Zakelly Lan created FLINK-29430:
-----------------------------------
Summary: Sanity check in
InternalKeyContextImpl#setCurrentKeyGroupIndex
Key: FLINK-29430
URL: https://issues.apache.org/jira/browse/FLINK-29430
Project: Flink
Issue Type: Bug
Components: Runtime / State Backends
Affects Versions: 1.15.2
Reporter: Zakelly Lan
Fix For: 1.16.0
Currently the HeapStateBackend check whether the current key group index is a
valid one while the RocksDBStateBackend will not. When using HeapStateBackend,
if the user uses a non-deterministic shuffle key, an exception is thrown as
follows:
{code:java}
java.lang.IllegalArgumentException: Key group 84 is not in
KeyGroupRange{startKeyGroup=32, endKeyGroup=63}. Unless you're directly using
low level state access APIs, this is most likely caused by non-deterministic
shuffle key (hashCode and equals implementation).
at
org.apache.flink.runtime.state.KeyGroupRangeOffsets.newIllegalKeyGroupException(KeyGroupRangeOffsets.java:37)
at
org.apache.flink.runtime.state.heap.StateTable.getMapForKeyGroup(StateTable.java:305)
at org.apache.flink.runtime.state.heap.StateTable.get(StateTable.java:261)
at org.apache.flink.runtime.state.heap.StateTable.get(StateTable.java:143)
at
org.apache.flink.runtime.state.heap.HeapValueState.value(HeapValueState.java:72)
at
com.alibaba.ververica.flink.state.benchmark.wordcount.WordCount$MixedFlatMapper.flatMap(WordCount.java:169)
at
com.alibaba.ververica.flink.state.benchmark.wordcount.WordCount$MixedFlatMapper.flatMap(WordCount.java:160)
at
org.apache.flink.streaming.api.operators.StreamFlatMap.processElement(StreamFlatMap.java:47)
at
org.apache.flink.streaming.runtime.tasks.OneInputStreamTask$StreamTaskNetworkOutput.emitRecord(OneInputStreamTask.java:233)
at
org.apache.flink.streaming.runtime.io.AbstractStreamTaskNetworkInput.processElement(AbstractStreamTaskNetworkInput.java:135)
at
org.apache.flink.streaming.runtime.io.AbstractStreamTaskNetworkInput.emitNext(AbstractStreamTaskNetworkInput.java:106)
at
org.apache.flink.streaming.runtime.io.StreamOneInputProcessor.processInput(StreamOneInputProcessor.java:65)
at
org.apache.flink.streaming.runtime.tasks.StreamTask.processInput(StreamTask.java:526)
at
org.apache.flink.streaming.runtime.tasks.mailbox.MailboxProcessor.runMailboxLoop(MailboxProcessor.java:203)
at
org.apache.flink.streaming.runtime.tasks.StreamTask.runMailboxLoop(StreamTask.java:811)
at
org.apache.flink.streaming.runtime.tasks.StreamTask.invoke(StreamTask.java:760)
at
org.apache.flink.runtime.taskmanager.Task.runWithSystemExitMonitoring(Task.java:954)
at org.apache.flink.runtime.taskmanager.Task.restoreAndInvoke(Task.java:933)
at org.apache.flink.runtime.taskmanager.Task.doRun(Task.java:746)
at org.apache.flink.runtime.taskmanager.Task.run(Task.java:568)
at java.lang.Thread.run(Thread.java:750)
{code}
However, the RocksDBStateBackend will run without an exception. The wrong key
group index will cause a state correctness problem, so it is better to do a
check in {_}InternalKeyContextImpl#{_}{_}setCurrentKeyGroupIndex{_}, and throw
an exception immediately.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
