<?xml version="1.0" encoding="UTF-8"?>
<!--
Licensed to the Apache Software Foundation (ASF) under one
or more contributor license agreements.  See the NOTICE file
distributed with this work for additional information
regarding copyright ownership.  The ASF licenses this file
to you under the Apache License, Version 2.0 (the
"License"); you may not use this file except in compliance
with the License.  You may obtain a copy of the License at

  http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing,
software distributed under the License is distributed on an
"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
KIND, either express or implied.  See the License for the
specific language governing permissions and limitations
under the License.
-->
<!DOCTYPE concept PUBLIC "-//OASIS//DTD DITA Concept//EN" "concept.dtd">
<concept rev="1.2" id="create_function">

  <title>CREATE FUNCTION Statement</title>
  <titlealts audience="PDF"><navtitle>CREATE FUNCTION</navtitle></titlealts>
  <prolog>
    <metadata>
      <data name="Category" value="Impala"/>
      <data name="Category" value="SQL"/>
      <data name="Category" value="DDL"/>
      <data name="Category" value="Schemas"/>
      <data name="Category" value="Impala Functions"/>
      <data name="Category" value="UDFs"/>
      <data name="Category" value="Developers"/>
      <data name="Category" value="Data Analysts"/>
    </metadata>
  </prolog>

  <conbody>

    <p>
      <indexterm audience="hidden">CREATE FUNCTION statement</indexterm>
      Creates a user-defined function (UDF), which you can use to implement custom logic during
      <codeph>SELECT</codeph> or <codeph>INSERT</codeph> operations.
    </p>

    <p conref="../shared/impala_common.xml#common/syntax_blurb"/>

    <p>
      The syntax is different depending on whether you create a scalar UDF, which is called once for each row and
      implemented by a single function, or a user-defined aggregate function (UDA), which is implemented by
      multiple functions that compute intermediate results across sets of rows.
    </p>

    <p rev="2.5.0 IMPALA-2843">
      In <keyword keyref="impala25_full"/> and higher, the syntax is also different for creating or dropping scalar Java-based UDFs.
      The statements for Java UDFs use a new syntax, without any argument types or return type specified. Java-based UDFs
      created using the new syntax persist across restarts of the Impala catalog server, and can be shared transparently
      between Impala and Hive.
    </p>

    <p>
      To create a persistent scalar C++ UDF with <codeph>CREATE FUNCTION</codeph>:
    </p>

<codeblock>CREATE FUNCTION [IF NOT EXISTS] [<varname>db_name</varname>.]<varname>function_name</varname>([<varname>arg_type</varname>[, <varname>arg_type</varname>...])
  RETURNS <varname>return_type</varname>
  LOCATION '<varname>hdfs_path_to_dot_so</varname>'
  SYMBOL='<varname>symbol_name</varname>'</codeblock>

    <p rev="2.5.0 IMPALA-2843">
      To create a persistent Java UDF with <codeph>CREATE FUNCTION</codeph>:
<codeblock>CREATE FUNCTION [IF NOT EXISTS] [<varname>db_name</varname>.]<varname>function_name</varname>
  LOCATION '<varname>hdfs_path_to_jar</varname>'
  SYMBOL='<varname>class_name</varname>'</codeblock>
    </p>

<!--
Examples:
CREATE FUNCTION IF NOT EXISTS foo location '/path/to/jar' SYMBOL='TestUdf';
CREATE FUNCTION bar location '/path/to/jar' SYMBOL='TestUdf2';
DROP FUNCTION foo;
DROP FUNCTION IF EXISTS bar;
-->

    <p>
      To create a persistent UDA, which must be written in C++, issue a <codeph>CREATE AGGREGATE FUNCTION</codeph> statement:
    </p>

