org.apache.drill.exec.physical.resultSet.impl

Class TestResultSetLoaderProtocol

java.lang.Object

DrillTest

org.apache.drill.test.SubOperatorTest

org.apache.drill.exec.physical.resultSet.impl.TestResultSetLoaderProtocol

public class TestResultSetLoaderProtocol
extends SubOperatorTest

Tests of the overall result set loader protocol focusing on which operations are valid in each state, basics of column lookup, basics of adding columns and so on. Uses the simplest possible type: a required int.

Run this test first to do a sanity check of the result set loader after making changes.

You will find that the result set loader creates a very complex tree of objects that can be quite hard to understand and debug. Please read the material in the various subsystems to see how the classes fit together to implement Drill's rich JSON-like data model.

To aid in debugging, you can also dump the result set loader, and all its child objects as follows:

 ((ResultSetLoaderImpl) rsLoader).dump(new HierarchicalPrinter());
 

Simply insert that line into these tests anywhere you want to visualize the structure. The object tree will show all the components and their current state.

Field Summary

Fields inherited from class org.apache.drill.test.SubOperatorTest

dirTestWatcher, fixture

Constructor Summary

Constructors

Constructor and Description
TestResultSetLoaderProtocol()

Method Summary

Modifier and Type	Method and Description
void	testBasics()
void	testCaseInsensitiveSchema() Schemas are case insensitive by default.
void	testCloseWithoutHarvest() Test that memory is released if the loader is closed with an active batch (that is, before the batch is harvested.)
void	testInitialSchema() Provide a schema up front to the loader; schema is built before the first row.
void	testOverwriteRow() The writer protocol allows a client to write to a row any number of times before invoking save().

Methods inherited from class org.apache.drill.test.SubOperatorTest

classSetup, classTeardown

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

TestResultSetLoaderProtocol

public TestResultSetLoaderProtocol()

Method Detail

testBasics

public void testBasics()

testCaseInsensitiveSchema

public void testCaseInsensitiveSchema()

Schemas are case insensitive by default. Verify that the schema mechanism works, with emphasis on the case insensitive case.

The tests here and elsewhere build columns from a MaterializedField. Doing so is rather old-school; better to use the newer ColumnMetadata which provides additional information. The code here simply uses the MaterializedField to create a ColumnMetadata implicitly.

testInitialSchema

public void testInitialSchema()

Provide a schema up front to the loader; schema is built before the first row.

Also verifies the test-time method to set a row of values using a single method.

testOverwriteRow

public void testOverwriteRow()

The writer protocol allows a client to write to a row any number of times before invoking save(). In this case, each new value simply overwrites the previous value. Here, we test the most basic case: a simple, flat tuple with no arrays. We use a very large Varchar that would, if overwrite were not working, cause vector overflow.

The ability to overwrite rows is seldom needed except in one future use case: writing a row, then applying a filter "in-place" to discard unwanted rows, without having to send the row downstream.

Because of this use case, specific rules apply when discarding row or overwriting values.

• Values can be written once per row. Fixed-width columns actually allow multiple writes. But, because of the way variable-width columns work, multiple writes will cause undefined results.
• To overwrite a row, call start() without calling save() on the previous row. Doing so ignores data for the previous row and starts a new row in place of the old one.

Note that there is no explicit method to discard a row. Instead, the rule is that a row is not saved until save() is called.

testCloseWithoutHarvest

public void testCloseWithoutHarvest()

Test that memory is released if the loader is closed with an active batch (that is, before the batch is harvested.)