When remote partitioning a batch job you can specify a Executor like ThreadPoolTaskExecutor and thus the partitions are launched asynchronously and each partition launches a pod using the same instance of KubernetesTaskLauncher. What I noticed is that the same pod name was being re-used on some of the launch requests and thus an exception was thrown from Fabrik8 saying pod with x name already exists.
There is a PR 422 that I created that will synchronize the launch method and prevents the name from being overwritten.
When remote partitioning a batch job you can specify a Executor like ThreadPoolTaskExecutor and thus the partitions are launched asynchronously and each partition launches a pod using the same instance of KubernetesTaskLauncher. What I noticed is that the same pod name was being re-used on some of the launch requests and thus an exception was thrown from Fabrik8 saying pod with x name already exists.
There is a PR 422 that I created that will synchronize the launch method and prevents the name from being overwritten.