Search in sources :

Example 11 with Input

use of org.apache.tez.runtime.api.Input in project hive by apache.

the class MergeFileRecordProcessor method getMRInput.

private MRInputLegacy getMRInput(Map<String, LogicalInput> inputs) throws Exception {
    LOG.info("The inputs are: " + inputs);
    // start the mr input and wait for ready event. number of MRInput is expected to be 1
    List<Input> li = Lists.newArrayList();
    int numMRInputs = 0;
    for (LogicalInput inp : inputs.values()) {
        if (inp instanceof MRInputLegacy) {
            numMRInputs++;
            if (numMRInputs > 1) {
                throw new IllegalArgumentException("Only one MRInput is expected");
            }
            inp.start();
            li.add(inp);
        } else {
            throw new IllegalArgumentException("Expecting only one input of type MRInputLegacy." + " Found type: " + inp.getClass().getCanonicalName());
        }
    }
    // typically alter table .. concatenate is run on only one partition/one table,
    // so it doesn't matter if we wait for all inputs or any input to be ready.
    processorContext.waitForAnyInputReady(li);
    final MRInputLegacy theMRInput;
    if (li.size() == 1) {
        theMRInput = (MRInputLegacy) li.get(0);
        theMRInput.init();
    } else {
        throw new IllegalArgumentException("MRInputs count is expected to be 1");
    }
    return theMRInput;
}
Also used : LogicalInput(org.apache.tez.runtime.api.LogicalInput) Input(org.apache.tez.runtime.api.Input) LogicalInput(org.apache.tez.runtime.api.LogicalInput) MRInputLegacy(org.apache.tez.mapreduce.input.MRInputLegacy)

Example 12 with Input

use of org.apache.tez.runtime.api.Input in project hive by apache.

the class ReduceRecordProcessor method init.

