Technology Blogs by SAP
Learn how to extend and personalize SAP applications. Follow the SAP technology blog for insights into SAP BTP, ABAP, SAP Analytics Cloud, SAP HANA, and more.
Showing results for 
Search instead for 
Did you mean: 


After a couple of projects I developed a kind of a best practise for creating custom operators. For replacing the manual steps I have written several helper scripts that finally funnelled into one. These I want to share with fellow Data Intelligence users. This blog starts with some general remarks on how to develop on the SAP Data Intelligence platform before it shows a way to locally code custom operators. I tried to explain all the rationales of my approach in some detail so you might easily tweak it into your way of doing things.

General Remarks about Programming with SAP Data Intelligence

The core development environment for SAP Data Intelligence is the Modeler. Here you plug together all the processing steps you need for your data pipeline. You have hundreds of operators available that support all kinds of requirements that you face in your daily business. Basically you can develop pipelines without writing scripts by just selecting and configuring operators.

Admittedly, this is not the whole truth. You mostly cannot avoid creating your own custom operators but SAP Data Intelligence is designed in such a way that there are multiple ways for doing this. You just have to select which is the one that is the most appropriate.

The architecture of SAP Data Intelligence is based on kubernetes and containers. The first is only of concern for developers when questions of performance and scaleability has a direct impact on how to design a solution. The latter defines the development environment more directly and some basic understanding could be of great help.

Most importantly, all pipelines run within one or more containers that are built when a pipeline has been created or modified. SAP Data Intelligence tries to put all operators within one container for obvious resource saving reasons. Which base image is being chosen is determined by the tags of the operators.

With the grouping feature

  1. Right-click on one operator

  2. Select 'Group' and

  3. Finally draw the new blue rectangle to encompass all operators for grouping

you can force that all operators run within one container that complies to the added tags.

If there is no image that complies to all tags, you tell an error message when the pipeline is started. The tagging and grouping is an important and powerful tool but needs a careful planning. Currently there is no support offered by the Modeler that helps understand the dependencies of operators, pipelines and available base images.

Trying to run all operators in one container is an important feature because it enables you to access the same memory space and therefore enables you to send data/messages of sizes that are only constrained by the available memory resources. Messages that are forced to bridge containers have currently a size limit of 10MB.

Generally you have three options of creating you own operators

  1. Using the Modeler Wizard for creating within SAP Data Intelligence

  2. Develop locally and import the operators as solution

  3. Adding your code in a container (e.g. you have legacy code that you do not want to touch by any means)

The sequence of the enumaration roughly coincides with the complexity the operator covers. As first rules I suggest the following :

  1. A couple of script lines (<15 lines) with a few (0-3) configuration parameters

  2. More than 15 lines of code with some logic based on more than 3 configuration parameters

  3. Compiled code or complex scripts that have already been developed, e.g. using programming languages that are currently not supported (FORTRAN, C, Java,..).

After a short overview of how to use the Modeler Wizard and describing the basics paradigms the main focus of this blog is the local development of operators.

Wizard-based Creation of new Operators

With SAP Data Intelligence you can create new custom Python operators using the Modeler Wizard. When being in the operator context (Operators Tab) of the tools panel on the left part of the Modeler screen, click the '+'-Button.

and the wizard leads through the five steps of defining a new operator:

  • Ports for the input and output of data

  • Tags for selecting the best base container

  • Configuration for defining the parameters

  • Script for the Python-code

  • Documentation for the manual

This is the straightforward way of creating operators with only a few parameters and limited number of code lines where you do not need any debugging. As soon as you expect some syntax errors creeping in and you are not sure foreseeing all logical implications, then the development in the Modeler-environment will become a nuisance. It takes a minute to run the code and there is no debugging except for adding print statements. Of course you can upload a script that you have written locally. This is of tremendous help if your operator has limited complexity and your first aim is to avoid syntax errors. For the more sophisticated operators you need another approach that supports

  • testing of operators that simulate the behaviour when running in a DI pipeline very closely

  • maintaining the lifecycle in github

  • controlling all the configuration parameters of an operator without the need to use the modeler wizard

  • Uploading the locally developed operators to a SAP Data Intelligence instance (not cut and paste)

All these items will be addressed in the following chapters.