<codeblock>CREATE [AGGREGATE] FUNCTION [IF NOT EXISTS] [<varname>db_name</varname>.]<varname>function_name</varname>([<varname>arg_type</varname>[, <varname>arg_type</varname>...])
  RETURNS <varname>return_type</varname>
  LOCATION '<varname>hdfs_path</varname>'
  [INIT_FN='<varname>function</varname>]
  UPDATE_FN='<varname>function</varname>
  MERGE_FN='<varname>function</varname>
  [PREPARE_FN='<varname>function</varname>]
  [CLOSEFN='<varname>function</varname>]
  <ph rev="2.0.0">[SERIALIZE_FN='<varname>function</varname>]</ph>
  [FINALIZE_FN='<varname>function</varname>]
  <ph rev="2.3.0 IMPALA-1829">[INTERMEDIATE <varname>type_spec</varname>]</ph></codeblock>

    <p conref="../shared/impala_common.xml#common/ddl_blurb"/>

    <p>
      <b>Varargs notation:</b>
    </p>

    <note rev="">
      <p rev="">
        Variable-length argument lists are supported for C++ UDFs, but currently not for Java UDFs.
      </p>
    </note>

    <p>
      If the underlying implementation of your function accepts a variable number of arguments:
    </p>

    <ul>
      <li>
        The variable arguments must go last in the argument list.
      </li>

      <li>
        The variable arguments must all be of the same type.
      </li>

      <li>
        You must include at least one instance of the variable arguments in every function call invoked from SQL.
      </li>

      <li>
        You designate the variable portion of the argument list in the <codeph>CREATE FUNCTION</codeph> statement
        by including <codeph>...</codeph> immediately after the type name of the first variable argument. For
        example, to create a function that accepts an <codeph>INT</codeph> argument, followed by a
        <codeph>BOOLEAN</codeph>, followed by one or more <codeph>STRING</codeph> arguments, your <codeph>CREATE
        FUNCTION</codeph> statement would look like:
<codeblock>CREATE FUNCTION <varname>func_name</varname> (INT, BOOLEAN, STRING ...)
  RETURNS <varname>type</varname> LOCATION '<varname>path</varname>' SYMBOL='<varname>entry_point</varname>';
</codeblock>
      </li>
    </ul>

    <p rev="">
      See <xref href="impala_udf.xml#udf_varargs"/> for how to code a C++ UDF to accept
      variable-length argument lists.
    </p>

    <p>
      <b>Scalar and aggregate functions:</b>
    </p>

    <p>
      The simplest kind of user-defined function returns a single scalar value each time it is called, typically
      once for each row in the result set. This general kind of function is what is usually meant by UDF.
      User-defined aggregate functions (UDAs) are a specialized kind of UDF that produce a single value based on
      the contents of multiple rows. You usually use UDAs in combination with a <codeph>GROUP BY</codeph> clause to
      condense a large result set into a smaller one, or even a single row summarizing column values across an
      entire table.
    </p>

    <p>
      You create UDAs by using the <codeph>CREATE AGGREGATE FUNCTION</codeph> syntax. The clauses
      <codeph>INIT_FN</codeph>, <codeph>UPDATE_FN</codeph>, <codeph>MERGE_FN</codeph>,
      <ph rev="2.0.0"><codeph>SERIALIZE_FN</codeph>,</ph> <codeph>FINALIZE_FN</codeph>, and
      <codeph>INTERMEDIATE</codeph> only apply when you create a UDA rather than a scalar UDF.
    </p>

    <p>
      The <codeph>*_FN</codeph> clauses specify functions to call at different phases of function processing.
    </p>

    <ul>
      <li>
        <b>Initialize:</b> The function you specify with the <codeph>INIT_FN</codeph> clause does any initial
        setup, such as initializing member variables in internal data structures. This function is often a stub for
        simple UDAs. You can omit this clause and a default (no-op) function will be used.
      </li>

      <li>
        <b>Update:</b> The function you specify with the <codeph>UPDATE_FN</codeph> clause is called once for each
        row in the original result set, that is, before any <codeph>GROUP BY</codeph> clause is applied. A separate
        instance of the function is called for each different value returned by the <codeph>GROUP BY</codeph>
        clause. The final argument passed to this function is a pointer, to which you write an updated value based
        on its original value and the value of the first argument.
      </li>

