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

Example 46 with BaseWork

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

the class TezTask method execute.

@Override
public int execute(DriverContext driverContext) {
    int rc = 1;
    boolean cleanContext = false;
    Context ctx = null;
    Ref<TezSessionState> sessionRef = Ref.from(null);
    try {
        // Get or create Context object. If we create it we have to clean it later as well.
        ctx = driverContext.getCtx();
        if (ctx == null) {
            ctx = new Context(conf);
            cleanContext = true;
            // some DDL task that directly executes a TezTask does not setup Context and hence TriggerContext.
            // Setting queryId is messed up. Some DDL tasks have executionId instead of proper queryId.
            String queryId = HiveConf.getVar(conf, HiveConf.ConfVars.HIVEQUERYID);
            WmContext wmContext = new WmContext(System.currentTimeMillis(), queryId);
            ctx.setWmContext(wmContext);
        }
        // Need to remove this static hack. But this is the way currently to get a session.
        SessionState ss = SessionState.get();
        // Note: given that we return pool sessions to the pool in the finally block below, and that
        // we need to set the global to null to do that, this "reuse" may be pointless.
        TezSessionState session = sessionRef.value = ss.getTezSession();
        if (session != null && !session.isOpen()) {
            LOG.warn("The session: " + session + " has not been opened");
        }
        // We only need a username for UGI to use for groups; getGroups will fetch the groups
        // based on Hadoop configuration, as documented at
        // https://hadoop.apache.org/docs/r2.8.0/hadoop-project-dist/hadoop-common/GroupsMapping.html
        String userName = ss.getUserName();
        List<String> groups = null;
        if (userName == null) {
            userName = "anonymous";
        } else {
            groups = UserGroupInformation.createRemoteUser(ss.getUserName()).getGroups();
        }
        MappingInput mi = new MappingInput(userName, groups, ss.getHiveVariables().get("wmpool"), ss.getHiveVariables().get("wmapp"));
        WmContext wmContext = ctx.getWmContext();
        // jobConf will hold all the configuration for hadoop, tez, and hive
        JobConf jobConf = utils.createConfiguration(conf);
        // Get all user jars from work (e.g. input format stuff).
        String[] allNonConfFiles = work.configureJobConfAndExtractJars(jobConf);
        // DAG scratch dir. We get a session from the pool so it may be different from Tez one.
        // TODO: we could perhaps reuse the same directory for HiveResources?
        Path scratchDir = utils.createTezDir(ctx.getMRScratchDir(), conf);
        CallerContext callerContext = CallerContext.create("HIVE", queryPlan.getQueryId(), "HIVE_QUERY_ID", queryPlan.getQueryStr());
        perfLogger.PerfLogBegin(CLASS_NAME, PerfLogger.TEZ_GET_SESSION);
        session = sessionRef.value = WorkloadManagerFederation.getSession(sessionRef.value, conf, mi, getWork().getLlapMode(), wmContext);
        perfLogger.PerfLogEnd(CLASS_NAME, PerfLogger.TEZ_GET_SESSION);
        try {
            ss.setTezSession(session);
            LOG.info("Subscribed to counters: {} for queryId: {}", wmContext.getSubscribedCounters(), wmContext.getQueryId());
            // Ensure the session is open and has the necessary local resources.
            // This would refresh any conf resources and also local resources.
            ensureSessionHasResources(session, allNonConfFiles);
            // This is a combination of the jar stuff from conf, and not from conf.
            List<LocalResource> allNonAppResources = session.getLocalizedResources();
            logResources(allNonAppResources);
            Map<String, LocalResource> allResources = DagUtils.createTezLrMap(session.getAppJarLr(), allNonAppResources);
            // next we translate the TezWork to a Tez DAG
            DAG dag = build(jobConf, work, scratchDir, ctx, allResources);
            dag.setCallerContext(callerContext);
            // Check isShutdown opportunistically; it's never unset.
            if (this.isShutdown) {
