Unsupported change: Broadcom currently says that resource resizing for VCF 9.0.2 Automation components is not supported. Their KB is here: Broadcom KB 412468. Do not treat this as production guidance. This was a home lab experiment.
I wanted to see whether the VCF Automation appliance in my lab could be reduced from the default sizing. The appliance was originally running with 24 vCPU and 96 GB RAM. At idle it was not using anything close to 24 vCPU, but the memory request profile inside Kubernetes was a different story.
What did not work
8 vCPU / 64 GB RAM failed quickly. Kubernetes reported multiple pods as OutOfcpu, including core prelude services. That was enough evidence to stop testing that size.
12 vCPU / 64 GB RAM also failed. CPU was no longer the main problem, but memory was. The node was already at about 95% of requested memory before tenant-manager-0 could schedule. That pod requests another 8 GiB, so it stayed pending with insufficient memory.
What did work
12 vCPU / 96 GB RAM did work in my lab. After resizing and clearing stale Unknown pods left by the reboot cycle, the appliance settled with 133 running pods and 27 completed pods. There were no remaining non-running pods, and no OutOfcpu or memory scheduling failures.
At the end of the test, the node was using roughly 1.4 vCPU and 36.9 GiB RAM. That makes the CPU reduction useful for a home lab where vCPU count and CPU ready time matter, but it does not support reducing RAM to 64 GB for this deployment.
In Summary
Bottom line for my lab: 12 vCPU and 96 GB RAM appears to be a workable unsupported lab configuration. 8 vCPU / 64 GB and 12 vCPU / 64 GB did not work. For anything supported, leave the appliance at the documented size and follow Broadcom KB 412468.