PUBLIC

SAP HANA Platform 2.0 SPS 05

Document Version: 1.0 – 2021-07-09

SAP HANA External Machine Learning Library for

SAP HANA Platform
© 2022 SAP SE or an SAP affiliate company. All rights reserved.

THE BEST RUN

Content

1 SAP HANA External Machine Learning Library. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Getting Started. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1 Prerequisites. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 Components. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.3 Technical Architecture. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.4 Deployment Scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.5 Users and Roles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.6 Checking the EML Installation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.7 Calling EML Procedures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 Models and Call Signatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12

3.1 TensorFlow Model Example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2 Preparing SavedModels. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.3 Calling Models with PREDICT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.4 PREDICT Parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.5 Multiple Inferences with PREDICTM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 Deploying and Mapping Models on a Landscape. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4.1 Deploying SavedModels in a TMS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4.2 Creating Remote Sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.3 Mapping Models onto Remote Sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.4 Updating Model and Landscape Definitions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4.5 Validating Model and Landscape Definitions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

5 CloudHook. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5.1 CLOUDHOOK_COMMAND. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .26

5.2 CLOUDHOOK_SENDBLOB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .27

5.3 CLOUDHOOK_GETBLOB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29

6 Important Disclaimer for Features in SAP HANA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

SAP HANA External Machine Learning Library for SAP HANA Platform
2 PUBLIC Content
1 SAP HANA External Machine Learning
Library

This guide describes how to integrate an external machine learning framework, Google TensorFlow, with SAP
HANA.

The integration of Google TensorFlow with SAP HANA is based on the SAP HANA Application Function Library
(AFL) as well as Google’s gRPC remote procedure call package. It also involves a separate server process that
hosts the actual machine learning functionality.

The guide is structured as follows:

● Getting Started
This section describes the basic requirements for using the External Machine Learning (EML) Library, the
main components it involves, the technical architecture, the supported deployment scenarios, and the
users and authorizations needed.
It also describes how to verify that the EML objects were successfully installed and introduces the general
syntax for calling EML procedures.
● Models and Call Signatures
This section introduces the main concepts associated with TensorFlow machine learning models, such as
signatures, tensors, shapes, and data types. It explains how input and output tensors on the TensorFlow
side are mapped to SAP HANA columns and tables and how the TensorFlow model can be called from SAP
HANA using the PREDICT or PREDICTM procedure.
● Deploying and Mapping Models on a Landscape
This section explains how to deploy a model in a TensorFlow ModelServer (TMS), how to create a remote
source in SAP HANA to contact the remote TMS, and how to map the model in the TMS to the remote
source.
It also describes how to update the model and landscape definitions and how to validate them, that is,
verify that a given model can be reached and has a valid serving state.
● CloudHook
This topic describes how the CLOUDHOOK functions can be used to integrate external services such as
SAP Data Hub or external cloud providers.

SAP HANA External Machine Learning Library for SAP HANA Platform
SAP HANA External Machine Learning Library PUBLIC 3
2 Getting Started

Familiarize yourself with the External Machine Learning (EML) Library and its specific requirements.

Related Information

Prerequisites [page 4]
Components [page 5]
Technical Architecture [page 6]
Deployment Scenarios [page 7]
Users and Roles [page 7]

2.1 Prerequisites

To use the SAP HANA External Machine Learning (EML) features, you need to do the following:

● Check that you are using a supported operating system version:

○ Minimum SUSE Linux Enterprise Server version: SLES 12 SP 2
○ Minimum Red Hat Enterprise Linux version: RHEL 7.4
These are needed due to an internal upgrade to gRPC 1.14.2 and its dependency on more modern
cryptography libraries.
● Install SAP HANA Platform 2.0 SPS 05.
● Install the External Machine Learning Library (SAP HANA EML AFL).
For information about how to install or update the SAP HANA EML AFL, see the section on installing or
updating SAP HANA components in the SAP HANA Server Installation and Update Guide.

 Note

The revision number of the SAP HANA EML AFL must match the revision number of SAP HANA. For
more information, see SAP Note 1898497.

● Enable the Script Server on the SAP HANA instance. For more information, see SAP Note 1650957.

Related Information

SAP HANA Server Installation and Update Guide

SAP Note 1898497
SAP Note 1650957

SAP HANA External Machine Learning Library for SAP HANA Platform
4 PUBLIC Getting Started
2.2 Components

The main components involved in the integration between SAP HANA and Google TensorFlow are described
below.

TensorFlow

TensorFlow provides a set of open-source machine-learning libraries together with a supporting infrastructure
that allows you to experiment with a wide range of machine learning algorithms. You can execute the
algorithms efficiently across different types of hardware environments, ranging from mobile phones to
distributed clusters of servers, with additional GPUs.

As a data scientist, you can use TensorFlow to create, train, and evaluate machine learning models. When you
are satisfied with the performance of the finalized model, you can prepare additional metadata and export the
model in the SavedModel format for consumption through TensorFlow Serving.

For more information, see the TensorFlow documentation at https://www.tensorflow.org/ .

The following TensorFlow versions are supported:

● Minimum version: 1.0

● Maximum and recommended version: 1.13.1

TensorFlow ModelServer (TMS)

TensorFlow Serving is a flexible, high-performance serving system for machine learning models and is
designed for production environments. A TensorFlow ModelServer (TMS) makes TensorFlow models exported
in SavedModel format accessible for execution through the gRPC remote procedure call mechanism. An
effective SAP HANA EML AFL installation needs to have access to at least one running TMS in order to execute
TensorFlow calculations.

For more information, see TensorFlow Serving .

The following TensorFlow Serving versions are supported:

● Minimum version: 0.5.1

● Minimum version for using dynamic model management: 1.7
● Minimum version for using transport layer security: 1.11
● Maximum and recommended version: 1.13.0

The recommended way to obtain and run the TMS is using Google’s official Docker images. For more
information, see TensorFlow Serving with Docker .

Alternatively, if your infrastructure does not support Docker images, the official TMS binaries can be used
instead. For more information, see Installation: Installing ModelServer .

SAP HANA External Machine Learning Library for SAP HANA Platform
Getting Started PUBLIC 5
External Machine Learning (EML) AFL

The EML AFL is a TensorFlow Serving client implementation for SAP HANA. It makes TensorFlow inferences on
models deployed in a TensorFlow ModelServer available as local AFL procedure calls in SAP HANA.

Supported gRPC versions: 1.0.0 or higher

The gRPC version currently used in EML is 1.14.2.

2.3 Technical Architecture

The basic architecture underlying the integration of Google TensorFlow with SAP HANA is shown below.

