Examples with KeyValueTable co.cask.cdap.api.dataset.lib.KeyValueTable used on opensource projects

Example 6 with KeyValueTable

use of co.cask.cdap.api.dataset.lib.KeyValueTable in project cdap by caskdata.

the class SparkTestRun method testScalaSparkWithObjectStore.

@Test
public void testScalaSparkWithObjectStore() throws Exception {
    ApplicationManager applicationManager = deploy(SparkAppUsingObjectStore.class);
    DataSetManager<ObjectStore<String>> keysManager = getDataset("keys");
    prepareInputData(keysManager);
    SparkManager sparkManager = applicationManager.getSparkManager(ScalaCharCountProgram.class.getSimpleName()).start();
    sparkManager.waitForRun(ProgramRunStatus.COMPLETED, 1, TimeUnit.MINUTES);
    DataSetManager<KeyValueTable> countManager = getDataset("count");
    checkOutputData(countManager);
}

Example 7 with KeyValueTable

use of co.cask.cdap.api.dataset.lib.KeyValueTable in project cdap by caskdata.

the class DynamicPartitionerWithAvroTest method runDynamicPartitionerMapReduce.

private void runDynamicPartitionerMapReduce(final List<? extends GenericRecord> records, boolean allowConcurrentWriters, boolean expectedStatus) throws Exception {
    ApplicationWithPrograms app = deployApp(AppWithMapReduceUsingAvroDynamicPartitioner.class);
    final long now = System.currentTimeMillis();
    final Multimap<PartitionKey, GenericRecord> keyToRecordsMap = groupByPartitionKey(records, now);
    // write values to the input kvTable
    final KeyValueTable kvTable = datasetCache.getDataset(INPUT_DATASET);
    Transactions.createTransactionExecutor(txExecutorFactory, kvTable).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() {
            // the keys are not used; it matters that they're unique though
            for (int i = 0; i < records.size(); i++) {
                kvTable.write(Integer.toString(i), records.get(i).toString());
            }
        }
    });
    String allowConcurrencyKey = "dataset." + OUTPUT_DATASET + "." + PartitionedFileSetArguments.DYNAMIC_PARTITIONER_ALLOW_CONCURRENCY;
    // run the partition writer m/r with this output partition time
    ImmutableMap<String, String> arguments = ImmutableMap.of(OUTPUT_PARTITION_KEY, Long.toString(now), allowConcurrencyKey, Boolean.toString(allowConcurrentWriters));
    long startTime = System.currentTimeMillis();
    boolean status = runProgram(app, AppWithMapReduceUsingAvroDynamicPartitioner.DynamicPartitioningMapReduce.class, new BasicArguments(arguments));
    Assert.assertEquals(expectedStatus, status);
    if (!expectedStatus) {
        // if we expect the program to fail, no need to check the output data for expected results
        return;
    }
    // Verify notifications
    List<Notification> notifications = getDataNotifications(startTime);
    Assert.assertEquals(1, notifications.size());
    Assert.assertEquals(NamespaceId.DEFAULT.dataset(OUTPUT_DATASET), DatasetId.fromString(notifications.get(0).getProperties().get("datasetId")));
    // this should have created a partition in the pfs
    final PartitionedFileSet pfs = datasetCache.getDataset(OUTPUT_DATASET);
    final Location pfsBaseLocation = pfs.getEmbeddedFileSet().getBaseLocation();
    Transactions.createTransactionExecutor(txExecutorFactory, (TransactionAware) pfs).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws IOException {
            Map<PartitionKey, PartitionDetail> partitions = new HashMap<>();
            for (PartitionDetail partition : pfs.getPartitions(null)) {
                partitions.put(partition.getPartitionKey(), partition);
                // check that the mapreduce wrote the output partition metadata to all the output partitions
                Assert.assertEquals(AppWithMapReduceUsingAvroDynamicPartitioner.DynamicPartitioningMapReduce.METADATA, partition.getMetadata().asMap());
            }
            Assert.assertEquals(3, partitions.size());
            Assert.assertEquals(keyToRecordsMap.keySet(), partitions.keySet());
            // Check relative paths of the partitions. Also check that their location = pfs baseLocation + relativePath
            for (Map.Entry<PartitionKey, PartitionDetail> partitionKeyEntry : partitions.entrySet()) {
                PartitionDetail partitionDetail = partitionKeyEntry.getValue();
                String relativePath = partitionDetail.getRelativePath();
                int zip = (int) partitionKeyEntry.getKey().getField("zip");
                Assert.assertEquals(Long.toString(now) + Path.SEPARATOR + zip, relativePath);
                Assert.assertEquals(pfsBaseLocation.append(relativePath), partitionDetail.getLocation());
            }
            for (Map.Entry<PartitionKey, Collection<GenericRecord>> keyToRecordsEntry : keyToRecordsMap.asMap().entrySet()) {
                Set<GenericRecord> genericRecords = new HashSet<>(keyToRecordsEntry.getValue());
                Assert.assertEquals(genericRecords, readOutput(partitions.get(keyToRecordsEntry.getKey()).getLocation()));
            }
        }
    });
}

Example 8 with KeyValueTable

use of co.cask.cdap.api.dataset.lib.KeyValueTable in project cdap by caskdata.

the class ETLWorker method run.

