Examples with AggExeInfo org.apache.spark.sql.execution.ndp.AggExeInfo used on opensource projects

Example 1 with AggExeInfo

use of org.apache.spark.sql.execution.ndp.AggExeInfo in project boostkit-bigdata by kunpengcompute.

the class DataIoAdapter method extractAggAndGroupExpression.

private Optional<AggregationInfo> extractAggAndGroupExpression(List<AggExeInfo> aggExecutionList) {
    Optional<AggregationInfo> resAggregationInfo = Optional.empty();
    for (AggExeInfo aggExeInfo : aggExecutionList) {
        List<AggregateFunction> aggregateExpressions = JavaConverters.seqAsJavaList(aggExeInfo.aggregateExpressions());
        List<NamedExpression> namedExpressions = JavaConverters.seqAsJavaList(aggExeInfo.groupingExpressions());
        resAggregationInfo = createAggregationInfo(aggregateExpressions, namedExpressions);
    }
    return resAggregationInfo;
}

Example 2 with AggExeInfo

use of org.apache.spark.sql.execution.ndp.AggExeInfo in project boostkit-bigdata by kunpengcompute.

the class NdpUtils method getColumnOffsetByAggExeInfo.

public static int getColumnOffsetByAggExeInfo(StructType dataSchema, Seq<AggExeInfo> aggExeInfo) {
    String columnName = "";
    int columnTempId = 0;
    if (aggExeInfo != null && aggExeInfo.size() > 0) {
        List<AggExeInfo> aggExecutionList = JavaConverters.seqAsJavaList(aggExeInfo);
        for (AggExeInfo aggExeInfoTemp : aggExecutionList) {
            List<AggregateFunction> aggregateExpressions = JavaConverters.seqAsJavaList(aggExeInfoTemp.aggregateExpressions());
            for (AggregateFunction aggregateFunction : aggregateExpressions) {
                List<Expression> expressions = JavaConverters.seqAsJavaList(aggregateFunction.children());
                for (Expression expression : expressions) {
                    columnName = expression.toString().split("#")[0].replaceAll("\\(", "");
                    Pattern pattern = Pattern.compile(columnName + "#(\\d+)");
                    Matcher matcher = pattern.matcher(expression.toString());
                    if (matcher.find()) {
                        columnTempId = Integer.parseInt(matcher.group(1));
                        break;
                    }
                }
                break;
            }
            List<NamedExpression> namedExpressions = JavaConverters.seqAsJavaList(aggExeInfoTemp.groupingExpressions());
            for (NamedExpression namedExpression : namedExpressions) {
                columnName = namedExpression.toString().split("#")[0];
                columnTempId = NdpUtils.getColumnId(namedExpression.toString());
                break;
            }
        }
    }
    Map<String, Integer> columnMap = new HashMap<>();
    scala.collection.Iterator<StructField> allTableSchemas = dataSchema.iterator();
    int dataSchemaColumnNum = 0;
    while (allTableSchemas.hasNext()) {
        StructField structField = allTableSchemas.next();
        columnMap.put(structField.name(), dataSchemaColumnNum++);
    }
    int columnOffset = columnTempId - columnMap.getOrDefault(columnName, columnMap.size());
    return Math.abs(columnOffset);
}

Example 3 with AggExeInfo

use of org.apache.spark.sql.execution.ndp.AggExeInfo in project boostkit-bigdata by kunpengcompute.

the class DataIoAdapter method getPageIterator.