Component Description

EML AFL The TensorFlow Serving client implementation for SAP HANA. It allows
predefined TensorFlow models to be remotely invoked through gRPC
calls encapsulated inside AFL procedures.

TensorFlow ModelServer (TMS) Makes TensorFlow models exported in SavedModel format accessible for
execution through gRPC remote procedure calls

Active Models The models currently served and therefore available for execution

gRPC Server The gRPC server interface for communication with the TMS client

SAP HANA External Machine Learning Library for SAP HANA Platform
6 PUBLIC Getting Started
 Component Description

Model Persistence Externally stored models in SavedModel format available for activation in
the TMS

CloudHook Server A building block for integrating external services

Other Cloud Services and Infrastructure Includes, for example, SAP Data Hub and external cloud providers

2.4 Deployment Scenarios

The TensorFlow ModelServer (TMS) can be co-deployed on the SAP HANA landscape or deployed on separate
hardware resources.

TMS Co-deployed on the SAP HANA Landscape

This is the simplest setup since there is no additional hardware required. This type of setup is supported for
test, development, and other non-production deployments. For additional security and robustness, the TMS
instances should be configured to run in an appropriately configured container on the host machines.

TMS Deployed on Separate Hardware Resources

For production environments, TMS resources must be kept separate from SAP HANA.

2.5 Users and Roles

You require the following users in the SAP HANA and TMS systems.

SAP HANA EML Administrator

An EML administrator manages the system landscape by setting up and monitoring the connections to the
TMS servers hosting the models.

An administrator user requires the following SAP HANA privileges:

● MONITORING
● AFL__SYS_AFL_EML_EXECUTE
● CREATE REMOTE SOURCE
● SELECT ON _SYS_AFL.EML_MODEL_CONFIGURATION
● UPDATE ON _SYS_AFL.EML_MODEL_CONFIGURATION
● INSERT ON _SYS_AFL.EML_MODEL_CONFIGURATION
● DELETE ON _SYS_AFL.EML_MODEL_CONFIGURATION

SAP HANA External Machine Learning Library for SAP HANA Platform
Getting Started PUBLIC 7
SAP HANA EML User

An EML user, typically the application developer, defines which models are available and how they can be
invoked.

An EML user requires the following SAP HANA privilege:

● AFL__SYS_AFL_EML_EXECUTE

Note that to use the previous versions of the EML PREDICT and PREDICTM functions that required wrapper
procedures to be generated, the AFLPM_CREATOR_ERASER_EXECUTE privilege is also required to create the
wrappers.

TensorFlow ModelServer Owner

This is the operating system user that runs the TMS instances. There can be multiple operating system users
on all target landscapes. For best isolation, this user should be distinct from the SAP HANA system
administrator user.

2.6 Checking the EML Installation

To confirm that the EML functions were installed successfully, you can check the following public views:

● sys.afl_areas
● sys.afl_packages
● sys.afl_functions

These views are granted to the PUBLIC role and can be accessed by anyone.

To check the views, run the following SQL statements:

SELECT * FROM "SYS"."AFL_AREAS" WHERE AREA_NAME = 'EML';

SELECT * FROM "SYS"."AFL_PACKAGES" WHERE AREA_NAME = 'EML';
SELECT * FROM "SYS"."AFL_FUNCTIONS" WHERE AREA_NAME = 'EML';

The result will indicate whether the EML functions were successfully installed on your system.

SAP HANA External Machine Learning Library for SAP HANA Platform
8 PUBLIC Getting Started
To verify that the EML procedures were generated, open the _SYS_AFL schema in the SAP HANA Catalog, as
shown below:

SAP HANA External Machine Learning Library for SAP HANA Platform
Getting Started PUBLIC 9
2.7 Calling EML Procedures

After the installation of the SAP HANA Application Function Library (AFL), the generated EML procedures are
available in the _SYS_AFL schema. All EML procedure names start with EML_.

CTL, CHECKDESTINATION, and CLOUDHOOK

The calling syntax for the CTL, CHECKDESTINATION, CLOUDHOOK_COMMAND, CLOUDHOOK_SENDBLOB,

and CLOUDHOOK_GETBLOB procedures is as follows:

CALL _SYS_AFL.EML_<function_name>_PROC(
{input_table1,…}, <output_tab>) WITH OVERVIEW;

● EML_<function_name>_PROC: The procedure name automatically generated by the AFL.

● <input_table1,…>: The user-defined names of the procedure's input tables.
● <output_table>: The user-defined name of the procedure's output table.

 Sample Code

CALL _SYS_AFL.EML_CHECKDESTINATION_PROC(CALL_PARAMS, RESULTS) WITH OVERVIEW;

CALL _SYS_AFL.EML_CLOUDHOOK_COMMAND_PROC(CALL_PARAMS,INPUT, OUTPUT) WITH

OVERVIEW;

PREDICT and PREDICTM

The EML PREDICT and PREDICTM procedures take a variable number of input tables. Unlike other SAP HANA
stored procedures, which are statically typed and require pre-defined input and output table signatures, the
EML PREDICT and PREDICTM procedures also allow you to pass input tables with different column structures.
The PREDICT and PREDICTM procedures are based on the PREDICT_ANY and PREDICTM_ANY functions.

The calling syntax is as follows:

CALL _SYS_AFL.EML_<procedure_name>(
{input_table1,…}, <output_table>) WITH OVERVIEW;

● <procedure_name>: The appropriate PREDICT or PREDICTM procedure name.

● <input_table1,…>: The user-defined names of the procedure's input tables.
● <output_table>: The user-defined name of the procedure's output table.

 Sample Code

CALL _SYS_AFL.EML_PREDICT_2(CALL_PARAMS, FEATURES1, FEATURES2, RESULTS) WITH

OVERVIEW;

SAP HANA External Machine Learning Library for SAP HANA Platform
10 PUBLIC Getting Started
  Note

The previous versions of the EML PREDICT and PREDICTM functions that required wrapper procedures to
be generated are still supported in existing code.

Related Information

SAP Note 2046767

PREDICT Parameters [page 15]
Multiple Inferences with PREDICTM [page 16]
CloudHook [page 26]

SAP HANA External Machine Learning Library for SAP HANA Platform
Getting Started PUBLIC 11
3 Models and Call Signatures

A TensorFlow machine learning model implements the actual machine learning functionality (typically
inference), which involves mapping a set of input data to a set of output data (for example, images to textual
descriptions, historical data to predictions). The actual artifact representing such a model in the TensorFlow
Serving environment is a SavedModel.

TensorFlow models are created and trained by data scientists, typically in a Python-based programming and
training environment. The inner workings of the finished model are hidden within a SavedModel, while the
details of the inputs and outputs for using the model are exposed in the SavedModel signature.