      <li>
        <b>Merge:</b> The function you specify with the <codeph>MERGE_FN</codeph> clause is called an arbitrary
        number of times, to combine intermediate values produced by different nodes or different threads as Impala
        reads and processes data files in parallel. The final argument passed to this function is a pointer, to
        which you write an updated value based on its original value and the value of the first argument.
      </li>

      <li rev="2.0.0">
        <b>Serialize:</b> The function you specify with the <codeph>SERIALIZE_FN</codeph> clause frees memory
        allocated to intermediate results. It is required if any memory was allocated by the Allocate function in
        the Init, Update, or Merge functions, or if the intermediate type contains any pointers. See
        <xref keyref="uda-sample.cc">the UDA code samples</xref> for details.
      </li>

      <li>
        <b>Finalize:</b> The function you specify with the <codeph>FINALIZE_FN</codeph> clause does any required
        teardown for resources acquired by your UDF, such as freeing memory, closing file handles if you explicitly
        opened any files, and so on. This function is often a stub for simple UDAs. You can omit this clause and a
        default (no-op) function will be used. It is required in UDAs where the final return type is different than
        the intermediate type. or if any memory was allocated by the Allocate function in the Init, Update, or
        Merge functions. See <xref keyref="uda-sample.cc">the UDA code samples</xref> for details.
      </li>
    </ul>

    <p>
      If you use a consistent naming convention for each of the underlying functions, Impala can automatically
      determine the names based on the first such clause, so the others are optional.
    </p>

    <p audience="hidden">
      The <codeph>INTERMEDIATE</codeph> clause specifies the data type of intermediate values passed from the
      <q>update</q> phase to the <q>merge</q> phase, and from the <q>merge</q> phase to the <q>finalize</q> phase.
      You can use any of the existing Impala data types, or the special notation
      <codeph>CHAR(<varname>n</varname>)</codeph> to allocate a scratch area of <varname>n</varname> bytes for the
      intermediate result. For example, if the different phases of your UDA pass strings to each other but in the
      end the function returns a <codeph>BIGINT</codeph> value, you would specify <codeph>INTERMEDIATE
      STRING</codeph>. Likewise, if the different phases of your UDA pass 2 separate <codeph>BIGINT</codeph> values
      between them (8 bytes each), you would specify <codeph>INTERMEDIATE CHAR(16)</codeph> so that each function
      could read from and write to a 16-byte buffer.
    </p>

    <p>
      For end-to-end examples of UDAs, see <xref href="impala_udf.xml#udfs"/>.
    </p>

    <p conref="../shared/impala_common.xml#common/complex_types_blurb"/>

    <p conref="../shared/impala_common.xml#common/udfs_no_complex_types"/>

    <p conref="../shared/impala_common.xml#common/usage_notes_blurb"/>

    <ul>
      <li>
        You can write Impala UDFs in either C++ or Java. C++ UDFs are new to Impala, and are the recommended format
        for high performance utilizing native code. Java-based UDFs are compatible between Impala and Hive, and are
        most suited to reusing existing Hive UDFs. (Impala can run Java-based Hive UDFs but not Hive UDAs.)
      </li>

      <li rev="2.5.0 IMPALA-1748 IMPALA-2843">
        <keyword keyref="impala25_full"/> introduces UDF improvements to persistence for both C++ and Java UDFs,
        and better compatibility between Impala and Hive for Java UDFs.
        See <xref href="impala_udf.xml#udfs"/> for details.
      </li>

      <li>
        The body of the UDF is represented by a <codeph>.so</codeph> or <codeph>.jar</codeph> file, which you store
        in HDFS and the <codeph>CREATE FUNCTION</codeph> statement distributes to each Impala node.
      </li>

      <li>
        Impala calls the underlying code during SQL statement evaluation, as many times as needed to process all
        the rows from the result set. All UDFs are assumed to be deterministic, that is, to always return the same
        result when passed the same argument values. Impala might or might not skip some invocations of a UDF if
        the result value is already known from a previous call. Therefore, do not rely on the UDF being called a
        specific number of times, and do not return different result values based on some external factor such as
        the current time, a random number function, or an external data source that could be updated while an
        Impala query is in progress.
      </li>

