@westfox35 On paper, that math makes total sense! And you're right, OpenZFS defaults to 50% max. The hidden trap in production is allocation lag. When a VM suddenly demands physical memory pages, ZFS doesn't always prune the ARC fast enough to hand them over. The Linux kernel panics under that instant pressure, wakes up the OOM killer, and typically murders the largest user-space process (the VM) because it can't kill the kernel-level ZFS cache. That's why hard-limiting is a safety net!