All inputs and outputs of TensorFlow models are described in terms of tensors, which denote arrays of various
data types and dimensionality, referred to as the tensor's shape. Every TensorFlow model takes a number of
input tensors and transforms them into a number of output tensors. Since TensorFlow models are general
dataflow graphs, there may be multiple different inputs and outputs possible that make available different parts
of the overall transformation flow encapsulated in the model.

A SavedModel has a number of signatures, each describing one meaningful combination of input and output
tensors. Each SavedModel should have exactly one serving_default signature and can have an arbitrary
number of additional named signatures that provide alternative uses of the same graph.

The External Machine Learning AFL supports only a subset of TensorFlow's tensor shapes and data types,
namely those that can be relatively easily mapped between standard SQL tables and data types and tensors. Of
the possible combinations defined in the TensorFlow documentation Tensor Ranks, Shapes, and Types, the
following are supported in particular:

TensorFlow Data Types SAP HANA Data Types Supported Shapes

DT_FLOAT, DT_DOUBLE FLOAT, DOUBLE Rank 2

DT_STRING All string and BLOB data types Rank 1 and Rank 2

DT_INT32, DT_INT64 INTEGER Rank 2

Related Information

SavedModel README
Module: tf.saved_model
Module: tf.saved_model.signature_def_utils
Tensor Ranks, Shapes, and Types

SAP HANA External Machine Learning Library for SAP HANA Platform
12 PUBLIC Models and Call Signatures
3.1 TensorFlow Model Example

This example shows a simple linear regression TensorFlow model that tries to fit a vector of input values to a
predict regressed output. The model has an input tensor named features and an output tensor named
results.

The signature is as follows (based on the saved_model_cli utility from TensorFlow r1.2.0 and higher):

% saved_model_cli show --dir linear_regression --all

MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:
signature_def['serving_default']:
The given SavedModel SignatureDef contains the following input(s):
inputs['features'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 1)
name: x1:0
The given SavedModel SignatureDef contains the following output(s):
outputs['results'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 1)
name: y:0
Method name is: tensorflow/serving/predict

The shapes (features and results are tensors of width 1 and indefinite length, indicated by the -1 value) fit
SAP HANA tables of the following shape:

CREATE TYPE FEATURES_INPUT_TT AS TABLE ("X" FLOAT);

CREATE TYPE RESULTS_OUTPUT_TT AS TABLE ("Y" FLOAT);

Related Information

SavedModel CLI (Command-Line Interface)

3.2 Preparing SavedModels

It is important to include complete signature information and, in particular, complete tensor shape information
using the utilities provided by TensorFlow (for example, tf.saved_model.utils.build_tensor_info).

If you have multiple input and output tensors, they are mapped to tables and columns in SAP HANA in the
order established by sorting their key values in the C locale. Although this might appear arbitrary, it needs to
be done because TensorFlow represents the tensors in a position-independent manner in a map keyed by
name, while in SAP HANA some of the relationships have to be represented by the relative position of the tables
and columns in the argument order. Note that tensor names and SAP HANA column and table names are
otherwise completely independent of each other.

SAP HANA External Machine Learning Library for SAP HANA Platform
Models and Call Signatures PUBLIC 13
The following example illustrates the mapping between tensor names and argument order. On the TensorFlow
side, the tensors are listed as follows:

signature_def['serving_default']:
The given SavedModel SignatureDef contains the following input(s):
inputs['second'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 1)
name: x1:0
inputs['first'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 1)
name: x2:0
The given SavedModel SignatureDef contains the following output(s):
outputs['y'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 1)
name: y:0

On the SAP HANA side, the "x2" tensor maps to the first input table and the "x1" tensor to the second input
table, despite being listed in the order shown above. This is because their key values sort in the order first - >
second.

Unlike input tensors, which are mapped one by one to separate input tables in the sorted order of their keys,
output tensors have a very limited set of types and shapes and are currently all combined into a single output
table. This means that output tensors must all have the same shape and can differ only in their data type.

Related Information

tf.saved_model.utils.build_tensor_info
Serving a TensorFlow Model

3.3 Calling Models with PREDICT

Once you have the information about the number, type, and positions of the model inputs and outputs at hand,
you can map this information to the required tables in SAP HANA and then invoke the TensorFlow model by
calling the PREDICT procedure provided by the EML AFL. The example below illustrates this process.

The example uses the simple linear regression model that is specified as follows:

The given SavedModel SignatureDef contains the following input(s):

inputs['features'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 1)
name: x1:0
The given SavedModel SignatureDef contains the following output(s):
outputs['results'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 1)
name: y:0

SAP HANA External Machine Learning Library for SAP HANA Platform
14 PUBLIC Models and Call Signatures
 1. Prepare the tables:

-- Define table types for linear regression

CREATE TYPE CALL_PARAMS_TT AS TABLE ("Parameter" VARCHAR(100), "Value"
VARCHAR(100));
CREATE TYPE FEATURES_TT AS TABLE ("X" FLOAT);
CREATE TYPE RESULTS_TT AS TABLE ("Y" FLOAT);

2. Invoke the TensorFlow model, in the same way as a normal AFL function:

-- The value for "Model" must match the name used by the TensorFlow
ModelServer.
CREATE TABLE CALL_PARAMS LIKE CALL_PARAMS_TT;
INSERT INTO CALL_PARAMS VALUES ('Model', 'linearregression');
CREATE TABLE FEATURES LIKE FEATURES_TT;
INSERT INTO FEATURES VALUES (1.1);
INSERT INTO FEATURES VALUES (2.5);
INSERT INTO FEATURES VALUES (2.9);
INSERT INTO FEATURES VALUES (8.9);
CREATE TABLE RESULTS LIKE RESULTS_TT;
-- Perform linear regression
CALL _SYS_AFL.EML_PREDICT(CALL_PARAMS, FEATURES, RESULTS) WITH OVERVIEW;
SELECT * FROM RESULTS;

Related Information

PREDICT Parameters [page 15]

3.4 PREDICT Parameters

The general form of PREDICT is as follows:

PREDICT(CALL_PARAMS,INPUT1, OUTPUT)
PREDICT_2(CALL_PARAMS,INPUT1,INPUT2, OUTPUT)
PREDICT_3(CALL_PARAMS,INPUT1,INPUT2,INPUT3, OUTPUT)
PREDICT_4(CALL_PARAMS,INPUT1, INPUT2, INPUT3, INPUT4, OUTPUT)

The EML AFL currently supports up to four INPUT tables, each of which can have an arbitrary number of
columns of FLOAT, DOUBLE, INTEGER, and various SAP HANA string and BLOB data types.

