/**
* Gets a set of input splits for a MapReduce job running over a Kiji table. One split is created
* per region in the input Kiji table.
*
* @param configuration of the job using the splits. The configuration should specify the input
* Kiji table being used, through the configuration variable {@link
* KijiConfKeys#KIJI_INPUT_TABLE_URI}.
* @param numSplits desired for the job. This framework hint is ignored by this method.
* @return an array of input splits to be operated on in the MapReduce job.
* @throws IOException if an I/O error occurs while communicating with HBase to determine the
* regions in the Kiji table.
*/
@Override
public InputSplit[] getSplits(JobConf configuration, int numSplits) throws IOException {
final String uriString =
Preconditions.checkNotNull(configuration.get(KijiConfKeys.KIJI_INPUT_TABLE_URI));
final KijiURI inputTableURI = KijiURI.newBuilder(uriString).build();
final Kiji kiji = Kiji.Factory.open(inputTableURI, configuration);
try {
final KijiTable table = kiji.openTable(inputTableURI.getTable());
try {
final HTableInterface htable = HBaseKijiTable.downcast(table).getHTable();
final List<InputSplit> splits = Lists.newArrayList();
for (KijiRegion region : table.getRegions()) {
final byte[] startKey = region.getStartKey();
// TODO(KIJIMR-65): For now pick the first available location (ie. region server), if any.
final String location =
region.getLocations().isEmpty() ? null : region.getLocations().iterator().next();
final TableSplit tableSplit =
new TableSplit(htable.getTableName(), startKey, region.getEndKey(), location);
splits.add(new KijiTableSplit(tableSplit));
}
return splits.toArray(new InputSplit[0]);
} finally {
table.release();
}
} finally {
kiji.release();
}
}
/**
* Generates a split for a given table.
*
* @param tableURI URI of the Kiji table to split.
* @param nsplits Number of splits.
* @param conf Base Hadoop configuration used to open the Kiji instance.
* @return a list of split start keys, as HFileKeyValue (with no value, just the keys).
* @throws IOException on I/O error.
*/
private static List<HFileKeyValue> makeTableKeySplit(
KijiURI tableURI, int nsplits, Configuration conf) throws IOException {
final Kiji kiji = Kiji.Factory.open(tableURI, conf);
try {
final KijiTable table = kiji.openTable(tableURI.getTable());
try {
if (NUM_SPLITS_AUTO == nsplits) {
final List<HFileKeyValue> startKeys = Lists.newArrayList();
for (KijiRegion region : table.getRegions()) {
startKeys.add(HFileKeyValue.createFromRowKey(region.getStartKey()));
}
return startKeys;
} else {
switch (KijiTableLayout.getEncoding(table.getLayout().getDesc().getKeysFormat())) {
case RAW:
{
// The user has explicitly specified how many HFiles to create, but this is not
// possible when row key hashing is disabled.
throw new JobConfigurationException(
String.format(
"Table '%s' has row key hashing disabled, so the number of HFile splits must be"
+ "determined by the number of HRegions in the HTable. "
+ "Use an HFileMapReduceJobOutput constructor that enables auto splitting.",
table.getName()));
}
case FORMATTED:
case HASH:
case HASH_PREFIX:
{
// Those cases are supported:
break;
}
default:
throw new RuntimeException(
"Unhandled row key encoding: "
+ KijiTableLayout.getEncoding(table.getLayout().getDesc().getKeysFormat()));
}
return generateEvenStartKeys(nsplits);
}
} finally {
ResourceUtils.releaseOrLog(table);
}
} finally {
ResourceUtils.releaseOrLog(kiji);
}
}
/**
* Initializes a new table-wide record writer.
*
* @param oformat KijiHFileOutputFormat this writer is built from.
* @param context Context of the task.
* @throws IOException on I/O error.
*/
public TableRecordWriter(KijiHFileOutputFormat oformat, TaskAttemptContext context)
throws IOException {
mContext = Preconditions.checkNotNull(context);
mConf = mContext.getConfiguration();
mLatestTimestamp = mConf.getLong(CONF_LATEST_TIMESTAMP, System.currentTimeMillis());
mLatestTimestampBytes = toBytes(mLatestTimestamp);
mOutputDir = oformat.getDefaultWorkFile(mContext, OUTPUT_EXTENSION);
mFileSystem = mOutputDir.getFileSystem(mConf);
mTableURI = KijiURI.newBuilder(mConf.get(KijiConfKeys.KIJI_OUTPUT_TABLE_URI)).build();
final Kiji kiji = Kiji.Factory.open(mTableURI, mConf);
final KijiTable table = kiji.openTable(mTableURI.getTable());
mLayout = table.getLayout();
ResourceUtils.releaseOrLog(table);
ResourceUtils.releaseOrLog(kiji);
}
/**
* Creates a new record reader that scans over a subset of rows from a Kiji table. The record
* reader will scan over rows in the table specified in the provided input split, subject to row
* limits specified in the data request serialized into the specified configuration.
*
* @param split for the MapReduce task that will use this record reader. The split specifies a
* subset of rows from a Kiji table.
* @param configuration for the MapReduce job using this record reader. The configuration should
* specify the input Kiji table through the configuration variable {@link
* KijiConfKeys#KIJI_INPUT_TABLE_URI} and a serialized {@link KijiDataRequest} through the
* configuration variable {@link KijiConfKeys#KIJI_INPUT_DATA_REQUEST}.
* @throws IOException if there is a problem constructing the record reader and opening the
* resources it requires.
*/
public KijiTableRecordReader(InputSplit split, Configuration configuration) throws IOException {
// Get data request from the job configuration.
final String dataRequestB64 = configuration.get(KijiConfKeys.KIJI_INPUT_DATA_REQUEST);
Preconditions.checkNotNull(dataRequestB64, "Missing data request in job configuration.");
final byte[] dataRequestBytes = Base64.decodeBase64(Bytes.toBytes(dataRequestB64));
mDataRequest = (KijiDataRequest) SerializationUtils.deserialize(dataRequestBytes);
// Open connections to Kiji.
assert split instanceof KijiTableSplit;
mSplit = (KijiTableSplit) split;
final KijiURI inputURI =
KijiURI.newBuilder(configuration.get(KijiConfKeys.KIJI_INPUT_TABLE_URI)).build();
final KijiScannerOptions scannerOptions =
new KijiScannerOptions()
.setStartRow(new HBaseEntityId(mSplit.getStartRow()))
.setStopRow(new HBaseEntityId(mSplit.getEndRow()));
mKiji = Kiji.Factory.open(inputURI, configuration);
mTable = mKiji.openTable(inputURI.getTable());
mReader = mTable.openTableReader();
mScanner = mReader.getScanner(mDataRequest, scannerOptions);
mIterator = mScanner.iterator();
}