      <li>
        The names of the function arguments in the UDF are not significant, only their number, positions, and data
        types.
      </li>

      <li>
        You can overload the same function name by creating multiple versions of the function, each with a
        different argument signature. For security reasons, you cannot make a UDF with the same name as any
        built-in function.
      </li>

      <li>
        In the UDF code, you represent the function return result as a <codeph>struct</codeph>. This
        <codeph>struct</codeph> contains 2 fields. The first field is a <codeph>boolean</codeph> representing
        whether the value is <codeph>NULL</codeph> or not. (When this field is <codeph>true</codeph>, the return
        value is interpreted as <codeph>NULL</codeph>.) The second field is the same type as the specified function
        return type, and holds the return value when the function returns something other than
        <codeph>NULL</codeph>.
      </li>

      <li>
        In the UDF code, you represent the function arguments as an initial pointer to a UDF context structure,
        followed by references to zero or more <codeph>struct</codeph>s, corresponding to each of the arguments.
        Each <codeph>struct</codeph> has the same 2 fields as with the return value, a <codeph>boolean</codeph>
        field representing whether the argument is <codeph>NULL</codeph>, and a field of the appropriate type
        holding any non-<codeph>NULL</codeph> argument value.
      </li>

      <li>
        For sample code and build instructions for UDFs,
        see <xref keyref="udf_samples">the sample UDFs in the Impala github repo</xref>.
      </li>

      <li>
        Because the file representing the body of the UDF is stored in HDFS, it is automatically available to all
        the Impala nodes. You do not need to manually copy any UDF-related files between servers.
      </li>

      <li>
        Because Impala currently does not have any <codeph>ALTER FUNCTION</codeph> statement, if you need to rename
        a function, move it to a different database, or change its signature or other properties, issue a
        <codeph>DROP FUNCTION</codeph> statement for the original function followed by a <codeph>CREATE
        FUNCTION</codeph> with the desired properties.
      </li>

      <li>
        Because each UDF is associated with a particular database, either issue a <codeph>USE</codeph> statement
        before doing any <codeph>CREATE FUNCTION</codeph> statements, or specify the name of the function as
        <codeph><varname>db_name</varname>.<varname>function_name</varname></codeph>.
      </li>
    </ul>

    <p conref="../shared/impala_common.xml#common/sync_ddl_blurb"/>

    <p conref="../shared/impala_common.xml#common/compatibility_blurb"/>

    <p>
      Impala can run UDFs that were created through Hive, as long as they refer to Impala-compatible data types
      (not composite or nested column types). Hive can run Java-based UDFs that were created through Impala, but
      not Impala UDFs written in C++.
    </p>

    <p conref="../shared/impala_common.xml#common/current_user_caveat"/>

    <p><b>Persistence:</b></p>

    <p conref="../shared/impala_common.xml#common/udf_persistence_restriction"/>

    <p conref="../shared/impala_common.xml#common/cancel_blurb_no"/>

    <p conref="../shared/impala_common.xml#common/permissions_blurb_no"/>

    <p conref="../shared/impala_common.xml#common/example_blurb"/>

    <p>
      For additional examples of all kinds of user-defined functions, see <xref href="impala_udf.xml#udfs"/>.
    </p>