@Override
void init(MRTaskReporter mrReporter, Map<String, LogicalInput> inputs, Map<String, LogicalOutput> outputs) throws Exception {
    perfLogger.perfLogBegin(CLASS_NAME, PerfLogger.TEZ_INIT_OPERATORS);
    super.init(mrReporter, inputs, outputs);
    MapredContext.init(false, new JobConf(jconf));
    List<LogicalInput> shuffleInputs = getShuffleInputs(inputs);
    // TODO HIVE-14042. Move to using a loop and a timed wait once TEZ-3302 is fixed.
    checkAbortCondition();
    if (shuffleInputs != null) {
        LOG.info("Waiting for ShuffleInputs to become ready");
        processorContext.waitForAllInputsReady(new ArrayList<Input>(shuffleInputs));
    }
    connectOps.clear();
    ReduceWork redWork = reduceWork;
    LOG.info("Main work is " + reduceWork.getName());
    List<HashTableDummyOperator> workOps = reduceWork.getDummyOps();
    Set<HashTableDummyOperator> dummyOps = workOps == null ? new HashSet<>() : new HashSet<>(workOps);
    tagToReducerMap.put(redWork.getTag(), redWork);
    if (mergeWorkList != null) {
        for (BaseWork mergeWork : mergeWorkList) {
            LOG.debug("Additional work {}", mergeWork.getName());
            workOps = mergeWork.getDummyOps();
            if (workOps != null) {
                dummyOps.addAll(workOps);
            }
            ReduceWork mergeReduceWork = (ReduceWork) mergeWork;
            reducer = mergeReduceWork.getReducer();
            // Check immediately after reducer is assigned, in cae the abort came in during
            checkAbortCondition();
            DummyStoreOperator dummyStoreOp = getJoinParentOp(reducer);
            connectOps.put(mergeReduceWork.getTag(), dummyStoreOp);
            tagToReducerMap.put(mergeReduceWork.getTag(), mergeReduceWork);
        }
        ((TezContext) MapredContext.get()).setDummyOpsMap(connectOps);
    }
    checkAbortCondition();
    bigTablePosition = (byte) reduceWork.getTag();
    ObjectInspector[] mainWorkOIs = null;
    ((TezContext) MapredContext.get()).setInputs(inputs);
    ((TezContext) MapredContext.get()).setTezProcessorContext(processorContext);
    int numTags = reduceWork.getTagToValueDesc().size();
    reducer = reduceWork.getReducer();
    // Check immediately after reducer is assigned, in cae the abort came in during
    checkAbortCondition();
    // set memory available for operators
    long memoryAvailableToTask = processorContext.getTotalMemoryAvailableToTask();
    if (reducer.getConf() != null) {
        reducer.getConf().setMaxMemoryAvailable(memoryAvailableToTask);
        LOG.info("Memory available for operators set to {}", LlapUtil.humanReadableByteCount(memoryAvailableToTask));
    }
    OperatorUtils.setMemoryAvailable(reducer.getChildOperators(), memoryAvailableToTask);
    // Setup values registry
    String valueRegistryKey = DynamicValue.DYNAMIC_VALUE_REGISTRY_CACHE_KEY;
    DynamicValueRegistryTez registryTez = dynamicValueCache.retrieve(valueRegistryKey, () -> new DynamicValueRegistryTez());
    dynamicValueCacheKeys.add(valueRegistryKey);
    RegistryConfTez registryConf = new RegistryConfTez(jconf, reduceWork, processorContext, inputs);
    registryTez.init(registryConf);
    checkAbortCondition();
    if (numTags > 1) {
        sources = new ReduceRecordSource[numTags];
        mainWorkOIs = new ObjectInspector[numTags];
        initializeMultipleSources(reduceWork, numTags, mainWorkOIs, sources);
        ((TezContext) MapredContext.get()).setRecordSources(sources);
        reducer.initialize(jconf, mainWorkOIs);
    } else {
        numTags = tagToReducerMap.keySet().size();
        sources = new ReduceRecordSource[numTags];
        mainWorkOIs = new ObjectInspector[numTags];
        for (int i : tagToReducerMap.keySet()) {
            redWork = tagToReducerMap.get(i);
            reducer = redWork.getReducer();
            // Check immediately after reducer is assigned, in cae the abort came in during
            checkAbortCondition();
            initializeSourceForTag(redWork, i, mainWorkOIs, sources, redWork.getTagToValueDesc().get(0), redWork.getTagToInput().get(0));
            reducer.initializeLocalWork(jconf);
        }
        reducer = reduceWork.getReducer();
        // Check immediately after reducer is assigned, in cae the abort came in during
        checkAbortCondition();
        ((TezContext) MapredContext.get()).setRecordSources(sources);
        reducer.initialize(jconf, new ObjectInspector[] { mainWorkOIs[bigTablePosition] });
        for (int i : tagToReducerMap.keySet()) {
            if (i == bigTablePosition) {
                continue;
            }
            redWork = tagToReducerMap.get(i);
            reducer = redWork.getReducer();
            // Check immediately after reducer is assigned, in cae the abort came in during
            checkAbortCondition();
            reducer.initialize(jconf, new ObjectInspector[] { mainWorkOIs[i] });
        }
    }
    checkAbortCondition();
    reducer = reduceWork.getReducer();
    // initialize reduce operator tree
    try {
        LOG.info(reducer.dump(0));
        // dummy parent operators as well.
        for (HashTableDummyOperator dummyOp : dummyOps) {
            // TODO HIVE-14042. Propagating abort to dummyOps.
            dummyOp.initialize(jconf, null);
            checkAbortCondition();
        }
        // set output collector for any reduce sink operators in the pipeline.
        List<Operator<?>> children = new ArrayList<>();
        children.add(reducer);
        children.addAll(dummyOps);
        createOutputMap();
        OperatorUtils.setChildrenCollector(children, outMap);
        checkAbortCondition();
        reducer.setReporter(reporter);
        MapredContext.get().setReporter(reporter);
    } catch (Throwable e) {
        super.setAborted(true);
        if (e instanceof OutOfMemoryError) {
            // Don't create a new object if we are already out of memory
            throw (OutOfMemoryError) e;
        } else if (e instanceof InterruptedException) {
            LOG.info("Hit an interrupt while initializing ReduceRecordProcessor. Message={}", e.getMessage());
            throw (InterruptedException) e;
        } else {
            throw new RuntimeException(redWork.getName() + " operator initialization failed", e);
        }
    }
    perfLogger.perfLogEnd(CLASS_NAME, PerfLogger.TEZ_INIT_OPERATORS);
}
Also used : Operator(org.apache.hadoop.hive.ql.exec.Operator) DummyStoreOperator(org.apache.hadoop.hive.ql.exec.DummyStoreOperator) HashTableDummyOperator(org.apache.hadoop.hive.ql.exec.HashTableDummyOperator) ArrayList(java.util.ArrayList) LogicalInput(org.apache.tez.runtime.api.LogicalInput) Input(org.apache.tez.runtime.api.Input) JobConf(org.apache.hadoop.mapred.JobConf) BaseWork(org.apache.hadoop.hive.ql.plan.BaseWork) ObjectInspector(org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector) DummyStoreOperator(org.apache.hadoop.hive.ql.exec.DummyStoreOperator) ReduceWork(org.apache.hadoop.hive.ql.plan.ReduceWork) HashTableDummyOperator(org.apache.hadoop.hive.ql.exec.HashTableDummyOperator) LogicalInput(org.apache.tez.runtime.api.LogicalInput) RegistryConfTez(org.apache.hadoop.hive.ql.exec.tez.DynamicValueRegistryTez.RegistryConfTez)

