In 10 years of running a production service, the only two long duration outages were due to btrfs quotas (unwanted in the first place) slowing things to a crawl.