Examples with MapWork org.apache.hadoop.hive.ql.plan.MapWork used on opensource projects

Example 1 with MapWork

use of org.apache.hadoop.hive.ql.plan.MapWork in project hive by apache.

the class GenericUDTFGetSplits method getSplits.

public InputSplit[] getSplits(JobConf job, int numSplits, TezWork work, Schema schema) throws IOException {
    DAG dag = DAG.create(work.getName());
    dag.setCredentials(job.getCredentials());
    DagUtils utils = DagUtils.getInstance();
    Context ctx = new Context(job);
    MapWork mapWork = (MapWork) work.getAllWork().get(0);
    // bunch of things get setup in the context based on conf but we need only the MR tmp directory
    // for the following method.
    JobConf wxConf = utils.initializeVertexConf(job, ctx, mapWork);
    // TODO: should we also whitelist input formats here? from mapred.input.format.class
    Path scratchDir = utils.createTezDir(ctx.getMRScratchDir(), job);
    FileSystem fs = scratchDir.getFileSystem(job);
    try {
        LocalResource appJarLr = createJarLocalResource(utils.getExecJarPathLocal(), utils, job);
        Vertex wx = utils.createVertex(wxConf, mapWork, scratchDir, appJarLr, new ArrayList<LocalResource>(), fs, ctx, false, work, work.getVertexType(mapWork));
        String vertexName = wx.getName();
        dag.addVertex(wx);
        utils.addCredentials(mapWork, dag);
        // we have the dag now proceed to get the splits:
        Preconditions.checkState(HiveConf.getBoolVar(wxConf, ConfVars.HIVE_TEZ_GENERATE_CONSISTENT_SPLITS));
        Preconditions.checkState(HiveConf.getBoolVar(wxConf, ConfVars.LLAP_CLIENT_CONSISTENT_SPLITS));
        HiveSplitGenerator splitGenerator = new HiveSplitGenerator(wxConf, mapWork);
        List<Event> eventList = splitGenerator.initialize();
        InputSplit[] result = new InputSplit[eventList.size() - 1];
        InputConfigureVertexTasksEvent configureEvent = (InputConfigureVertexTasksEvent) eventList.get(0);
        List<TaskLocationHint> hints = configureEvent.getLocationHint().getTaskLocationHints();
        Preconditions.checkState(hints.size() == eventList.size() - 1);
        if (LOG.isDebugEnabled()) {
            LOG.debug("NumEvents=" + eventList.size() + ", NumSplits=" + result.length);
        }
        LlapCoordinator coordinator = LlapCoordinator.getInstance();
        if (coordinator == null) {
            throw new IOException("LLAP coordinator is not initialized; must be running in HS2 with " + ConfVars.LLAP_HS2_ENABLE_COORDINATOR.varname + " enabled");
        }
        // See the discussion in the implementation as to why we generate app ID.
        ApplicationId applicationId = coordinator.createExtClientAppId();
        // This assumes LLAP cluster owner is always the HS2 user.
        String llapUser = UserGroupInformation.getLoginUser().getShortUserName();
        String queryUser = null;
        byte[] tokenBytes = null;
        LlapSigner signer = null;
        if (UserGroupInformation.isSecurityEnabled()) {
            signer = coordinator.getLlapSigner(job);
            // 1. Generate the token for query user (applies to all splits).
            queryUser = SessionState.getUserFromAuthenticator();
            if (queryUser == null) {
                queryUser = UserGroupInformation.getCurrentUser().getUserName();
                LOG.warn("Cannot determine the session user; using " + queryUser + " instead");
            }
            LlapTokenLocalClient tokenClient = coordinator.getLocalTokenClient(job, llapUser);
            // We put the query user, not LLAP user, into the message and token.
            Token<LlapTokenIdentifier> token = tokenClient.createToken(applicationId.toString(), queryUser, true);
            LOG.info("Created the token for remote user: {}", token);
            bos.reset();
            token.write(dos);
            tokenBytes = bos.toByteArray();
        } else {
            queryUser = UserGroupInformation.getCurrentUser().getUserName();
        }
        LOG.info("Number of splits: " + (eventList.size() - 1));
        SignedMessage signedSvs = null;
        for (int i = 0; i < eventList.size() - 1; i++) {
            TaskSpec taskSpec = new TaskSpecBuilder().constructTaskSpec(dag, vertexName, eventList.size() - 1, applicationId, i);
            // 2. Generate the vertex/submit information for all events.
            if (i == 0) {
                // The queryId could either be picked up from the current request being processed, or
                // generated. The current request isn't exactly correct since the query is 'done' once we
                // return the results. Generating a new one has the added benefit of working once this
                // is moved out of a UDTF into a proper API.
                // Setting this to the generated AppId which is unique.
                // Despite the differences in TaskSpec, the vertex spec should be the same.
                signedSvs = createSignedVertexSpec(signer, taskSpec, applicationId, queryUser, applicationId.toString());
            }
            SubmitWorkInfo submitWorkInfo = new SubmitWorkInfo(applicationId, System.currentTimeMillis(), taskSpec.getVertexParallelism(), signedSvs.message, signedSvs.signature);
            byte[] submitWorkBytes = SubmitWorkInfo.toBytes(submitWorkInfo);
            // 3. Generate input event.
            SignedMessage eventBytes = makeEventBytes(wx, vertexName, eventList.get(i + 1), signer);
            // 4. Make location hints.
            SplitLocationInfo[] locations = makeLocationHints(hints.get(i));
            result[i] = new LlapInputSplit(i, submitWorkBytes, eventBytes.message, eventBytes.signature, locations, schema, llapUser, tokenBytes);
        }
        return result;
    } catch (Exception e) {
        throw new IOException(e);
    }
}