Example 13 with Input

use of org.apache.tez.runtime.api.Input in project hive by apache.

the class MapRecordProcessor method getMRInput.

private MRInputLegacy getMRInput(Map<String, LogicalInput> inputs) throws Exception {
    // there should be only one MRInput
    MRInputLegacy theMRInput = null;
    // start all mr/multi-mr inputs
    Set<Input> li = new HashSet<>();
    for (LogicalInput inp : inputs.values()) {
        if (inp instanceof MRInputLegacy || inp instanceof MultiMRInput) {
            inp.start();
            li.add(inp);
        }
    }
    // TODO: HIVE-14042. Potential blocking call. MRInput handles this correctly even if an interrupt is swallowed.
    // MultiMRInput may not. Fix once TEZ-3302 is resolved.
    processorContext.waitForAllInputsReady(li);
    LOG.info("The input names are: {}", String.join(",", inputs.keySet()));
    for (Entry<String, LogicalInput> inp : inputs.entrySet()) {
        if (inp.getValue() instanceof MRInputLegacy) {
            if (theMRInput != null) {
                throw new IllegalArgumentException("Only one MRInput is expected");
            }
            // a better logic would be to find the alias
            theMRInput = (MRInputLegacy) inp.getValue();
        } else if (inp.getValue() instanceof MultiMRInput) {
            multiMRInputMap.put(inp.getKey(), (MultiMRInput) inp.getValue());
        }
    }
    if (theMRInput != null) {
        theMRInput.init();
    } else {
        String alias = mapWork.getAliasToWork().keySet().iterator().next();
        if (inputs.get(alias) instanceof MultiMRInput) {
            mainWorkMultiMRInput = (MultiMRInput) inputs.get(alias);
        } else {
            throw new IOException("Unexpected input type found: " + inputs.get(alias).getClass().getCanonicalName());
        }
    }
    return theMRInput;
}
Also used : LogicalInput(org.apache.tez.runtime.api.LogicalInput) Input(org.apache.tez.runtime.api.Input) MultiMRInput(org.apache.tez.mapreduce.input.MultiMRInput) LogicalInput(org.apache.tez.runtime.api.LogicalInput) IOException(java.io.IOException) MRInputLegacy(org.apache.tez.mapreduce.input.MRInputLegacy) HashSet(java.util.HashSet) MultiMRInput(org.apache.tez.mapreduce.input.MultiMRInput)

Example 14 with Input

use of org.apache.tez.runtime.api.Input in project tez by apache.

the class LogicalIOProcessorRuntimeTask method createInput.

private LogicalInput createInput(InputSpec inputSpec, InputContext inputContext) throws TezException {
    InputDescriptor inputDesc = inputSpec.getInputDescriptor();
    Input input = ReflectionUtils.createClazzInstance(inputDesc.getClassName(), new Class[] { InputContext.class, Integer.TYPE }, new Object[] { inputContext, inputSpec.getPhysicalEdgeCount() });
    if (!(input instanceof LogicalInput)) {
        throw new TezUncheckedException(inputDesc.getClass().getName() + " is not a sub-type of LogicalInput." + " Only LogicalInput sub-types supported by LogicalIOProcessor.");
    }
    return (LogicalInput) input;
}
Also used : InputDescriptor(org.apache.tez.dag.api.InputDescriptor) LogicalInput(org.apache.tez.runtime.api.LogicalInput) Input(org.apache.tez.runtime.api.Input) MergedLogicalInput(org.apache.tez.runtime.api.MergedLogicalInput) TezUncheckedException(org.apache.tez.dag.api.TezUncheckedException) LogicalInput(org.apache.tez.runtime.api.LogicalInput) MergedLogicalInput(org.apache.tez.runtime.api.MergedLogicalInput)

Example 15 with Input

use of org.apache.tez.runtime.api.Input in project tez by apache.

the class TestInputReadyTracker method testGrouped.