@Override
public void run() {
    final SourceState currentState = new SourceState();
    final SourceState nextState = new SourceState();
    final Map<String, List<Object>> dataToSink = new HashMap<>();
    boolean hasData = false;
    final Map<String, List<InvalidEntry>> transformIdToErrorRecords = intializeTransformIdToErrorsList();
    final WorkerContext context = getContext();
    Set<String> transformErrorsWithoutDataset = Sets.newHashSet();
    // Fetch SourceState from State Table.
    // Only required at the beginning since we persist the state if there is a change.
    Transactionals.execute(context, new TxRunnable() {

        @Override
        public void run(DatasetContext context) throws Exception {
            KeyValueTable stateTable = context.getDataset(ETLRealtimeApplication.STATE_TABLE);
            byte[] stateBytes = stateTable.read(stateStoreKeyBytes);
            if (stateBytes != null) {
                SourceState state = GSON.fromJson(Bytes.toString(stateBytes), SourceState.class);
                currentState.setState(state);
            }
        }
    });
    DefaultEmitter<Object> sourceEmitter = new DefaultEmitter<>();
    TrackedEmitter<Object> trackedSourceEmitter = new TrackedEmitter<>(sourceEmitter, new DefaultStageMetrics(metrics, sourceStageName), TrackedTransform.RECORDS_OUT, context.getDataTracer(sourceStageName));
    while (!stopped) {
        // Invoke poll method of the source to fetch data
        try {
            SourceState newState = source.poll(trackedSourceEmitter, new SourceState(currentState));
            if (newState != null) {
                nextState.setState(newState);
            }
        } catch (Exception e) {
            // Continue since the source threw an exception. No point in processing records and state is not changed.
            LOG.warn("Exception thrown during polling of Source for data", e);
            sourceEmitter.reset();
            continue;
        }
        // to be persisted in the sink.
        for (Object sourceData : sourceEmitter.getEntries()) {
            try {
                TransformResponse transformResponse = transformExecutor.runOneIteration(sourceData);
                for (Map.Entry<String, Collection<Object>> transformedValues : transformResponse.getSinksResults().entrySet()) {
                    dataToSink.put(transformedValues.getKey(), new ArrayList<>());
                    Iterator emitterIterator = transformedValues.getValue().iterator();
                    while (emitterIterator.hasNext()) {
                        if (!hasData) {
                            hasData = true;
                        }
                        dataToSink.get(transformedValues.getKey()).add(emitterIterator.next());
                    }
                }
                for (Map.Entry<String, Collection<InvalidEntry<Object>>> transformErrorsEntry : transformResponse.getMapTransformIdToErrorEmitter().entrySet()) {
                    if (!transformErrorsWithoutDataset.contains(transformErrorsEntry.getKey())) {
                        if (!tranformIdToDatasetName.containsKey(transformErrorsEntry.getKey()) && !transformErrorsEntry.getValue().isEmpty()) {
                            transformErrorsWithoutDataset.add(transformErrorsEntry.getKey());
                            LOG.warn("Error records were emitted in transform {}, " + "but error dataset is not configured for this transform", transformErrorsEntry.getKey());
                        }
                        if (tranformIdToDatasetName.containsKey(transformErrorsEntry.getKey()) && !transformErrorsEntry.getValue().isEmpty()) {
                            // add the errors
                            if (!hasData && transformErrorsEntry.getValue().size() > 0) {
                                hasData = true;
                            }
                            transformIdToErrorRecords.get(transformErrorsEntry.getKey()).addAll(transformErrorsEntry.getValue());
                        }
                    }
                }
            } catch (Exception e) {
                LOG.warn("Exception thrown while processing data {}", sourceData, e);
            }
        }
        sourceEmitter.reset();
        // Start a Transaction if there is data to persist or if the Source state has changed.
        try {
            if (hasData || (!nextState.equals(currentState))) {
                getContext().execute(new TxRunnable() {

