[mindhash]

I worked for an infra provider. We used token buckets for network usage. We didn't have actual cpu usage but you can use zabbix for that. Memory is hard to capture because of paging etc

For database use, it's better to create separate dB if you want to charge by storage. If the tables are shared, use the (count of rows for customer/ total rows)* total table storage. Ideally this should be possible at tablespace level