org.apache.drill.exec.physical.resultSet

Interface RowSetLoader

All Superinterfaces:

ColumnWriter, TupleWriter

All Known Implementing Classes:

RowSetLoaderImpl

public interface RowSetLoader
extends TupleWriter

Interface for writing values to a row set. Only available for newly-created single row sets.

Typical usage:

 void writeABatch() {
   RowSetLoader writer = ...
   while (! writer.isFull()) {
     writer.start();
     writer.scalar(0).setInt(10);
     writer.scalar(1).setString("foo");
     ...
     writer.save();
   }
 }

Alternative usage:

 void writeABatch() {
   RowSetLoader writer = ...
   while (writer.start()) {
     writer.scalar(0).setInt(10);
     writer.scalar(1).setString("foo");
     ...
     writer.save();
   }
 }

The above writes until the batch is full, based on size or vector overflow. That is, the details of vector overflow are hidden from the code that calls the writer.

Nested Class Summary

Nested classes/interfaces inherited from interface org.apache.drill.exec.vector.accessor.TupleWriter

TupleWriter.UndefinedColumnException

Method Summary

Modifier and Type	Method and Description
RowSetLoader	addRow(Object... values) Write a row of values, given by Java objects.
RowSetLoader	addSingleCol(Object value) Similar to addRow(Object...), but for the odd case in which a row consists of a single column that is an object array (such as for a list or map) and so is ambiguous.
boolean	isFull() Indicates that no more rows fit into the current row batch and that the row batch should be harvested and sent downstream.
boolean	limitReached(int maxRecords) Used to push a limit down to the file reader.
ResultSetLoader	loader()
int	rowCount() The number of rows in the current row set.
int	rowIndex() The index of the current row.
void	save() Saves the current row and moves to the next row.
boolean	start() Prepare a new row for writing.

Methods inherited from interface org.apache.drill.exec.vector.accessor.TupleWriter

addColumn, addColumn, array, array, column, column, dict, dict, isProjected, scalar, scalar, set, size, tuple, tuple, tupleSchema, type, type, variant, variant

Methods inherited from interface org.apache.drill.exec.vector.accessor.ColumnWriter

copy, isProjected, nullable, schema, setNull, setObject, type

Method Detail

loader

ResultSetLoader loader()

addRow

RowSetLoader addRow(Object... values)

Write a row of values, given by Java objects. Object type must match expected column type. Stops writing, and returns false, if any value causes vector overflow. Value format:

• For scalars, the value as a suitable Java type (int or Integer, say, for INTEGER values.)
• For scalar arrays, an array of a suitable Java primitive type for scalars. For example, int[] for an INTEGER column.
• For a Map, an Object array with values encoded as above. (In fact, the list here is the same as the map format.
• For a list (repeated map, list of list), an Object array with values encoded as above. (So, for a repeated map, an outer Object map encodes the array, an inner one encodes the map members.

Parameters:

values - variable-length argument list of column values

Returns:

this writer

addSingleCol

RowSetLoader addSingleCol(Object value)

Similar to addRow(Object...), but for the odd case in which a row consists of a single column that is an object array (such as for a list or map) and so is ambiguous.

Parameters:

value - value of the one and only column

Returns:

this writer

isFull

boolean isFull()

Indicates that no more rows fit into the current row batch and that the row batch should be harvested and sent downstream. Any overflow row is automatically saved for the next cycle. The value is undefined when a batch is not active.

Will be false on the first row, and all subsequent rows until either the maximum number of rows are written, or a vector overflows. After that, will return true. The method returns false as soon as any column writer overflows even in the middle of a row write. That is, this writer does not automatically handle overflow rows because that added complexity is seldom needed for tests.

Returns:

true if another row can be written, false if not

limitReached

boolean limitReached(int maxRecords)

Used to push a limit down to the file reader. This method checks to see whether the maxRecords parameter is not zero (for no limit) and is not greater than the current record count.

Parameters:

maxRecords - Maximum rows to be returned. (From the limit clause of the query)

Returns:

True if the row count exceeds the maxRecords, false if not.

rowCount

int rowCount()

The number of rows in the current row set. Does not count any overflow row saved for the next batch.

Returns:

number of rows to be sent downstream

rowIndex

int rowIndex()

The index of the current row. Same as the row count except in an overflow row in which case the row index will revert to zero as soon as any vector overflows. Note: this means that the index can change between columns in a single row. Applications usually don't use this index directly; rely on the writers to write to the proper location.

Returns:

the current write index

start

boolean start()

Prepare a new row for writing. Call this before each row.

Handles a very special case: that of discarding the last row written. A reader can read a row into vectors, then "sniff" the row to check, for example, against a filter. If the row is not wanted, simply omit the call to save() and the next all to start() will discard the unsaved row.

Note that the vectors still contain values in the discarded position; just the various pointers are unset. If the batch ends before the discarded values are overwritten, the discarded values just exist at the end of the vector. Since vectors start with garbage contents, the discarded values are simply a different kind of garbage. But, if the client writes a new row, then the new row overwrites the discarded row. This works because we only change the tail part of a vector; never the internals.

Returns:

true if another row can be added, false if the batch is full

save

void save()

Saves the current row and moves to the next row. Failing to call this method effectively abandons the in-flight row; something that may be useful to recover from partially-written rows that turn out to contain errors. Done automatically if using setRow().