Previously, I thought this was only a problem on mongod. However, it seems like there's a race condition on mongos where audit_synchronize_job is executing and fetching the old audit config from the cluster, and right after the cluster parameter refresher is executing and fetches the new audit config, but the refresher finishes first and the old audit config is then set and returned by getAuditConfig. We should block audit_synchronize_job from running on transitional cluster versions to prevent this.