Example 2 with MapWork

use of org.apache.hadoop.hive.ql.plan.MapWork in project hive by apache.

the class TestTezTask method setUp.

@SuppressWarnings("unchecked")
@Before
public void setUp() throws Exception {
    utils = mock(DagUtils.class);
    fs = mock(FileSystem.class);
    path = mock(Path.class);
    when(path.getFileSystem(any(Configuration.class))).thenReturn(fs);
    when(utils.getTezDir(any(Path.class))).thenReturn(path);
    when(utils.createVertex(any(JobConf.class), any(BaseWork.class), any(Path.class), any(LocalResource.class), any(List.class), any(FileSystem.class), any(Context.class), anyBoolean(), any(TezWork.class), any(VertexType.class))).thenAnswer(new Answer<Vertex>() {

        @Override
        public Vertex answer(InvocationOnMock invocation) throws Throwable {
            Object[] args = invocation.getArguments();
            return Vertex.create(((BaseWork) args[1]).getName(), mock(ProcessorDescriptor.class), 0, mock(Resource.class));
        }
    });
    when(utils.createEdge(any(JobConf.class), any(Vertex.class), any(Vertex.class), any(TezEdgeProperty.class), any(VertexType.class))).thenAnswer(new Answer<Edge>() {

        @Override
        public Edge answer(InvocationOnMock invocation) throws Throwable {
            Object[] args = invocation.getArguments();
            return Edge.create((Vertex) args[1], (Vertex) args[2], mock(EdgeProperty.class));
        }
    });
    work = new TezWork("", null);
    mws = new MapWork[] { new MapWork(), new MapWork() };
    rws = new ReduceWork[] { new ReduceWork(), new ReduceWork() };
    work.addAll(mws);
    work.addAll(rws);
    int i = 0;
    for (BaseWork w : work.getAllWork()) {
        w.setName("Work " + (++i));
    }
    op = mock(Operator.class);
    LinkedHashMap<String, Operator<? extends OperatorDesc>> map = new LinkedHashMap<String, Operator<? extends OperatorDesc>>();
    map.put("foo", op);
    mws[0].setAliasToWork(map);
    mws[1].setAliasToWork(map);
    LinkedHashMap<Path, ArrayList<String>> pathMap = new LinkedHashMap<>();
    ArrayList<String> aliasList = new ArrayList<String>();
    aliasList.add("foo");
    pathMap.put(new Path("foo"), aliasList);
    mws[0].setPathToAliases(pathMap);
    mws[1].setPathToAliases(pathMap);
    rws[0].setReducer(op);
    rws[1].setReducer(op);
    TezEdgeProperty edgeProp = new TezEdgeProperty(EdgeType.SIMPLE_EDGE);
    work.connect(mws[0], rws[0], edgeProp);
    work.connect(mws[1], rws[0], edgeProp);
    work.connect(rws[0], rws[1], edgeProp);
    task = new TezTask(utils);
    task.setWork(work);
    task.setConsole(mock(LogHelper.class));
    QueryPlan mockQueryPlan = mock(QueryPlan.class);
    doReturn(UUID.randomUUID().toString()).when(mockQueryPlan).getQueryId();
    task.setQueryPlan(mockQueryPlan);
    conf = new JobConf();
