I think any solution here is inherently fragile since future versions of
flink can have different abstract classes or interfaces you won't know it
has to support. But for a given release, something you can consider is a
wrapper class that extends/implements the ones you support. Then, during
the me
Hey all,
I've been trying to debug a job recovery performance issue and I'm noticing
some interesting events in the timeline that seem unexpected to me. Here's
a brief outline of the first checkpoint following a job restart:
1. All tasks are deployed and transition into the RUNNING state.
2. I se
It seems like the JobManager is treating this as a job failure.
A FAILED JobStatus is a globally terminal state so everything gets deleted
with zookeeper HA.
https://github.com/apache/flink/blob/master/flink-core/src/main/java/org/apache/flink/api/common/JobStatus.java#L39
https://github.com/apache
Teng Fei Liao created FLINK-17355:
-
Summary: Exactly once kafka checkpointing sensitive to single node
failures
Key: FLINK-17355
URL: https://issues.apache.org/jira/browse/FLINK-17355
Project: Flink
Teng Fei Liao created FLINK-17195:
-
Summary: Worker threads for source function have wrong ClassLoader
Key: FLINK-17195
URL: https://issues.apache.org/jira/browse/FLINK-17195
Project: Flink