                throw new HiveException("Operation cancelled");
            }
            DAGClient dagClient = submit(jobConf, dag, sessionRef);
            session = sessionRef.value;
            boolean wasShutdown = false;
            synchronized (dagClientLock) {
                assert this.dagClient == null;
                wasShutdown = this.isShutdown;
                if (!wasShutdown) {
                    this.dagClient = dagClient;
                }
            }
            if (wasShutdown) {
                closeDagClientOnCancellation(dagClient);
                throw new HiveException("Operation cancelled");
            }
            // finally monitor will print progress until the job is done
            TezJobMonitor monitor = new TezJobMonitor(work.getAllWork(), dagClient, conf, dag, ctx);
            rc = monitor.monitorExecution();
            if (rc != 0) {
                this.setException(new HiveException(monitor.getDiagnostics()));
            }
            // fetch the counters
            try {
                Set<StatusGetOpts> statusGetOpts = EnumSet.of(StatusGetOpts.GET_COUNTERS);
                counters = dagClient.getDAGStatus(statusGetOpts).getDAGCounters();
            } catch (Exception err) {
                // Don't fail execution due to counters - just don't print summary info
                LOG.warn("Failed to get counters. Ignoring, summary info will be incomplete. " + err, err);
                counters = null;
            }
        } finally {
            // Note: due to TEZ-3846, the session may actually be invalid in case of some errors.
            // Currently, reopen on an attempted reuse will take care of that; we cannot tell
            // if the session is usable until we try.
            // We return this to the pool even if it's unusable; reopen is supposed to handle this.
            wmContext = ctx.getWmContext();
            try {
                if (sessionRef.value != null) {
                    sessionRef.value.returnToSessionManager();
                }
            } catch (Exception e) {
                LOG.error("Failed to return session: {} to pool", session, e);
                throw e;
            }
            if (!conf.getVar(HiveConf.ConfVars.TEZ_SESSION_EVENTS_SUMMARY).equalsIgnoreCase("none") && wmContext != null) {
                if (conf.getVar(HiveConf.ConfVars.TEZ_SESSION_EVENTS_SUMMARY).equalsIgnoreCase("json")) {
                    wmContext.printJson(console);
                } else if (conf.getVar(HiveConf.ConfVars.TEZ_SESSION_EVENTS_SUMMARY).equalsIgnoreCase("text")) {
                    wmContext.print(console);
                }
            }
        }
        if (LOG.isInfoEnabled() && counters != null && (HiveConf.getBoolVar(conf, HiveConf.ConfVars.TEZ_EXEC_SUMMARY) || Utilities.isPerfOrAboveLogging(conf))) {
            for (CounterGroup group : counters) {
                LOG.info(group.getDisplayName() + ":");
                for (TezCounter counter : group) {
                    LOG.info("   " + counter.getDisplayName() + ": " + counter.getValue());
                }
            }
        }
    } catch (Exception e) {
        LOG.error("Failed to execute tez graph.", e);
    // rc will be 1 at this point indicating failure.
    } finally {
        Utilities.clearWork(conf);
        // Clear gWorkMap
        for (BaseWork w : work.getAllWork()) {
            JobConf workCfg = workToConf.get(w);
            if (workCfg != null) {
                Utilities.clearWorkMapForConf(workCfg);
            }
        }
        if (cleanContext) {
            try {
                ctx.clear();
            } catch (Exception e) {
                /*best effort*/
                LOG.warn("Failed to clean up after tez job", e);
            }
        }
        // need to either move tmp files or remove them
        DAGClient dagClient = null;
        synchronized (dagClientLock) {
            dagClient = this.dagClient;
            this.dagClient = null;
        }
        // DagClient as such should have no bearing on jobClose.
        if (dagClient != null) {
            // rc will only be overwritten if close errors out
            rc = close(work, rc, dagClient);
        }
    }
    return rc;
}

