You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 14 Next »

 

DUnit tests are JUnit tests that extend CacheTestCase or DistributedTestCase and run code in more than one VM. Because these tests tend to involve multiple members in a distributed system, these tests are more prone to potential race conditions. Here are some tips for tracking down failures and fixing these tests.

Running a single junit test

 

./gradlew --no-daemon geode-core:test -Dtest.single=BucketRegionJUnitTest

 

 

Running a single dunit test

You can use the standard Gradle properties to run a single test. Or, better yet, run the test in your IDE to help you debug it.

 

./gradlew geode-core:distributedTest -DdistributedTest.single=PartitionedRegionTestUtilsDUnitTest

 

To specify a logLevl:

./gradlew geode-core:distributedTest -DlogLevel=debug -DdistributedTest.single=PartitionedRegionTestUtilsDUnitTest

 

Running a test multiple times

One way to run a test a lot of times is to add a new method to your case test that runs a problematic method many times. Here's an example

 

  public void testLoop() throws Exception {
    for(int i=0; i < 200; i++) {
      testGetBucketOwners();
      tearDown();
      setUp();
    }
  }

 

Running a list of tests

 

To save time, the JVMs are reused between test cases. This can cause issues if previous tests leave the JVM in a bad state. You can see the list of previously executed tests as the first step in the stdout of a test, if you are looking at a failure from Jenkins:

 

Previously run tests: [PartitionedRegionBucketCreationDistributionDUnitTest, PartitionedRegionTestUtilsDUnitTest, DistTXDebugDUnitTest]

 

To run the same tests in order, you can add a suite to your JUnit test

 


public static Test suite() {
  final TestSuite suite = new TestSuite();
  suite.addTestSuite(PartitionedRegionBucketCreationDistributionDUnitTest.class);
  suite.addTestSuite(PartitionedRegionTestUtilsDUnitTest.class);
  //Use the JUnit4TestAdapter one of your test has been converted to Junit4 syntax.
  suite.addTest(new JUnit4TestAdapter(DistTXDebugDUnitTest.class));
  return suite;
}

 

 

A complete example: Note, it mixed dunit tests and junit tests.
 
package com.gemstone.gemfire.internal.cache.wan;
import com.gemstone.gemfire.internal.cache.DistributedRegionJUnitTest;
import com.gemstone.gemfire.internal.cache.PartitionedRegionRedundancyZoneDUnitTest;
import com.gemstone.gemfire.internal.cache.PartitionedRegionTestUtilsDUnitTest;
import com.gemstone.gemfire.internal.cache.wan.misc.ShutdownAllPersistentGatewaySenderDUnitTest;
import junit.framework.Test;
import junit.framework.TestCase;
import junit.framework.TestSuite;
@Category(UnitTest.class)
public class MyTestCase extends TestCase {
 
  public static Test suite() {
    Class[] classes = new Class[] {
        CacheClientNotifierDUnitTest.class, 
        ShutdownAllPersistentGatewaySenderDUnitTest.class, 
        PartitionedRegionRedundancyZoneDUnitTest.class, 
        PartitionedRegionTestUtilsDUnitTest.class,
        DistributedRegionJUnitTest.class
        };
    return new TestSuite(classes);
  }
}
 
Or a new version:
@RunWith(Suite.class)
@Suite.SuiteClasses({ CacheClientNotifierDUnitTest.class, DistributedRegionJUnitTest.class })
//@Category(UnitTest.class)
public class MyTestCase2 extends TestCase {
}
 
 
 

Fixing suspect strings

Distributed tests check the log output from each test for "suspect strings" which include any warnings, errors, or exceptions. If your test is supposed to log one of these strings, you can add an expected exception to the beginning of your test method, or even in your setUp method if all test cases are expected to log this message. There is no need to remove the expected exception, it is automatically removed in tearDown().

