Capacity is just arithmetic (done early)

Every VM was sized by doing the math first, against the one resource that actually runs out here: memory.

• CPU overcommits gracefully; disk is plentiful; RAM is the hard wall, especially with no swap on the host (the right call for a hypervisor, but it removes your safety net).
• Hidden consumers count: an uncapped ZFS ARC will quietly eat the budget you meant for VMs. Cap it, then plan against what's left.
• Most VMs are over-provisioned out of habit. Half this lab's growth was paid for by reclaiming RAM that was allocated but never actually used — and "used" memory on Linux lies, because free RAM becomes cache. Size against available, not used.

The mindset: "can we fit this?" should be answerable with arithmetic before you click create, not discovered by an OOM kill afterwards. Find your binding constraint, account for the hidden tenants (caches, the host itself), leave a margin, and never trust the headline "used" number. Capacity planning is unglamorous and it's most of the job.