Examples with SMBMapJoinOperator org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator used on opensource projects

Example 16 with SMBMapJoinOperator

use of org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator in project hive by apache.

the class SortedMergeBucketMapjoinProc method process.

@Override
public Object process(Node nd, Stack<Node> stack, NodeProcessorCtx procCtx, Object... nodeOutputs) throws SemanticException {
    if (nd instanceof SMBMapJoinOperator) {
        return null;
    }
    MapJoinOperator mapJoinOp = (MapJoinOperator) nd;
    SortBucketJoinProcCtx smbJoinContext = (SortBucketJoinProcCtx) procCtx;
    boolean convert = canConvertBucketMapJoinToSMBJoin(mapJoinOp, stack, smbJoinContext, nodeOutputs);
    // and sort merge bucketed mapjoin cannot be performed
    if (!convert && pGraphContext.getConf().getBoolVar(HiveConf.ConfVars.HIVEENFORCESORTMERGEBUCKETMAPJOIN)) {
        throw new SemanticException(ErrorMsg.SORTMERGE_MAPJOIN_FAILED.getMsg());
    }
    if (convert) {
        convertBucketMapJoinToSMBJoin(mapJoinOp, smbJoinContext);
    }
    return null;
}

Example 17 with SMBMapJoinOperator

use of org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator in project hive by apache.

the class SemanticAnalyzer method analyzeInternal.

