Currently, the Flink job client runs as a Kubernetes job, which uses the same container as the Flink JobManager and TaskManager. There is a benefit of separate the two. Some functionalities such as downloading remote Job jar can be supported in the job client image, while using the official Flink image for JM/TM.
We could create a dedicated Docker image for the job client, we could also use the operator image (by adding the capability of job submission) for the job client. The latter approach seems to be easier to manage for end users.
Currently, the Flink job client runs as a Kubernetes job, which uses the same container as the Flink JobManager and TaskManager. There is a benefit of separate the two. Some functionalities such as downloading remote Job jar can be supported in the job client image, while using the official Flink image for JM/TM.
We could create a dedicated Docker image for the job client, we could also use the operator image (by adding the capability of job submission) for the job client. The latter approach seems to be easier to manage for end users.