0
votes

I am facing the following IO exception while running a simple wordcount application over apache flink in cluster mode.

IOException: Size of the state is larger than the maximum permitted memory-backed state. Size=34521369 , maxSize=5242880 . Consider using a different state backend, like the File System State backend.

I have already using filesystem state backend and configured the checkpoint and state backend directories.

How can I fix this issue?

1
How did you configure the state backend? - Dawid Wysakowicz
Can you share the stack trace, please? This should actually not happen in the case of the FsStateBackend. - Stephan Ewen

1 Answers

1
votes

Note that the FSStateBackend is a bit misnamed in that it actually uses the heap for state, only using the file system for checkpoints. If you have a large amount of state you should instead consider:

  • Increasing the heap allocated to the JVM
  • Using the RocksDB state backend, which is off-heap and uses storage
  • Distribute your job across additional Task Managers on other machines