According to the the notes in the code of BSplineApprox, the option knotVecType = :Uniform is not recommended if it is used with the chord length method for global interpolation. However, the alternative knotVecType = :Average does not produce a correct set of knots. Namely:

julia> x = 0:0.1:10;

julia> y = randn(101); # or anything else

julia> sp = BSplineApprox(y, x, 3, 20, :ArcLen, :Average)
BSplineApprox with 101 points, with degree 3, number of control points 20

[... snip ...]

julia> sp.k
24-element Vector{Float64}:
 0.0
 0.0
 0.0
 0.0
 0.041544922773446444
 0.05312314970883667
 0.062325767284134394
 0.07163762461960164
 0.0816032331679999
 0.09584179757523949
 ⋮
 0.16196790599023062
 0.16692261841638856
 0.17258753155720719
 0.17997404351857807
 0.18633410292440233
 0.19189122073158726
 1.0
 1.0
 1.0
 1.0

As can be seen, the "internal" knots start at 0.04 (close to the starting point), but end at 0.19 (very far from the endpoint that is 1.0). Therefore, all the knots are concentrated at the beginning of the fitted series, instead of spread along the whole domain of x.

This is happening with DataInterpolations v8.1.0, in a clean environment.