Examples with TransactionCodec org.apache.tephra.TransactionCodec used on opensource projects

Example 1 with TransactionCodec

use of org.apache.tephra.TransactionCodec in project cdap by caskdata.

the class HBaseTableTest method testCachedEncodedTransaction.

@Test
public void testCachedEncodedTransaction() throws Exception {
    String tableName = "testEncodedTxTable";
    DatasetProperties props = DatasetProperties.EMPTY;
    getTableAdmin(CONTEXT1, tableName, props).create();
    DatasetSpecification tableSpec = DatasetSpecification.builder(tableName, HBaseTable.class.getName()).build();
    // use a transaction codec that counts the number of times encode() is called
    final AtomicInteger encodeCount = new AtomicInteger();
    final TransactionCodec codec = new TransactionCodec() {

        @Override
        public byte[] encode(Transaction tx) throws IOException {
            encodeCount.incrementAndGet();
            return super.encode(tx);
        }
    };
    // use a table util that creates an HTable that validates the encoded tx on each get
    final AtomicReference<Transaction> txRef = new AtomicReference<>();
    HBaseTableUtil util = new DelegatingHBaseTableUtil(hBaseTableUtil) {

        @Override
        public HTable createHTable(Configuration conf, TableId tableId) throws IOException {
            HTable htable = super.createHTable(conf, tableId);
            return new MinimalDelegatingHTable(htable) {

                @Override
                public Result get(org.apache.hadoop.hbase.client.Get get) throws IOException {
                    Assert.assertEquals(txRef.get().getTransactionId(), codec.decode(get.getAttribute(TxConstants.TX_OPERATION_ATTRIBUTE_KEY)).getTransactionId());
                    return super.get(get);
                }

                @Override
                public Result[] get(List<org.apache.hadoop.hbase.client.Get> gets) throws IOException {
                    for (org.apache.hadoop.hbase.client.Get get : gets) {
                        Assert.assertEquals(txRef.get().getTransactionId(), codec.decode(get.getAttribute(TxConstants.TX_OPERATION_ATTRIBUTE_KEY)).getTransactionId());
                    }
                    return super.get(gets);
                }

                @Override
                public ResultScanner getScanner(org.apache.hadoop.hbase.client.Scan scan) throws IOException {
                    Assert.assertEquals(txRef.get().getTransactionId(), codec.decode(scan.getAttribute(TxConstants.TX_OPERATION_ATTRIBUTE_KEY)).getTransactionId());
                    return super.getScanner(scan);
                }
            };
        }
    };
    HBaseTable table = new HBaseTable(CONTEXT1, tableSpec, Collections.<String, String>emptyMap(), cConf, TEST_HBASE.getConfiguration(), util, codec);
    DetachedTxSystemClient txSystemClient = new DetachedTxSystemClient();
    // test all operations: only the first one encodes
    Transaction tx = txSystemClient.startShort();
    txRef.set(tx);
    table.startTx(tx);
    table.put(b("row1"), b("col1"), b("val1"));
    Assert.assertEquals(0, encodeCount.get());
    table.get(b("row"));
    Assert.assertEquals(1, encodeCount.get());
    table.get(ImmutableList.of(new Get("a"), new Get("b")));
    Assert.assertEquals(1, encodeCount.get());
    Scanner scanner = table.scan(new Scan(null, null));
    Assert.assertEquals(1, encodeCount.get());
    scanner.close();
    table.increment(b("z"), b("z"), 0L);
    Assert.assertEquals(1, encodeCount.get());
    table.commitTx();
    table.postTxCommit();
    // test that for the next tx, we encode again
    tx = txSystemClient.startShort();
    txRef.set(tx);
    table.startTx(tx);
    table.get(b("row"));
    Assert.assertEquals(2, encodeCount.get());
    table.commitTx();
    // test that we encode again, even of postTxCommit was not called
    tx = txSystemClient.startShort();
    txRef.set(tx);
    table.startTx(tx);
    table.get(b("row"));
    Assert.assertEquals(3, encodeCount.get());
    table.commitTx();
    table.rollbackTx();
    // test that rollback does not encode the tx
    Assert.assertEquals(3, encodeCount.get());
    // test that we encode again if the previous tx rolled back
    tx = txSystemClient.startShort();
    txRef.set(tx);
    table.startTx(tx);
    table.get(b("row"));
    Assert.assertEquals(4, encodeCount.get());
    table.commitTx();
    table.close();
    Assert.assertEquals(4, encodeCount.get());
}

Example 2 with TransactionCodec

use of org.apache.tephra.TransactionCodec in project cdap by caskdata.

the class HBaseTableExporter method createSubmittableJob.

