There are different ways to solve this.

• Pure FME: Put the DateTimeStamper in the very beginning of the workspace, initiated by a Creator. Then you have only one datetime at the start of the workspace. You can then continue the workspace using FeatureReaders to merge this attribute to all features. As an alternative, you can add this attribute to all feature just before writing using a FeatureMerger set with Requestor = 1 and Supplier = 1.
• Python: Use a Scripted User Parameter. This is a easier way to get it done and will be faster as not all features will have to drag the datetime attribute through the process. Working sample:

import datetime 
return datetime.datetime.now().strftime("%Y_%m_%d__%H_%M_%S")

import datetime 
return datetime.datetime.now().strftime("%Y_%m_%d__%H_%M_%S")

I generally use (and recommend) the Pure FME option,  but I like the Python one too, I'm going to try that out!

Here's another pure FME pattern that avoids using a FeatureMerger. Only downside is that the VariableSetter/Retriever doesn't always play nice with feature caching.

My preferred solution is using the Python scripted parameter, however. It's very efficient, as it's only evaluated a single time when the workspace starts.

import datetime 
return datetime.datetime.now().strftime("%Y_%m_%d__%H_%M_%S")

Creator -> VariableSetter -> FeatureReader -> VariableRetriever is my solution

Thanks all!

I totally forgot about the ability to set a global variables within a workspace, this is what I ended up going for.

The DateTimeStamp transformer was an option as well, but as I had multiple output streams within the workspace that already required a different DateTime for their output name I was hoping not to clutter the workflow with more joins than required.