private List<ConnectorSplit> getSplitsByTokenRange(CassandraTable table, String partitionId) {
String schema = table.getTableHandle().getSchemaName();
String tableName = table.getTableHandle().getTableName();
String tokenExpression = table.getTokenExpression();
ImmutableList.Builder<ConnectorSplit> builder = ImmutableList.builder();
List<CassandraTokenSplitManager.TokenSplit> tokenSplits;
try {
tokenSplits = tokenSplitMgr.getSplits(schema, tableName);
} catch (IOException e) {
throw new RuntimeException(e);
}
for (CassandraTokenSplitManager.TokenSplit tokenSplit : tokenSplits) {
String condition =
buildTokenCondition(
tokenExpression, tokenSplit.getStartToken(), tokenSplit.getEndToken());
List<HostAddress> addresses =
new HostAddressFactory().AddressNamesToHostAddressList(tokenSplit.getHosts());
CassandraSplit split =
new CassandraSplit(connectorId, schema, tableName, partitionId, condition, addresses);
builder.add(split);
}
return builder.build();
}
private static Set<List<Comparable<?>>> getPartitionKeysSet(
CassandraTable table, TupleDomain<ColumnHandle> tupleDomain) {
ImmutableList.Builder<Set<Comparable<?>>> partitionColumnValues = ImmutableList.builder();
for (CassandraColumnHandle columnHandle : table.getPartitionKeyColumns()) {
Domain domain = tupleDomain.getDomains().get(columnHandle);
// if there is no constraint on a partition key, return an empty set
if (domain == null) {
return ImmutableSet.of();
}
// todo does cassandra allow null partition keys?
if (domain.isNullAllowed()) {
return ImmutableSet.of();
}
ImmutableSet.Builder<Comparable<?>> columnValues = ImmutableSet.builder();
for (Range range : domain.getRanges()) {
// if the range is not a single value, we can not perform partition pruning
if (!range.isSingleValue()) {
return ImmutableSet.of();
}
Comparable<?> value = range.getSingleValue();
CassandraType valueType = columnHandle.getCassandraType();
columnValues.add(valueType.getValueForPartitionKey(value));
}
partitionColumnValues.add(columnValues.build());
}
return Sets.cartesianProduct(partitionColumnValues.build());
}
@Override
public ConnectorPartitionResult getPartitions(
ConnectorSession session,
ConnectorTableHandle tableHandle,
TupleDomain<ColumnHandle> tupleDomain) {
CassandraTableHandle cassandraTableHandle =
checkType(tableHandle, CassandraTableHandle.class, "tableHandle");
checkNotNull(tupleDomain, "tupleDomain is null");
CassandraTable table = schemaProvider.getTable(cassandraTableHandle);
List<CassandraColumnHandle> partitionKeys = table.getPartitionKeyColumns();
// fetch the partitions
List<CassandraPartition> allPartitions = getCassandraPartitions(table, tupleDomain);
log.debug(
"%s.%s #partitions: %d",
cassandraTableHandle.getSchemaName(),
cassandraTableHandle.getTableName(),
allPartitions.size());
// do a final pass to filter based on fields that could not be used to build the prefix
List<ConnectorPartition> partitions =
allPartitions
.stream()
.filter(partition -> tupleDomain.overlaps(partition.getTupleDomain()))
.collect(toList());
// All partition key domains will be fully evaluated, so we don't need to include those
TupleDomain<ColumnHandle> remainingTupleDomain = TupleDomain.none();
if (!tupleDomain.isNone()) {
if (partitions.size() == 1 && ((CassandraPartition) partitions.get(0)).isUnpartitioned()) {
remainingTupleDomain = tupleDomain;
} else {
@SuppressWarnings({"rawtypes", "unchecked"})
List<ColumnHandle> partitionColumns = (List) partitionKeys;
remainingTupleDomain =
TupleDomain.withColumnDomains(
Maps.filterKeys(tupleDomain.getDomains(), not(in(partitionColumns))));
}
}
// push down indexed column fixed value predicates only for unpartitioned partition which uses
// token range query
if (partitions.size() == 1 && ((CassandraPartition) partitions.get(0)).isUnpartitioned()) {
Map<ColumnHandle, Domain> domains = tupleDomain.getDomains();
List<ColumnHandle> indexedColumns = new ArrayList<>();
// compose partitionId by using indexed column
StringBuilder sb = new StringBuilder();
for (Map.Entry<ColumnHandle, Domain> entry : domains.entrySet()) {
CassandraColumnHandle column = (CassandraColumnHandle) entry.getKey();
Domain domain = entry.getValue();
if (column.isIndexed() && domain.isSingleValue()) {
sb.append(CassandraCqlUtils.validColumnName(column.getName()))
.append(" = ")
.append(
CassandraCqlUtils.cqlValue(
toCQLCompatibleString(entry.getValue().getSingleValue()),
column.getCassandraType()));
indexedColumns.add(column);
// Only one indexed column predicate can be pushed down.
break;
}
}
if (sb.length() > 0) {
CassandraPartition partition = (CassandraPartition) partitions.get(0);
TupleDomain<ColumnHandle> filterIndexedColumn =
TupleDomain.withColumnDomains(
Maps.filterKeys(remainingTupleDomain.getDomains(), not(in(indexedColumns))));
partitions = new ArrayList<>();
partitions.add(
new CassandraPartition(partition.getKey(), sb.toString(), filterIndexedColumn, true));
return new ConnectorPartitionResult(partitions, filterIndexedColumn);
}
}
return new ConnectorPartitionResult(partitions, remainingTupleDomain);
}
