There are cases when we cannot use cloud service as in the book. Even if in theory the mapping between the technical needs and the functionality of the cloud service is perfect, we could have external factors that influence the way how we design the solution.
You should remember all the time to check all current context, team shape, and client preferences before building a solution.
Imagine that you need to build a solution that would fetch data from an Azure SQL Database, do some transformation and push content to an on-premises database over HTTPS.
When you read the high-level reqs you already see an ETL solution, that would be translated into an Azure Data Factory that load data from SQL Database, to some transformation and make HTTPS requests.
Azure Data Factory integration with SSIS packages enables us to build an ETL seamless, using the team knowledge that already exists on SQL Server and SSIS. When the team technology stack is Java and Oracle, you might look at a different solution.
The logic components (e.g. data transformation) are moved from SSIS packages to Azure Functions that are automatically triggered by Azure Data Factory Pipelines. Even if the solution is changed a little bit, Azure Data Factory remains a strong candidate.
To be able to do HTTP requests to on-premises systems, Azure Data Factory requires to have an IR (Integration Runtime). You just find out that even if you have HTTPS, REST API, and Web Table connector when you need to do HTTPS requests that are not pointing to Azure Services you need to configure a self-hosted IR.
The self-hosted IR runs on an Azure Virtual Machine that you need to deploy and install an agent. Data Factory can create automatically the self-hosted IR by itself, but even so, you end up with additional VMs. The IR is used by Azure Data Factory do execute the HTTPS requests to on-premises applications.
At this moment in time, Azure Data Factory plays the role of the orchestrator between Azure Functions, IR and data movement.
Data Source and Data Content
Things become more complex when the data sources from where content needs to be pushed to the on-premises system are not only Azure SQL Database. There places in code, that need to trigger data to be pushed to on-premises.
In addition to this, specific changes to data that it is stored in Azure SQL Database needs to trigger such a push to on-premises. Some of them are hard to catch from SQL and in most cases, the ''signal' comes from application logic. It is not enough to specify to the Pipeline to push content that was modified from the last run. You need also additional logic that needs to decide what data needs to be pushed.
Most of the transformation logic is inside Azure Functions. The pipelines need to fetch data from multiple sources and near-real-time becomes a requirement. Taking this into account, we can redesign the solution as you can see below.
The part of the logic that do the transformation and calls the on-premises system is stored inside Azure Functions. Azure Event Hub is used to collect all the data that needs to be pushed to the on-premises systems.
There a gate logic implemented at the application level that automatically pushes data to Azure Event Hub when specific content needs to deliver to the on-premises system.
As we can see the end solution looks different in comparison with the first one. Even so, it does its job well and is a perfect match with team shape, requirements and future enhancements.