Examples with HiveOptions org.apache.nifi.util.hive.HiveOptions used on opensource projects

Example 1 with HiveOptions

use of org.apache.nifi.util.hive.HiveOptions in project nifi by apache.

the class PutHiveStreaming method onTrigger.

private void onTrigger(ProcessContext context, ProcessSession session, FunctionContext functionContext) throws ProcessException {
    FlowFile flowFile = session.get();
    if (flowFile == null) {
        return;
    }
    final String dbName = context.getProperty(DB_NAME).evaluateAttributeExpressions(flowFile).getValue();
    final String tableName = context.getProperty(TABLE_NAME).evaluateAttributeExpressions(flowFile).getValue();
    // Only allow one thread to work on a DB/table at a time
    final Semaphore newSemaphore = new Semaphore(1);
    Semaphore semaphore = tableSemaphoreMap.putIfAbsent(dbName + "." + tableName, newSemaphore);
    if (semaphore == null) {
        semaphore = newSemaphore;
    }
    boolean gotSemaphore = false;
    try {
        gotSemaphore = semaphore.tryAcquire(0, TimeUnit.SECONDS);
    } catch (InterruptedException ie) {
    // Nothing to do, gotSemaphore defaults to false
    }
    if (!gotSemaphore) {
        // We didn't get a chance to acquire, so rollback the session and try again next time
        session.rollback();
        return;
    }
    final ComponentLog log = getLogger();
    final String metastoreUri = context.getProperty(METASTORE_URI).evaluateAttributeExpressions(flowFile).getValue();
    final boolean autoCreatePartitions = context.getProperty(AUTOCREATE_PARTITIONS).asBoolean();
    final Integer maxConnections = context.getProperty(MAX_OPEN_CONNECTIONS).asInteger();
    final Integer heartbeatInterval = context.getProperty(HEARTBEAT_INTERVAL).evaluateAttributeExpressions().asInteger();
    final Integer txnsPerBatch = context.getProperty(TXNS_PER_BATCH).evaluateAttributeExpressions(flowFile).asInteger();
    final Integer recordsPerTxn = context.getProperty(RECORDS_PER_TXN).evaluateAttributeExpressions(flowFile).asInteger();
    final Map<HiveEndPoint, HiveWriter> myWriters = new ConcurrentHashMap<>();
    threadWriterList.add(myWriters);
    HiveOptions o = new HiveOptions(metastoreUri, dbName, tableName).withTxnsPerBatch(txnsPerBatch).withAutoCreatePartitions(autoCreatePartitions).withMaxOpenConnections(maxConnections).withHeartBeatInterval(heartbeatInterval).withCallTimeout(callTimeout);
    if (SecurityUtil.isSecurityEnabled(hiveConfig)) {
        final String explicitPrincipal = context.getProperty(kerberosProperties.getKerberosPrincipal()).evaluateAttributeExpressions().getValue();
        final String explicitKeytab = context.getProperty(kerberosProperties.getKerberosKeytab()).evaluateAttributeExpressions().getValue();
        final KerberosCredentialsService credentialsService = context.getProperty(KERBEROS_CREDENTIALS_SERVICE).asControllerService(KerberosCredentialsService.class);
        final String resolvedPrincipal;
        final String resolvedKeytab;
