[
https://issues.apache.org/jira/browse/IMPALA-10371?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17282350#comment-17282350
]
Daniel Becker commented on IMPALA-10371:
----------------------------------------
[~tarmstrong], do you have any suggestions as to how this problem could be
solved? I guess [~rizaon]'s hack could solve the current problem but maybe it's
not future proof. Do you think we could somehow get rid of the thread local
JNIEnv variable, for example by explicitly managing a thread-safe map that
stores JNIEnv instances for each thread that asks for it?
> test_java_udfs crash impalad if result spooling is enabled
> ----------------------------------------------------------
>
> Key: IMPALA-10371
> URL: https://issues.apache.org/jira/browse/IMPALA-10371
> Project: IMPALA
> Issue Type: Bug
> Components: Backend
> Affects Versions: Impala 3.4.0
> Reporter: Riza Suminto
> Assignee: Daniel Becker
> Priority: Major
> Attachments: 46a19881-resolved.txt, hs_err_pid12878.log
>
>
> The following test query from TestUdfExecution::test_java_udfs crash impalad
> when result spooling is enabled.
> {code:java}
> select throws_exception() from functional.alltypestiny{code}
> The following is a truncated JVM crash log related to the crash
> {code:java}
> --------------- T H R E A D ---------------Current thread
> (0x000000000fb4c000): JavaThread "Thread-700" [_thread_in_native, id=30853,
> stack(0x00007f79715ff000,0x00007f7971dff000)]Stack:
> [0x00007f79715ff000,0x00007f7971dff000], sp=0x00007f7971dfa280, free
> space=8172k
> Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native
> code)
> V [libjvm.so+0xb6b032]
> V [libjvm.so+0x4f14bd]
> V [libjvm.so+0x80fa8f]
> V [libjvm.so+0x7e0991]
> V [libjvm.so+0x69fa10]
> j
> org.apache.impala.TestUdfException.evaluate()Lorg/apache/hadoop/io/BooleanWritable;+9
> v ~StubRoutines::call_stub
> V [libjvm.so+0x6af9ba]
> V [libjvm.so+0xa1def8]
> V [libjvm.so+0xa1f8d5]
> V [libjvm.so+0x7610f8] JVM_InvokeMethod+0x128
> J 2286
> sun.reflect.NativeMethodAccessorImpl.invoke0(Ljava/lang/reflect/Method;Ljava/lang/Object;[Ljava/lang/Object;)Ljava/lang/Object;
> (0 bytes) @ 0x00007f7acb553ced [0x00007f7acb553c00+0xed]
> J 6921 C2
> sun.reflect.NativeMethodAccessorImpl.invoke(Ljava/lang/Object;[Ljava/lang/Object;)Ljava/lang/Object;
> (104 bytes) @ 0x00007f7acbd1de38 [0x00007f7acbd1ddc0+0x78]
> J 3645 C2 org.apache.impala.hive.executor.UdfExecutor.evaluate()V (396 bytes)
> @ 0x00007f7acaf6e894 [0x00007f7acaf6e640+0x254]
> v ~StubRoutines::call_stub
> V [libjvm.so+0x6af9ba]
> V [libjvm.so+0x72c046]
> V [libjvm.so+0x730523]
> C 0x00007f7ab4c5d0d0
> C [impalad+0x26a2648]
> impala::ScalarExprEvaluator::GetValue(impala::ScalarExpr const&,
> impala::TupleRow const*)+0x7a
> C [impalad+0x26a25cb]
> impala::ScalarExprEvaluator::GetValue(impala::TupleRow const*)+0x2b
> C [impalad+0x21f4f78]
> impala::AsciiQueryResultSet::AddRows(std::vector<impala::ScalarExprEvaluator*,
> std::allocator<impala::ScalarExprEvaluator*> > const&, impala::RowBatch*,
> int, int)+0x4c2
> C [impalad+0x25c5862]
> impala::BufferedPlanRootSink::GetNext(impala::RuntimeState*,
> impala::QueryResultSet*, int, bool*, long)+0x70c
> C [impalad+0x296cf17] impala::Coordinator::GetNext(impala::QueryResultSet*,
> int, bool*, long)+0x557
> C [impalad+0x219f5fe] impala::ClientRequestState::FetchRowsInternal(int,
> impala::QueryResultSet*, long)+0x6b2
> C [impalad+0x219d98e] impala::ClientRequestState::FetchRows(int,
> impala::QueryResultSet*, long)+0x46
> C [impalad+0x21c1d29]
> impala::ImpalaServer::FetchInternal(impala::TUniqueId, bool, int,
> beeswax::Results*)+0x717
> C [impalad+0x21bbde9] impala::ImpalaServer::fetch(beeswax::Results&,
> beeswax::QueryHandle const&, bool, int)+0x577
> {code}
> If result spooling is enabled, BufferedPlanRootSink will be used and
> ScalarExprEvaluation will be called in BufferedPlanRootSink::GetNext, leading
> to this crash.
> Without result spooling, BlockingPlanRootSink will be used and
> ScalarExprEvaluation is called in BlockingPlanRootSink::Send. No crash happen
> when result spooling is disabled.
> Attached is the full JVM crash log and resolved minidump.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org
For additional commands, e-mail: issues-all-h...@impala.apache.org
