Estimating the variance in the many Normal means model

Question

Is there a Bayesian way to estimate $\sigma^2$ given data $Z_i \sim N(\theta_i, \sigma^2)$, $i = 1, 2, \dots, n$? The MLE is $\hat\sigma^2 = 0$ which is unfortunate, so I'm considering a Bayesian approach. Wasserman's "All of Nonparametric Statistics" doesn't say much about it; all of Chapter 7 assumes $\sigma^2$ is known.