今天有同事来问一个我写过的UDF的问题,想起之前貌似写过一篇这样的文章,草稿箱里找了下,确实有,躺了一年半了,发出来,也许对某些同学有帮助~
HIVE允许用户使用UDF(user defined function)对数据进行处理。
用户可以使用‘show functions’ 查看function list,可以使用‘describe function function-name‘查看函数说明。
hive> show functions; OK ! != ...... Time taken: 0.275 seconds hive> desc function substr; OK substr(str, pos[, len]) - returns the substring of str that starts at pos and is of length len orsubstr(bin, pos[, len]) - returns the slice of byte array that starts at pos and is of length len Time taken: 0.095 seconds
hive提供的build-in函数包括以下几类:
1. 关系操作符:包括 = 、 <> 、 <= 、>=等
2. 算数操作符:包括 + 、 - 、 *、/等
3. 逻辑操作符:包括AND 、 && 、 OR 、 || 等
4. 复杂类型构造函数:包括map、struct、create_union等
5. 复杂类型操作符:包括A[n]、Map[key]、S.x
6. 数学操作符:包括ln(double a)、sqrt(double a)等
7. 集合操作符:包括size(Array<T>)、sort_array(Array<T>)等
8. 类型转换函数: binary(string|binary)、cast(expr as <type>)
9. 日期函数:包括from_unixtime(bigint unixtime[, string format])、unix_timestamp()等
10.条件函数:包括if(boolean testCondition, T valueTrue, T valueFalseOrNull)等
11. 字符串函数:包括acat(string|binary A, string|binary B...)等
12. 其他:xpath、get_json_objectscii(string str)、con
编写Hive UDF有两种方式:
1. extends UDF , 重写evaluate方法
2. extends GenericUDF,重写initialize、getDisplayString、evaluate方法
编写UDF代码实例(更多例子参考https://svn.apache.org/repos/asf/hive/tags/release-0.8.1/ql/src/java/org/apache/hadoop/hive/ql/udf/):
功能:大小转小写
ToLowerCase.java:
package test.udf; import org.apache.hadoop.hive.ql.exec.UDF; import org.apache.hadoop.io.Text; public class ToLowerCase extends UDF { public Text evaluate(final Text s) { if (s == null) { return null; } return new Text(s.toString().toLowerCase()); } }
功能:计算array中去重后元素个数
UDFArrayUniqElementNumber .java
package test.udf; import org.apache.hadoop.hive.ql.exec.Description; import org.apache.hadoop.hive.ql.exec.UDFArgumentException; import org.apache.hadoop.hive.ql.exec.UDFArgumentTypeException; import org.apache.hadoop.hive.ql.metadata.HiveException; import org.apache.hadoop.hive.ql.udf.generic.GenericUDF; import org.apache.hadoop.hive.serde2.objectinspector.ListObjectInspector; import org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector; import org.apache.hadoop.hive.serde2.objectinspector.ObjectInspectorUtils; import org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector.Category; import org.apache.hadoop.hive.serde2.objectinspector.primitive.PrimitiveObjectInspectorFactory; import org.apache.hadoop.io.IntWritable; /** * UDF: * Get nubmer of objects with duplicate elements eliminated * @author xiaomin.zhou */ @Description(name = "array_uniq_element_number", value = "_FUNC_(array) - Returns nubmer of objects with duplicate elements eliminated.", extended = "Example:\n" + " > SELECT _FUNC_(array(1, 2, 2, 3, 3)) FROM src LIMIT 1;\n" + " 3") public class UDFArrayUniqElementNumber extends GenericUDF { private static final int ARRAY_IDX = 0; private static final int ARG_COUNT = 1; // Number of arguments to this UDF private static final String FUNC_NAME = "ARRAY_UNIQ_ELEMENT_NUMBER"; // External Name private ListObjectInspector arrayOI; private ObjectInspector arrayElementOI; private final IntWritable result = new IntWritable(-1); public ObjectInspector initialize(ObjectInspector[] arguments) throws UDFArgumentException { // Check if two arguments were passed if (arguments.length != ARG_COUNT) { throw new UDFArgumentException("The function " + FUNC_NAME + " accepts " + ARG_COUNT + " arguments."); } // Check if ARRAY_IDX argument is of category LIST if (!arguments[ARRAY_IDX].getCategory().equals(Category.LIST)) { throw new UDFArgumentTypeException(ARRAY_IDX, "\"" + org.apache.hadoop.hive.serde.Constants.LIST_TYPE_NAME + "\" " + "expected at function ARRAY_CONTAINS, but " + "\"" + arguments[ARRAY_IDX].getTypeName() + "\" " + "is found"); } arrayOI = (ListObjectInspector) arguments[ARRAY_IDX]; arrayElementOI = arrayOI.getListElementObjectInspector(); return PrimitiveObjectInspectorFactory.writableIntObjectInspector; } public IntWritable evaluate(DeferredObject[] arguments) throws HiveException { result.set(0); Object array = arguments[ARRAY_IDX].get(); int arrayLength = arrayOI.getListLength(array); if (arrayLength <= 1) { result.set(arrayLength); return result; } //element compare; Algorithm complexity: O(N^2) int num = 1; int i, j; for(i = 1; i < arrayLength; i++) { Object listElement = arrayOI.getListElement(array, i); for(j = i - 1; j >= 0; j--) { if (listElement != null) { Object tmp = arrayOI.getListElement(array, j); if (ObjectInspectorUtils.compare(tmp, arrayElementOI, listElement, arrayElementOI) == 0) { break; } } } if(-1 == j) { num++; } } result.set(num); return result; } public String getDisplayString(String[] children) { assert (children.length == ARG_COUNT); return "array_uniq_element_number(" + children[ARRAY_IDX]+ ")"; } }
生成udf.jar
hive有三种方法使用自定义的UDF函数
1. 临时添加UDF
如下:
hive> select * from test; OK Hello wORLD ZXM ljz Time taken: 13.76 seconds hive> add jar /home/work/udf.jar; Added /home/work/udf.jar to class path Added resource: /home/work/udf.jar hive> create temporary function mytest as ‘test.udf.ToLowerCase‘; OK Time taken: 0.103 seconds hive> show functions; ...... mytest ...... hive> select mytest(test.name) from test; ...... OK hello world zxm ljz Time taken: 38.218 seconds这种方式在会话结束后,函数自动销毁,因此每次打开新的会话,都需要重新add jar并且create temporary function
2. 进入会话前自动创建
使用hive -i参数在进入hive时自动初始化
$ cat hive_init add jar /home/work/udf.jar; create temporary function mytest as ‘test.udf.ToLowerCase‘; $ hive -i hive_init Logging initialized using configuration in file:/home/work/hive/hive-0.8.1/conf/hive-log4j.properties Hive history file=/tmp/work/hive_job_log_work_201209200147_1951517527.txt hive> show functions; ...... mytest ...... hive> select mytest(test.name) from test; ...... OK hello world zxm ljz方法2和方法1本质上是相同的,区别在于方法2在会话初始化时自动完成
3. 自定义UDF注册为hive内置函数
可参考:hive利器 自定义UDF+重编译hive
和前两者相比,第三种方式直接将用户的自定义函数作为注册为内置函数,未来使用起来非常简单,但这种方式也非常危险,一旦出错,将是灾难性的,因此,建议如果不是特别通用,并且固化下来的函数,还是使用前两种方式比较靠谱。
Reference: