Hi, The problem is that most of the exceptions appear when my job has been running for some hours.
The only way for me to reproduce some of those errors is by using the web UI and hitting the cancel button of my job. So if I can find a way to generate this action locally in a test, maybe I can use a debugger to see where the code is invoking the ClassLoader. But in the case of the stack trace I sent you, when going into the source code of the Eventhub, it turns out that the class causing the exception is actually an anonymous class implementing the standard interface Runnable. The cluster is running on a linux distribution, so I thought at first that, as linux is flushing from time to time the tmp file, it could be the reason. But since then, I restarted the cluster with the configuration taskmanager.tmp.dirs and jobmanager.web.tmpdir set to another directory and it seems that it did not solve the issue. Best, Romain -- Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/