/**
 * Sets up the actual MapReduce job.
 * @param tx The transaction which needs to be passed to the Scan instance. This transaction is be used by
 *           coprocessors to filter out the data corresonding to the invalid transactions .
 * @param tableName Name of the table which need to be exported as HFiles.
 * @return the configured job
 * @throws IOException
 */
public Job createSubmittableJob(Transaction tx, String tableName) throws IOException {
    Job job = Job.getInstance(hConf, "HBaseTableExporter");
    job.setJarByClass(HBaseTableExporter.class);
    Scan scan = new Scan();
    scan.setCacheBlocks(false);
    // Set the transaction attribute for the scan.
    scan.setAttribute(TxConstants.TX_OPERATION_ATTRIBUTE_KEY, new TransactionCodec().encode(tx));
    job.setNumReduceTasks(0);
    TableMapReduceUtil.initTableMapperJob(tableName, scan, Import.KeyValueImporter.class, null, null, job);
    FileSystem fs = FileSystem.get(hConf);
    Random rand = new Random();
    Path root = new Path(fs.getWorkingDirectory(), "hbasetableexporter");
    fs.mkdirs(root);
    while (true) {
        bulkloadDir = new Path(root, "" + rand.nextLong());
        if (!fs.exists(bulkloadDir)) {
            break;
        }
    }
    HFileOutputFormat2.setOutputPath(job, bulkloadDir);
    HTable hTable = new HTable(hConf, tableName);
    HFileOutputFormat2.configureIncrementalLoad(job, hTable);
    return job;
}

Example 3 with TransactionCodec

use of org.apache.tephra.TransactionCodec in project cdap by caskdata.

the class MapReduceTaskContextProvider method createCacheLoader.

/**
 * Creates a {@link CacheLoader} for the task context cache.
 */
private CacheLoader<ContextCacheKey, BasicMapReduceTaskContext> createCacheLoader(final Injector injector) {
    final DiscoveryServiceClient discoveryServiceClient = injector.getInstance(DiscoveryServiceClient.class);
    final DatasetFramework datasetFramework = injector.getInstance(DatasetFramework.class);
    final SecureStore secureStore = injector.getInstance(SecureStore.class);
    final SecureStoreManager secureStoreManager = injector.getInstance(SecureStoreManager.class);
    final MessagingService messagingService = injector.getInstance(MessagingService.class);
    // Multiple instances of BasicMapReduceTaskContext can share the same program.
    final AtomicReference<Program> programRef = new AtomicReference<>();
    return new CacheLoader<ContextCacheKey, BasicMapReduceTaskContext>() {

