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

Example 11 with PartitionDetail

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

the class PartitionedFileSetDataset method consumePartitions.

// PartitionConsumerState consists of two things:
//   1) A list of transaction IDs representing the list of transactions in progress during the previous call.
//      Each of these transaction IDs need to be checked for new partitions because there may be partitions created by
//      those partitions since the previous call.
//   2) A transaction ID from which to start scanning for new partitions. This is an exclusive end range that the
//      previous call stopped scanning partitions at.
//   Note that each of the transactions IDs in (1) will be smaller than the transactionId in (2).
@ReadWrite
@Override
public PartitionConsumerResult consumePartitions(PartitionConsumerState partitionConsumerState, int limit, Predicate<PartitionDetail> predicate) {
    List<Long> previousInProgress = partitionConsumerState.getVersionsToCheck();
    Set<Long> noLongerInProgress = setDiff(previousInProgress, tx.getInProgress());
    List<PartitionDetail> partitions = Lists.newArrayList();
    Iterator<Long> iter = noLongerInProgress.iterator();
    while (iter.hasNext()) {
        Long txId = iter.next();
        if (partitions.size() >= limit) {
            break;
        }
        try (Scanner scanner = partitionsTable.readByIndex(WRITE_PTR_COL, Bytes.toBytes(txId))) {
            scannerToPartitions(scanner, partitions, limit, predicate);
        }
        // remove the txIds as they are added to the partitions list already
        // if they're not removed, they will be persisted in the state for the next scan
        iter.remove();
    }
    // exclusive scan end, to be used as the start for a next call to consumePartitions
    long scanUpTo;
    if (partitions.size() < limit) {
        // no read your own writes (partitions)
        scanUpTo = Math.min(tx.getWritePointer(), tx.getReadPointer() + 1);
        Long endTxId;
        try (Scanner scanner = partitionsTable.scanByIndex(WRITE_PTR_COL, Bytes.toBytes(partitionConsumerState.getStartVersion()), Bytes.toBytes(scanUpTo))) {
            endTxId = scannerToPartitions(scanner, partitions, limit, predicate);
        }
        if (endTxId != null) {
            // nonnull means that the scanner was not exhausted
            scanUpTo = endTxId;
        }
    } else {
        // if we have already hit the limit, don't scan; instead, use the startVersion as the startVersion to the next
        // call to consumePartitions
        scanUpTo = partitionConsumerState.getStartVersion();
    }
    List<Long> inProgressBeforeScanEnd = Lists.newArrayList(noLongerInProgress);
    for (long txId : tx.getInProgress()) {
        if (txId >= scanUpTo) {
            break;
        }
        inProgressBeforeScanEnd.add(txId);
    }
    return new PartitionConsumerResult(new PartitionConsumerState(scanUpTo, inProgressBeforeScanEnd), partitions);
}

Example 12 with PartitionDetail

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

the class PartitionedFileSetDataset method dropPartition.

@WriteOnly
@Override
public void dropPartition(PartitionKey key) {
    byte[] rowKey = generateRowKey(key, partitioning);
    PartitionDetail partition = getPartition(key);
    if (partition == null) {
        // silently ignore non-existing partitions
        return;
    }
    // TODO: make DDL operations transactional [CDAP-1393]
    dropPartitionFromExplore(key);
    partitionsTable.delete(rowKey);
    if (!isExternal) {
        Location partitionLocation = partition.getLocation();
        try {
            if (partitionLocation.exists()) {
                Location dstLocation = getQuarantineLocation().append(partition.getRelativePath());
                Location dstParent = Locations.getParent(dstLocation);
                // shouldn't be null, since dstLocation was created by appending to a location, so it must have a parent
                Preconditions.checkNotNull(dstParent);
                // before moving into quarantine, we need to ensure that parent location exists
                if (!dstParent.exists()) {
                    if (!dstParent.mkdirs()) {
                        throw new DataSetException(String.format("Failed to create parent directory %s", dstParent));
                    }
                }
                partitionLocation.renameTo(dstLocation);
            }
        } catch (IOException ioe) {
            throw new DataSetException(String.format("Failed to move location %s into quarantine", partitionLocation));
        }
        operationsInThisTx.add(new DropPartitionOperation(key, partition.getRelativePath()));
    }
}

Example 13 with PartitionDetail

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

the class SportResultsTest method testPartitionedCounting.

