Liang Geng created KYLIN-4263:
---------------------------------
Summary: Inapproprate exception handling causes job stuck on
running status
Key: KYLIN-4263
URL: https://issues.apache.org/jira/browse/KYLIN-4263
Project: Kylin
Issue Type: Bug
Components: Job Engine
Affects Versions: v2.6.4
Environment: Kylin-2.6.4-release, HBase-1.1.5
Reporter: Liang Geng
Recently, I notice there are several jobs stuck on running status at step 3:
"Extract Fact Table Distinct Columns".
After deep dive the kylin.log, I found that the RetriesExhaustedException was
raised when interactive with HBase.
When the sub-task - MapReduceExecutable.doWork is invoked, the method
"isDiscarded" (org/apache/kylin/engine/mr/common/MapReduceExecutable.java:151)
will check the status of the task which will trigger RetriesExhaustedException.
{code:java}
2019-11-15 03:03:00,827 ERROR [Scheduler 1760924857 Job
758cd282-4b8a-8c07-2a37-d42186b74fba-174] dao.ExecutableDao:400 : error get job
output id:758cd282-4b8a-8c07-2a37-d42186b74fba-02
org.apache.hadoop.hbase.client.RetriesExhaustedException: Failed after
attempts=1, exceptions:
Fri Nov 15 03:03:00 GMT+08:00 2019,
RpcRetryingCaller{globalStartTime=1573758175808, pause=100, retries=1},
java.io.IOException: Call to HOSTNAME failed on local exception:
org.apache.hadoop.hbase.ipc.CallTimeoutException: Call id=143230,
waitTime=5001, operationTimeout=5000 expired.
at
org.apache.hadoop.hbase.client.RpcRetryingCaller.callWithRetries(RpcRetryingCaller.java:157)
at org.apache.hadoop.hbase.client.HTable.get(HTable.java:865)
at org.apache.hadoop.hbase.client.HTable.get(HTable.java:831)
at
org.apache.kylin.storage.hbase.HBaseResourceStore.internalGetFromHTable(HBaseResourceStore.java:472)
at
org.apache.kylin.storage.hbase.HBaseResourceStore.getFromHTable(HBaseResourceStore.java:450)
at
org.apache.kylin.storage.hbase.HBaseResourceStore.getResourceImpl(HBaseResourceStore.java:273)
at
org.apache.kylin.common.persistence.ResourceStore.lambda$getResourceWithRetry$0(ResourceStore.java:289)
at
org.apache.kylin.common.persistence.ExponentialBackoffRetry.doWithRetry(ExponentialBackoffRetry.java:52)
at
org.apache.kylin.common.persistence.ResourceStore.getResourceWithRetry(ResourceStore.java:289)
at
org.apache.kylin.common.persistence.ResourceStore.getResource(ResourceStore.java:266)
at
org.apache.kylin.common.persistence.ResourceStore.getResource(ResourceStore.java:257)
at
org.apache.kylin.job.dao.ExecutableDao.readJobOutputResource(ExecutableDao.java:248)
at org.apache.kylin.job.dao.ExecutableDao.getJobOutput(ExecutableDao.java:392)
at
org.apache.kylin.job.execution.ExecutableManager.getOutput(ExecutableManager.java:169)
at
org.apache.kylin.job.execution.AbstractExecutable.getOutput(AbstractExecutable.java:373)
at
org.apache.kylin.job.execution.AbstractExecutable.isDiscarded(AbstractExecutable.java:499)
at
org.apache.kylin.engine.mr.common.MapReduceExecutable.doWork(MapReduceExecutable.java:151)
at
org.apache.kylin.job.execution.AbstractExecutable.execute(AbstractExecutable.java:167)
at
org.apache.kylin.job.execution.DefaultChainedExecutable.doWork(DefaultChainedExecutable.java:71)
at
org.apache.kylin.job.execution.AbstractExecutable.execute(AbstractExecutable.java:167)
at
org.apache.kylin.job.impl.threadpool.DefaultScheduler$JobRunner.run(DefaultScheduler.java:114)
at
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745)
{code}
Ideally, the exception will be caught by "catch (Exception e)
(org/apache/kylin/engine/mr/common/MapReduceExecutable.java:197)". Indeed, the
exception was caught, but the "logger.error("error execute " + this.toString(),
e)org/apache/kylin/engine/mr/common/MapReduceExecutable.java:198" raises a new
exception.
Let's take a look inside the "toString" method.
