completing missing data with knn

What it means is that almost all examples have missing values. The way the impute operator works is: it takes an attribute with missing values and treats it as a label, then it uses the remaining attributes as predictors. It then trains a model, in your case knn, using all examples where predictors have no missing values, the examples with missing values are rejected. Once the model is trained it is used to predict the missing values in the label. This is repeated for each attribute with missing values (to become a label). If at any stage your predictors have too many missing values, you may end up with no examples to train the moeel on, which happened in your case. 

To verify this out a break point before knn and see what you get on input.