FUNCTION

Config path: /<database>/<schema>/function/<name>(<dtypes>).yaml

Example:

arguments:
  input_val: OBJECT

returns: VARCHAR(1000)

body: |-
  GET(input_val, COALESCE('BOOKINGS_LANG', 'en'))::varchar(1000)

arguments:
  x: VARCHAR(10000)

returns: VARCHAR(255)
language: java

imports:
  - stage: test_internal_stage
    path: /lib/zero-allocation-hashing-0.15.jar

handler: SnowHash.xxHash

body: |-
  import net.openhft.hashing.LongHashFunction;

  class SnowHash {
      public static LongHashFunction hash_func = LongHashFunction.xx();

      public static String xxHash(String x) {
          return Long.toHexString(hash_func.hashChars(x));
      }
  }

language: python
runtime_version: "3.8"

returns: VARIANT

packages:
  - numpy
  - pandas
  - xgboost==1.5.0

handler: udf

body: |-
  import numpy as np
  import pandas as pd
  import xgboost as xgb

  def udf():
    return [np.__version__, pd.__version__, xgb.__version__]

Schema

• language (str) - language of function (default: SQL)

• runtime_version (str) - used to specify version of Python, Java, etc.

• arguments (dict)

  • {key} (ident) - argument name

  • {value} (str) - argument data type --- OR ---
  • {value} (dict)

    • type (str) - argument data type
    • default (str) - default SQL expression for optional argument

• returns (str) - for single return value, return data type --- OR ---

• returns (dict) - for table return values

  • {key} (ident) - return column name

  • {value} (str) - return column data type

• body (str) - function body

• is_secure (bool) - is function SECURE

• is_aggregate (bool) - is function AGGREGATE

• is_strict (bool) - is function STRICT (always returns NULL on NULL input)

• is_immutable (bool) - is function IMMUTABLE (same input always produced the same output)

• is_memoizable (bool)

• imports (list) - files to import (usually JAR packages)

  • {items} (dict)

    • stage (ident) - name of stage

    • path (str) - path to file

• packages (list) - Snowflake system packages to import as dependencies

  • {items} (str) - name of package, with optional version of package

• handler (str) - name of class and method to be called

• external_access_integrations (list)

  • {items} (ident) - name of external access integration
• secrets (dict)

  • {key} (str) - secret variable name used in function code

  • {value} (ident) - name of secret object
• comment (str)

Usage notes

1. Snowflake supports overloading of function names. Multiple functions may have the same name as long as they have different arguments. It is required to use comma-separated base data types of arguments in config names. For example: my_function(number).yaml, my_function(varchar,number).yaml
2. Files for imports should be maintained using STAGE FILES.
3. If function body is empty, handler and imports with pre-compiled JAR or Python code are required.

4. runtime_version should be specified as string with explicit double-quotes (e.g. "3.8"). Otherwise YAML parser may confuse it with number, which may cause some unwanted effects.

5. You may use custom YAML tag !include to store function body in a separate file instead of storing it inside YAML.

Links

• CREATE FUNCTION
• SHOW USER FUNCTIONS
• Parser & JSON Schema (GitHub)