        if (credentialsService == null) {
            resolvedPrincipal = explicitPrincipal;
            resolvedKeytab = explicitKeytab;
        } else {
            resolvedPrincipal = credentialsService.getPrincipal();
            resolvedKeytab = credentialsService.getKeytab();
        }
        o = o.withKerberosPrincipal(resolvedPrincipal).withKerberosKeytab(resolvedKeytab);
    }
    final HiveOptions options = o;
    // Store the original class loader, then explicitly set it to this class's classloader (for use by the Hive Metastore)
    ClassLoader originalClassloader = Thread.currentThread().getContextClassLoader();
    Thread.currentThread().setContextClassLoader(this.getClass().getClassLoader());
    final List<String> partitionColumnList;
    final String partitionColumns = context.getProperty(PARTITION_COLUMNS).evaluateAttributeExpressions().getValue();
    if (partitionColumns == null || partitionColumns.isEmpty()) {
        partitionColumnList = Collections.emptyList();
    } else {
        String[] partitionCols = partitionColumns.split(",");
        partitionColumnList = new ArrayList<>(partitionCols.length);
        for (String col : partitionCols) {
            partitionColumnList.add(col.trim());
        }
    }
    final AtomicReference<List<HiveStreamingRecord>> successfulRecords = new AtomicReference<>();
    successfulRecords.set(new ArrayList<>());
    final FlowFile inputFlowFile = flowFile;
    final RoutingResult result = new RoutingResult();
    final ExceptionHandler<FunctionContext> exceptionHandler = new ExceptionHandler<>();
    exceptionHandler.mapException(s -> {
        try {
            if (s == null) {
                return ErrorTypes.PersistentFailure;
            }
            throw s;
        } catch (IllegalArgumentException | HiveWriter.WriteFailure | SerializationError inputError) {
            return ErrorTypes.InvalidInput;
        } catch (HiveWriter.CommitFailure | HiveWriter.TxnBatchFailure | HiveWriter.TxnFailure writerTxError) {
            return ErrorTypes.TemporalInputFailure;
        } catch (ConnectionError | HiveWriter.ConnectFailure connectionError) {
            // Can't connect to Hive endpoint.
            log.error("Error connecting to Hive endpoint: table {} at {}", new Object[] { options.getTableName(), options.getMetaStoreURI() });
            return ErrorTypes.TemporalFailure;
        } catch (IOException | InterruptedException tempError) {
            return ErrorTypes.TemporalFailure;
        } catch (Exception t) {
            return ErrorTypes.UnknownFailure;
        }
    });
    final BiFunction<FunctionContext, ErrorTypes, ErrorTypes.Result> adjustError = RollbackOnFailure.createAdjustError(getLogger());
    exceptionHandler.adjustError(adjustError);
    // Create output flow files and their Avro writers
    functionContext.setFlowFiles(session.create(inputFlowFile), session.create(inputFlowFile));
    try {
        session.read(inputFlowFile, new InputStreamCallback() {

