This should speed up push rule calculations for rooms with large numbers of local users when the main push rule cache fails. Co-authored-by: reivilibre <oliverw@matrix.org>