Workflowless

MidPoint ditched workflow engine. Scandal! How dare they? IDM without a workflow engine? Blasphemy! Abomination!

Workflow had been a holly cow of provisioning for almost 20 years. Hordes of IDM systems were born (and died) with integrated workflow engine. From the cumbersome proprietary workflow engine of Waveset Lighthouse to the open source BPMN engine of Apache Syncope, workflow was a natural part of identity management. Except one pesky little detail: it does not make sense.

Workflow engines are designed to govern flow of work among humans. It made a lot of sense to integrate workflow engine in IDM solutions in early 2000s. Lots of IDM tasks were manual at that time. And customers usually did not have a company-wide workflow system where IDM could simply be integrated. And even if they did, the infighting of software vendors made integration with workflow engines a complete nightmare. Therefore, any practical IDM solution was supposed to bring its own workflow engine. Otherwise it could not be deployed in a reasonable time. Fortunately, those times are over.

Character of IDM deployments was changing during 2000s and early 2010s. There was more automation and less manual work. And even if there was a manual work, it was largely limited to two areas: approvals and manual provisioning. Workflow engines were still in use as those were often the only places where behavior of an IDM system could be customized. However, the job of the workflow engine was no longer focused on interaction with humans. Workflow engines were (ab)used to run quite complex provisioning algorithms, evaluate policies and so on. But they were never designed to do that. It was a major pain to set up these processes. And it was even harder to maintain them. If you want to scare old IDM engineer, just whisper a word “upgrade” into his ears.

MidPoint was born in 2011. It was designed by engineers who went through the first age of IDM deployments in 2000s. Therefore, workflow engine had to be part of midPoint. Other products had it. Analysts wanted it. So we integrated workflow into midPoint without a huge amount of thinking. But we have realized quite soon that the workflow engine was reduced to do just a single job: approvals. The engine was not even processing the request and selecting the approvers. MidPoint did all of that. The engine just executed the approvals. That was pretty boring job for one big engine. It was an overkill. Therefore, we have jettisoned the workflow engine in midPoint 4.0. That was one of the best decisions that have ever made.

Now, approvals and manual provisioning are not the only things in IDM that require manual interaction, are they? Of course, there is a lot of things that cannot be automatized. However, many of those things are not really processes. They cannot be described by an algorithm, they do not have a prescribed flow of actors, forks and joins. These things tend to be “cases”. Something that needs to be solved, but for which an algorithmic solution is not available. It still needs human interaction, but that interaction is not constrained by a process. It is more like an improvised dance. Like a semi-structured teamwork. Workflow engine is not going to help with that.

But we cannot get rid of processes completely, can we? There is still few of them left. Maybe there is an enrolment process for a new employee. Maybe that employee needs to get company badge, keys to the office, attend health&safety training and so on. We may need an algorithmic process which is full of human interaction. That is still a very valid requirement. Process is needed. The point is that it does not make any sense to drive that process in the workflow engine which is integrated into an IDM system. There is usually a company-wide workflow system these days. Company physical security staff will not enjoy logging into the IDM system to work with employee enrolment process and then log into another workflow system to request a time off and do all the other stuff. It does not make sense. Workflow engine embedded in an IDM system is a bad idea.

What makes sense is the ability for an IDM system to integrate with existing company-wide workflow engine. IDM system should be able to forward process to the workflow system and continue the process when workflow engine is finished. IDM should not include its own workflow engine. IDM should cooperate with an existing engine. That is the right way to do it in 2020s.

Leave a Reply

Your email address will not be published.