            @Override
            public void process(InputStream in) throws IOException {
                try (final DataFileStream<GenericRecord> reader = new DataFileStream<>(in, new GenericDatumReader<GenericRecord>())) {
                    GenericRecord currRecord = null;
                    // Copy codec and schema information to all writers
                    final String codec = reader.getMetaString(DataFileConstants.CODEC) == null ? DataFileConstants.NULL_CODEC : reader.getMetaString(DataFileConstants.CODEC);
                    functionContext.initAvroWriters(session, codec, reader);
                    Runnable flushSuccessfulRecords = () -> {
                        // Now send the records to the successful FlowFile and update the success count
                        functionContext.appendRecordsToSuccess(session, successfulRecords.get());
                        // Clear the list of successful records, we'll use it at the end when we flush whatever records are left
                        successfulRecords.set(new ArrayList<>());
                    };
                    while (reader.hasNext()) {
                        // We can NOT reuse currRecord here, because currRecord is accumulated in successful records.
                        // If we use the same GenericRecord instance, every record ends up having the same contents.
                        // To avoid this, we need to create a brand new GenericRecord instance here each time.
                        currRecord = reader.next();
                        functionContext.recordCount.incrementAndGet();
                        // Extract the partition values (they must be put separately into the Hive Streaming API)
                        List<String> partitionValues = new ArrayList<>();
                        if (!exceptionHandler.execute(functionContext, currRecord, input -> {
                            for (String partition : partitionColumnList) {
                                Object partitionValue = input.get(partition);
                                if (partitionValue == null) {
                                    throw new IllegalArgumentException("Partition column '" + partition + "' not found in Avro record");
                                }
                                partitionValues.add(partitionValue.toString());
                            }
                        }, onRecordError(context, session, myWriters))) {
                            continue;
                        }
                        final HiveStreamingRecord record = new HiveStreamingRecord(partitionValues, currRecord);
                        final AtomicReference<HiveWriter> hiveWriterRef = new AtomicReference<>();
                        // Write record to Hive streaming
                        if (!exceptionHandler.execute(functionContext, record, input -> {
                            final HiveEndPoint endPoint = makeHiveEndPoint(record.getPartitionValues(), options);
                            final HiveWriter hiveWriter = getOrCreateWriter(myWriters, options, endPoint);
                            hiveWriterRef.set(hiveWriter);
                            hiveWriter.write(record.getRecord().toString().getBytes(StandardCharsets.UTF_8));
                            successfulRecords.get().add(record);
                        }, onHiveRecordError(context, session, myWriters))) {
                            continue;
                        }
                        // If we've reached the records-per-transaction limit, flush the Hive Writer and update the Avro Writer for successful records
                        final HiveWriter hiveWriter = hiveWriterRef.get();
                        if (hiveWriter.getTotalRecords() >= recordsPerTxn) {
                            exceptionHandler.execute(functionContext, successfulRecords.get(), input -> {
                                hiveWriter.flush(true);
                                // Proceed function context. Process session can't be rollback anymore.
                                functionContext.proceed();
                                // Now send the records to the success relationship and update the success count
                                flushSuccessfulRecords.run();
                            }, onHiveRecordsError(context, session, myWriters).andThen((fc, input, res, commitException) -> {
                                // Reset hiveWriter for succeeding records.
                                switch(res.destination()) {
                                    case Retry:
                                    case Failure:
                                        try {
                                            // Abort current tx and move to next.
                                            hiveWriter.abort();
                                        } catch (Exception e) {
                                            // Can't even abort properly, throw a process exception
                                            throw new ProcessException(e);
                                        }
                                }
                            }));
                        }
                    }
                    exceptionHandler.execute(functionContext, successfulRecords.get(), input -> {
                        // Finish any transactions
                        flushAllWriters(myWriters, true);
                        closeAllWriters(myWriters);
                        // Now send any remaining records to the success relationship and update the count
                        flushSuccessfulRecords.run();
                    // Append successfulRecords on failure.
                    }, onHiveRecordsError(context, session, myWriters));
                } catch (IOException ioe) {
                    // The Avro file is invalid (or may not be an Avro file at all), send it to failure
                    final ErrorTypes.Result adjusted = adjustError.apply(functionContext, ErrorTypes.InvalidInput);
                    final String msg = "The incoming flow file can not be read as an Avro file";
                    switch(adjusted.destination()) {
                        case Failure:
                            log.error(msg, ioe);
                            result.routeTo(inputFlowFile, REL_FAILURE);
                            break;
                        case ProcessException:
                            throw new ProcessException(msg, ioe);
                    }
                }
            }
        });
        // If we got here, we've processed the outgoing flow files correctly, so remove the incoming one if necessary
        if (result.getRoutedFlowFiles().values().stream().noneMatch(routed -> routed.contains(inputFlowFile))) {
            session.remove(inputFlowFile);
        }
    } catch (DiscontinuedException e) {
        // The input FlowFile processing is discontinued. Keep it in the input queue.
        getLogger().warn("Discontinued processing for {} due to {}", new Object[] { flowFile, e }, e);
        result.routeTo(flowFile, Relationship.SELF);
    } catch (ShouldRetryException e) {
        // This exception is already a result of adjusting an error, so simply transfer the FlowFile to retry.
        getLogger().error(e.getMessage(), e);
        flowFile = session.penalize(flowFile);
        result.routeTo(flowFile, REL_RETRY);
    } finally {
        threadWriterList.remove(myWriters);
        functionContext.transferFlowFiles(session, result, options);
        // Restore original class loader, might not be necessary but is good practice since the processor task changed it
        Thread.currentThread().setContextClassLoader(originalClassloader);
        semaphore.release();
    }
}