@SuppressWarnings("checkstyle:methodlength")
void analyzeInternal(ASTNode ast, Supplier<PlannerContext> pcf) throws SemanticException {
    LOG.info("Starting Semantic Analysis");
    // 1. Generate Resolved Parse tree from syntax tree
    boolean needsTransform = needsTransform();
    // change the location of position alias process here
    processPositionAlias(ast);
    cacheTableHelper.populateCache(ctx.getParsedTables(), conf, getTxnMgr());
    PlannerContext plannerCtx = pcf.get();
    if (!genResolvedParseTree(ast, plannerCtx)) {
        return;
    }
    if (HiveConf.getBoolVar(conf, ConfVars.HIVE_REMOVE_ORDERBY_IN_SUBQUERY)) {
        for (String alias : qb.getSubqAliases()) {
            removeOBInSubQuery(qb.getSubqForAlias(alias));
        }
    }
    final String llapIOETLSkipFormat = HiveConf.getVar(conf, ConfVars.LLAP_IO_ETL_SKIP_FORMAT);
    if (qb.getParseInfo().hasInsertTables() || qb.isCTAS()) {
        if (llapIOETLSkipFormat.equalsIgnoreCase("encode")) {
            conf.setBoolean(ConfVars.LLAP_IO_ENCODE_ENABLED.varname, false);
            LOG.info("Disabling LLAP IO encode as ETL query is detected");
        } else if (llapIOETLSkipFormat.equalsIgnoreCase("all")) {
            conf.setBoolean(ConfVars.LLAP_IO_ENABLED.varname, false);
            LOG.info("Disabling LLAP IO as ETL query is detected");
        }
    }
    // Check query results cache.
    // If no masking/filtering required, then we can check the cache now, before
    // generating the operator tree and going through CBO.
    // Otherwise we have to wait until after the masking/filtering step.
    boolean isCacheEnabled = isResultsCacheEnabled();
    QueryResultsCache.LookupInfo lookupInfo = null;
    if (isCacheEnabled && !needsTransform && queryTypeCanUseCache()) {
        lookupInfo = createLookupInfoForQuery(ast);
        if (checkResultsCache(lookupInfo, false)) {
            return;
        }
    }
    ASTNode astForMasking;
    if (isCBOExecuted() && needsTransform && (qb.isCTAS() || forViewCreation || qb.isMaterializedView() || qb.isMultiDestQuery())) {
        // If we use CBO and we may apply masking/filtering policies, we create a copy of the ast.
        // The reason is that the generation of the operator tree may modify the initial ast,
        // but if we need to parse for a second time, we would like to parse the unmodified ast.
        astForMasking = (ASTNode) ParseDriver.adaptor.dupTree(ast);
    } else {
        astForMasking = ast;
    }
    // 2. Gen OP Tree from resolved Parse Tree
    sinkOp = genOPTree(ast, plannerCtx);
    boolean usesMasking = false;
    if (!forViewCreation && ast.getToken().getType() != HiveParser.TOK_CREATE_MATERIALIZED_VIEW && (tableMask.isEnabled() && analyzeRewrite == null)) {
        // Here we rewrite the * and also the masking table
        ParseResult rewrittenResult = rewriteASTWithMaskAndFilter(tableMask, astForMasking, ctx.getTokenRewriteStream(), ctx, db);
        ASTNode rewrittenAST = rewrittenResult.getTree();
        if (astForMasking != rewrittenAST) {
            usesMasking = true;
            plannerCtx = pcf.get();
            ctx.setSkipTableMasking(true);
            ctx.setTokenRewriteStream(rewrittenResult.getTokenRewriteStream());
            init(true);
            // change the location of position alias process here
            processPositionAlias(rewrittenAST);
            genResolvedParseTree(rewrittenAST, plannerCtx);
            if (this instanceof CalcitePlanner) {
                ((CalcitePlanner) this).resetCalciteConfiguration();
            }
            sinkOp = genOPTree(rewrittenAST, plannerCtx);
        }
    }
    // validate if this sink operation is allowed for non-native tables
    if (sinkOp instanceof FileSinkOperator) {
        FileSinkOperator fileSinkOperator = (FileSinkOperator) sinkOp;
        Optional<HiveStorageHandler> handler = Optional.ofNullable(fileSinkOperator).map(FileSinkOperator::getConf).map(FileSinkDesc::getTable).map(Table::getStorageHandler);
        if (handler.isPresent()) {
            handler.get().validateSinkDesc(fileSinkOperator.getConf());
        }
    }
    // TODO: Enable caching for queries with masking/filtering
    if (isCacheEnabled && needsTransform && !usesMasking && queryTypeCanUseCache()) {
        lookupInfo = createLookupInfoForQuery(ast);
        if (checkResultsCache(lookupInfo, false)) {
            return;
        }
    }
    // 3. Deduce Resultset Schema
    if ((forViewCreation || createVwDesc != null) && !this.ctx.isCboSucceeded()) {
        resultSchema = convertRowSchemaToViewSchema(opParseCtx.get(sinkOp).getRowResolver());
    } else {
        // succeeds.
        if (resultSchema == null) {
            resultSchema = convertRowSchemaToResultSetSchema(opParseCtx.get(sinkOp).getRowResolver(), HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_RESULTSET_USE_UNIQUE_COLUMN_NAMES));
        }
    }
    // 4. Generate Parse Context for Optimizer & Physical compiler
    copyInfoToQueryProperties(queryProperties);
    ParseContext pCtx = new ParseContext(queryState, opToPartPruner, opToPartList, topOps, new HashSet<JoinOperator>(joinContext.keySet()), new HashSet<SMBMapJoinOperator>(smbMapJoinContext.keySet()), loadTableWork, loadFileWork, columnStatsAutoGatherContexts, ctx, idToTableNameMap, destTableId, uCtx, listMapJoinOpsNoReducer, prunedPartitions, tabNameToTabObject, opToSamplePruner, globalLimitCtx, nameToSplitSample, inputs, rootTasks, opToPartToSkewedPruner, viewAliasToInput, reduceSinkOperatorsAddedByEnforceBucketingSorting, analyzeRewrite, tableDesc, createVwDesc, materializedViewUpdateDesc, queryProperties, viewProjectToTableSchema);
    // Set the semijoin hints in parse context
    pCtx.setSemiJoinHints(parseSemiJoinHint(getQB().getParseInfo().getHintList()));
    // Set the mapjoin hint if it needs to be disabled.
    pCtx.setDisableMapJoin(disableMapJoinWithHint(getQB().getParseInfo().getHintList()));
    if (forViewCreation) {
        // Generate lineage info if LineageLogger hook is configured.
        // Add the transformation that computes the lineage information.
        Set<String> postExecHooks = Sets.newHashSet(Splitter.on(",").trimResults().omitEmptyStrings().split(Strings.nullToEmpty(HiveConf.getVar(conf, HiveConf.ConfVars.POSTEXECHOOKS))));
        if (postExecHooks.contains("org.apache.hadoop.hive.ql.hooks.PostExecutePrinter") || postExecHooks.contains("org.apache.hadoop.hive.ql.hooks.LineageLogger") || postExecHooks.contains("org.apache.atlas.hive.hook.HiveHook")) {
            List<Transform> transformations = new ArrayList<Transform>();
            transformations.add(new HiveOpConverterPostProc());
            transformations.add(new Generator(postExecHooks));
            for (Transform t : transformations) {
                pCtx = t.transform(pCtx);
            }
        }
    }
    // 5. Take care of view creation
    if (createVwDesc != null) {
        if (ctx.getExplainAnalyze() == AnalyzeState.RUNNING) {
            return;
        }
        if (!ctx.isCboSucceeded()) {
            saveViewDefinition();
        }
        // validate the create view statement at this point, the createVwDesc gets
        // all the information for semanticcheck
        validateCreateView();
        createVwDesc.setTablesUsed(pCtx.getTablesUsed());
    }
    // it means that in step 2, the ColumnAccessInfo was already created
    if (!forViewCreation || getColumnAccessInfo() == null) {
        // 6. Generate table access stats if required
        if (HiveConf.getBoolVar(this.conf, HiveConf.ConfVars.HIVE_STATS_COLLECT_TABLEKEYS)) {
            TableAccessAnalyzer tableAccessAnalyzer = new TableAccessAnalyzer(pCtx);
            setTableAccessInfo(tableAccessAnalyzer.analyzeTableAccess());
        }
        AuxOpTreeSignature.linkAuxSignatures(pCtx);
        // 7. Perform Logical optimization
        if (LOG.isDebugEnabled()) {
            LOG.debug("Before logical optimization\n" + Operator.toString(pCtx.getTopOps().values()));
        }
        Optimizer optm = new Optimizer();
        optm.setPctx(pCtx);
        optm.initialize(conf);
        pCtx = optm.optimize();
        if (pCtx.getColumnAccessInfo() != null) {
            // set ColumnAccessInfo for view column authorization
            setColumnAccessInfo(pCtx.getColumnAccessInfo());
        }
        if (LOG.isDebugEnabled()) {
            LOG.debug("After logical optimization\n" + Operator.toString(pCtx.getTopOps().values()));
        }
        // 8. Generate column access stats if required - wait until column pruning
        // takes place during optimization
        boolean isColumnInfoNeedForAuth = SessionState.get().isAuthorizationModeV2() && HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_AUTHORIZATION_ENABLED);
        if (isColumnInfoNeedForAuth || HiveConf.getBoolVar(this.conf, HiveConf.ConfVars.HIVE_STATS_COLLECT_SCANCOLS)) {
            ColumnAccessAnalyzer columnAccessAnalyzer = new ColumnAccessAnalyzer(pCtx);
            // view column access info is carried by this.getColumnAccessInfo().
            setColumnAccessInfo(columnAccessAnalyzer.analyzeColumnAccess(this.getColumnAccessInfo()));
        }
    }
    if (forViewCreation) {
        return;
    }
    // 9. Optimize Physical op tree & Translate to target execution engine (MR,
    // TEZ..)
    compilePlan(pCtx);
    // find all Acid FileSinkOperatorS
    new QueryPlanPostProcessor(rootTasks, acidFileSinks, ctx.getExecutionId());
    // 10. Attach CTAS/Insert-Commit-hooks for Storage Handlers
    final Optional<TezTask> optionalTezTask = rootTasks.stream().filter(task -> task instanceof TezTask).map(task -> (TezTask) task).findFirst();
    if (optionalTezTask.isPresent()) {
        final TezTask tezTask = optionalTezTask.get();
        rootTasks.stream().filter(task -> task.getWork() instanceof DDLWork).map(task -> (DDLWork) task.getWork()).filter(ddlWork -> ddlWork.getDDLDesc() instanceof PreInsertTableDesc).map(ddlWork -> (PreInsertTableDesc) ddlWork.getDDLDesc()).map(desc -> new InsertCommitHookDesc(desc.getTable(), desc.isOverwrite())).forEach(insertCommitHookDesc -> tezTask.addDependentTask(TaskFactory.get(new DDLWork(getInputs(), getOutputs(), insertCommitHookDesc), conf)));
    }
    LOG.info("Completed plan generation");
    // 11. put accessed columns to readEntity
    if (HiveConf.getBoolVar(this.conf, HiveConf.ConfVars.HIVE_STATS_COLLECT_SCANCOLS)) {
        putAccessedColumnsToReadEntity(inputs, columnAccessInfo);
    }
    if (isCacheEnabled && lookupInfo != null) {
        if (queryCanBeCached()) {
            // requires SemanticAnalyzer state to be reset.
            if (checkResultsCache(lookupInfo, true)) {
                LOG.info("Cached result found on second lookup");
            } else {
                QueryResultsCache.QueryInfo queryInfo = createCacheQueryInfoForQuery(lookupInfo);
                // Specify that the results of this query can be cached.
                setCacheUsage(new CacheUsage(CacheUsage.CacheStatus.CAN_CACHE_QUERY_RESULTS, queryInfo));
            }
        }
    }
}

