can mllib Logistic Regression package handle 10 million sparse features?

[Yang]

anybody had actual experience applying it to real problems of this scale?

thanks