/**
* 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);
}
}