@Test
public void testPartitionedCounting() throws Exception {
    // deploy the application and start the upload service
    ApplicationManager appManager = deployApplication(SportResults.class);
    ServiceManager serviceManager = appManager.getServiceManager("UploadService").start();
    serviceManager.waitForStatus(true);
    // upload a few dummy results
    URL url = serviceManager.getServiceURL();
    uploadResults(url, "fantasy", 2014, FANTASY_2014);
    uploadResults(url, "fantasy", 2015, FANTASY_2015);
    uploadResults(url, "critters", 2014, CRITTERS_2014);
    // start a map/reduce that counts all seasons for the fantasy league
    MapReduceManager mrManager = appManager.getMapReduceManager("ScoreCounter").start(ImmutableMap.of("league", "fantasy"));
    // should be much faster, though
    mrManager.waitForRun(ProgramRunStatus.COMPLETED, 5, TimeUnit.MINUTES);
    // validate the output by reading directly from the file set
    DataSetManager<PartitionedFileSet> dataSetManager = getDataset("totals");
    PartitionedFileSet totals = dataSetManager.get();
    PartitionDetail partitionDetail = totals.getPartition(PartitionKey.builder().addStringField("league", "fantasy").build());
    Assert.assertNotNull(partitionDetail);
    Location location = partitionDetail.getLocation();
    // find the part file that has the actual results
    Assert.assertTrue(location.isDirectory());
    for (Location file : location.list()) {
        if (file.getName().startsWith("part")) {
            location = file;
        }
    }
    BufferedReader reader = new BufferedReader(new InputStreamReader(location.getInputStream(), "UTF-8"));
    // validate each line
    Map<String, String[]> expected = ImmutableMap.of("My Team", new String[] { "My Team", "2", "0", "1", "53", "65" }, "Your Team", new String[] { "Your Team", "1", "0", "2", "63", "60" }, "Other Team", new String[] { "Other Team", "1", "0", "1", "40", "31" });
    while (true) {
        String line = reader.readLine();
        if (line == null) {
            break;
        }
        String[] fields = line.split(",");
        Assert.assertArrayEquals(expected.get(fields[0]), fields);
    }
    // verify using SQL
    // query with SQL
    Connection connection = getQueryClient();
    ResultSet results = connection.prepareStatement("SELECT wins, ties, losses, scored, conceded " + "FROM totals WHERE team = 'My Team' AND league = 'fantasy'").executeQuery();
    // should return only one row, with correct time fields
    Assert.assertTrue(results.next());
    Assert.assertEquals(2, results.getInt(1));
    Assert.assertEquals(0, results.getInt(2));
    Assert.assertEquals(1, results.getInt(3));
    Assert.assertEquals(53, results.getInt(4));
    Assert.assertEquals(65, results.getInt(5));
    Assert.assertFalse(results.next());
}

Example 14 with PartitionDetail

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

the class SparkFileSetTestRun method testSparkWithPartitionedFileSet.

private void testSparkWithPartitionedFileSet(ApplicationManager applicationManager, String sparkProgram) throws Exception {
    DataSetManager<PartitionedFileSet> pfsManager = getDataset("pfs");
    PartitionedFileSet pfs = pfsManager.get();
    PartitionOutput partitionOutput = pfs.getPartitionOutput(PartitionKey.builder().addStringField("x", "nn").build());
    Location location = partitionOutput.getLocation();
    prepareFileInput(location);
    partitionOutput.addPartition();
    pfsManager.flush();
    Map<String, String> inputArgs = new HashMap<>();
    PartitionedFileSetArguments.setInputPartitionFilter(inputArgs, PartitionFilter.builder().addRangeCondition("x", "na", "nx").build());
    Map<String, String> outputArgs = new HashMap<>();
    final PartitionKey outputKey = PartitionKey.builder().addStringField("x", "xx").build();
    PartitionedFileSetArguments.setOutputPartitionKey(outputArgs, outputKey);
    Map<String, String> args = new HashMap<>();
    args.putAll(RuntimeArguments.addScope(Scope.DATASET, "pfs", inputArgs));
    args.putAll(RuntimeArguments.addScope(Scope.DATASET, "pfs", outputArgs));
    args.put("input", "pfs");
    args.put("output", "pfs");
    SparkManager sparkManager = applicationManager.getSparkManager(sparkProgram).start(args);
    sparkManager.waitForRun(ProgramRunStatus.COMPLETED, 10, TimeUnit.MINUTES);
    pfsManager.flush();
    PartitionDetail partition = pfs.getPartition(outputKey);
    Assert.assertNotNull(partition);
    validateFileOutput(partition.getLocation());
    // Cleanup after test completed
    location.delete(true);
    partition.getLocation().delete(true);
    pfs.dropPartition(partitionOutput.getPartitionKey());
    pfs.dropPartition(partition.getPartitionKey());
    pfsManager.flush();
}

