In Booking last year payments visibility became for us one of core goals. Being SOA environment we have plenty of services calling SAP via integration platform like Mulesoft. We have reviewed - how can we see end-to-end traces from the caller to the end document as a single span tree. I have explored available options and found several possible solutions. One of solutions - is telemetry implementation by Lars Hvam exposing data in protobuf format.

Technically it's a right thing to do - all code in ABAP, direct integration. But practically HTTP calls on a scale appeared to be a problem. Using AMC to expose events from ABAP sessions also failed - we were stuck with "Handshake" lock. So we looked at native abap capabilities to allow processes to stream events with minimal costs and it appeared that MQTT protocol fits the best and our performance tests showed exceptional performance comparing to HTTP calls. Another point is a message serialization. Identity JSON transformation is significantly faster that custom-bult procedures and doesn't require support.

So at the end our solution consists of:
- ABAP SDK allowing to create traces, spans, raise span events and streaming them using MQTT in JSON format ( also supports plugins, so it's very extendable )
- NodeJS proxy service which is coupled with a format produced by ABAP Trace SDK, proxying messages via MQTT to telemetry providers ( Honeycomb in our case )

May be in future we will consider also usage of a message broker, such as event Mesh. Currently I'm working on more general use case of streaming not only telemetry , but also Kafka events from SAP using similar approach. We consider currently Solace as our target environment for PoC

Since ABAP version 7.57 so-called Data Dictionary (DDIC) ABAP CDS views get obsolete and SAP recommends the usage of entity views instead. During this session I will present their advantages: compliance with RAP and data buffering on application server. I have also discovered during the implementation of the complex scenario that RAP and data buffering does not work within update task process. I will share the examples of the code for more precise explanations of these topics.