 

  public void testClientPutWithInterrupt() throws Throwable {
    addExpectedException("InterruptedException");
   //...
  }

 

Turn on debug level in dunit tests

There are many ways to run a test case with a specific debug level; the easiest way is to temporarily add the following code to the test program.

 

  @Override
  public Properties getDistributedSystemProperties() {
    Properties props = new Properties();
    props.setProperty("log-level", "debug");
    return props;
  }

 

Turn on trace level using log4j2.xml

Product code contains some trace code similar to the following: 

 

      if (logger.isTraceEnabled(LogMarker.TOMBSTONE)) {
        logger.trace(LogMarker.TOMBSTONE, "Destroyed entries sweeper starting with default sleep interval={}", this.expiryTime);
      }

This will not be displayed in a debug level log. To specify a customized log4j2.xml, there are also several ways, but the easiest one is to copy your log4j2.xml into open/gemfire-core/src/main/resources/log4j2.xml. Here is an example of log4j2.xml that turns on LogMarker.TOMBSTONE, LogMarker.DISTRIBUTION, LogMarker.DISK_STORE_MONITOR:

 

<?xml version="1.0" encoding="UTF-8"?>
<Configuration status="ERROR" shutdownHook="disable" packages="com.gemstone.gemfire.internal.logging.log4j">
  <Properties>
    <Property name="gemfire-pattern">[%level{lowerCase=true} %date{yyyy/MM/dd HH:mm:ss.SSS z} &lt;%thread&gt; tid=%tid] %message%n%throwable%n</Property>
  </Properties>
  <filters>
    <MarkerFilter marker="DISTRIBUTION" onMatch="ACCEPT" onMismatch="NEUTRAL"/>
    <MarkerFilter marker="DISK_STORE_MONITOR" onMatch="ACCEPT" onMismatch="NEUTRAL"/>
    <MarkerFilter marker="TOMBSTONE" onMatch="ACCEPT" onMismatch="NEUTRAL"/>
    <!--MarkerFilter marker="PERSIST_RECOVERY" onMatch="ACCEPT" onMismatch="NEUTRAL"/-->
    <!--MarkerFilter marker="PERSIST_WRITES" onMatch="ACCEPT" onMismatch="NEUTRAL"/-->
    <!-- have to explicitly DENY all the markers' log, then explicitly add back one by one -->
    <MarkerFilter marker="GEMFIRE_MARKER" onMatch="DENY" onMismatch="NEUTRAL"/>
  </filters>
  <Appenders>
    <Console name="STDOUT" target="SYSTEM_OUT">
      <PatternLayout pattern="${gemfire-pattern}"/>
    </Console>
    <Console name="STDERR" target="SYSTEM_ERR">
      <PatternLayout pattern="${gemfire-pattern}"/>
    </Console>
    <!--RollingFile name="RollingFile" fileName="${filename}"-->
    <File name="Log" fileName="system.log" bufferedIO="true">
      <PatternLayout pattern="${gemfire-pattern}"/>
    </File>
  </Appenders>
  <Loggers>
    <Logger name="com.gemstone" level="INFO" additivity="false">
      <AppenderRef ref="Log"/>
    </Logger>
    <Logger name="com.gemstone.gemfire.distributed.internal" level="TRACE" additivity="false">
      <AppenderRef ref="STDOUT" level="DEBUG"/>
    </Logger>
    <Logger name="com.gemstone.gemfire.cache.client.internal" level="TRACE" additivity="false">
      <AppenderRef ref="STDOUT" level="DEBUG"/>
    </Logger>
    <Logger name="com.gemstone.gemfire.internal.cache" level="TRACE" additivity="false">
      <AppenderRef ref="STDOUT" level="TRACE"/>
    </Logger>
    <!--Logger name="com.gemstone.gemfire.distributed.internal.DistributionManager" level="TRACE" additivity="false">
      <AppenderRef ref="Log" level="DEBUG"/>
    </Logger-->
    <!--Logger name="com.gemstone.gemfire.distributed.internal.DistributionMessage" level="TRACE" additivity="false">
      <AppenderRef ref="Log" level="DEBUG"/>
    </Logger-->
    <Root level="ERROR">
      <!--AppenderRef ref="STDOUT"/-->
    </Root>
  </Loggers>
</Configuration>

 

Specify bigger heap size for dunit tests

The default heap size for a DUnit test is 768m. We can modify open/build.gradle to use a larger heap size in case the DUnit test runs out of memory. 

