[IMPALA-11342] Hive UDFs are unable to use classes from the same Jar in catalogd

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 4.1.0
• Fix Version/s: Impala 4.2.0, Impala 4.1.1
• Component/s: Catalog
• Epic Color: ghx-label-14
• Assignee: Csaba Ringhofer
• Reporter: Csaba Ringhofer
• Created: 08/Jun/22 13:25
• Updated: 05/Dec/22 09:40
• Resolved: 01/Sep/22 12:57

Description

While it is a known limitations that Hive UDFs in Impala cannot load dependencies from classpath, they should be able to load classes from the same jar (and thus creating an shaded jar with the dependencies should be a valid workaround). Meanwhile I got an error when the parent class of the UDF was a class defined in another package in the jar.

The exception occurred when catalogd tried to instantiate the UDF class:

I0608 14:50:21.520296 22464 jni-util.cc:286] java.lang.NoClassDefFoundError: com/esri/core/geometry/Point
        at java.lang.Class.getDeclaredConstructors0(Native Method)
        at java.lang.Class.privateGetDeclaredConstructors(Class.java:2671)
        at java.lang.Class.getConstructor0(Class.java:3075)
        at java.lang.Class.getConstructor(Class.java:1825)
        at org.apache.impala.hive.executor.HiveLegacyJavaFunction.instantiateUDFInstance(HiveLegacyJavaFunction.java:115)
        at org.apache.impala.hive.executor.HiveLegacyJavaFunction.<init>(HiveLegacyJavaFunction.java:74)
        at org.apache.impala.hive.executor.HiveJavaFunctionFactoryImpl.create(HiveJavaFunctionFactoryImpl.java:50)
        at org.apache.impala.hive.executor.HiveJavaFunctionFactoryImpl.create(HiveJavaFunctionFactoryImpl.java:67)
        at org.apache.impala.service.CatalogOpExecutor.createFunction(CatalogOpExecutor.java:2117)
        at org.apache.impala.service.CatalogOpExecutor.execDdlRequest(CatalogOpExecutor.java:453)
        at org.apache.impala.service.JniCatalog.execDdl(JniCatalog.java:245)
Caused by: java.lang.ClassNotFoundException: com.esri.core.geometry.Point
        at java.net.URLClassLoader.findClass(URLClassLoader.java:381)
        at java.lang.ClassLoader.loadClass(ClassLoader.java:424)
        at java.lang.ClassLoader.loadClass(ClassLoader.java:357)
        ... 11 more

The issue seems to be cause by closing the UrlClassLoader too early, after loading the class but before doing the actual instantiation:
https://github.com/apache/impala/blob/13bbff4e4e5fc5d459cc6f7a5512f84ceba897cd/fe/src/main/java/org/apache/impala/hive/executor/HiveUdfLoader.java#L71

Note that this is an issue only in the catalogd, not the executors, as the class loader is kept open longer in that case.

Issue Links

• is caused by: IMPALA-10997 Refactor code for UDF implementation
• is related to: IMPALA-11774 CatalogD failed to start due to loading UDF failure
• relates to: IMPALA-8016 UDFs unable to use other functions defined in same jar