Package org.apache.hadoop.hbase.regionserver

Class TestScannerBlockSizeLimits

java.lang.Object

org.apache.hadoop.hbase.regionserver.TestScannerBlockSizeLimits

public class TestScannerBlockSizeLimits
extends Object

Field Summary

Fields

Modifier and Type	Field	Description
static final HBaseClassTestRule	CLASS_RULE
private static final byte[]	COLUMN1
private static final byte[]	COLUMN2
private static final byte[]	COLUMN3
private static final byte[]	COLUMN5
private static final byte[]	DATA
private static final byte[][]	FAMILIES
private static final byte[]	FAMILY1
private static final byte[]	FAMILY2
private static final org.apache.hadoop.hbase.TableName	TABLE
private static final HBaseTestingUtil	TEST_UTIL

Constructor Summary

Constructors

Constructor	Description
TestScannerBlockSizeLimits()

Method Summary

Modifier and Type	Method	Description
private static void	createTestData()
private org.apache.hadoop.hbase.client.Scan	getBaseScan()	We enable cursors and partial results to give us more granularity over counting of results, and we enable STREAM so that no auto switching from pread to stream occurs -- this throws off the rpc counts.
static void	setUp()
void	setupEach()
void	testCheckLimitAfterFilteringCell()	At the end of the loop in StoreScanner, we do one more check of size limits.
void	testCheckLimitAfterFilteringRowCells()	After RegionScannerImpl.populateResults, row filters are run.
void	testCheckLimitAfterFilteringRowCellsDueToFilterRow()	After RegionScannerImpl.populateResults, row filters are run.
void	testCheckLimitAfterFilterRowKey()	Tests that we check size limit after filterRowKey.
void	testSeekNextUsingHint()	Tests that when we seek over blocks we dont include them in the block size of the request
void	testSingleBlock()	Simplest test that ensures we don't count block sizes too much.

Methods inherited from class java.lang.Object

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

Field Details

CLASS_RULE

public static final HBaseClassTestRule CLASS_RULE

TEST_UTIL

private static final HBaseTestingUtil TEST_UTIL

TABLE

private static final org.apache.hadoop.hbase.TableName TABLE

FAMILY1

private static final byte[] FAMILY1

FAMILY2

private static final byte[] FAMILY2

DATA

private static final byte[] DATA

FAMILIES

private static final byte[][] FAMILIES

COLUMN1

private static final byte[] COLUMN1

COLUMN2

private static final byte[] COLUMN2

COLUMN3

private static final byte[] COLUMN3

COLUMN5

private static final byte[] COLUMN5

Constructor Details

TestScannerBlockSizeLimits

public TestScannerBlockSizeLimits()

Method Details

setUp

public static void setUp()
                  throws Exception

Throws:

Exception

setupEach

public void setupEach()
               throws Exception

Throws:

Exception

createTestData

private static void createTestData()
                            throws IOException,
InterruptedException

Throws:

IOException

InterruptedException

testSingleBlock

public void testSingleBlock()
                     throws IOException

Simplest test that ensures we don't count block sizes too much. These 2 requested cells are in the same block, so should be returned in 1 request. If we mis-counted blocks, it'd come in 2 requests.

Throws:

IOException

testCheckLimitAfterFilterRowKey

public void testCheckLimitAfterFilterRowKey()
                                     throws IOException

Tests that we check size limit after filterRowKey. When filterRowKey, we call nextRow to skip to next row. This should be efficient in this case, but we still need to check size limits after each row is processed. So in this test, we accumulate some block IO reading row 1, then skip row 2 and should return early at that point. The next rpc call starts with row3 blocks loaded, so can return the whole row in one rpc. If we were not checking size limits, we'd have been able to load an extra row 3 cell into the first rpc and thus split row 3 across multiple Results.

Throws:

IOException

testCheckLimitAfterFilteringRowCellsDueToFilterRow

public void testCheckLimitAfterFilteringRowCellsDueToFilterRow()
                                                        throws IOException

After RegionScannerImpl.populateResults, row filters are run. If row is excluded due to filter.filterRow(), nextRow() is called which might accumulate more block IO. Validates that in this case we still honor block limits.

Throws:

IOException

testCheckLimitAfterFilteringCell

public void testCheckLimitAfterFilteringCell()
                                      throws IOException

At the end of the loop in StoreScanner, we do one more check of size limits. This is to catch block size being exceeded while filtering cells within a store. Test to ensure that we do that, otherwise we'd see no cursors below.

Throws:

IOException

testCheckLimitAfterFilteringRowCells

public void testCheckLimitAfterFilteringRowCells()
                                          throws IOException

After RegionScannerImpl.populateResults, row filters are run. If row is excluded due to filter.filterRowCells(), we fall through to a final results.isEmpty() check near the end of the method. If results are empty at this point (which they are), nextRow() is called which might accumulate more block IO. Validates that in this case we still honor block limits.

Throws:

IOException

testSeekNextUsingHint

public void testSeekNextUsingHint()
                           throws IOException

Tests that when we seek over blocks we dont include them in the block size of the request

Throws:

IOException

getBaseScan

private org.apache.hadoop.hbase.client.Scan getBaseScan()

We enable cursors and partial results to give us more granularity over counting of results, and we enable STREAM so that no auto switching from pread to stream occurs -- this throws off the rpc counts.