 

        maxHeapSize '768m'
        jvmArgs = ['-XX:+HeapDumpOnOutOfMemoryError', '-ea']
        systemProperties = [
          'gemfire.DEFAULT_MAX_OPLOG_SIZE' : '10',
          'gemfire.disallowMcastDefaults'  : 'true',
          'jline.terminal'                 : 'jline.UnsupportedTerminal',
        ]

 

Use new await() to wait for an async event to finish

We've introduced com.jayway.awaitility.Awaitility.await in Geode. There's a better way to replace our old WaitCriterion. await() can also be used in product code, while WaitCriterion is only in DistributedTestCase. 

 
 

              Cache cache = getCache();
              DistributedTestCase.disconnectFromDS();
              await().atMost(30, SECONDS).until(() -> {return (cache == null || cache.isClosed());});

is equivalent to:               
              WaitCriterion waitForCacheClose = new WaitCriterion() {
                @Override
                public boolean done() {
                  return cache == null || cache.isClosed();
                }

                @Override
                public String description() {
                  return "Wait for Cache to be closed";
                }
              };
              
              DistributedTestCase.waitForCriterion(waitForCacheClose, 30000, 500, true);

Using Mockito to test region class

Our region classes (LocalRegion, DistributedRegion, PartitionedRegion, BucketRegion, etc) and some other legacy gemfire classes heaily rely on cache, distribute system, distribution manager and other modules to function. Thus it's difficult to mock the class and do the junit test.

However, since it's a common issue for everyone, we introduced a template and 2 sample junit test classes to show how to write junit tests for region class.

The fix is GEODE-774: changed virtualPut(), basicDestroy(), basicInvalidate(), basicUpdateEntryVersion(), not to distribute the event if hasSeen()==true and the event does not contain a valid version tag. 

Let's see how we do junit test for above fix. 

The test logic is simple:

  • create a region
  • create an event with version tag
  • call the methods of the region
  • verify the distribution is triggered or not

 

The challenge is how to mock the required components to create region, event, version tag. 

The com.gemstone.gemfire.test.fake.Fakes class has done the major mocking for us and also defined the expected behaviors between these mock objects. You only need to use it: 

 

GemFireCacheImpl cache = Fakes.cache();

 

In DistributedRegionJUnitTest, it gives an example on how to create a DistributedRegion using the mocking. 

  protected DistributedRegion prepare(boolean isConcurrencyChecksEnabled) {
    GemFireCacheImpl cache = Fakes.cache(); // using the Fakes to define mock objects and their behavior
    
    // create region attributes and internal region arguments
    RegionAttributes ra = createRegionAttributes(isConcurrencyChecksEnabled);
    InternalRegionArguments ira = new InternalRegionArguments();
    
    doPRSpecific(ira); // for BucketRegion, we need to config some PR attributes
    
    // create a region object
    DistributedRegion region = createAndDefineRegion(isConcurrencyChecksEnabled, ra, ira, cache);
    if (isConcurrencyChecksEnabled) {
      region.enableConcurrencyChecks();
    }
    
    // some additional behaviors to be define for the test
    doNothing().when(region).notifyGatewaySender(any(), any());
    doReturn(true).when(region).hasSeenEvent(any(EntryEventImpl.class));

    return region;
  }

 

The region created here is not a mock object, but a real region. However, we need to monitor its behavior in mocking. 