Example 47 with BaseWork

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

the class TezProgressMonitor method rows.

public List<List<String>> rows() {
    try {
        List<List<String>> results = new ArrayList<>();
        for (BaseWork baseWork : topSortedWork) {
            String vertexName = baseWork.getName();
            VertexProgress progress = progressCountsMap.get(vertexName);
            if (progress != null) {
                // Map 1 .......... container  SUCCEEDED      7          7        0        0       0       0
                results.add(Arrays.asList(getNameWithProgress(vertexName, progress.succeededTaskCount, progress.totalTaskCount), getMode(baseWork), progress.vertexStatus(vertexStatusMap.get(vertexName)), progress.total(), progress.completed(), progress.running(), progress.pending(), progress.failed(), progress.killed()));
            }
        }
        return results;
    } catch (Exception e) {
        console.printInfo("Getting  Progress Bar table rows failed: " + e.getMessage() + " stack trace: " + Arrays.toString(e.getStackTrace()));
    }
    return Collections.emptyList();
}

Example 48 with BaseWork

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

the class OperatorStatsReaderHook method run.

@Override
public void run(HookContext hookContext) throws Exception {
    if (hookContext.getHookType() == HookType.PRE_EXEC_HOOK) {
        return;
    }
    if (hookContext.getHookType() == HookType.POST_EXEC_HOOK && !isCollectOnSuccess()) {
        return;
    }
    HiveConf conf = hookContext.getConf();
    QueryPlan plan = hookContext.getQueryPlan();
    List<TezTask> rootTasks = Utilities.getTezTasks(plan.getRootTasks());
    for (TezTask tezTask : rootTasks) {
        List<BaseWork> baseWorks = tezTask.getWork().getAllWork();
        for (BaseWork baseWork : baseWorks) {
            String vertexName = baseWork.getName();
            LOG.debug("Reading runtime statistics for tez vertex task: {}", vertexName);
            TezCounters counters = tezTask.getTezCounters();
            if (counters != null) {
                String groupName = HiveConf.getVar(conf, HiveConf.ConfVars.HIVECOUNTERGROUP);
                for (Operator<? extends OperatorDesc> op : baseWork.getAllOperators()) {
                    String operatorId = op.getOperatorId();
                    OperatorStats operatorStats = null;
                    String counterName = Operator.Counter.RECORDS_OUT_OPERATOR.toString() + "_" + operatorId;
                    TezCounter tezCounter = counters.getGroup(groupName).findCounter(counterName, false);
                    if (tezCounter != null) {
                        if (operatorStats == null) {
                            operatorStats = new OperatorStats(operatorId);
                        }
                        operatorStats.setOutputRecords(tezCounter.getValue());
                    }
                    if (operatorStats != null) {
                        ((PrivateHookContext) hookContext).getContext().getPlanMapper().link(op, operatorStats);
                    } else {
                        LOG.debug("Unable to get statistics for vertex: {} opId: {} groupName: {}", vertexName, operatorId, groupName);
                    }
                }
            }
        }
    }
}

Example 49 with BaseWork

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

the class TestTezTask method testBuildDag.

@Test
public void testBuildDag() throws IllegalArgumentException, IOException, Exception {
    DAG dag = task.build(conf, work, path, new Context(conf), DagUtils.createTezLrMap(appLr, null));
    for (BaseWork w : work.getAllWork()) {
        Vertex v = dag.getVertex(w.getName());
        assertNotNull(v);
        List<Vertex> outs = v.getOutputVertices();
        for (BaseWork x : work.getChildren(w)) {
            boolean found = false;
            for (Vertex u : outs) {
                if (u.getName().equals(x.getName())) {
                    found = true;
                    break;
                }
            }
            assertTrue(found);
        }
    }
}

Example 50 with BaseWork

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

the class LlapRecordReader method findMapWork.

private static MapWork findMapWork(JobConf job) throws HiveException {
    String inputName = job.get(Utilities.INPUT_NAME, null);
    if (LOG.isDebugEnabled()) {
        LOG.debug("Initializing for input " + inputName);
    }
    String prefixes = job.get(DagUtils.TEZ_MERGE_WORK_FILE_PREFIXES);
    if (prefixes != null && !StringUtils.isBlank(prefixes)) {
        // So, we don't use the below code that would get the correct MapWork. See HIVE-16985.
        return null;
    }
    BaseWork work = null;
    // HIVE-16985: try to find the fake merge work for SMB join, that is really another MapWork.
    if (inputName != null) {
        if (prefixes == null || !Lists.newArrayList(prefixes.split(",")).contains(inputName)) {
            inputName = null;
        }
    }
    if (inputName != null) {
        work = Utilities.getMergeWork(job, inputName);
    }
    if (work == null || !(work instanceof MapWork)) {
        work = Utilities.getMapWork(job);
    }
    return (MapWork) work;
}