private List<CassandraPartition> getCassandraPartitions(
CassandraTable table, TupleDomain<ColumnHandle> tupleDomain) {
if (tupleDomain.isNone()) {
return ImmutableList.of();
}
Set<List<Comparable<?>>> partitionKeysSet = getPartitionKeysSet(table, tupleDomain);
// empty filter means, all partitions
if (partitionKeysSet.isEmpty()) {
return schemaProvider.getAllPartitions(table);
}
ImmutableList.Builder<ListenableFuture<List<CassandraPartition>>> getPartitionResults =
ImmutableList.builder();
for (List<Comparable<?>> partitionKeys : partitionKeysSet) {
getPartitionResults.add(
executor.submit(() -> schemaProvider.getPartitions(table, partitionKeys)));
}
ImmutableList.Builder<CassandraPartition> partitions = ImmutableList.builder();
for (ListenableFuture<List<CassandraPartition>> result : getPartitionResults.build()) {
try {
partitions.addAll(result.get());
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw Throwables.propagate(e);
} catch (ExecutionException e) {
throw new PrestoException(EXTERNAL, "Error fetching cassandra partitions", e);
}
}
return partitions.build();
}
@Override
public ConnectorSplitSource getPartitionSplits(
ConnectorSession session,
ConnectorTableHandle tableHandle,
List<ConnectorPartition> partitions) {
checkNotNull(tableHandle, "tableHandle is null");
CassandraTableHandle cassandraTableHandle =
checkType(tableHandle, CassandraTableHandle.class, "tableHandle");
checkNotNull(partitions, "partitions is null");
if (partitions.isEmpty()) {
return new FixedSplitSource(connectorId, ImmutableList.<ConnectorSplit>of());
}
// if this is an unpartitioned table, split into equal ranges
if (partitions.size() == 1) {
ConnectorPartition partition = partitions.get(0);
CassandraPartition cassandraPartition =
checkType(partition, CassandraPartition.class, "partition");
if (cassandraPartition.isUnpartitioned()
|| cassandraPartition.isIndexedColumnPredicatePushdown()) {
CassandraTable table = schemaProvider.getTable(cassandraTableHandle);
List<ConnectorSplit> splits =
getSplitsByTokenRange(table, cassandraPartition.getPartitionId());
return new FixedSplitSource(connectorId, splits);
}
}
return new FixedSplitSource(
connectorId, getSplitsForPartitions(cassandraTableHandle, partitions));
}
@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);
}
