Uploaded image for project: 'Flink'
  1. Flink
  2. FLINK-20606

sql cli with hive catalog cannot create function using user classes from jar which specified by -j option




      with flink version 1.12.0(versions before also affected)

      I started a sql cli  with a hive catalog and specified a user jar file with -j option like this:

      bin/sql-client.sh embedded -j /Users/akis/Desktop/flink-func/myfunc.jar

      when i tried to create a custom function using class from myfunc.jar,cli reported ClassNotFoundException.


      Flink SQL> use catalog myhive;
      Flink SQL> create function myfunc1 as 'me.aki.flink.flinkudf.MyFunc';
      [ERROR] Could not execute SQL statement. Reason:
      java.lang.ClassNotFoundException: me.aki.flink.flinkudf.MyFunc



      me.aki.flink.flinkudf.MyFunc is the identifier of udf,which defined like this


      package me.aki.flink.flinkudf;
      import org.apache.flink.table.functions.ScalarFunction;
      public class MyFunc extends ScalarFunction {
          public String eval(String s) {
              return "myfunc_" + s;




      after walking through the related code, I believe this is a bug caused by wrong classloader


      when using a hive catalog, flink will use  CatalogFunctionImpl  to wrap the function。 The

      isGeneric() methed  uses Class.forName(String clazzName) which will use a current classloader(classloader loads flink/lib) to determine the class。


      however with -j option, user jar is set to the ExecutionContext and loaded by another userClassLoader


      and the fix can be easy to pass a classloader to the Class.forName method.

      ClassLoader cl = Thread.currentThread().getContextClassLoader();
      Class c = Class.forName(className, true, cl);

      after do such fix and build a new flink dist,create function behaves right


      Flink SQL> select myfunc1('1');
      // output









          Issue Links



              • Assignee:
                akisaya akisaya
                akisaya akisaya
              • Votes:
                0 Vote for this issue
                5 Start watching this issue


                • Created: