Semprini

Having systems of record is a widely accepted architectural pattern used to provide clear delineation of responsibilities and guidance to end-to-end solution architectures.

It is an architectural pattern of it's time, suited to an IT landscape of large monolithic applications and was the dominant strategy since the 90s across almost all corporate IT. If this it your IT strategy, you have good governance and patterns for change and an integration architecture which is for keeping these monoliths in-sync then go forth and prosper - you will have mature IT and a reasonable cadence of change.

More likely however, is that significant data is created in what is considered systems of record, half migrated new systems of record plus micro-services and analytic / machine learning flows so the misalignment of systems of record to the business view has become more obvious and material.

There is not a good semantic fit between an applications view of data and the business view of data so lets stop pretending.

The famous dig from Jamie Zawinski:

Some people, when confronted with a problem, think "I know, I'll use versioning." Now they have 2.1.0 problems.

My assertion is that integration versioning should be a last resort, used in specific situations. If you own both sides of the integration contract then do not version.

Versioning is in effect kicking the can down the road for your future self, or worse - leaving it for someone with no knowledge of the interface.

By following Postel's Law, using abstraction layers, micro-service boilerplate and not being afraid of regression testing (I.e. modern IT practices) we have simple to change components which we can track and upgrade.

A couple of contracts back I was consulting as a solution architect at a national retail organisation I ran an experiment as a proof of the cadence which is possible using Data Autonomy.

Shortly after the project went live, I decided to build it again myself using Data Autonomy as realistically as possible. The result was a far better solution in a third of the time & cost.

BI & analytics loves Data Autonomy and event driven architecture. In the operational side of Semantic Hub / Data Mesh, data is already clean and in business form. Data engineering can subscribe to all significant business object changes and metrics can be automatically calculated live. Dimensional modelling also becomes a much simpler process as canonical data is available near real time.

The data engineering and integration teams can become an active part of data governance and data stewardship. Working closely with the business domain SMEs, everyone is on the same page and reporting is simplified.

Data Autonomy is a collaboration of A Holistic Data Fabric/Mesh, RESTful interfaces, RESTful events and enterprise micro-services. Focused around Resource Oriented Architecture (ROA) rather than Service Oriented Architecture (SOA), its purpose is to provide the ability to change rapidly by keeping the level of dependencies constant across the IT landscape.

Standard event driven integration practice is to take data from a system, transform into a middle model and then transform to each destination. Data Autonomy simply says that while we have the data in the middle form, lets save it.

Each business domain must be free to evolve and mature independently as the view of significant business objects evolves. The data governance group with data stewards and SMEs should be responsible for producing domain aligned data product definitions which are then realized into data products. The features of these data products should cover interfaces and persistence for both operational and analytical data.

If you have a "microservice" which exposes an API calling a back end system, you don't have a microservice - you have an API. Your API has wisely used some of the concepts of microservice architecture. Microservice architecture is not an integration architecture, it's an application architecture and applications are responsible for business logic. This distinction is important because the beauty of microservice architecture is both technical and behavioural.

However, I'll extend an olive branch to the API people, no doubt aghast by my blasphemy, by calling them "integration microservices".

Integration microservices are fine I guess, but too often they are used, just as integration is often used, to sweep the legacy monster under the carpet so we don't have to see it - out of sight, out of mind. We implement integration microservices containerised, auto-scaling, auto-healing etc, IT management has an orgasm, we all add something to our CVs and the legacy monster waits.