Example 18 with SMBMapJoinOperator

use of org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator in project hive by apache.

the class SparkSortMergeJoinOptimizer method convertJoinToSMBJoinAndReturn.

protected SMBMapJoinOperator convertJoinToSMBJoinAndReturn(JoinOperator joinOp, SortBucketJoinProcCtx smbJoinContext) throws SemanticException {
    MapJoinOperator mapJoinOp = convertJoinToBucketMapJoin(joinOp, smbJoinContext);
    SMBMapJoinOperator smbMapJoinOp = convertBucketMapJoinToSMBJoin(mapJoinOp, smbJoinContext);
    smbMapJoinOp.setConvertedAutomaticallySMBJoin(true);
    return smbMapJoinOp;
}

Example 19 with SMBMapJoinOperator

use of org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator in project hive by apache.

the class SortMergeJoinTaskDispatcher method convertSMBWorkToJoinWork.

/*
   * Convert the work containing to sort-merge join into a work, as if it had a regular join.
   * Note that the operator tree is not changed - is still contains the SMB join, but the
   * plan is changed (aliasToWork etc.) to contain all the paths as if it was a regular join.
   */
private MapredWork convertSMBWorkToJoinWork(MapredWork currWork, SMBMapJoinOperator oldSMBJoinOp) throws SemanticException {
    try {
        // deep copy a new mapred work
        MapredWork currJoinWork = SerializationUtilities.clonePlan(currWork);
        SMBMapJoinOperator newSMBJoinOp = getSMBMapJoinOp(currJoinWork);
        // change the newly created map-red plan as if it was a join operator
        genSMBJoinWork(currJoinWork.getMapWork(), newSMBJoinOp);
        return currJoinWork;
    } catch (Exception e) {
        throw new SemanticException("Generate Map Join Task Error", e);
    }
}