Search in sources :

Example 1 with HiveSplitGenerator

use of org.apache.hadoop.hive.ql.exec.tez.HiveSplitGenerator 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);
    }
}
Also used : Vertex(org.apache.tez.dag.api.Vertex) SubmitWorkInfo(org.apache.hadoop.hive.llap.SubmitWorkInfo) LlapTokenIdentifier(org.apache.hadoop.hive.llap.security.LlapTokenIdentifier) SplitLocationInfo(org.apache.hadoop.mapred.SplitLocationInfo) HiveSplitGenerator(org.apache.hadoop.hive.ql.exec.tez.HiveSplitGenerator) TaskSpecBuilder(org.apache.tez.dag.api.TaskSpecBuilder) LlapSigner(org.apache.hadoop.hive.llap.security.LlapSigner) TaskLocationHint(org.apache.tez.dag.api.TaskLocationHint) LlapTokenLocalClient(org.apache.hadoop.hive.llap.security.LlapTokenLocalClient) DagUtils(org.apache.hadoop.hive.ql.exec.tez.DagUtils) LlapInputSplit(org.apache.hadoop.hive.llap.LlapInputSplit) FileSystem(org.apache.hadoop.fs.FileSystem) JobConf(org.apache.hadoop.mapred.JobConf) LlapInputSplit(org.apache.hadoop.hive.llap.LlapInputSplit) InputSplit(org.apache.hadoop.mapred.InputSplit) Context(org.apache.hadoop.hive.ql.Context) Path(org.apache.hadoop.fs.Path) TaskSpec(org.apache.tez.runtime.api.impl.TaskSpec) SignedMessage(org.apache.hadoop.hive.llap.security.LlapSigner.SignedMessage) DAG(org.apache.tez.dag.api.DAG) IOException(java.io.IOException) LlapCoordinator(org.apache.hadoop.hive.llap.coordinator.LlapCoordinator) TaskLocationHint(org.apache.tez.dag.api.TaskLocationHint) LoginException(javax.security.auth.login.LoginException) URISyntaxException(java.net.URISyntaxException) UDFArgumentLengthException(org.apache.hadoop.hive.ql.exec.UDFArgumentLengthException) FileNotFoundException(java.io.FileNotFoundException) HiveException(org.apache.hadoop.hive.ql.metadata.HiveException) UDFArgumentException(org.apache.hadoop.hive.ql.exec.UDFArgumentException) UDFArgumentTypeException(org.apache.hadoop.hive.ql.exec.UDFArgumentTypeException) CommandNeedRetryException(org.apache.hadoop.hive.ql.CommandNeedRetryException) IOException(java.io.IOException) LocalResource(org.apache.hadoop.yarn.api.records.LocalResource) MapWork(org.apache.hadoop.hive.ql.plan.MapWork) Event(org.apache.tez.runtime.api.Event) InputConfigureVertexTasksEvent(org.apache.tez.runtime.api.events.InputConfigureVertexTasksEvent) InputDataInformationEvent(org.apache.tez.runtime.api.events.InputDataInformationEvent) InputConfigureVertexTasksEvent(org.apache.tez.runtime.api.events.InputConfigureVertexTasksEvent) ApplicationId(org.apache.hadoop.yarn.api.records.ApplicationId)

Aggregations

FileNotFoundException (java.io.FileNotFoundException)1 IOException (java.io.IOException)1 URISyntaxException (java.net.URISyntaxException)1 LoginException (javax.security.auth.login.LoginException)1 FileSystem (org.apache.hadoop.fs.FileSystem)1 Path (org.apache.hadoop.fs.Path)1 LlapInputSplit (org.apache.hadoop.hive.llap.LlapInputSplit)1 SubmitWorkInfo (org.apache.hadoop.hive.llap.SubmitWorkInfo)1 LlapCoordinator (org.apache.hadoop.hive.llap.coordinator.LlapCoordinator)1 LlapSigner (org.apache.hadoop.hive.llap.security.LlapSigner)1 SignedMessage (org.apache.hadoop.hive.llap.security.LlapSigner.SignedMessage)1 LlapTokenIdentifier (org.apache.hadoop.hive.llap.security.LlapTokenIdentifier)1 LlapTokenLocalClient (org.apache.hadoop.hive.llap.security.LlapTokenLocalClient)1 CommandNeedRetryException (org.apache.hadoop.hive.ql.CommandNeedRetryException)1 Context (org.apache.hadoop.hive.ql.Context)1 UDFArgumentException (org.apache.hadoop.hive.ql.exec.UDFArgumentException)1 UDFArgumentLengthException (org.apache.hadoop.hive.ql.exec.UDFArgumentLengthException)1 UDFArgumentTypeException (org.apache.hadoop.hive.ql.exec.UDFArgumentTypeException)1 DagUtils (org.apache.hadoop.hive.ql.exec.tez.DagUtils)1 HiveSplitGenerator (org.apache.hadoop.hive.ql.exec.tez.HiveSplitGenerator)1