use of io.confluent.ksql.execution.util.ExpressionTypeManager in project ksql by confluentinc.
the class SelectionUtil method buildProjectionSchema.
/*
* The algorithm behind this method feels unnecessarily complicated and is begging
* for someone to come along and improve it, but until that time here is
* a description of what's going on.
*
* Essentially, we need to build a logical schema that mirrors the physical
* schema until https://github.com/confluentinc/ksql/issues/6374 is addressed.
* That means that the keys must be ordered in the same way as the parent schema
* (e.g. if the source schema was K1 INT KEY, K2 INT KEY and the projection is
* SELECT K2, K1 this method will produce an output schema that is K1, K2
* despite the way that the keys were ordered in the projection) - see
* https://github.com/confluentinc/ksql/pull/7477 for context on the bug.
*
* But we cannot simply select all the keys and then the values, we must maintain
* the interleaving of key and values because transient queries return all columns
* to the user as "value columns". If someone issues a SELECT VALUE, * FROM FOO
* it is expected that VALUE shows up _before_ the key fields. This means we need to
* reorder the key columns within the list of projections without affecting the
* relative order the keys/values.
*
* To spice things up even further, there's the possibility that the same key is
* aliased multiple times (SELECT K1 AS X, K2 AS Y FROM ...), which is not supported
* but is verified later when building the final projection - so we maintain it here.
*
* Now on to the algorithm itself: we make two passes through the list of projections.
* The first pass builds a mapping from source key to all the projections for that key.
* We will use this mapping to sort the keys in the second pass. This mapping is two
* dimensional to address the possibility of the same key with multiple aliases.
*
* The second pass goes through the list of projections again and builds the logical schema,
* but this time if we encounter a projection that references a key column, we instead take
* it from the list we built in the first pass (in order defined by the parent schema).
*/
public static LogicalSchema buildProjectionSchema(final LogicalSchema parentSchema, final List<SelectExpression> projection, final FunctionRegistry functionRegistry) {
final ExpressionTypeManager expressionTypeManager = new ExpressionTypeManager(parentSchema, functionRegistry);
// keyExpressions[i] represents the expressions found in projection
// that are associated with parentSchema's key at index i
final List<List<SelectExpression>> keyExpressions = new ArrayList<>(parentSchema.key().size());
for (int i = 0; i < parentSchema.key().size(); i++) {
keyExpressions.add(new ArrayList<>());
}
// first pass to construct keyExpressions, keyExpressionMembership
// is just a convenience data structure so that we don't have to do
// the isKey check in the second iteration below
final Set<SelectExpression> keyExpressionMembership = new HashSet<>();
for (final SelectExpression select : projection) {
final Expression expression = select.getExpression();
if (expression instanceof ColumnReferenceExp) {
final ColumnName name = ((ColumnReferenceExp) expression).getColumnName();
parentSchema.findColumn(name).filter(c -> c.namespace() == Namespace.KEY).ifPresent(c -> {
keyExpressions.get(c.index()).add(select);
keyExpressionMembership.add(select);
});
}
}
// second pass, which iterates the projections but ignores any key expressions,
// instead taking them from the ordered keyExpressions list
final Builder builder = LogicalSchema.builder();
int currKeyIdx = 0;
for (final SelectExpression select : projection) {
if (keyExpressionMembership.contains(select)) {
while (keyExpressions.get(currKeyIdx).isEmpty()) {
currKeyIdx++;
}
final SelectExpression keyExp = keyExpressions.get(currKeyIdx).remove(0);
final SqlType type = expressionTypeManager.getExpressionSqlType(keyExp.getExpression());
builder.keyColumn(keyExp.getAlias(), type);
} else {
final Expression expression = select.getExpression();
final SqlType type = expressionTypeManager.getExpressionSqlType(expression);
if (type == null) {
throw new IllegalArgumentException("Can't infer a type of null. Please explicitly cast " + "it to a required type, e.g. CAST(null AS VARCHAR).");
}
builder.valueColumn(select.getAlias(), type);
}
}
return builder.build();
}
use of io.confluent.ksql.execution.util.ExpressionTypeManager in project ksql by confluentinc.
the class PartitionByParamsFactory method buildSchema.
private static LogicalSchema buildSchema(final LogicalSchema sourceSchema, final List<Expression> partitionBys, final FunctionRegistry functionRegistry, final List<PartitionByColumn> partitionByCols) {
final ExpressionTypeManager expressionTypeManager = new ExpressionTypeManager(sourceSchema, functionRegistry);
final List<SqlType> keyTypes = partitionBys.stream().map(expressionTypeManager::getExpressionSqlType).collect(Collectors.toList());
if (isPartitionByNull(partitionBys)) {
final Builder builder = LogicalSchema.builder();
builder.valueColumns(sourceSchema.value());
return builder.build();
} else {
final Builder builder = LogicalSchema.builder();
for (int i = 0; i < partitionBys.size(); i++) {
builder.keyColumn(partitionByCols.get(i).name, keyTypes.get(i));
}
builder.valueColumns(sourceSchema.value());
for (int i = 0; i < partitionBys.size(); i++) {
if (partitionByCols.get(i).shouldAppend) {
// New key column added, copy in to value schema:
builder.valueColumn(partitionByCols.get(i).name, keyTypes.get(i));
}
}
return builder.build();
}
}
use of io.confluent.ksql.execution.util.ExpressionTypeManager in project ksql by confluentinc.
the class StreamFlatMapBuilder method buildSchema.
public static LogicalSchema buildSchema(final LogicalSchema inputSchema, final List<FunctionCall> tableFunctions, final FunctionRegistry functionRegistry) {
final LogicalSchema.Builder schemaBuilder = LogicalSchema.builder();
final List<Column> cols = inputSchema.value();
// We copy all the original columns to the output schema
schemaBuilder.keyColumns(inputSchema.key());
for (final Column col : cols) {
schemaBuilder.valueColumn(col);
}
final ExpressionTypeManager expressionTypeManager = new ExpressionTypeManager(inputSchema, functionRegistry);
// And add new columns representing the exploded values at the end
for (int i = 0; i < tableFunctions.size(); i++) {
final FunctionCall functionCall = tableFunctions.get(i);
final ColumnName colName = ColumnNames.synthesisedSchemaColumn(i);
final SqlType fieldType = expressionTypeManager.getExpressionSqlType(functionCall);
schemaBuilder.valueColumn(colName, fieldType);
}
return schemaBuilder.build();
}
Aggregations