Hi DashShen, Thanks for trying the native K8s integration.
Just like you say, currently K8s application mode could only support local jar. That means you need to bundle your user jar in the image. I know it is not very convenient to use since users have to build their own image. This is already in plan and could be resolved by some new features(e.g. init container, pod template). You could have your own init container to download the user jar. Once the JobManager launched, it already exists locally. Since you may download your jars from http server, distribute filesystem(HDFS), cloud storage(e.g. S3, Aliyun OSS, etc.), we choose user-defined init container to support the downloading instead of doing it in Flink internally. It is more flexible. > Why Yarn could support remote user jar? For Yarn application mode, we leverage the Yarn local resource feature to make the Yarn NodeManger help us to downloading the user jar from HDFS. So we do not have any additional cost to make this done. Best, Yang DashShen <382778...@qq.com> 于2020年7月8日周三 下午5:12写道: > Hi, > When I use ApplicationMode on k8s to submit a new flink application, it's > not capable for me to using remote jar like what yarn does and I have to > pack the jar into the image before. So I want know why the application mode > for k8s doesn't support to reading remote jar. > > > > > -- > Sent from: http://apache-flink-mailing-list-archive.1008284.n3.nabble.com/ >