When do conditional probability measures not exist?

Question

Keener's Theoretical Statistics defines conditional distributions for random vectors $X$, $Y$ as follows.

The function $Q$ is a conditional distribution for $Y$ given $X$, written $Y \mid X = x \sim Q_x$, if

1. $Q_x(\cdot)$ is a probability measure for all $x$,
2. $Q_x(B)$ is a measurable function of $x$ for any Borel set $B$, and
3. for any Borel sets $A$ and $B$, $$ P(X \in A, Y \in B) = \int_A Q_x(B) dP_X(x). $$

He then includes a statement that I don't quite understand:

"When $X$ and $Y$ are random vectors, conditional distributions will always exist. Conditional probabilities can be defined in more general settings, but assignments so that $Q_x(\cdot)$ is a probability measure may not be possible."

What more general settings is Keener referring to where conditional probability measures may not be possible? $X$ and $Y$ being arbitrary random vectors already seems pretty general to me.