    appLr = mock(LocalResource.class);
    HiveConf hiveConf = new HiveConf();
    hiveConf.setVar(HiveConf.ConfVars.HIVE_AUTHORIZATION_MANAGER, "org.apache.hadoop.hive.ql.security.authorization.plugin.sqlstd.SQLStdHiveAuthorizerFactory");
    SessionState.start(hiveConf);
    session = mock(TezClient.class);
    sessionState = mock(TezSessionState.class);
    when(sessionState.getSession()).thenReturn(session);
    when(session.submitDAG(any(DAG.class))).thenThrow(new SessionNotRunning("")).thenReturn(mock(DAGClient.class));
}

Example 3 with MapWork

use of org.apache.hadoop.hive.ql.plan.MapWork in project hive by apache.

the class TestUtilities method runTestGetInputSummary.

private ContentSummary runTestGetInputSummary(JobConf jobConf, Properties properties, int numOfPartitions, int bytesPerFile, Class<? extends InputFormat> inputFormatClass) throws IOException {
    // creates scratch directories needed by the Context object
    SessionState.start(new HiveConf());
    MapWork mapWork = new MapWork();
    Context context = new Context(jobConf);
    LinkedHashMap<Path, PartitionDesc> pathToPartitionInfo = new LinkedHashMap<>();
    LinkedHashMap<Path, ArrayList<String>> pathToAliasTable = new LinkedHashMap<>();
    TableScanOperator scanOp = new TableScanOperator();
    PartitionDesc partitionDesc = new PartitionDesc(new TableDesc(inputFormatClass, null, properties), null);
    String testTableName = "testTable";
    Path testTablePath = new Path(testTableName);
    Path[] testPartitionsPaths = new Path[numOfPartitions];
    for (int i = 0; i < numOfPartitions; i++) {
        String testPartitionName = "p=" + 1;
        testPartitionsPaths[i] = new Path(testTablePath, "p=" + i);
        pathToPartitionInfo.put(testPartitionsPaths[i], partitionDesc);
        pathToAliasTable.put(testPartitionsPaths[i], Lists.newArrayList(testPartitionName));
        mapWork.getAliasToWork().put(testPartitionName, scanOp);
    }
    mapWork.setPathToAliases(pathToAliasTable);
    mapWork.setPathToPartitionInfo(pathToPartitionInfo);
    FileSystem fs = FileSystem.getLocal(jobConf);
    try {
        fs.mkdirs(testTablePath);
        byte[] data = new byte[bytesPerFile];
        for (int i = 0; i < numOfPartitions; i++) {
            fs.mkdirs(testPartitionsPaths[i]);
            FSDataOutputStream out = fs.create(new Path(testPartitionsPaths[i], "test1.txt"));
            out.write(data);
            out.close();
        }
        return Utilities.getInputSummary(context, mapWork, null);
    } finally {
        if (fs.exists(testTablePath)) {
            fs.delete(testTablePath, true);
        }
    }
}

Example 4 with MapWork

use of org.apache.hadoop.hive.ql.plan.MapWork in project hive by apache.

the class TestUtilities method testGetInputPathsWithEmptyTables.

