That TestMultiSlaveReplication doesn't pass means something pretty
fundamental was broken...

J-D

On Tue, Jul 17, 2012 at 10:08 AM, Andrew Purtell <[email protected]> wrote:
> I dispatched each unit test individually to 20 EC2 c1.mediums (64 bit
> system, 2 VCPUs, kind of slow on purpose but still allowing some
> thread concurrency). On the instance each test was run 100 times or
> until failure. For each iteration after Maven exited the process table
> was checked to see if any surefire processes lingered, and if so the
> test would also be reported failed.
>
> OS: Amazon Linux AMI release 2012.03
> uname: Linux 3.2.21-1.32.6.amzn1.x86_64 #1 SMP Sat Jun 23 02:32:15 UTC
> 2012 x86_64 x86_64 x86_64 GNU/Linux
> JVM: java version "1.6.0_24"
>     OpenJDK Runtime Environment (IcedTea6 1.11.3)
> (amazon-52.1.11.3.45.amzn1-x86_64)
>     OpenJDK 64-Bit Server VM (build 20.0-b12, mixed mode)
>
> Here are the tests that failed to complete successful runs in the above:
>
> TestCatalogTracker
>     hangs on a join in testServerNotRunningIOException waiting on a CT
> that is stuck on CatalogTracker.waitForMeta and will linger in the
> background
>
> TestColumnSeeking
>     
> testDuplicateVersions(org.apache.hadoop.hbase.regionserver.TestColumnSeeking):
> expected:<0> but was:<200>
>
> TestAtomicOperation
>     
> testMultiRowMutationMultiThreads(org.apache.hadoop.hbase.regionserver.TestAtomicOperation):
> expected:<0> but was:<1>
>
> TestSplitLogManager
>     
> testOrphanTaskAcquisition(org.apache.hadoop.hbase.master.TestSplitLogManager):
> java.lang.AssertionError
>
> TestRegionRebalancing
>     
> testRebalanceOnRegionServerNumberChange(org.apache.hadoop.hbase.TestRegionRebalancing):
> After 5 attempts, region assignments were not balanced.
>
> TestDrainingServer
>     junit.framework.AssertionFailedError from
> org.apache.hadoop.hbase.TestDrainingServer.setUpBeforeClass
>
> TestMasterObserver
>     
> testTableOperations(org.apache.hadoop.hbase.coprocessor.TestMasterObserver):
> org.apache.hadoop.hbase.InvalidFamilyOperationException: Column family
> 'fam2' does not exist
>     
> testRegionTransitionOperations(org.apache.hadoop.hbase.coprocessor.TestMasterObserver):
> org.apache.hadoop.hbase.TableExistsException: observed_table
>
> TestServerCustomProtocol
>     
> testSingleMethod(org.apache.hadoop.hbase.regionserver.TestServerCustomProtocol):
> Results should contain region
> test,bbb,1342509423473.2c0326188f899f3e91ec5eb623959c13. for row 'bbb'
>
> TestFromClientSide
>     testPoolBehavior(org.apache.hadoop.hbase.client.TestFromClientSide):
> expected:<3> but was:<4>
>
> TestZooKeeper
>     testClientSessionExpired(org.apache.hadoop.hbase.TestZooKeeper)
>
> TestReplication
>     
> testDisableInactivePeer(org.apache.hadoop.hbase.replication.TestReplication):
> Shutting down
>
> TestMasterReplication
>     
> testSimplePutDelete(org.apache.hadoop.hbase.replication.TestMasterReplication):
> Waited too much time for put replication
>
> TestMultiSlaveReplication
>     
> testMultiSlaveReplication(org.apache.hadoop.hbase.replication.TestMultiSlaveReplication):
> Unable to add peer
>
> TestReplicationPeer
>     
> testResetZooKeeperSession(org.apache.hadoop.hbase.replication.TestReplicationPeer):
> ReplicationPeer ZooKeeper session was not properly expired.
>
> I didn't get to them all before AWS yanked back my spot instances but
> I ordered the list from most likely to least, the remaining tests were
> in io.hfile.*, thrift.*, and util.*
>
> I'll circle back, confirm each individually, and open JIRAs with more detail.
>
> The cluster of replication test failures are a concern, but I've seen
> in other environments such as this one that the tests are timing
> dependent. On a slow or busy test system they can fail with "waited
> too much time ...". So a solution for this is to not use the system
> clock but instead EnvironmentEdge or whatever incremented only when
> the test process has CPU time. I haven't looked into this in detail
> yet.
>
> Best regards,
>
>    - Andy
>
> Problems worthy of attack prove their worth by hitting back. - Piet
> Hein (via Tom White)

Reply via email to