[IMPALA-1456] Hive udfs cause impalad to crash if they use String type arguments

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 2.1
• Fix Version/s: Impala 2.1
• Labels: ramp-up
• Assignee: casey
• Reporter: casey
• Created: 06/Nov/14 06:31
• Updated: 04/Jan/17 23:58
• Resolved: 10/Nov/14 19:23

Description

The precondition below fails if a Hive udf uses a String as an argument type.

  /**                                                                                     
   * evaluate function called by the backend. The inputs to the UDF have                  
   * been serialized to 'input'                                                           
   */                                                                                     
  public void evaluate() throws ImpalaRuntimeException {                                  
    try {                                                                                 
      for (int i = 0; i < argTypes_.length; ++i) {                                        
        if (UnsafeUtil.UNSAFE.getByte(inputNullsPtr_ + i) == 0) {                         
          if (isArgString_[i]) {                                                          
            Preconditions.checkState(inputArgs_[i] instanceof ImpalaBytesWritable);

the value in inputArgs_ that should be a string is null.

This came up while adding tests in tests/test-hive-udfs/src/main/java/com/cloudera/impala/TestUdf.java for IMPALA-1134. If something like

  public String evaluate(String a, String b, String c) {
    if (a == null || b == null || c == null) return null;
    return a + b + c;
  }

is added, impalad crashes during execution.