Hi vtygoss, Have you taken a look at Zeppelin ? Zeppelin support Flink UDF in several different approaches, and as you said, it is not a good practice to put udf jars under $FLINK_HOME/lib. In Zeppelin you don't do that, you can dynamically load udf jars. Check the following doc for more details.
https://zeppelin.apache.org/docs/0.10.0/interpreter/flink.html#flink-udf vtygoss <vtyg...@126.com> 于2021年10月16日周六 下午4:54写道： > Hi, community! > > > I am working on building a stream processing platform using Flink 1.12.0. > I met a problem in the scenario of SQL Application migration from > SparkSQL/HiveSQL to FlinkSQL. > > > How to dynamically modify the classloader of the JobManager already > launched to handle "ADD JAR HDFS://..." statement in sql text? > > > I already found the way to modify the classloader of all TaskManager > dynamically through "Catalog#getHiveConf#setAuxJar" or > "pipeline.classpath"; But i can't find the way to modify the classloader of > JobManager dynamically, then the Application will fail because of UDF > ClassNotFoundException. And i don't want put the udf-jar into > $FLINK_HOME/lib, that's too heavy. > > > Thanks for your any suggestions or replies! > > > Best Regards! > > > > -- Best Regards Jeff Zhang
