[
https://issues.apache.org/jira/browse/HBASE-16980?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yu Li updated HBASE-16980:
--------------------------
Assignee: Yu Li
Ok, here comes the analysis.
I could reproduce the failure in my local environment, but not consistently.
{{testMultipleRows}} fails more frequently than {{testReadModifyWrite}}, and
each time it fails, I could see {{RetriesExhaustedException}} caused by
{{CallQueueTooBigException}}, like below:
{noformat}
Caused by: org.apache.hadoop.hbase.client.RetriesExhaustedException: Failed
after attempts=2, exceptions:
Tue Nov 01 14:53:14 CST 2016, RpcRetryingCaller{globalStartTime=1477983194439,
pause=100, retries=2},
org.apache.hadoop.hbase.ipc.RemoteWithExtrasException(org.apache.hadoop.hbase.CallQueueTooBigException):
Call queue is full on localhost,59045,1477983189616, too many items queued ?
Tue Nov 01 14:53:14 CST 2016, RpcRetryingCaller{globalStartTime=1477983194439,
pause=100, retries=2},
org.apache.hadoop.hbase.ipc.RemoteWithExtrasException(org.apache.hadoop.hbase.CallQueueTooBigException):
Call queue is full on localhost,59045,1477983189616, too many items queued ?
at
org.apache.hadoop.hbase.client.RpcRetryingCaller.callWithRetries(RpcRetryingCaller.java:157)
at
org.apache.hadoop.hbase.ipc.RegionCoprocessorRpcChannel.callExecService(RegionCoprocessorRpcChannel.java:108)
at
org.apache.hadoop.hbase.ipc.CoprocessorRpcChannel.callBlockingMethod(CoprocessorRpcChannel.java:73)
... 6 more
Caused by:
org.apache.hadoop.hbase.ipc.RemoteWithExtrasException(org.apache.hadoop.hbase.CallQueueTooBigException):
Call queue is full on localhost,59045,1477983189616, too many items queued ?
at
org.apache.hadoop.hbase.ipc.RpcClientImpl.call(RpcClientImpl.java:1267)
at
org.apache.hadoop.hbase.ipc.AbstractRpcClient.callBlockingMethod(AbstractRpcClient.java:227)
at
org.apache.hadoop.hbase.ipc.AbstractRpcClient$BlockingRpcChannelImplementation.callBlockingMethod(AbstractRpcClient.java:336)
at
org.apache.hadoop.hbase.protobuf.generated.ClientProtos$ClientService$BlockingStub.execService(ClientProtos.java:34118)
at
org.apache.hadoop.hbase.protobuf.ProtobufUtil.execService(ProtobufUtil.java:1631)
at
org.apache.hadoop.hbase.ipc.RegionCoprocessorRpcChannel$1.call(RegionCoprocessorRpcChannel.java:104)
at
org.apache.hadoop.hbase.ipc.RegionCoprocessorRpcChannel$1.call(RegionCoprocessorRpcChannel.java:1)
at
org.apache.hadoop.hbase.client.RpcRetryingCaller.callWithRetries(RpcRetryingCaller.java:136)
... 8 more
com.google.protobuf.ServiceException: Error calling method
RowProcessorService.Process
at
org.apache.hadoop.hbase.ipc.CoprocessorRpcChannel.callBlockingMethod(CoprocessorRpcChannel.java:75)
at
org.apache.hadoop.hbase.protobuf.generated.RowProcessorProtos$RowProcessorService$BlockingStub.process(RowProcessorProtos.java:1631)
at
org.apache.hadoop.hbase.coprocessor.TestRowProcessorEndpoint.swapRows(TestRowProcessorEndpoint.java:272)
at
org.apache.hadoop.hbase.coprocessor.TestRowProcessorEndpoint.access$3(TestRowProcessorEndpoint.java:265)
at
org.apache.hadoop.hbase.coprocessor.TestRowProcessorEndpoint$SwapRowsRunner.run(TestRowProcessorEndpoint.java:258)
at
org.apache.hadoop.hbase.coprocessor.TestRowProcessorEndpoint$1.run(TestRowProcessorEndpoint.java:225)
at java.lang.Thread.run(Thread.java:745)
{noformat}
And when such exception happens, the design of the test cases could not assure
the correctness, let's see them one by one:
For {{testMultipleRows}}, it will launch 100 threads to swap two rows in
parallel, and since the thread number is even, finally the two rows will not be
swapped, but only if *all operations succeeds* or *even number of operations
failed*.
For {{testReadModifyWrite}} the similar reason, if any operation fails because
of RetriesExhaustedException, the final check of {{assertEquals(numThreads + 1,
finalCounter)}} will fail.
Currently there's a {{failures}} counter but in either {{IncrementRunner}} or
{{SwapRowsRunner}} we catch {{Throwable}} but never increase it...
To enforce the UT cases, we should
1) don't assert failures count to be zero;
2) count the failures for {{testReadModifyWrite}}
3) take {{swapped}} flag into account when assert result for
{{testMultipleRows}}
Regarding why HBASE-16195 makes the case failed more frequently, I've no much
clue... It seems to me below change is relative
{code}
- this.chunkQueue.add(c);
+ if (chunkQueue != null && !this.closed && !this.chunkQueue.offer(c)) {
+ if (LOG.isTraceEnabled()) {
+ LOG.trace("Chunk queue is full, won't reuse this new chunk.
Current queue size: "
+ + chunkQueue.size());
+ }
+ }
{code}
After HBASE-16195 it won't add the chunk into {{chunkQueue}} anymore, so it
seems the {{chunkQueue!=null}} check is more expensive than
{{this.chunkQueue.add(c)}}? Unlikely by theory though, right?...
Anyway I believe this is some UT case design issue and not that relative to
HBASE-16195 change. Will upload a patch soon to reinforce the UT.
[~apurtell] and [~busbey] please let me know your thoughts. Thanks.
Assigning the issue to myself, btw.
> TestRowProcessorEndpoint failing consistently
> ---------------------------------------------
>
> Key: HBASE-16980
> URL: https://issues.apache.org/jira/browse/HBASE-16980
> Project: HBase
> Issue Type: Bug
> Affects Versions: 1.2.4
> Reporter: Andrew Purtell
> Assignee: Yu Li
>
> Found while evaluating 1.2.4 RC1
> {noformat}
> TestRowProcessorEndpoint.testMultipleRows:246 expected:<3> but was:<2>
> TestRowProcessorEndpoint.testReadModifyWrite:184 expected:<101> but was:<91>
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