                    @Override
                    public void run(DatasetContext context) throws Exception {
                        // Invoke the sink's write method if there is any object to be written.
                        if (!dataToSink.isEmpty()) {
                            DefaultDataWriter defaultDataWriter = new DefaultDataWriter(getContext(), context);
                            for (Map.Entry<String, List<Object>> sinkEntry : dataToSink.entrySet()) {
                                sinks.get(sinkEntry.getKey()).write(sinkEntry.getValue(), defaultDataWriter);
                            }
                        }
                        for (Map.Entry<String, List<InvalidEntry>> errorRecordEntry : transformIdToErrorRecords.entrySet()) {
                            String transformId = errorRecordEntry.getKey();
                            final String datasetName = tranformIdToDatasetName.get(transformId);
                            Table errorTable = context.getDataset(datasetName);
                            long timeInMillis = System.currentTimeMillis();
                            byte[] currentTime = Bytes.toBytes(timeInMillis);
                            String transformIdentifier = appName + SEPARATOR + transformId;
                            for (InvalidEntry invalidEntry : errorRecordEntry.getValue()) {
                                // using random uuid as we want to write each record uniquely,
                                // but we are not concerned about the uuid while scanning later.
                                byte[] rowKey = Bytes.concat(currentTime, Bytes.toBytes(transformIdentifier), Bytes.toBytes(UUID.randomUUID()));
                                Put errorPut = constructErrorPut(rowKey, invalidEntry, timeInMillis);
                                errorTable.write(rowKey, errorPut);
                            }
                        }
                        // Persist nextState if it is different from currentState
                        if (!nextState.equals(currentState)) {
                            KeyValueTable stateTable = context.getDataset(ETLRealtimeApplication.STATE_TABLE);
                            stateTable.write(stateStoreKey, GSON.toJson(nextState));
                        }
                        // after running one iteration and succesfully writing to sinks and error datasets, reset the emitters.
                        transformExecutor.resetEmitter();
                    }
                });
                // Update the in-memory copy of the state only if the transaction succeeded.
                currentState.setState(nextState);
            }
        } catch (Exception e) {
            LOG.warn("Exception thrown during persisting of data", e);
        } finally {
            // Clear the persisted sink data (in case transaction failure occurred, we will poll the source with old state)
            hasData = false;
            dataToSink.clear();
            for (List<InvalidEntry> invalidEntryList : transformIdToErrorRecords.values()) {
                invalidEntryList.clear();
            }
        }
    }
}

Example 9 with KeyValueTable

use of co.cask.cdap.api.dataset.lib.KeyValueTable in project cdap by caskdata.

the class ETLWorkerTest method testLookup.

@Test
public void testLookup() throws Exception {
    addDatasetInstance(KeyValueTable.class.getName(), "lookupTable");
    DataSetManager<KeyValueTable> lookupTable = getDataset("lookupTable");
    lookupTable.get().write("Bob".getBytes(Charsets.UTF_8), "123".getBytes(Charsets.UTF_8));
    lookupTable.flush();
    File outDir = TMP_FOLDER.newFolder();
    ETLRealtimeConfig etlConfig = ETLRealtimeConfig.builder().addStage(new ETLStage("source", LookupSource.getPlugin(ImmutableSet.of("Bob", "Bill"), "lookupTable"))).addStage(new ETLStage("sink", MockSink.getPlugin(outDir))).addConnection("source", "sink").build();
    ApplicationId appId = NamespaceId.DEFAULT.app("lookupTestApp");
    AppRequest<ETLRealtimeConfig> appRequest = new AppRequest<>(APP_ARTIFACT, etlConfig);
    ApplicationManager appManager = deployApplication(appId, appRequest);
    WorkerManager workerManager = appManager.getWorkerManager(ETLWorker.NAME);
    workerManager.start();
    workerManager.waitForStatus(true, 10, 1);
    Schema schema = Schema.recordOf("bobbill", Schema.Field.of("Bob", Schema.nullableOf(Schema.of(Schema.Type.STRING))), Schema.Field.of("Bill", Schema.nullableOf(Schema.of(Schema.Type.STRING))));
    List<StructuredRecord> expected = new ArrayList<>();
    expected.add(StructuredRecord.builder(schema).set("Bob", "123").build());
    try {
        List<StructuredRecord> actual = MockSink.getRecords(outDir, 0, 10, TimeUnit.SECONDS);
        Assert.assertEquals(expected, actual);
    } finally {
        stopWorker(workerManager);
    }
    validateMetric(1, appId, "source.records.out");
    validateMetric(1, appId, "sink.records.in");
}

Example 10 with KeyValueTable

use of co.cask.cdap.api.dataset.lib.KeyValueTable in project cdap by caskdata.

the class AuthorizationTest method verifyDummyData.

private void verifyDummyData(NamespaceId namespaceId, String datasetName) throws Exception {
    DataSetManager<KeyValueTable> outTableManager = getDataset(namespaceId.dataset(datasetName));
    KeyValueTable outputTable = outTableManager.get();
    Assert.assertEquals("world", Bytes.toString(outputTable.read("hello")));
}