If $\eta$ is a Grassmann variable, due to invariance under translations we get that,
$$\int d\eta\ \eta = 1 \tag1$$
Nevertheless, for being Grassmann's, $\eta$ satisfies $\eta^2 = 0$. Differentiating this condition you get,
$$d(\eta^2) = 2\eta d\eta \equiv 0 \Rightarrow \int d\eta\ \eta = 0 \tag2$$
So, Eq. (2) obtained just via definition of Grassmann variable goes against Eq. (1) that comes out from translation invariance. But I've seen the use of Eq. (1) in all books about fermions' path integral, so what is the thing that I'm misunderstanding?