        @Override
        public BasicMapReduceTaskContext load(ContextCacheKey key) throws Exception {
            TaskAttemptID taskAttemptId = key.getTaskAttemptID();
            // taskAttemptId could be null if used from a org.apache.hadoop.mapreduce.Partitioner or
            // from a org.apache.hadoop.io.RawComparator, in which case we can get the JobId from the conf. Note that the
            // JobId isn't in the conf for the OutputCommitter#setupJob method, in which case we use the taskAttemptId
            Path txFile = MainOutputCommitter.getTxFile(key.getConfiguration(), taskAttemptId != null ? taskAttemptId.getJobID() : null);
            FileSystem fs = txFile.getFileSystem(key.getConfiguration());
            Preconditions.checkArgument(fs.exists(txFile));
            Transaction tx;
            try (FSDataInputStream txFileInputStream = fs.open(txFile)) {
                byte[] txByteArray = ByteStreams.toByteArray(txFileInputStream);
                tx = new TransactionCodec().decode(txByteArray);
            }
            MapReduceContextConfig contextConfig = new MapReduceContextConfig(key.getConfiguration());
            MapReduceClassLoader classLoader = MapReduceClassLoader.getFromConfiguration(key.getConfiguration());
            Program program = programRef.get();
            if (program == null) {
                // Creation of program is relatively cheap, so just create and do compare and set.
                programRef.compareAndSet(null, createProgram(contextConfig, classLoader.getProgramClassLoader()));
                program = programRef.get();
            }
            WorkflowProgramInfo workflowInfo = contextConfig.getWorkflowProgramInfo();
            DatasetFramework programDatasetFramework = workflowInfo == null ? datasetFramework : NameMappedDatasetFramework.createFromWorkflowProgramInfo(datasetFramework, workflowInfo, program.getApplicationSpecification());
            // Setup dataset framework context, if required
            if (programDatasetFramework instanceof ProgramContextAware) {
                ProgramRunId programRunId = program.getId().run(ProgramRunners.getRunId(contextConfig.getProgramOptions()));
                ((ProgramContextAware) programDatasetFramework).setContext(new BasicProgramContext(programRunId));
            }
            MapReduceSpecification spec = program.getApplicationSpecification().getMapReduce().get(program.getName());
            MetricsCollectionService metricsCollectionService = null;
            MapReduceMetrics.TaskType taskType = null;
            String taskId = null;
            ProgramOptions options = contextConfig.getProgramOptions();
            // from a org.apache.hadoop.io.RawComparator
            if (taskAttemptId != null) {
                taskId = taskAttemptId.getTaskID().toString();
                if (MapReduceMetrics.TaskType.hasType(taskAttemptId.getTaskType())) {
                    taskType = MapReduceMetrics.TaskType.from(taskAttemptId.getTaskType());
                    // if this is not for a mapper or a reducer, we don't need the metrics collection service
                    metricsCollectionService = injector.getInstance(MetricsCollectionService.class);
                    options = new SimpleProgramOptions(options.getProgramId(), options.getArguments(), new BasicArguments(RuntimeArguments.extractScope("task", taskType.toString().toLowerCase(), contextConfig.getProgramOptions().getUserArguments().asMap())), options.isDebug());
                }
            }
            CConfiguration cConf = injector.getInstance(CConfiguration.class);
            TransactionSystemClient txClient = injector.getInstance(TransactionSystemClient.class);
            return new BasicMapReduceTaskContext(program, options, cConf, taskType, taskId, spec, workflowInfo, discoveryServiceClient, metricsCollectionService, txClient, tx, programDatasetFramework, classLoader.getPluginInstantiator(), contextConfig.getLocalizedResources(), secureStore, secureStoreManager, authorizationEnforcer, authenticationContext, messagingService, mapReduceClassLoader);
        }
    };
}

Example 4 with TransactionCodec

use of org.apache.tephra.TransactionCodec in project cdap by caskdata.

the class MainOutputCommitter method setupJob.

@Override
public void setupJob(JobContext jobContext) throws IOException {
    Configuration configuration = jobContext.getConfiguration();
    MapReduceClassLoader classLoader = MapReduceClassLoader.getFromConfiguration(configuration);
    MapReduceTaskContextProvider taskContextProvider = classLoader.getTaskContextProvider();
    Injector injector = taskContextProvider.getInjector();
    cConf = injector.getInstance(CConfiguration.class);
    MapReduceContextConfig contextConfig = new MapReduceContextConfig(jobContext.getConfiguration());
    ProgramId programId = contextConfig.getProgramId();
    LOG.info("Setting up for MapReduce job: namespaceId={}, applicationId={}, program={}, runid={}", programId.getNamespace(), programId.getApplication(), programId.getProgram(), ProgramRunners.getRunId(contextConfig.getProgramOptions()));
    RetryStrategy retryStrategy = SystemArguments.getRetryStrategy(contextConfig.getProgramOptions().getUserArguments().asMap(), contextConfig.getProgramId().getType(), cConf);
    this.txClient = new RetryingLongTransactionSystemClient(injector.getInstance(TransactionSystemClient.class), retryStrategy);
    // We start long-running tx to be used by mapreduce job tasks.
    this.transaction = txClient.startLong();
    // Write the tx somewhere, so that we can re-use it in mapreduce tasks
    Path txFile = getTxFile(configuration, jobContext.getJobID());
    FileSystem fs = txFile.getFileSystem(configuration);
    try (FSDataOutputStream fsDataOutputStream = fs.create(txFile, false)) {
        fsDataOutputStream.write(new TransactionCodec().encode(transaction));
    }
    // we can instantiate the TaskContext after we set the tx above. It's used by the operations below
    taskContext = taskContextProvider.get(taskAttemptContext);
    this.outputs = Outputs.transform(contextConfig.getOutputs(), taskContext);
    super.setupJob(jobContext);
}