What is mean accuracy and why is it a harsh metric for multi-label validation?

The score method docs for scikit-learn's SGDClassifier have the following description:

Return the mean accuracy on the given test data and labels.

In multi-label classification, this is the subset accuracy which is a harsh metric since you require for each sample that each label set be correctly predicted.

What is meant by the terms mean accuracy and subset accuracy?

Could you please elaborate on why this is a harsh metric, perhaps with an example?