Develop Custom Operator Locally

Custom Operator Structure

When adding a Python custom operator to the pipeline there are commented examples of how to access the port and configuration parameters and see the general structure a custom operator has to follow.

Most simple example:

def on_input(data):     
global counter
counter += 1
api.send("output", str(counter))api.set_port_callback("input", on_input)

You will find a documentation of all methods that can be used at SAP Help Portal

The most important function is
 api.set_port_callback(<portname>, <callback function>)​

which is called when a new message arrives at the named port. If input from more than one port is needed an array of ports can be passed. The callback function needs the same number of arguments as the number of ports in the ports array.

For sending messages out the function is used:

For accessing the configuration parameters the class api.config provides a direct access to the configuration variables: api.config.<variable>

Local Testing of Custom Operators

Of course you can test the syntax of the script when using a Python development environment but the missing definitions of the global variables api.message and api.config will be complained. To avoid importing any SAP Data Intelligence packages you can mimic these classes. But you have to ensure that your local definition is only used when there are no global definitions. Otherwise you always have to delete the class definition before you upload the script. The most common pythonic way to test the existence of a class is the try - except NameError - trick:

except NameError:
class api:
class config:
var1 = 'foo'
var2 = 'bar'

class Message:
def __init__(self,body = None,attributes = ""):
self.body = body
self.attributes = attributes

def send(port,msg) :
# example code for testing result
print('Port: ', port)
print('Attributes: ', msg.attributes)
print('Body: ', str(msg.body))

def set_port_callback(port, callback) :
test_msg = api.Message(attributes={'name':'doit'},body = 3)

def process(msg):
# start custom code
result = ''
for i in range (0,msg.body) :
result += str(i) + ':' + api.config.var1 + ' - ' + api.config.var2 + ' '
msg = api.Message(attributes={'name':'concat','type':'str'},body=result)
# end custom code
api.send("output", msg)

api.set_port_callback('input', process)

If you want to do more elaborated tests with different messages and configurations, you would rather add a main()-function and decouple the process()-function and the api.send() by adding an intermediate function-call call_on_input(msg). This enables you to define an additional function, msg) that returns the results directly and thus can be better compared with the expected results.

The drawback is that you have to care to comment the lines starting with '#if __name__ == '__main__': and maybe uncomment the line  api.set_port_callback('input', process).

After the "class api"-definition you might instead add

def process(msg):
result = ''
for i in range (0,msg.body) :
result += str(i) + ':' + api.config.var1 + ' - ' + api.config.var2 + ' '
return api.Message(attributes={'name':'concat','type':'str'},body=result)

def call_on_input(msg) :
new_msg = process(msg)

#api.set_port_callback('input', call_on_input)

def main() :
print('Test: Default')
api.set_port_callback('input', call_on_input)
print('Test: config')
config = api.config
config.var1 = 'own foo'
config.var12 = 'own bar'
test_msg = api.Message(attributes={'name':'test1'},body =4)
new_msg =,test_msg)
print('Attributes: ', new_msg.attributes)
print('Body: ', str(new_msg.body))

if __name__ == '__main__':

With this code template you can test your code locally but you still have to use the Modeler operator wizard to create the operator, adding all the tags, ports and config parameters manually and finally upload your code.

Solution Generation of Locally Developed Custom Operator

Of the four requirements that I have stated 2 chapters ago for a satifying local development of custom operators we have talked about the first. Two of the left overs will be tackled now. For that it helps to understand where and how the operators are stored in SAP Data Intelligence.

Operators are stored like packages using folders as namespaces. So you will find the custom Python operators in the /vrep-file system that you can examine using the System Management -> Files in 'My Workspace'. Under /files/vflow/subengines/com/sap/python36/operators you are going to see all your operators. It is always a good idea to group your operators in packages, that means you have at least one parent folder before you see the folders of your operators. You will have at least 4 files defining your operator:

  1. - documentation

  2. configSchema.json - defining the configuration parameters

  3. operator.json - general information about ports, default values, etc.

  4. <customOperatorName>.py - containing the script

  5. <icon-name>.svg - (optional) own operator icon

The task is to create these files mostly automatically and to have only one source of truth: the operator script. So we have to add some further information to the custom code that enables the generation of all the necessary files. Fortunately there is not very much to add.