CALL_PARAMS

CALL_PARAMS is a table of the following type:

CREATE CALL_PARAMS AS TABLE ("Parameter" VARCHAR(100), "Value"

VARCHAR(100));

It can contain the following:

SAP HANA External Machine Learning Library for SAP HANA Platform
Models and Call Signatures PUBLIC 15
 Parameter Value Mandatory Comment

Model Name of the model as Yes Must map to at least

deployed in a TMS, op­ one running TMS to
tionally qualified with a execute
named signature

Deadline String representing an No Maximum number of

integer number milliseconds to wait at
most for a result

The syntax for accessing named signatures is <model>%<signaturename>, so to

access a named signature second_sig on a model samplemodel, you would specify
samplemodel%second_sig as the value. Specifying just samplemodel is equivalent
to specifying samplemodel%serving_default.

 Note

The keywords for the Parameter column (that is, "Model" and "Deadline") are case
sensitive.

INPUT

INPUT denotes the tables that match the input tensors of the specified signature of the
model according to the following mapping rules, in the order given by sorting the input
tensor keys in the C locale:

Tensor Shape Tensor Type Table

(-1) or (N) DT_STRING Single column of an SAP

HANA string type

(-1,M) or (N,M) M columns of a matching SAP

DT_FLOAT
HANA data type, at least N
DT_DOUBLE
rows or an arbitrary number of
DT_STRING rows when N == -1
DT_INT32
DT_INT64

For PREDICT and any definite value of N (that is, not "-1", which means arbitrary),
PREDICT ignores rows beyond N. For PREDICTM, the number of rows under each tag
must match N exactly.
OUTPUT

The OUTPUT table has N rows of M columns matching M output tensors of shape (-1,N)
and appropriate data types.

3.5 Multiple Inferences with PREDICTM

The PREDICTM procedure enables multiple inferences to be processed within a single AFL call.

To distinguish the different sets of inputs and results, each input table has an INTEGER "tag" column as its first
column. Spans of adjacent rows with the same INTEGER tag value are formatted into a separate inference call.

SAP HANA External Machine Learning Library for SAP HANA Platform
16 PUBLIC Models and Call Signatures
The results are tagged with the corresponding INTEGER tag in the output table’s "tag" column. The results in
the output table may appear in a different order than the input ranges, but they are guaranteed to be in
consecutive rows per partial result set. Where possible, calls are processed in parallel.

The general form of PREDICTM is as follows:

PREDICTM(CALL_PARAMS,INPUT1, OUTPUT)
PREDICTM_2(CALL_PARAMS,INPUT1,INPUT2, OUTPUT)
PREDICTM_3(CALL_PARAMS,INPUT1,INPUT2,INPUT3, OUTPUT)
PREDICTM_4(CALL_PARAMS,INPUT1, INPUT2, INPUT3, INPUT4, OUTPUT)

The semantics of CALL_PARAMS and OUTPUT are identical to PREDICT, with the exception of the first
additional "tag" column. So a parallel version involving multiple calls to the linear regression from the PREDICT
section looks like this (reusing CALL_PARAMS from the PREDICT example):

-- Create tagged feature table

CREATE TYPE MULTI_FEATURES_TT AS TABLE ("Tag" INTEGER, "X" DOUBLE);
CREATE TABLE MULTI_FEATURES LIKE MULTI_FEATURES_TT;
INSERT INTO MULTI_FEATURES VALUES (1,1.1);
INSERT INTO MULTI_FEATURES VALUES (1,2.5);
INSERT INTO MULTI_FEATURES VALUES (2,1.9);
INSERT INTO MULTI_FEATURES VALUES (2,3.3);
INSERT INTO MULTI_FEATURES VALUES (2,5.9);
INSERT INTO MULTI_FEATURES VALUES (2,8.9);

-- Create tagged result table

CREATE TYPE MULTI_RESULTS_TT AS TABLE ("Tag" INTEGER,"Y" DOUBLE);
CREATE TABLE MULTI_RESULTS LIKE MULTI_RESULTS_TT;

-- Perform multiple linear regressions

CALL _SYS_AFL.EML_PREDICTM(CALL_PARAMS, MULTI_FEATURES, MULTI_RESULTS) WITH
OVERVIEW;
SELECT * FROM MULTI_RESULTS;

 Note

In case of unrecoverable errors (either input formatting, remote processing, or result collection) in any
given tag subset, PREDICTM may partially fill the output table with results prior to the unrecoverable error.

SAP HANA External Machine Learning Library for SAP HANA Platform
Models and Call Signatures PUBLIC 17
4 Deploying and Mapping Models on a
Landscape

TensorFlow models you want to use need to be mapped to a system landscape that has at least one TensorFlow
ModelServer (TMS) hosting the models.

This task is performed by the system administrator and requires the appropriate roles and privileges. The
networking-related part of this mapping makes use of the remote source concept from the SAP HANA
federation framework, while the content part uses a system configuration table. All tasks are managed through
normal SQL commands on the configuration table and the CTL and CHECKDESTINATION procedures.

Related Information

Deploying SavedModels in a TMS [page 18]

Creating Remote Sources [page 19]
Mapping Models onto Remote Sources [page 21]
Updating Model and Landscape Definitions [page 22]
Validating Model and Landscape Definitions [page 24]

4.1 Deploying SavedModels in a TMS

A SavedModel is a directory tree containing a defined structure of special files. For the purposes of
deployment, the saved model directory tree has to be copied to a file system location that can be reached by
the running TMS instance.

In order to use a model, you need to expose the model as a service using TensorFlow serving. You can load
multiple models into a TMS at the same time by using the model_config_file method, which allows you to
list the individual models that you want to add to the TMS in a configuration file.

Example: Serve the two example models "testmodel" and "anothermodel" on a single TMS
1. Ensure that you have the following model directory structure:
○ The SavedModel for testmodel is in the directory /models/testmodel_1.
○ The SavedModel for anothermodel is in the directory /models/testmodel_2.
2. Create a file called mymodelconfig.conf and add the models as follows:

model_config_list: {
config: {
name: "testmodel",
base_path: "/models/testmodel_1",
model_platform: "tensorflow"
},

SAP HANA External Machine Learning Library for SAP HANA Platform
18 PUBLIC Deploying and Mapping Models on a Landscape
 config: {
name: "anothermodel",
base_path: "/models/testmodel_2",
model_platform: "tensorflow"
}
}

3. Run TensorFlow Serving, either using Docker or directly.

 Note

It is recommended to use Docker to run TensorFlow Serving.

