Hi, ShengKai Thanks for pushing for this meaningful job. I have some problems:
Firstly, the outline design show the overall design in k8s. I wonder what is the scope of this FLIP, only aim for k8s, not including yarn? Secondly, > Flink offers great flexibility by permitting users to pass custom implementations(catalogs/connectors/udfs) through JAR files. When talking about "custom", you mean these also will have some builtin implementations? If it exists, how to get their location in dfs based on SQL? Depending on some configuration or just convention over configuration. Thirdly, why the FLIP-316 uses json is because "JSON plan eliminates the need to serialize the whole TableEnvironment". Based on this pr, it is no need to serialize between gateway and flink JM. Is the FLIP-316 still in need later? Best, Hongshun On Thu, Oct 24, 2024 at 4:27 PM Shengkai Fang <fskm...@gmail.com> wrote: > Hi, everyone. > > I'd like to initiate a discussion about FLIP-480: Support to deploy script > in application mode[1]. > > FLIP-480 supports to solve the problem that table program can not run in > application mode. Comparing to FLIP-316[2], FLIP-480 tries to compile the > script in the JM side, which is free from the limitation of the JSON > plan(JSON plan only serialize the identifier for temporary object) . > > For more details, please refer to the FLIP[1]. Welcome any feedback and > suggestions for improvement. > > Best, > Shengkai > > [1] > > https://cwiki.apache.org/confluence/display/FLINK/FLIP-480%3A+Support+to+deploy+SQL+script+in+application+mode > [2] > > https://cwiki.apache.org/confluence/display/FLINK/FLIP-316%3A+Support+application+mode+for+SQL+Gateway?src=contextnavpagetreemode >
