Part 1’s design only helps if it runs every minute, for ever, and says when it stops. Self-monitoring through Telegraf, batched pushes, the four alerts for day one, and an offline test mode.
vLLM exposes about 200 Prometheus series per model, and pushing them raw into VCF Operations is a cardinality bomb. A pipeline that computes the useful numbers client-side and drops what can’t be graphed.
A self-contained HTML build report, down to a Gantt chart that spots the reboots. Plus the Session 0 traps, like mangled vmtoolsd arguments and ejecting an ISO with no Explorer, that cost real hours.
One script runs at every startup until the build is done. Flag files make each step run once, a state file keeps true timings across reboots, and at the end it cleans up, reports and deletes itself.
One catalog request, one fully built, domain-joined Windows Server 2025 with its agents, a validation report, and no trace of the tooling that built it. The three-layer design, and why each seam is where it is.
A 400-line menu-driven PowerShell script became a cloud template, two vRO actions and two subscriptions. Where each behaviour belongs, four non-obvious decisions, and the ‘yes’ that isn’t ’true’.