○ Using Docker:
1. Copy mymodelconfig.conf to /models/mymodelconfig.conf.
2. Run the following:

docker run -d –name "MyTMS" -p 5000:8500 -u nobody --mount

type=bind,source=<path-to-"models"-directory>,target=/models/ \
-t tensorflow/serving --model_config_file=/models/mymodelconfig.conf

○ Directly:
On the TMS host, invoke the tensorflow_model_server with the following command:

tensorflow_model_server --port=5000 --model_config_file=mymodelconfig.conf

Note that the file paths do not have to be absolute. Relative file paths are interpreted relative to the
working directory of the TMS.
For the full set of arguments for the tensorflow_model_server, see the Google documentation.

Related Information

TensorFlow Serving with Docker

4.2 Creating Remote Sources

To contact a remote TMS or CloudHook server, SAP HANA needs network addresses, port numbers, and
optionally security-related information about the network transport to be established. Because Tensorflow
Serving is a protocol defined on top of gRPC, remote sources need to be defined on the grpc adapter.

 Example

A TMS is deployed on a host reachable under the name tmsserver.mydomain.com, port number 5000.
The model testmodel is deployed to that TMS. The connection is unauthenticated and unencrypted. The
only transport setup currently understood by the TMS code as published by Google is as follows:

CREATE REMOTE SOURCE TMS_1

ADAPTER "grpc"
CONFIGURATION 'server=tmsserver.mydomain.com;port=5000';

SAP HANA External Machine Learning Library for SAP HANA Platform
Deploying and Mapping Models on a Landscape PUBLIC 19
The full set of options for the grpc adapter is given below:

Property Property Description Values Comment

server Server DNS name or IP address

port Port Port number

deadline How long to wait at most for Number of milliseconds be­ Maximum timeout is 10 mi­
a response from the TMS tween 1 and 600000 nutes

ssl_mode SSL mode "use SSL" or "no SSL"

client_crt Certificate X509 client certificate Only needed when the TMS
is set to certificate authenti­
cation

client_key Private key X509 client private key Only needed when the TMS
is set to certificate authenti­
cation

ca Certification Authority X509 Certification Authority Only needed if any of the

certificate keys on the client or server
are not in the default CA trust
list

proxyhost HTTP CONNECT capable DNS name and port Example:

proxy server host.domain.com:
8080

proxyuser Optional proxy credentials Proxy user and optional Example:

proxy password proxyuser
proxyuser:proxypassword

cloudhooktype Command vocabulary type of Simple valid identifier string Describes the type of com­
the CloudHook server mand vocabulary and pur­
pose of the CloudHook
server represented by the re­
mote source

cloudhookargs Optional additional argu­ String Interpreted by the remote

ments for the CloudHook CloudHook server

Note that it is not possible to enter the correct values for the client_crt, client_key, and ca fields in any of the
graphical interfaces to SAP HANA, because the certificates are either binary or have a defined multiline
encoding. The only way to create a remote source with these values is through a textual definition.

 Example

Create a remote source to a TMS on the host tmshost behind the proxy proxy.corp:

CREATE REMOTE SOURCE PROXYTMS

ADAPTER "grpc"
CONFIGURATION 'server=tmshost;port=50052;proxyhost=proxy.corp:8888';

SAP HANA External Machine Learning Library for SAP HANA Platform
20 PUBLIC Deploying and Mapping Models on a Landscape
  Example

Create a remote source with encryption enabled. The TMS server keys have been signed with a private
Certification Authority that is not in the default trust list and therefore needs to be explicitly specified:

CREATE REMOTE SOURCE SSL_SOURCE ADAPTER "grpc" CONFIGURATION

'server=tmsserver.cloud.com;port=50055;use_ssl=use SSL;ca=-----BEGIN
CERTIFICATE-----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-----END CERTIFICATE-----';

4.3 Mapping Models onto Remote Sources

With the TMS contact points defined, you can enter the mapping specifying which models are deployed where
in the _SYS_AFL.EML_MODEL_CONFIGURATION table.

The table has the following definition:

TABLE _SYS_AFL.EML_MODEL_CONFIGURATION (
"Model" VARCHAR(256),
"Parameter" VARCHAR(256),
"Value" VARCHAR(256)
);

Model

The name of the model as deployed in a given TMS. Model names must be unique
across the entire landscape and the underlying TensorFlow model must be the same
model on all TMS instances where it is deployed, otherwise the behavior is undefined.

SAP HANA External Machine Learning Library for SAP HANA Platform
Deploying and Mapping Models on a Landscape PUBLIC 21
Parameter and Value

An open extensible set of parameters that can be assigned to a given model. Currently
the only valid "Parameter" is RemoteSource , with a "Value" specifying the name of a
grpc remote source, as defined by the SAP HANA federation framework.

 Note

The keywords for the Parameter column (that is, RemoteSource) are case
sensitive.

 Example

The TMS is deployed on a host that can be reached through the TMS_1 remote source already defined in
the previous example. The model testmodel is deployed to that TMS. Insert the required values into the
EML model configuration table:

INSERT INTO _SYS_AFL.EML_MODEL_CONFIGURATION VALUES('testmodel',

'RemoteSource', 'TMS_1');

A single TMS can host multiple models and a single model can be hosted on multiple TMS instances. Typically,
it is better to deploy multiple models on a single TMS on a given host rather than deploying multiple TMS
instances on a single host with a single model in each TMS. In the latter case, this occupies multiple ports and
needlessly replicates the overhead of the underlying TensorFlow runtime in each TMS.

4.4 Updating Model and Landscape Definitions

In typical steady-state operation, SAP HANA reads the contents of the

_SYS_AFL.EML_MODEL_CONFIGURATION table during system startup and routes all inference calls to the
deployed models according to their state, as defined in SYS_AFL.EML_MODEL_CONFIGURATION, at that point
in time.

Any changes applied to SYS_AFL.EML_MODEL_CONFIGURATION at runtime (either through changes to the

table or changes to the underlying remote source definitions) are not immediately honored by the runtime. As
an administrator, you can prepare changes to the definitions with multiple independent statements and then
apply the entire set of changes to the EML AFL runtime through an explicit update call as follows:

SYS_AFL.EML_CTL_PROC(Request,CALL_PARAMS,RESULTS)

Request

Currently the only valid value for Request is UpdateModelConfiguration.

CALL_PARAMS

CALL_PARAMS is a table with the following format:

TABLE CALL_PARAMS ("Parameter" VARCHAR(100), "Value"

VARCHAR(100));

The contents are ignored for UpdateModelConfiguration.

