We changed the causal_clustering.log_shipping_max_lag from 256 to 512 and that seems to fixed the issue.