Example 15 with PartitionDetail

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

the class FileUploadServiceTestRun method testFileUploadService.

@Test
public void testFileUploadService() throws Exception {
    ApplicationManager appManager = deployApplication(FileUploadApp.class);
    // Start the service
    ServiceManager serviceManager = appManager.getServiceManager(FileUploadApp.SERVICE_NAME).start();
    try {
        // Upload URL is "base/upload/pfs/[partition_value], which the partition value is a long
        URI serviceURI = serviceManager.getServiceURL(10, TimeUnit.SECONDS).toURI();
        // Upload with wrong MD5, should get 400.
        byte[] content = Strings.repeat("0123456789 ", 100).getBytes(Charsets.UTF_8);
        Assert.assertEquals(HttpURLConnection.HTTP_BAD_REQUEST, upload(serviceURI.resolve("upload/" + FileUploadApp.PFS_NAME + "/1").toURL(), content, "123", 30));
        long beforeUploadTime = System.currentTimeMillis();
        // Upload with right MD5, should get 200
        Assert.assertEquals(HttpURLConnection.HTTP_OK, upload(serviceURI.resolve("upload/" + FileUploadApp.PFS_NAME + "/1").toURL(), content, BaseEncoding.base64().encode(Hashing.md5().hashBytes(content).asBytes()), 20));
        // Inspect the partitioned file set and verify the content
        PartitionedFileSet pfs = (PartitionedFileSet) getDataset(FileUploadApp.PFS_NAME).get();
        PartitionDetail partition = pfs.getPartition(PartitionKey.builder().addLongField("time", 1).build());
        Assert.assertNotNull(partition);
        // Verify a notification should have been published for the new partition
        List<Notification> notifications = getDataNotifications(beforeUploadTime);
        // Should have one message
        Assert.assertEquals(1, notifications.size());
        verifyDataNotification(notifications.get(0), NamespaceId.DEFAULT.dataset(FileUploadApp.PFS_NAME), Collections.singletonList(PartitionKey.builder().addLongField("time", 1L).build()));
        // There should be one file under the partition directory
        List<Location> locations = partition.getLocation().list();
        Assert.assertEquals(1, locations.size());
        Assert.assertArrayEquals(content, ByteStreams.toByteArray(Locations.newInputSupplier(locations.get(0))));
        // Verify the tracking table of chunks sizes
        KeyValueTable trackingTable = (KeyValueTable) getDataset(FileUploadApp.KV_TABLE_NAME).get();
        CloseableIterator<KeyValue<byte[], byte[]>> iter = trackingTable.scan(null, null);
        // Sum up all chunks sizes as being tracked by the tracking table.
        long sum = 0;
        int iterSize = 0;
        while (iter.hasNext()) {
            KeyValue<byte[], byte[]> kv = iter.next();
            sum += Bytes.toInt(kv.getKey()) * Bytes.toLong(kv.getValue());
            iterSize++;
        }
        // The iterator should have size >= 2, since we uses different chunk size for two different upload
        Assert.assertTrue(iterSize >= 2);
        // The sum of all chunks sizes should be the same as the
        // content size * 2 (since we have one failure, one success upload)
        Assert.assertEquals(content.length * 2, sum);
    } finally {
        serviceManager.stop();
    }
}