Adding Ports Information

For adding the ports you have to provide the name and the type for all the ports:

inports = [{"name":"input","type":"message","description":"Description of the inport"}]
outports = [{"name":"output","type":"message","description":"Description of the outport"}]

def call_on_input(msg) :
new_msg = process(msg)

api.send(outports[0]["name"],new_msg)api.set_port_callback(inports[0]["name"], call_on_input)

Adding Configuration Description

For the configuration parameters additional information to the variable name is required:

  • title

  • description

  • type (Data Intelligence Datatype)

That means that for all config variables a dictionary is needed containing the data.

config_params['filename'] = {'title': 'Filename', 'description':'Filename (path)', 'type':'string'}

Adding Meta Data Information

Finally for completing the operator definition the script needs to provide

  • tags of required packages/libraries with version

  • version of the operator for creating the manifest.json

  • operator description that is the used as the operator title in the **Modeler**

  • operator description_long that contains some more descriptive text used in the generated only

  • add_readme is a dictionary where additional information can be added that appears in the generated as additional sections.

tags = {'python36': ''}  # tags that helps to select the appropriate containeroperator_description = 'Read File from Container'
operator description_long = "This operator reads data directly from the container file system....
version = "0.0.2" # for creating the manifest.json
add_readme["Reference"] = ["more info"](http://..\

Although it might not belong to the configuration context, it is stored in the api.config-class for convenient purpose.

Finally we have all definitions that we need to generate a solution package. A template of this CustomOperator can be viewed/downloaded at GitHub.

Generating the Solution

Now all the information that is needed for generating a custom operator is available in the script. If you follow the advised structure of the previous chapter then you can use a scipt I have created: gensolution. This is currently a publicly available script under the MIT-licence. So you can use and enhance it at your will. The script is part of the python package sdi_utils where I put all my helper functions. GitHub Repository.

gensolution provides two ways to generate a solution package:

  • Calling it in the main()-function of a custom operators

  • Using the commandline to create a package consisting of a number of packages with more than one operator at once.


The installation could be done manually from the GitHub or using pip:

python3 -m pip install sdi_utils

I have tested and used it only on MacOS so far. For more information have a look on GitHub.


usage: gensolution [-h] [--project PROJECT] [--version VERSION] [--debug]
[--zip] [--force] [--reverse] [--package PACKAGE]
[--operator OPERATOR]

Generate SAP Data Intelligence solution for local operator code
optional arguments: -h, --help show this help message and exit
--project PROJECT <new project path> - creates new project with required folder structure
--version VERSION <version> in format <num.num.num>
--debug For debug-level information
--zip Zipping solution folder
--force Removes subdirectories from <solution/operators>
--reverse Ceates a custom operator script from solution package
--package PACKAGE <package name> for reverse custom operator creation
--operator OPERATOR <package.operator folder> for reverse custom operator creation

Setting up Project: Option --project

gensolution expects that not only the meta-data is provided as described in the chapter "Solution Generation of locally developed Custom Operator" but also a fixed structure of the project. This can be generated automatically with "gensolution --project <path/project name>". Alternatively you can create the path manually and change to the project folder and call "gensolution --project ." In the folder "src/<projectname>/ " you find a "" that can be used for a start.

|_ project

     |_ src

         |_ <package1>

          |_ <operator11>

          |_ <operator-script11>.py

          |_ <operator12>

          |_ <operator-script12>.py


         |_ <package2>

Reverse engineering: Option --reverse

It has been the case that I have already started the development of a new operator in DI Modeler and only later realised that the script becomes rather complicated. Therefore I added the option to read the configSchema.json and operator.json and create a local script accordingly. Please be aware that it is not doing any code inspection of the operator script. This script is copied to the new local folder with the prefix 'old_' and the code snippets have to cut and paste to the right text locations.

For more and updated information about gensolution have a look at GitHub.

Call in main() of

If you only like to develop just one operator for a package, then you can also add a "gensolution"-call in the main()-call:

sd_utils.gensolution.gensolution(os.path.realpath(__file__), config, inports, outports)


After some general remarks about how to develop custom operators we delved into a kind of 'my'-best practise for developing custom operators locally that enables you to develop and test operators locally outside of SAP Data Intelligence and easily import them as a solution to your DI instance.

Please let me know if this helps you or what you think should be improved or enhanced.