rowIndex;

        private String aggregateName;

        private Map<Object, Object> expected;

        private Set<Object> keys;

        @Override
        public void inspectRow(Object row, int tag) throws HiveException {
            assertTrue(row instanceof Object[]);
            Object[] fields = (Object[]) row;
            assertEquals(columnTypes.length, fields.length);
            ArrayList<Object> keyValue = new ArrayList<Object>(columnTypes.length - 1);
            for (int i = 0; i < columnTypes.length - 1; ++i) {
                Object key = fields[i];
                if (null == key) {
                    keyValue.add(null);
                } else if (key instanceof Text) {
                    Text txKey = (Text) key;
                    keyValue.add(txKey.toString());
                } else if (key instanceof ByteWritable) {
                    ByteWritable bwKey = (ByteWritable) key;
                    keyValue.add(bwKey.get());
                } else if (key instanceof ShortWritable) {
                    ShortWritable swKey = (ShortWritable) key;
                    keyValue.add(swKey.get());
                } else if (key instanceof IntWritable) {
                    IntWritable iwKey = (IntWritable) key;
                    keyValue.add(iwKey.get());
                } else if (key instanceof LongWritable) {
                    LongWritable lwKey = (LongWritable) key;
                    keyValue.add(lwKey.get());
                } else if (key instanceof TimestampWritable) {
                    TimestampWritable twKey = (TimestampWritable) key;
                    keyValue.add(twKey.getTimestamp());
                } else if (key instanceof DoubleWritable) {
                    DoubleWritable dwKey = (DoubleWritable) key;
                    keyValue.add(dwKey.get());
                } else if (key instanceof FloatWritable) {
                    FloatWritable fwKey = (FloatWritable) key;
                    keyValue.add(fwKey.get());
                } else if (key instanceof BooleanWritable) {
                    BooleanWritable bwKey = (BooleanWritable) key;
                    keyValue.add(bwKey.get());
                } else {
                    Assert.fail(String.format("Not implemented key output type %s: %s", key.getClass().getName(), key));
                }
            }
            String keyAsString = Arrays.deepToString(keyValue.toArray());
            assertTrue(expected.containsKey(keyValue));
            Object expectedValue = expected.get(keyValue);
            Object value = fields[columnTypes.length - 1];
            Validator validator = getValidator(aggregateName);
            validator.validate(keyAsString, expectedValue, new Object[] { value });
            keys.add(keyValue);
        }

        private FakeCaptureVectorToRowOutputOperator.OutputInspector init(String aggregateName, Map<Object, Object> expected, Set<Object> keys) {
            this.aggregateName = aggregateName;
            this.expected = expected;
            this.keys = keys;
            return this;
        }
    }.init(aggregateName, expected, keys));
    for (VectorizedRowBatch unit : data) {
        vgo.process(unit, 0);
    }
    vgo.close(false);
    List<Object> outBatchList = out.getCapturedRows();
    assertNotNull(outBatchList);
    assertEquals(expected.size(), outBatchList.size());
    assertEquals(expected.size(), keys.size());
}

Example 10 with FakeCaptureVectorToRowOutputOperator

use of org.apache.hadoop.hive.ql.exec.vector.util.FakeCaptureVectorToRowOutputOperator in project hive by apache.

the class TestVectorGroupByOperator method testAggregateCountStarIterable.

public void testAggregateCountStarIterable(Iterable<VectorizedRowBatch> data, Object expected) throws HiveException {
    List<String> mapColumnNames = new ArrayList<String>();
    mapColumnNames.add("A");
    VectorizationContext ctx = new VectorizationContext("name", mapColumnNames);
    Pair<GroupByDesc, VectorGroupByDesc> pair = buildGroupByDescCountStar(ctx);
    GroupByDesc desc = pair.fst;
    VectorGroupByDesc vectorDesc = pair.snd;
    vectorDesc.setProcessingMode(ProcessingMode.HASH);
    CompilationOpContext cCtx = new CompilationOpContext();
    Operator<? extends OperatorDesc> groupByOp = OperatorFactory.get(cCtx, desc);
    VectorGroupByOperator vgo = (VectorGroupByOperator) Vectorizer.vectorizeGroupByOperator(groupByOp, ctx, vectorDesc);
    FakeCaptureVectorToRowOutputOperator out = FakeCaptureVectorToRowOutputOperator.addCaptureOutputChild(cCtx, vgo);
    vgo.initialize(hconf, null);
    for (VectorizedRowBatch unit : data) {
        vgo.process(unit, 0);
    }
    vgo.close(false);
    List<Object> outBatchList = out.getCapturedRows();
    assertNotNull(outBatchList);
    assertEquals(1, outBatchList.size());
    Object result = outBatchList.get(0);
    Validator validator = getValidator("count");
    validator.validate("_total", expected, result);
}