SAP HANA External Machine Learning Library for SAP HANA Platform
22 PUBLIC Deploying and Mapping Models on a Landscape
RESULTS

RESULTS is a table with the following format:

TABLE CTL_RESULTS ("Key" VARCHAR(100), "Value" INTEGER, "Text"

VARCHAR(100));

It is used to report the result of whatever action CTL has taken, by giving both the error
codes and long texts for the various items of a control request.

The only key for the UpdateModelConfiguration request is Status, with the error
code and error text shown in the Value and Text columns.

Example

Add a second model to the TMS created in the previous section:

INSERT INTO _SYS_AFL.EML_MODEL_CONFIGURATION VALUES('anothermodel',

'RemoteSource', 'TMS_1');

Update the runtime in an SAP HANA system with a single index server/script server:

CREATE TABLE CALL_PARAMS ("Parameter" VARCHAR(100), "Value" VARCHAR(100));

CREATE TABLE CTL_RESULTS ("Key" VARCHAR(100), "Value" INTEGER, "Text"
VARCHAR(100));
CALL _SYS_AFL.EML_CTL_PROC(''UpdateModelConfiguration'', CALL_PARAMS,
CTL_RESULTS);

Update the runtime in an SAP HANA system with multiple index servers/script servers:

CREATE TABLE CALL_PARAMS ("Parameter" VARCHAR(100), "Value" VARCHAR(100));

CREATE TABLE CTL_RESULTS ("Key" VARCHAR(100), "Value" INTEGER, "Text"
VARCHAR(100));
CREATE PROCEDURE UPDATE_MODEL_CONFIGURATION() AS
BEGIN
DECLARE CURSOR CUR FOR
SELECT VOLUME_ID FROM SYS.M_VOLUMES WHERE SERVICE_NAME = 'indexserver';
FOR CUR_ROW AS CUR DO
EXEC 'CALL _SYS_AFL.EML_CTL_PROC(''UpdateModelConfiguration'',
CALL_PARAMS, CTL_RESULTS)'
|| ' WITH OVERVIEW WITH HINT(ROUTE_TO(' || :CUR_ROW.VOLUME_ID ||
'))';
END FOR;
END;
CALL UPDATE_MODEL_CONFIGURATION();

SAP HANA External Machine Learning Library for SAP HANA Platform
Deploying and Mapping Models on a Landscape PUBLIC 23
4.5 Validating Model and Landscape Definitions

The CHECKDESTINATION call allows you to validate that a given model or CloudHook server can be actively
called under a specified set of landscape definitions. It can be used to verify that the model or CloudHook is
reachable and, for models, that it has a valid serving state.

You can use the CHECKDESTINATION call as follows:

CREATE TABLE CALL_PARAMS_TT ("Parameter" VARCHAR(100), "Value" VARCHAR(100));

TABLE RESULTS ("Code" VARCHAR(10), "Longtext" VARCHAR(100));
_SYS_AFL.EML_CHECKDESTINATION_PROC(CALL_PARAMS, RESULTS);

CALL_PARAMS

The following is valid for CALL_PARAMS:

Parameter Value Mandatory Comment

Model Name of the model as No Must map to at least

deployed in a TMS one running TMS to
execute. It does not
have to be in
SYS_AFL.EML_MODE
L_CONFIGURATION.

RemoteSource Name of a remote No Does not have to be in

source SYS_AFL.EML_MODE
L_CONFIGURATION

Deadline String representing an No Maximum number of

integer number milliseconds to wait at
most for a result, as
specified by Remote­
Source or 5000 ms as
default

The "RemoteSource" parameter allows TMS hosts to be checked that have not yet been
deployed into the active runtime service. Actual model invocations through PREDICT-
style calls need to have a remote source defined in
SYS_AFL.EML_MODEL_CONFIGURATION.

 Note

The keywords for the Parameter column (that is, "Model", "RemoteSource", and
"Deadline") are case sensitive.

 Example

Check whether "testmodel" can be reached and is actively deployed on a given

remote source:

CREATE TABLE CALL_PARAMS ("Parameter" VARCHAR(100), "Value"

VARCHAR(100));
CREATE TABLE RESULTS ("Code" VARCHAR(10), "Longtext"
VARCHAR(100));
INSERT INTO CALL_PARAMS VALUES ('Model', 'testmodel');

SAP HANA External Machine Learning Library for SAP HANA Platform
24 PUBLIC Deploying and Mapping Models on a Landscape
 CALL _SYS_AFL.EML_CHECKDESTINATION_PROC(CALL_PARAMS,
RESULTS);

● If the "Model" parameter is omitted but a "RemoteSource" is explicitly specified,

the CHECKDESTINATION function tries to check basic connectivity and credentials
on a TMS by trying to invoke a dummy model request. Depending on the settings of
the TMS, this may lead to (informational) log messages for non-existent "dummy-
model-check" messages by the TMS servers.
● If a "Model" parameter is specified but not a "RemoteSource", the
CHECKDESTINATION function uses the currently active mapping to remote
sources and checks the status of every TMS the model is listed on.
● If both "Model" and "RemoteSource" are omitted, the CHECKDESTINATION
function tries to do a complete check of all models and remote sources specified in
SYS_AFL.EML_MODEL_CONFIGURATION. That is, it tries to determine what the
status of all models and target destinations would be if a CTL call were made to
push the contents of SYS_AFL.EML_MODEL_CONFIGURATION into the runtime
service.

RESULTS

The return values in the RESULTS table after successful invocation (that is, the call to
CHECKDESTINATION was technically correct, but not necessarily the underlying
destination being checked) are as follows:

Code Long Text Comment

0 Ok The model was reached at the given desti­

nation.

1 Cannot determine destination Either the host is missing or there is no

for model <model> matching model. Check the parameters
and SYS_AFL.EML_MODEL_CONFIGURA­
TION.

2 Cannot create channel creden­ Check the authorization settings of the

tials for model remote sources registered for the model

3 Messages describing problems The remote source (either derived from

with the remote source defini- _SYS_AFL.EML_MODEL_CONFIGURA­
tion TION or explicitly given) either doesn't ex­
ist in SAP HANA or the configuration
string is ill-formed.

4 Destination <destination> ex­ The host could not be resolved or reached

ceeded deadline within the configured timeout (default: 5
seconds). Check the name resolution and
IP reachability.

5 Server <destination> ok but no A TMS was found but did not have a
model <model> matching model. Check the TMS configu-
ration and the model name.

14 Server <destination> is alive The host was reached but no TMS, Cloud­
but refused connection Hook, or proxy was found at the given
port number. Check the proxy or TMS/
CloudHook (depending on whether a
proxy is used).