    <p rev="2.5.0 IMPALA-2843">
      The following example shows how to take a Java jar file and make all the functions inside one of its classes
      into UDFs under a single (overloaded) function name in Impala. Each <codeph>CREATE FUNCTION</codeph> or
      <codeph>DROP FUNCTION</codeph> statement applies to all the overloaded Java functions with the same name.
      This example uses the signatureless syntax for <codeph>CREATE FUNCTION</codeph> and <codeph>DROP FUNCTION</codeph>,
      which is available in <keyword keyref="impala25_full"/> and higher.
    </p>
    <p rev="2.5.0 IMPALA-2843">
      At the start, the jar file is in the local filesystem. Then it is copied into HDFS, so that it is
      available for Impala to reference through the <codeph>CREATE FUNCTION</codeph> statement and
      queries that refer to the Impala function name.
    </p>
<codeblock rev="2.5.0 IMPALA-2843">
$ jar -tvf udf-examples.jar
     0 Mon Feb 22 04:06:50 PST 2016 META-INF/
   122 Mon Feb 22 04:06:48 PST 2016 META-INF/MANIFEST.MF
     0 Mon Feb 22 04:06:46 PST 2016 org/
     0 Mon Feb 22 04:06:46 PST 2016 org/apache/
     0 Mon Feb 22 04:06:46 PST 2016 org/apache/impala/
  2460 Mon Feb 22 04:06:46 PST 2016 org/apache/impala/IncompatibleUdfTest.class
   541 Mon Feb 22 04:06:46 PST 2016 org/apache/impala/TestUdfException.class
  3438 Mon Feb 22 04:06:46 PST 2016 org/apache/impala/JavaUdfTest.class
  5872 Mon Feb 22 04:06:46 PST 2016 org/apache/impala/TestUdf.class
...
$ hdfs dfs -put udf-examples.jar /user/impala/udfs
$ hdfs dfs -ls /user/impala/udfs
Found 2 items
-rw-r--r--   3 jrussell supergroup        853 2015-10-09 14:05 /user/impala/udfs/hello_world.jar
-rw-r--r--   3 jrussell supergroup       7366 2016-06-08 14:25 /user/impala/udfs/udf-examples.jar
</codeblock>
    <p rev="2.5.0 IMPALA-2843">
      In <cmdname>impala-shell</cmdname>, the <codeph>CREATE FUNCTION</codeph> refers to the HDFS path of the jar file
      and the fully qualified class name inside the jar. Each of the functions inside the class becomes an
      Impala function, each one overloaded under the specified Impala function name.
    </p>
<codeblock rev="2.5.0 IMPALA-2843">
[localhost:21000] > create function testudf location '/user/impala/udfs/udf-examples.jar' symbol='org.apache.impala.TestUdf';
[localhost:21000] > show functions;
+-------------+---------------------------------------+-------------+---------------+
| return type | signature                             | binary type | is persistent |
+-------------+---------------------------------------+-------------+---------------+
| BIGINT      | testudf(BIGINT)                       | JAVA        | true          |
| BOOLEAN     | testudf(BOOLEAN)                      | JAVA        | true          |
| BOOLEAN     | testudf(BOOLEAN, BOOLEAN)             | JAVA        | true          |
| BOOLEAN     | testudf(BOOLEAN, BOOLEAN, BOOLEAN)    | JAVA        | true          |
| DOUBLE      | testudf(DOUBLE)                       | JAVA        | true          |
| DOUBLE      | testudf(DOUBLE, DOUBLE)               | JAVA        | true          |
| DOUBLE      | testudf(DOUBLE, DOUBLE, DOUBLE)       | JAVA        | true          |
| FLOAT       | testudf(FLOAT)                        | JAVA        | true          |
| FLOAT       | testudf(FLOAT, FLOAT)                 | JAVA        | true          |
| FLOAT       | testudf(FLOAT, FLOAT, FLOAT)          | JAVA        | true          |
| INT         | testudf(INT)                          | JAVA        | true          |
| DOUBLE      | testudf(INT, DOUBLE)                  | JAVA        | true          |
| INT         | testudf(INT, INT)                     | JAVA        | true          |
| INT         | testudf(INT, INT, INT)                | JAVA        | true          |
| SMALLINT    | testudf(SMALLINT)                     | JAVA        | true          |
| SMALLINT    | testudf(SMALLINT, SMALLINT)           | JAVA        | true          |
| SMALLINT    | testudf(SMALLINT, SMALLINT, SMALLINT) | JAVA        | true          |
| STRING      | testudf(STRING)                       | JAVA        | true          |
| STRING      | testudf(STRING, STRING)               | JAVA        | true          |
| STRING      | testudf(STRING, STRING, STRING)       | JAVA        | true          |
| TINYINT     | testudf(TINYINT)                      | JAVA        | true          |
+-------------+---------------------------------------+-------------+---------------+
</codeblock>
    <p rev="2.5.0 IMPALA-2843">
      These are all simple functions that return their single arguments, or
      sum, concatenate, and so on their multiple arguments. Impala determines which
      overloaded function to use based on the number and types of the arguments.
    </p>
<codeblock rev="2.5.0 IMPALA-2843">
insert into bigint_x values (1), (2), (4), (3);
select testudf(x) from bigint_x;
+-----------------+
| udfs.testudf(x) |
+-----------------+
| 1               |
| 2               |
| 4               |
| 3               |
+-----------------+