/**
   * Check that calling {@link Utilities#getInputPaths(JobConf, MapWork, Path, Context, boolean)}
   * can process two different empty tables without throwing any exceptions.
   */
@Test
public void testGetInputPathsWithEmptyTables() throws Exception {
    String alias1Name = "alias1";
    String alias2Name = "alias2";
    MapWork mapWork1 = new MapWork();
    MapWork mapWork2 = new MapWork();
    JobConf jobConf = new JobConf();
    Path nonExistentPath1 = new Path(UUID.randomUUID().toString());
    Path nonExistentPath2 = new Path(UUID.randomUUID().toString());
    PartitionDesc mockPartitionDesc = mock(PartitionDesc.class);
    TableDesc mockTableDesc = mock(TableDesc.class);
    when(mockTableDesc.isNonNative()).thenReturn(false);
    when(mockTableDesc.getProperties()).thenReturn(new Properties());
    when(mockPartitionDesc.getProperties()).thenReturn(new Properties());
    when(mockPartitionDesc.getTableDesc()).thenReturn(mockTableDesc);
    doReturn(HiveSequenceFileOutputFormat.class).when(mockPartitionDesc).getOutputFileFormatClass();
    mapWork1.setPathToAliases(new LinkedHashMap<>(ImmutableMap.of(nonExistentPath1, Lists.newArrayList(alias1Name))));
    mapWork1.setAliasToWork(new LinkedHashMap<String, Operator<? extends OperatorDesc>>(ImmutableMap.of(alias1Name, (Operator<?>) mock(Operator.class))));
    mapWork1.setPathToPartitionInfo(new LinkedHashMap<>(ImmutableMap.of(nonExistentPath1, mockPartitionDesc)));
    mapWork2.setPathToAliases(new LinkedHashMap<>(ImmutableMap.of(nonExistentPath2, Lists.newArrayList(alias2Name))));
    mapWork2.setAliasToWork(new LinkedHashMap<String, Operator<? extends OperatorDesc>>(ImmutableMap.of(alias2Name, (Operator<?>) mock(Operator.class))));
    mapWork2.setPathToPartitionInfo(new LinkedHashMap<>(ImmutableMap.of(nonExistentPath2, mockPartitionDesc)));
    List<Path> inputPaths = new ArrayList<>();
    try {
        Path scratchDir = new Path(HiveConf.getVar(jobConf, HiveConf.ConfVars.LOCALSCRATCHDIR));
        inputPaths.addAll(Utilities.getInputPaths(jobConf, mapWork1, scratchDir, mock(Context.class), false));
        inputPaths.addAll(Utilities.getInputPaths(jobConf, mapWork2, scratchDir, mock(Context.class), false));
        assertEquals(inputPaths.size(), 2);
    } finally {
        File file;
        for (Path path : inputPaths) {
            file = new File(path.toString());
            if (file.exists()) {
                file.delete();
            }
        }
    }
}

Example 5 with MapWork

use of org.apache.hadoop.hive.ql.plan.MapWork in project hive by apache.

the class TestDynamicPartitionPruner method testSingleSourceMultipleFiltersOrdering2.

@Test(timeout = 5000)
public void testSingleSourceMultipleFiltersOrdering2() throws InterruptedException, SerDeException {
    InputInitializerContext mockInitContext = mock(InputInitializerContext.class);
    doReturn(2).when(mockInitContext).getVertexNumTasks("v1");
    MapWork mapWork = createMockMapWork(new TestSource("v1", 2));
    DynamicPartitionPruner pruner = new DynamicPartitionPrunerForEventTesting(mockInitContext, mapWork);
    PruneRunnable pruneRunnable = new PruneRunnable(pruner);
    Thread t = new Thread(pruneRunnable);
    t.start();
    try {
        pruneRunnable.start();
        InputInitializerEvent event = InputInitializerEvent.create("FakeTarget", "TargetInput", ByteBuffer.allocate(0));
        event.setSourceVertexName("v1");
        pruner.processVertex("v1");
        pruner.addEvent(event);
        pruner.addEvent(event);
        pruner.addEvent(event);
        pruner.addEvent(event);
        pruneRunnable.awaitEnd();
        assertFalse(pruneRunnable.inError.get());
    } finally {
        t.interrupt();
        t.join();
    }
}