SAP HANA External Machine Learning Library for SAP HANA Platform
Deploying and Mapping Models on a Landscape PUBLIC 25
5 CloudHook

The CLOUDHOOK functions allow generic external management functionality to be invoked using SQLScript.
They serve as a building block for integrating external services such as SAP Data Hub or external cloud
providers.

For this purpose, an intermediate CloudHook server is positioned between EML and the external services. The
management functionality itself is therefore not directly embedded in EML.

The following CLOUDHOOK functions are available:

● CLOUDHOOK_COMMAND
● CLOUDHOOK_SENDBLOB
● CLOUDHOOK_GETBLOB

Each function has its own specific signature, but the parameters are identical.

Related Information

CLOUDHOOK_COMMAND [page 26]

CLOUDHOOK_SENDBLOB [page 27]
CLOUDHOOK_GETBLOB [page 29]

5.1 CLOUDHOOK_COMMAND

The CLOUDHOOK_COMMAND function can be used to manage externally stored models from within SAP
HANA. Typical use cases include listing stored models, deploying and undeploying models, listing active
models, and activating and deactivating model versions.

The general form of CLOUDHOOK_COMMAND is as follows:

CLOUDHOOK_COMMAND_PROC(CALL_PARAMS,INPUT, OUTPUT)

CALL_PARAMS

CALL_PARAMS is a table of the following type:

CREATE CALL_PARAMS AS TABLE ("Parameter" VARCHAR(100), "Value"

VARCHAR(100));

It can contain the following:

SAP HANA External Machine Learning Library for SAP HANA Platform
26 PUBLIC CloudHook
 Parameter Value Mandatory Comment

RemoteSource A valid remote source Yes Must map to at least

with a "cloudhook­ one running Cloud­
type" set Hook server to exe­
cute

Command String representing a Yes Valid values depend on

command the "cloudhooktype",
with two reserved val­
ues "SendBlob" and
"GetBlob"

Deadline String representing an No Maximum number of

integer number milliseconds to wait at
most for a result

Note that the "Command" values "SendBlob" and "GetBlob" are reserved for the
CLOUDHOOK_SENDBLOB and CLOUDHOOK_GETBLOB functions.
INPUT

INPUT is a table of a type compatible with the following:

CREATE INPUT AS TABLE ("Tag" VARCHAR(100), "Value"

VARCHAR(5000));

OUTPUT

OUTPUT is a table of a type compatible with the following:

CREATE OUTPUT AS TABLE ("Tag" VARCHAR(100), "Value"

VARCHAR(5000));

Which "Tag" values are valid and whether they are mandatory, optional, unique, or can appear multiple times
depends on the "cloudhooktype".

5.2 CLOUDHOOK_SENDBLOB

The CLOUDHOOK_SENDBLOB function can be used to move large binary objects from BLOB columns in SAP
HANA to the CloudHook server. It has the reserved "Command" value "SendBlob".

The general form of CLOUDHOOK_SENDBLOB is as follows:

CLOUDHOOK_SENDBLOB_PROC(CALL_PARAMS,INPUT, OUTPUT)

CALL_PARAMS

CALL_PARAMS is a table of the following type:

CREATE CALL_PARAMS AS TABLE ("Parameter" VARCHAR(100), "Value"

VARCHAR(100));

It can contain the following:

SAP HANA External Machine Learning Library for SAP HANA Platform
CloudHook PUBLIC 27
 Parameter Value Mandatory Comment

RemoteSource A valid remote source Yes Must map to at least

with a "cloudhook­ one running Cloud­
type" set Hook server to exe­
cute

Command String representing a Yes Must be "SendBlob"

command

Deadline String representing an No Maximum number of

integer number milliseconds to wait at
most for a result

INPUT

INPUT is a table of a type compatible with the following:

CREATE INPUT AS TABLE ("Tag" VARCHAR(100), "Value" BLOB);

OUTPUT

OUTPUT is a table of a type compatible with the following:

CREATE OUTPUT AS TABLE ("Tag" VARCHAR(100), "Value"

VARCHAR(100));

The CLOUDHOOK_SENDBLOB function has the following usage pattern:

1. For every "Tag"/"Value" pair in "INPUT", the contents of "Value" are sent to the remote CloudHook server.
2. The remote CloudHook server stores the values and returns a string handle in "OUTPUT" under the same
"Tag".

For example:

DROP REMOTE SOURCE MYDOCKER;

CREATE REMOTE SOURCE MYDOCKER
ADAPTER "grpc"
CONFIGURATION 'cloudhooktype=TMSControl;server=mydockerhost;port=50200';
CREATE TYPE CH_PARAMS_TT AS TABLE ("Parameter" VARCHAR(100), "Value"
VARCHAR(100));
CREATE TYPE CH_ARG_INPUT_TT AS TABLE ("Tag" VARCHAR(100), "Value" BLOB);
CREATE TYPE CH_ARG_OUTPUT_TT AS TABLE ("Tag" VARCHAR(100), "Value" VARCHAR(100));
CREATE TABLE CH_PARAMS LIKE CH_PARAMS_TT;
CREATE TABLE CH_INPUT LIKE CH_ARG_INPUT_TT;
CREATE TABLE CH_OUTPUT LIKE CH_ARG__OUTPUT_TT;
INSERT INTO CH_PARAMS VALUES ('Command', 'SendBlob');
INSERT INTO CH_PARAMS VALUES ('RemoteSource', 'MYDOCKER');
INSERT INTO CH_INPUT VALUES('SavedModel',’<the-large-savedmodel-binary-here’);
CALL _SYS_AFL.EML_CLOUDHOOK_SENDBLOB_PROC(CH_PARAMS, CH_INPUT, CH_OUTPUT) WITH
OVERVIEW;

SELECT * FROM CH_OUTPUT;

"Tag" "Value"

"SavedModel" "a7f8f184-81dd-4495-8169-d1b9434f5ba4"

SAP HANA External Machine Learning Library for SAP HANA Platform
28 PUBLIC CloudHook
5.3 CLOUDHOOK_GETBLOB
The CLOUDHOOK_GETBLOB function can be used to move large binary objects from the CloudHook server to
BLOB columns in SAP HANA. It has the reserved "Command" value "GetBlob".