/**
 * Contact with Omni-Data-Server
 * @param pageCandidate file split info
 * @param sparkOutPut data schema
 * @param partitionColumn partition column
 * @param filterOutPut filter schema
 * @param pushDownOperators push down expressions
 * @return WritableColumnVector data result info
 * @throws TaskExecutionException connect to omni-data-server failed exception
 * @notice 3rd parties api throws Exception, function has to catch basic Exception
 */
public Iterator<WritableColumnVector[]> getPageIterator(PageCandidate pageCandidate, Seq<Attribute> sparkOutPut, Seq<Attribute> partitionColumn, Seq<Attribute> filterOutPut, PushDownInfo pushDownOperators) throws TaskExecutionException, UnknownHostException {
    // initCandidates
    initCandidates(pageCandidate, filterOutPut);
    // create AggregationInfo
    // init agg candidates
    List<Attribute> partitionColumnBatch = JavaConverters.seqAsJavaList(partitionColumn);
    for (Attribute attribute : partitionColumnBatch) {
        partitionColumnName.add(attribute.name());
    }
    List<AggExeInfo> aggExecutionList = JavaConverters.seqAsJavaList(pushDownOperators.aggExecutions());
    if (aggExecutionList.size() == 0) {
        initColumnInfo(sparkOutPut);
    }
    DataSource dataSource = initDataSource(pageCandidate);
    RowExpression rowExpression = initFilter(pushDownOperators.filterExecutions());
    Optional<RowExpression> prestoFilter = rowExpression == null ? Optional.empty() : Optional.of(rowExpression);
    Optional<AggregationInfo> aggregations = initAggAndGroupInfo(aggExecutionList);
    // create limitLong
    OptionalLong limitLong = NdpUtils.convertLimitExeInfo(pushDownOperators.limitExecution());
    Predicate predicate = new Predicate(omnidataTypes, omnidataColumns, prestoFilter, omnidataProjections, ImmutableMap.of(), ImmutableMap.of(), aggregations, limitLong);
    TaskSource taskSource = new TaskSource(dataSource, predicate, 1048576);
    SparkDeserializer deserializer = initSparkDeserializer();
    WritableColumnVector[] page = null;
    int failedTimes = 0;
    String[] sdiHostArray = pageCandidate.getSdiHosts().split(",");
    int randomIndex = (int) (Math.random() * sdiHostArray.length);
    Iterator<String> sdiHosts = Arrays.stream(sdiHostArray).iterator();
    Set<String> sdiHostSet = new HashSet<>();
    sdiHostSet.add(sdiHostArray[randomIndex]);
    while (sdiHosts.hasNext()) {
        String sdiHost;
        if (failedTimes == 0) {
            sdiHost = sdiHostArray[randomIndex];
        } else {
            sdiHost = sdiHosts.next();
            if (sdiHostSet.contains(sdiHost)) {
                continue;
            }
        }
        String ipAddress = InetAddress.getByName(sdiHost).getHostAddress();
        Properties properties = new Properties();
        properties.put("omnidata.client.target.list", ipAddress);
        LOG.info("Push down node info: [hostname :{} ,ip :{}]", sdiHost, ipAddress);
        try {
            orcDataReader = new DataReaderImpl<SparkDeserializer>(properties, taskSource, deserializer);
            hasNextPage = true;
            page = (WritableColumnVector[]) orcDataReader.getNextPageBlocking();
            if (orcDataReader.isFinished()) {
                orcDataReader.close();
                hasNextPage = false;
            }
            break;
        } catch (OmniDataException omniDataException) {
            OmniErrorCode errorCode = omniDataException.getErrorCode();
            switch(errorCode) {
                case OMNIDATA_INSUFFICIENT_RESOURCES:
                    LOG.warn("OMNIDATA_INSUFFICIENT_RESOURCES: " + "OmniData-server's push down queue is full, " + "begin to find next OmniData-server");
                    break;
                case OMNIDATA_UNSUPPORTED_OPERATOR:
                    LOG.warn("OMNIDATA_UNSUPPORTED_OPERATOR: " + "OmniDataException: exist unsupported operator");
                    break;
                case OMNIDATA_GENERIC_ERROR:
                    LOG.warn("OMNIDATA_GENERIC_ERROR: Current OmniData-server unavailable, " + "begin to find next OmniData-server");
                    break;
                case OMNIDATA_NOT_FOUND:
                    LOG.warn("OMNIDATA_NOT_FOUND: Current OmniData-Server not found, " + "begin to find next OmniData-server");
                    break;
                case OMNIDATA_INVALID_ARGUMENT:
                    LOG.warn("OMNIDATA_INVALID_ARGUMENT: INVALID_ARGUMENT, " + "exist unsupported operator or dataType");
                    break;
                case OMNIDATA_IO_ERROR:
                    LOG.warn("OMNIDATA_IO_ERROR: Current OmniData-Server io exception, " + "begin to find next OmniData-server");
                    break;
                default:
                    LOG.warn("OmniDataException: OMNIDATA_ERROR.");
            }
            LOG.warn("Push down failed node info [hostname :{} ,ip :{}]", sdiHost, ipAddress);
            ++failedTimes;
        } catch (Exception e) {
            LOG.warn("Push down failed node info [hostname :{} ,ip :{}]", sdiHost, ipAddress, e);
            ++failedTimes;
        }
    }
    int retryTime = Math.min(TASK_FAILED_TIMES, sdiHostArray.length);
    if (failedTimes >= retryTime) {
        LOG.warn("No Omni-data-server to Connect, Task has tried {} times.", retryTime);
        throw new TaskExecutionException("No Omni-data-server to Connect");
    }
    List<WritableColumnVector[]> l = new ArrayList<>();
    l.add(page);
    return l.iterator();
}