My early VCF 9 migration plans opened with the maintenance window schedule. It felt natural. Windows are what clients ask about first, and they're the part of the plan with a calendar attached.
The plan I'd write now opens with a spreadsheet of hosts and four columns: ESX build, lifecycle method, forward and reverse DNS, storage type. No dates anywhere on it. Whatever that sheet turns up decides whether the windows hold, and I've stopped pretending otherwise.
Inventory Before Schedule
Every blocker I've written about in this series is knowable before anyone touches a host. Clusters still on VUM baselines have to convert to vLCM images before they can be imported or converted, and for a 5.2 to 9.x path the ESX upgrade step depends on that conversion too. Standard vSwitch and Nexus 1000V are out; DVS only. Enhanced Linked Mode has to be split into single-vCenter domains. Storage type can't change after commissioning, and the ESX build has to match the VCF Bill of Materials, not come close to it. Missing PTR records stop the installer cold.
None of that requires a maintenance window to discover. It requires someone with read access and an afternoon. I used to find these during the work, under a clock. Now I find them in week one, when the only cost is an uncomfortable email.
Decide the Org Model on Paper
An upgrade from Aria Automation lands you in a VM Apps organization whether anyone chose it or not. If Kubernetes is on the roadmap at all, or ServiceNow is part of how requests get fulfilled, that choice deserves an actual meeting before the upgrade, because on 9.0 a vCenter belongs to one org type and changing your mind later means new infrastructure.
Stop Trusting Checklists That Don't State a Patch Level
Import checklists labeled only 9.0.x can disagree about whether NSX should be registered before a convert operation. Neither is wrong. They were captured at different patch levels, and the label never changed. I lost time on that exactly once before I started writing the source build at the top of every plan page, above the title.
The same discipline applies to upgrade sequencing. That order is documented at different levels of granularity depending on the document, and I no longer reproduce it from memory. I pull it from the runbook for the exact target version, every time, and paste the page reference into the plan.
A checklist without a version on it is a rumor with formatting.
Give Logs and Licensing Their Own Line Items
Aria Operations for Logs has no in-place upgrade path, so it gets a line item with its own owner, not a footnote under "management components." The delivery workbooks budget roughly eight hours for it: two to deploy, four to move data, two to verify. That's a baseline, not a promise for your log volume, but it's a lot better than zero. Licensing reverts to evaluation after every upgrade path, so relicensing goes on the calendar as a dated task with a name next to it, well inside the 90-day window.
Write Down the Things Nobody Asks For
Some items never come up in a kickoff and bite later. I put them in the handoff document now, in plain language, so the person operating the environment in a year isn't reverse-engineering them.
Handoff Items I Now Write Down Every Time
- NSX Global Manager certificate thumbprints have to be recaptured by hand after any Local Manager certificate rotation done through SDDC Manager. Miss it and Global-to-Local trust fails quietly.
- Global Manager sits outside SDDC Manager's lifecycle and certificate automation. Both are manual tracks.
- Edges deployed outside SDDC Manager still get patched, but their passwords can't be managed centrally afterward.
- Which functions have moved from SDDC Manager into VCF Operations for the version actually deployed.
What I Still Don't Have a Good Answer For
Timing. I can tell a client what the prerequisites are and roughly how long each piece takes, but the decision about when to start is mostly a licensing and support-window conversation, and that one has more to do with their contracts than my plan. Some environments I'd tell to move now. Some I'd tell to wait a year and be happier for it. I don't have a formula for which is which, and I'm wary of anyone who says they do.
Questions I Ask in Week One
Which exact build is each component on today, and where is that written down?
Which clusters are still on VUM baselines, and who owns converting them?
Who is relicensing this environment, and on what date?
If Logs history matters to anyone, who is moving it?
If you're staring down a VCF 9 migration and want a second set of eyes on the inventory before the plan gets dates attached to it, send me a DM. That's usually the cheapest hour in the whole project.