The general form of CLOUDHOOK_GETBLOB is as follows:

CLOUDHOOK_GETBLOB_PROC(CALL_PARAMS,INPUT, OUTPUT)

CALL_PARAMS

CALL_PARAMS is a table of the following type:

CREATE CALL_PARAMS AS TABLE ("Parameter" VARCHAR(100), "Value"

VARCHAR(100));

It can contain the following:

Parameter Value Mandatory Comment

RemoteSource A valid remote source Yes Must map to at least

with a "cloudhook­ one running Cloud­
type" set Hook server to exe­
cute

Command String representing a Yes Must be "GetBlob"

command

Deadline String representing an No Maximum number of

integer number milliseconds to wait at
most for a result

INPUT

INPUT is a table of a type compatible with the following:

CREATE INPUT AS TABLE ("Tag" VARCHAR(100), "Value"

VARCHAR(100));

OUTPUT

OUTPUT is a table of a type compatible with the following:

CREATE OUTPUT AS TABLE ("Tag" VARCHAR(100), "Value" BLOB);

The CLOUDHOOK_GETBLOB function has the following usage pattern:

1. For every "Tag"/"Value" pair in "INPUT", the contents are requested from the remote CloudHook server.
2. The remote CloudHook server retrieves the contents and places them in "OUTPUT" under the same "Tag".

For example:

DROP REMOTE SOURCE MYDOCKER;

CREATE REMOTE SOURCE MYDOCKER
ADAPTER "grpc"
CONFIGURATION 'cloudhooktype=TMSControl;server=mydockerhost;port=50200';
CREATE TYPE CH_PARAMS_TT AS TABLE ("Parameter" VARCHAR(100), "Value"
VARCHAR(100));
CREATE TYPE CH_ARG_INPUT_TT AS TABLE ("Tag" VARCHAR(100), "Value" VARCHAR(100));

SAP HANA External Machine Learning Library for SAP HANA Platform
CloudHook PUBLIC 29
 CREATE TYPE CH_ARG_OUTPUT_TT AS TABLE ("Tag" VARCHAR(100), "Value" BLOB);
CREATE TABLE CH_PARAMS LIKE CH_PARAMS_TT;
CREATE TABLE CH_INPUT LIKE CH_ARG_INPUT_TT;
CREATE TABLE CH_OUTPUT LIKE CH_ARG__OUTPUT_TT;
INSERT INTO CH_PARAMS VALUES ('Command', 'GetBlob');
INSERT INTO CH_PARAMS VALUES ('RemoteSource', 'MYDOCKER');
DELETE FROM CH_INPUT;
INSERT INTO CH_INPUT VALUES('SavedModel', 'a7f8f184-81dd-4495-8169-
d1b9434f5ba4');
DELETE FROM CH_OUTPUT;
CALL _SYS_AFL.EML_CLOUDHOOK_GETBLOB_PROC(CH_PARAMS, CH_INPUT, CH_OUTPUT) WITH
OVERVIEW;

As a result, CH_OUTPUT will have the previously written BLOB object (from the CLOUDHOOK_SENDBLOB
example) under the "Tag" "SavedModel".

SAP HANA External Machine Learning Library for SAP HANA Platform
30 PUBLIC CloudHook
6 Important Disclaimer for Features in SAP
HANA

For information about the capabilities available for your license and installation scenario, refer to the Feature
Scope Description for SAP HANA.

SAP HANA External Machine Learning Library for SAP HANA Platform
Important Disclaimer for Features in SAP HANA PUBLIC 31
Important Disclaimers and Legal Information

Hyperlinks
Some links are classified by an icon and/or a mouseover text. These links provide additional information.
About the icons:

● Links with the icon : You are entering a Web site that is not hosted by SAP. By using such links, you agree (unless expressly stated otherwise in your
agreements with SAP) to this:

● The content of the linked-to site is not SAP documentation. You may not infer any product claims against SAP based on this information.
● SAP does not agree or disagree with the content on the linked-to site, nor does SAP warrant the availability and correctness. SAP shall not be liable for any
damages caused by the use of such content unless damages have been caused by SAP's gross negligence or willful misconduct.

● Links with the icon : You are leaving the documentation for that particular SAP product or service and are entering a SAP-hosted Web site. By using such
links, you agree that (unless expressly stated otherwise in your agreements with SAP) you may not infer any product claims against SAP based on this
information.

Videos Hosted on External Platforms

Some videos may point to third-party video hosting platforms. SAP cannot guarantee the future availability of videos stored on these platforms. Furthermore, any
advertisements or other content hosted on these platforms (for example, suggested videos or by navigating to other videos hosted on the same site), are not within
the control or responsibility of SAP.

Beta and Other Experimental Features

Experimental features are not part of the officially delivered scope that SAP guarantees for future releases. This means that experimental features may be changed by
SAP at any time for any reason without notice. Experimental features are not for productive use. You may not demonstrate, test, examine, evaluate or otherwise use
the experimental features in a live operating environment or with data that has not been sufficiently backed up.
The purpose of experimental features is to get feedback early on, allowing customers and partners to influence the future product accordingly. By providing your
feedback (e.g. in the SAP Community), you accept that intellectual property rights of the contributions or derivative works shall remain the exclusive property of SAP.

Example Code
Any software coding and/or code snippets are examples. They are not for productive use. The example code is only intended to better explain and visualize the syntax
and phrasing rules. SAP does not warrant the correctness and completeness of the example code. SAP shall not be liable for errors or damages caused by the use of
example code unless damages have been caused by SAP's gross negligence or willful misconduct.

Bias-Free Language
SAP supports a culture of diversity and inclusion. Whenever possible, we use unbiased language in our documentation to refer to people of all cultures, ethnicities,
genders, and abilities.

SAP HANA External Machine Learning Library for SAP HANA Platform
32 PUBLIC Important Disclaimers and Legal Information
SAP HANA External Machine Learning Library for SAP HANA Platform
Important Disclaimers and Legal Information PUBLIC 33
www.sap.com/contactsap

© 2022 SAP SE or an SAP affiliate company. All rights reserved.

No part of this publication may be reproduced or transmitted in any form

or for any purpose without the express permission of SAP SE or an SAP
affiliate company. The information contained herein may be changed
without prior notice.

Some software products marketed by SAP SE and its distributors

contain proprietary software components of other software vendors.
National product specifications may vary.

These materials are provided by SAP SE or an SAP affiliate company for

informational purposes only, without representation or warranty of any
kind, and SAP or its affiliated companies shall not be liable for errors or
omissions with respect to the materials. The only warranties for SAP or
SAP affiliate company products and services are those that are set forth
in the express warranty statements accompanying such products and
services, if any. Nothing herein should be construed as constituting an
additional warranty.

SAP and other SAP products and services mentioned herein as well as
their respective logos are trademarks or registered trademarks of SAP
SE (or an SAP affiliate company) in Germany and other countries. All
other product and service names mentioned are the trademarks of their
respective companies.

Please see https://www.sap.com/about/legal/trademark.html for

additional trademark information and notices.

THE BEST RUN