Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
1.5.0, 1.6.0, 1.7.0
Description
We have a long running flink job that eventually fails and is shut down due to an internal serialization exception that we keep on getting. Here is the stack trace:
2018-08-23 18:39:48,199 INFO org.apache.flink.runtime.executiongraph.ExecutionGraph - Job NAV Estimation (4b5463d76167f9f5aac83a890e8a867d) switched from state FAILING to FAILED. java.io.IOException: Corrupt stream, found tag: 127 at org.apache.flink.streaming.runtime.streamrecord.StreamElementSerializer.deserialize(StreamElementSerializer.java:219) at org.apache.flink.streaming.runtime.streamrecord.StreamElementSerializer.deserialize(StreamElementSerializer.java:49) at org.apache.flink.runtime.plugable.NonReusingDeserializationDelegate.read(NonReusingDeserializationDelegate.java:55) at org.apache.flink.runtime.io.network.api.serialization.SpillingAdaptiveSpanningRecordDeserializer.getNextRecord(SpillingAdaptiveSpanningRecordDeserializer.java:140) at org.apache.flink.streaming.runtime.io.StreamTwoInputProcessor.processInput(StreamTwoInputProcessor.java:206) at org.apache.flink.streaming.runtime.tasks.TwoInputStreamTask.run(TwoInputStreamTask.java:115) at org.apache.flink.streaming.runtime.tasks.StreamTask.invoke(StreamTask.java:306) at org.apache.flink.runtime.taskmanager.Task.run(Task.java:703) at java.lang.Thread.run(Thread.java:748)
I think I have tracked down the issue to a mismatch in the serialization/deserialization/copy code in the StreamElementSerializer with regards to the LATENCY_MARKER.
The Serialization logic writes 3 LONGs and an INT. The copy logic only writes (and reads) a LONG and 2 INTs. Adding a test for the LatencyMarker throws an EOFException, and fixing the copy code causes the test to pass again.
I've written a unit test that highlights the problem, and have written the code to correct it.
I'll submit a PR that goes along with it.
Attachments
Issue Links
- links to