Tuesday, October 23, 2018

Even Aces make mistakes: Reuse

Every year Debrah Lilley is kind enough to round up a couple of Oracle Aces to talk about projects, and/or new features at Oracle Open World.
This year we present a session with 9(!) people about our 'favorite' mistake we made and what we learned from it.

From the summary:
"A quick diversion from our now traditional featured short talks, this year our EMEA ACE Directors will share their biggest errors and what they learned from it. A fast, fun session that will energize you on your Oracle OpenWorld 2018 experience."

I have five minutes to talk about my favorite mistake, so this blog post elaborates on the topic a bit more.

Reuse

In software, we like to reuse code. This enhances productivity, it minimizes the chance of introducing new mistakes and gives us a chance to focus on the new things we are trying to accomplish, instead of reinventing the wheel. This has been a very strong driver of Service Oriented Architecture (SOA). 
There are multiple ways of implementing or realizing reuse:
  • create a runtime component that you call from your code to do the job ('a service')
  • create a library or code that you share from a central location and import in your code ('a common artifact in SOA Suite in MDS, or a library in Java or Node.js)
  • a pattern ('template') that you apply as a best practice 
  • a copy (!?) of the code you want to reuse and adapt. 
One of the downsides of reuse, is flexibility loss. If you are reusing something, a change to it will impact all the consumer or users of that piece of code. service etc. This is why micro services architecture is all about bounded contexts and decoupling using events. 

In SOA there has been a very common misconception: the fact that reuse is good, no matter what, resulting in inflexible very complex code that is expensive to change and, in the end, hard to reuse (!)
This particularly happens a lot with canonical models, and yes, I made that mistake too.

Let's go back in time, to a project where we created an application for the province of Overijssel for permits and grants, using Oracle SOA Suite 11g, a content management. The solution basically consisted of 4 different composites (Apply, Process, Decide, Notify), the Oracle Servicebus, a content management system, SAP ERP, a permit application and a.NET user interface.

For the integration between the .NET interface and the BPEL process support, we designed a canonical model.

Canonical model side note


The picture below shows what a canonical mode is supposed to do: make sure that in the servicebus and between applications, a canonical model is used.


End of side note


However, we decided to use the canonical model inside of our BPELs as well. This had several advantages:
  1. we did not have to design the WSDL and XML Schemas for the BPEL
  2. we could assign the entire message to the BPEL input and output variables

Change

We were done very fast, but we found we needed a number of changes in the canonical model to facilitate the .NET application. They wanted different structures and different fields. None of these had any functional impact on process flow, since the BPEL was mostly driving the process, adding approvals, enrichments and fetching and storing data in the correct backend system.
However, because we used the canonical model in all of our BPELs, we had to redo all of our assignments and xslts to pick up this new version, making sure nothing changed in the flow and functionality of the component.
The change cost the same amount of time as the inital realization, because the work in BPEL is not in clicking together the scopes, but in building the logic in the assignments and xslts.

Lessons learned

We learned a number of things from this particular approach:

  1. Separation of concerns matters. This has been a long standing IT and computer science principle. See for example (https://en.wikipedia.org/wiki/Separation_of_concerns). There is a reason you put different groups in your electrical wiring in your house: if one of the groups fails, the other parts of the house are unaffected. When you have to make a change for a specific purpose, you should have to make that change in one place only. In case of the canonical model, in the servicebus and the components that are calling the servicebus.
  2. Reuse has an impact on changeability and maintainability. To make sure things can be changed, apply the tolerant reader pattern ((https://martinfowler.com/bliki/TolerantReader.html)
  3. Pick your reuse pattern carefully: a pattern you want to reuse (or a template), a library that you will apply, a service that you want to call, or creating a copy to be quick, but that you will be able to change independently.
  4. Don't send data to systems that don't care about it. In this case were were sending data and dragging it along in the BPEL that were not needed in the BPEL. Typically all you need in the BPEL are identifiers and dates. Then, when you need more data in a specific scope, you can fetch it from the appropriate source. 
We refactored the code and made specific WSDLs and XSDs for all BPELs and used the canonical model in the OSB, where it belongs. 
It became a lot easier to work with the messages in the BPEL and changes in the UI had no or little impact on the process, as they should!

Code smells

So how do you know if you are reusing the canonical model too much. These are some tell tales that point in that direction:

SOA Suite

  1. You have to redeploy 'MDS' everytime a new feature is defined
  2. All your artefacts (WSDLs, XSDs, DVMs) are in MDS, there are no 'local composite' artefact
  3. You have a lot of data in your messages in BPEL that are in none of the assigns or in only one of the assigns
  4. You have a lot of merge conflicts with other teams that are working on completely different topics
  5. You have a if statements in your BPEL that never merge back into 1 common flow, based on something in your canonical model

Java, Javascript

  1. Your ORM library gets updated all the time and your code needs to change because of it.
  2. You filter out values and fields in most of your code
  3. You are extending the common objects to add your own behavior and attributes and filter out the common behavior. 

Conclusion

Reuse is a powerful and important topic to be productive and to avoid mistakes. However, reuse is not a goal, it is a means to an end.  Modern systems need to be changeable in the first place. Reuse should be carried out within the context as appropriate. Global models should be treated carefully and be minimized to where they add value and are needed. 
We can all learn from domain driven design principles and microservices architectures in our other architectures as well to make sure we don't paint ourselves into a corner!

Happy coding 😎

No comments:

Post a Comment