insert into int_x values (1), (2), (4), (3);
select testudf(x, x+1, x*x) from int_x;
+-------------------------------+
| udfs.testudf(x, x + 1, x * x) |
+-------------------------------+
| 4                             |
| 9                             |
| 25                            |
| 16                            |
+-------------------------------+

select testudf(x) from string_x;
+-----------------+
| udfs.testudf(x) |
+-----------------+
| one             |
| two             |
| four            |
| three           |
+-----------------+
select testudf(x,x) from string_x;
+--------------------+
| udfs.testudf(x, x) |
+--------------------+
| oneone             |
| twotwo             |
| fourfour           |
| threethree         |
+--------------------+
</codeblock>

    <p rev="2.5.0 IMPALA-2843">
      The previous example used the same Impala function name as the name of the class.
      This example shows how the Impala function name is independent of the underlying
      Java class or function names. A second <codeph>CREATE FUNCTION</codeph> statement
      results in a set of overloaded functions all named <codeph>my_func</codeph>,
      to go along with the overloaded functions all named <codeph>testudf</codeph>.
    </p>
<codeblock rev="2.5.0 IMPALA-2843">
create function my_func location '/user/impala/udfs/udf-examples.jar'
  symbol='org.apache.impala.TestUdf';

show functions;
+-------------+---------------------------------------+-------------+---------------+
| return type | signature                             | binary type | is persistent |
+-------------+---------------------------------------+-------------+---------------+
| BIGINT      | my_func(BIGINT)                       | JAVA        | true          |
| BOOLEAN     | my_func(BOOLEAN)                      | JAVA        | true          |
| BOOLEAN     | my_func(BOOLEAN, BOOLEAN)             | JAVA        | true          |
...
| BIGINT      | testudf(BIGINT)                       | JAVA        | true          |
| BOOLEAN     | testudf(BOOLEAN)                      | JAVA        | true          |
| BOOLEAN     | testudf(BOOLEAN, BOOLEAN)             | JAVA        | true          |
...
</codeblock>
    <p rev="2.5.0 IMPALA-2843">
      The corresponding <codeph>DROP FUNCTION</codeph> statement with no signature
      drops all the overloaded functions with that name.
    </p>
<codeblock rev="2.5.0 IMPALA-2843">
drop function my_func;
show functions;
+-------------+---------------------------------------+-------------+---------------+
| return type | signature                             | binary type | is persistent |
+-------------+---------------------------------------+-------------+---------------+
| BIGINT      | testudf(BIGINT)                       | JAVA        | true          |
| BOOLEAN     | testudf(BOOLEAN)                      | JAVA        | true          |
| BOOLEAN     | testudf(BOOLEAN, BOOLEAN)             | JAVA        | true          |
...
</codeblock>
    <p rev="2.5.0 IMPALA-2843">
      The signatureless <codeph>CREATE FUNCTION</codeph> syntax for Java UDFs ensures that
      the functions shown in this example remain available after the Impala service
      (specifically, the Catalog Server) are restarted.
    </p>

    <p conref="../shared/impala_common.xml#common/related_info"/>

    <p>
      <xref href="impala_udf.xml#udfs"/> for more background information, usage instructions, and examples for
      Impala UDFs; <xref href="impala_drop_function.xml#drop_function"/>
    </p>
  </conbody>
</concept>