 

    DistributedRegion region = new DistributedRegion("testRegion", ra, null, cache, ira);
    if (isConcurrencyChecksEnabled) {
      region.enableConcurrencyChecks();
    }
    
    // since it is a real region object, we need to tell mockito to monitor it
    region = Mockito.spy(region);

    doNothing().when(region).distributeUpdate(any(), anyLong(), anyBoolean(), anyBoolean(), any(), anyBoolean());
    doNothing().when(region).distributeDestroy(any(), any());
    doNothing().when(region).distributeInvalidate(any());
    doNothing().when(region).distributeUpdateEntryVersion(any());

 

BucketRegionJUnitTest is a subclass of DistributedRegionJUnitTest to reuse most of the logic. However it needs to implement the doPRSpecific() for more mocking. 

  protected void doPRSpecific(InternalRegionArguments ira) {
    // PR specific
    PartitionedRegion pr = mock(PartitionedRegion.class);
    BucketAdvisor ba = mock(BucketAdvisor.class);
    ReadWriteLock primaryMoveLock = new ReentrantReadWriteLock();
    Lock activeWriteLock = primaryMoveLock.readLock();
    when(ba.getActiveWriteLock()).thenReturn(activeWriteLock);
    when(ba.isPrimary()).thenReturn(true);
    
    ira.setPartitionedRegion(pr)
      .setPartitionedRegionBucketRedundancy(1)
      .setBucketAdvisor(ba);
  }

 

Overall, the Fakes.java, DistributedRegionJUnitTest, BucketRegionJUnitTest  are a good start to build a shareable template to write mock-based test cases for gemfire legacy components. 

 

Summary of a junit test for BucketRegion:

    GemFireCacheImpl cache = Fakes.cache();
 
    RegionAttributes ra = createRegionAttributes(isConcurrencyChecksEnabled);
    InternalRegionArguments ira = new InternalRegionArguments();
 
    // specify more mock behaviors for ParitionedRegion and BucketRegion
    PartitionedRegion pr = mock(PartitionedRegion.class);
    BucketAdvisor ba = mock(BucketAdvisor.class);
    ReadWriteLock primaryMoveLock = new ReentrantReadWriteLock();
    Lock activeWriteLock = primaryMoveLock.readLock();
    when(ba.getActiveWriteLock()).thenReturn(activeWriteLock);
    when(ba.isPrimary()).thenReturn(true);
    ira.setPartitionedRegion(pr)
      .setPartitionedRegionBucketRedundancy(1)
      .setBucketAdvisor(ba);
 
    // create Bucket Region
    BucketRegion br = new BucketRegion("testRegion", ra, null, cache, ira);
    // since br is a real bucket region object, we need to tell mockito to monitor it
    br = Mockito.spy(br);
 
    doNothing().when(br).distributeUpdateOperation(any(), anyLong());
    doNothing().when(br).distributeDestroyOperation(any());
    doNothing().when(br).distributeInvalidateOperation(any());
    doNothing().when(br).distributeUpdateEntryVersionOperation(any());
    doNothing().when(br).checkForPrimary();
    doNothing().when(br).handleWANEvent(any());
    doReturn(false).when(br).needWriteLock(any());
    
    if (isConcurrencyChecksEnabled) {
      region.enableConcurrencyChecks();
    }
    
    doNothing().when(region).notifyGatewaySender(any(), any());
    doReturn(true).when(region).hasSeenEvent(any(EntryEventImpl.class));
 
    EntryEventImpl event = createDummyEvent(region);
    VersionTag tag = createVersionTag(false);
    event.setVersionTag(tag);
 
    br.virtualPut(event, false, false, null, false, 12345L, false);
    // verify the result
    if (cnt > 0) {
      verify(br, times(cnt)).distributeUpdateOperation(eq(event), eq(12345L));
    } else {
      verify(br, never()).distributeUpdateOperation(eq(event), eq(12345L));
    }


Running a single hydra test, continously if necessary

https://sites.google.com/a/pivotal.io/gem-engineering/process/issue-tracking/run-a-single-hydra-test


  • No labels