{code:java}
public String toString() {
return Objects.toStringHelper(this).add("id", getId()).add("name",
getName()).add("state", getStatus()).toString();
}{code}
The method "getStatus" is invoked which will cause RuntimeException again
(cause HBase timeout again). And then, the "RuntimeException" is caught by
"catch (Exception e)
org/apache/kylin/job/execution/AbstractExecutable.java:189" and raises a new
exception - "ExecuteException". Let's OMIT one step AbstractExecutable.execute,
it's too hard to describe cause the call stack is too deep. The
"org.apache.kylin.job.execution.AbstractExecutable#execute" is invoked by
"org.apache.kylin.job.execution.DefaultChainedExecutable#doWork", but there's
no exception handling inside the "doWork" method.
{code:java}
at
org.apache.hadoop.hbase.client.RpcRetryingCaller.callWithRetries(RpcRetryingCaller.java:157)
at org.apache.hadoop.hbase.client.HTable.get(HTable.java:865)
at org.apache.hadoop.hbase.client.HTable.get(HTable.java:831)
at
org.apache.kylin.storage.hbase.HBaseResourceStore.internalGetFromHTable(HBaseResourceStore.java:472)
at
org.apache.kylin.storage.hbase.HBaseResourceStore.getFromHTable(HBaseResourceStore.java:450)
at
org.apache.kylin.storage.hbase.HBaseResourceStore.getResourceImpl(HBaseResourceStore.java:273)
at
org.apache.kylin.common.persistence.ResourceStore.lambda$getResourceWithRetry$0(ResourceStore.java:289)
at
org.apache.kylin.common.persistence.ExponentialBackoffRetry.doWithRetry(ExponentialBackoffRetry.java:52)
at
org.apache.kylin.common.persistence.ResourceStore.getResourceWithRetry(ResourceStore.java:289)
at
org.apache.kylin.common.persistence.ResourceStore.getResource(ResourceStore.java:266)
at
org.apache.kylin.common.persistence.ResourceStore.getResource(ResourceStore.java:257)
at
org.apache.kylin.job.dao.ExecutableDao.readJobOutputResource(ExecutableDao.java:248)
at org.apache.kylin.job.dao.ExecutableDao.getJobOutput(ExecutableDao.java:392)
at
org.apache.kylin.job.execution.ExecutableManager.getOutput(ExecutableManager.java:169)
at
org.apache.kylin.job.execution.AbstractExecutable.getStatus(AbstractExecutable.java:267)
at
org.apache.kylin.job.execution.AbstractExecutable.toString(AbstractExecutable.java:527)
at
org.apache.kylin.engine.mr.common.MapReduceExecutable.doWork(MapReduceExecutable.java:198)
at
org.apache.kylin.job.execution.AbstractExecutable.execute(AbstractExecutable.java:167)
at
org.apache.kylin.job.execution.DefaultChainedExecutable.doWork(DefaultChainedExecutable.java:71)
at
org.apache.kylin.job.execution.AbstractExecutable.execute(AbstractExecutable.java:167)
at
org.apache.kylin.job.impl.threadpool.DefaultScheduler$JobRunner.run(DefaultScheduler.java:114)
at
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745){code}
Finally, the "ExecuteException" is passed to
"org.apache.kylin.job.impl.threadpool.DefaultScheduler.JobRunner#run". The
"ExecuteException" is caught by "catch (ExecuteException e)
" and prints out error log.
{code:java}
2019-11-15 03:03:04,667 ERROR [pool-9-thread-10]
threadpool.DefaultScheduler:116 : ExecuteException
job:52d3840d-02c7-f5cc-cb34-43ce3015d191
org.apache.kylin.job.exception.ExecuteException: java.lang.RuntimeException:
org.apache.kylin.job.exception.PersistentException:
org.apache.hadoop.hbase.client.RetriesExhaustedException: Failed after
attempts=1, exceptions:
Fri Nov 15 03:03:04 GMT+08:00 2019,
RpcRetryingCaller{globalStartTime=1573758184665, pause=100, retries=1},
org.apache.hadoop.hbase.CallQueueTooBigException: Call queue is full on
/0:0:0:0:0:0:0:0:16020, too many items queued ?
at
org.apache.kylin.job.execution.AbstractExecutable.execute(AbstractExecutable.java:189)
at
org.apache.kylin.job.impl.threadpool.DefaultScheduler$JobRunner.run(DefaultScheduler.java:114)
at
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745)
{code}
And then the "MapReduceExecutable" sub-task is removed by
"context.removeRunningJob(executable)". However the task is removed, BUT its
status still RUNNING. In this situation, the status of the task is inconsistent.
{code:java}
2019-11-15 03:03:04,669 INFO [FetcherRunner 2113735343-72]
threadpool.DefaultFetcherRunner:85 : Job Fetcher: 3 should running, 3 actual
running, 0 stopped, 0 ready, 367 already succeed, 9 error, 5 discarded, 1
others{code}
I think the is a BUG, caused by inapproprate exception handling. For a quick
fix, we should add a try-catch statement in the toString method. Or just kill
this kind of tasks?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