@Test(timeout = 20000)
public void testGrouped() throws InterruptedException {
    InputReadyTracker inputReadyTracker = new InputReadyTracker();
    ImmediatelyReadyInputForTest input1 = new ImmediatelyReadyInputForTest(inputReadyTracker);
    ControlledReadyInputForTest input2 = new ControlledReadyInputForTest(inputReadyTracker);
    ImmediatelyReadyInputForTest input3 = new ImmediatelyReadyInputForTest(inputReadyTracker);
    ControlledReadyInputForTest input4 = new ControlledReadyInputForTest(inputReadyTracker);
    List<Input> group1Inputs = new ArrayList<Input>();
    group1Inputs.add(input1);
    group1Inputs.add(input2);
    List<Input> group2Inputs = new ArrayList<Input>();
    group2Inputs.add(input3);
    group2Inputs.add(input4);
    Map<String, MergedLogicalInput> mergedInputMap = new HashMap<String, MergedLogicalInput>();
    MergedInputContext mergedInputContext1 = new TezMergedInputContextImpl(null, "group1", mergedInputMap, inputReadyTracker, null, null);
    MergedInputContext mergedInputContext2 = new TezMergedInputContextImpl(null, "group2", mergedInputMap, inputReadyTracker, null, null);
    AnyOneMergedInputForTest group1 = new AnyOneMergedInputForTest(mergedInputContext1, group1Inputs);
    AllMergedInputForTest group2 = new AllMergedInputForTest(mergedInputContext2, group2Inputs);
    mergedInputMap.put("group1", group1);
    mergedInputMap.put("group2", group2);
    // Register groups with tracker
    List<MergedLogicalInput> groups = Lists.newArrayList(group1, group2);
    inputReadyTracker.setGroupedInputs(groups);
    // Test for simple inputs
    List<Input> requestList;
    long startTime = 0l;
    long readyTime = 0l;
    requestList = new ArrayList<Input>();
    requestList.add(group1);
    Input readyInput = inputReadyTracker.waitForAnyInputReady(requestList);
    assertTrue(group1.isReady);
    assertTrue(input1.isReady);
    assertFalse(input2.isReady);
    assertEquals(group1, readyInput);
    requestList = new ArrayList<Input>();
    requestList.add(group2);
    startTime = System.nanoTime();
    setDelayedInputReady(input4);
    inputReadyTracker.waitForAllInputsReady(requestList);
    readyTime = System.nanoTime();
    // Should have moved into ready state - only happens when the setReady function is invoked.
    // Ensure the method returned only after the specific Input was told it is ready
    assertTrue(group2.isReady);
    assertTrue(input3.isReady);
    assertTrue(input4.isReady);
    assertTrue(readyTime >= startTime + SLEEP_TIME);
}
Also used : HashMap(java.util.HashMap) TezMergedInputContextImpl(org.apache.tez.runtime.api.impl.TezMergedInputContextImpl) MergedLogicalInput(org.apache.tez.runtime.api.MergedLogicalInput) ArrayList(java.util.ArrayList) MergedInputContext(org.apache.tez.runtime.api.MergedInputContext) Input(org.apache.tez.runtime.api.Input) AbstractLogicalInput(org.apache.tez.runtime.api.AbstractLogicalInput) MergedLogicalInput(org.apache.tez.runtime.api.MergedLogicalInput) Test(org.junit.Test)

Aggregations

Input (org.apache.tez.runtime.api.Input)20 Test (org.junit.Test)12 LinkedList (java.util.LinkedList)10 KeyValuesReader (org.apache.tez.runtime.library.api.KeyValuesReader)9 LogicalInput (org.apache.tez.runtime.api.LogicalInput)7 MergedLogicalInput (org.apache.tez.runtime.api.MergedLogicalInput)6 MergedInputContext (org.apache.tez.runtime.api.MergedInputContext)5 ArrayList (java.util.ArrayList)3 AbstractLogicalInput (org.apache.tez.runtime.api.AbstractLogicalInput)3 IOException (java.io.IOException)2 ObjectInspector (org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector)2 MRInputLegacy (org.apache.tez.mapreduce.input.MRInputLegacy)2 TezMergedInputContextImpl (org.apache.tez.runtime.api.impl.TezMergedInputContextImpl)2 KeyValueReader (org.apache.tez.runtime.library.api.KeyValueReader)2 HashMap (java.util.HashMap)1 HashSet (java.util.HashSet)1 DummyStoreOperator (org.apache.hadoop.hive.ql.exec.DummyStoreOperator)1 ExprNodeEvaluator (org.apache.hadoop.hive.ql.exec.ExprNodeEvaluator)1 HashTableDummyOperator (org.apache.hadoop.hive.ql.exec.HashTableDummyOperator)1 Operator (org.apache.hadoop.hive.ql.exec.Operator)1