Hi Paul, Let's add some hints about how submitting to yarn through Kyuubi in application mode differs from the flink run-application command. As far as I know, Kyuubi has made some optimizations for application mode[1].
If you agree with it, I’ll fix it. [1] https://kyuubi.readthedocs.io/en/v1.8.0/deployment/engine_on_yarn.html#yarn-application-mode Best regards, Xianxun > 2023年11月16日 16:34，Paul Lam <paullin3...@gmail.com> 写道： > > Hi Junjie, > > It could be resolved by the engine share level[1]. > > If set to connection level, each connection would have its own flink > application cluster. > > [1] https://kyuubi.readthedocs.io/en/v1.8.0/configuration/settings.html > > Best, > Paul Lam > >> 2023年11月14日 18:27，junjie.m...@goupwith.com 写道： >> >> hi,team >> kyuubi 1.8 supports the flink on yarn application mode, but does not >> generate an independent yarn application each time an sql is submitted to >> the spark thrift server. Therefore, resources between Flink SQL tasks cannot >> be isolated. >> Is there a solution? >> >> >> >> Junjie.M >
