Every time Argopy use the dm extension of the argo accessor and the merge method, which is basically the case for all DataFetchers, there is one specific line of code that is very slow:

idx = indexfs.copy(deep=True) if isinstance(indexfs, ArgoIndexStoreProto) else ArgoIndex()

The screenshot below illustrate this:

where we can see that 10% of the computation time is lost in this line, which appears way too much to me.

It would be nice to dig further in here to determine if it's the deep copy or the new ArgoIndex instanciation that is the culprit for this poor performance