Foundation services

Bring-up JSON — Generate, Validate, Review

The VCF Installer can deploy from a JSON specification file instead of the wizard. This guide covers the step between the filled planning workbook and the first bring-up: producing that file, proving it is well-formed, and checking it against your plan before it touches the hosts. It is story 3.2 of the deployment plan.

Status: TechDocs- and OpenAPI-sourced, not lab-verified. The wizard and JSON pages are the 9.1 TechDocs (Last Updated October 26, 2026, read 2026-10-29). Field names and endpoints come from the 9.1.0.0 VCF Installer OpenAPI specification; the 9.1.1 bundle was not read, so check a field against the bundle that matches your Installer build (§9). Every API snippet is untested until someone runs it against a lab.


1. Where this sits

intake  →  workbook (story 3.1)  →  JSON (this page, story 3.2)  →  validate  →  bring-up

The JSON route has one requirement on the Installer itself: the 9.1 TechDocs say to “Verify that you deployed VCF Installer version 9.1.1 or later” before using a JSON specification file. The same page lists what the file can deploy: a new VCF fleet or instance, a new vSphere Foundation platform, deferred components, convergence of existing infrastructure, and VCF management services plus the License Server for vSphere Foundation.


2. Pick the route

RouteWhat you doGood forWatch out
A. Wizard, then downloadFill the wizard, download the JSON template or the spec from the review page, edit it, upload it from the Installer homepageA first deployment; a second one that is mostly a copy of the firstThe downloaded file has no passwords (§3)
B. Generate from the workbookRun VCF.JSONGenerator over the filled P&P workbookMany values already live in the workbook; repeatable runsThe module is tightly coupled to the workbook version (§4)
C. Hand-built from the API schemaWrite the file against the Installer’s SddcSpecA scenario neither of the others expresses; reviewing what A or B producedField names depend on the Installer version (§5)

They combine well: generate (B) or download (A), then read the result as a route-C reviewer would.


3. Route A: the wizard as a spec editor

All of this is [documented] on the 9.1 deployment wizard page and the JSON specification page:

  • On the wizard’s prerequisites screen you can download a JSON template “based on your choices in the VCF Installer so far”, populate it, and deploy with it.
  • On the review screen you can “download the JSON specification file, make the necessary changes, and then upload the modified JSON specification file from the VCF Installer homepage”.
  • Upload it with Deploy Using JSON Spec, review the Summary and JSON Preview, then the validation results (§6).
  • Passwords: after a deployment, “Download the JSON specification file. The JSON specification file does not include passwords.” Anything you re-upload needs its passwords put back by you. The Installer’s Review Passwords page, or Deployments under the user menu, shows them for past deployments.

4. Route B: generate from the workbook

VCF.JSONGenerator is Ken Gould’s cross-platform PowerShell module, a companion to the Planning and Preparation workbook: GitHub · PowerShell Gallery. Credit and thanks to the author; this repo only points at it.

What its README says [documented]:

  • Run it from the folder that holds the workbook: Start-VCFJsonGeneration, then choose the option that discovers and loads a workbook. Set-VCFJsonGenerationPrequisites detects and installs the dependencies (ImportExcel and VMware.PowerCLI or VCF.PowerCLI). Let it install VCF.PowerCLI, not VMware.PowerCLI next to an existing VCF.PowerCLI: two PowerCLI versions on one host break the import (42 §6.1).
  • “very tight coupling between the version of the Planning & Preparation file” and the module version. Use the module release that matches the workbook revision you filled in; the README names VCF 9.0, so confirm in the module’s own release notes that your 9.1 / 9.1.1 workbook is supported before you trust its output.
  • It only generates the files: it does not configure prerequisites and does not submit anything to an API endpoint. You still validate and upload.

The workbook fields it reads are the named ranges mapped in workbook-cell-mapping.md, which is why filling the workbook from the intake first (story 3.1) matters.


5. Route C: hand-built from the API schema

TechDocs sends you to the VMware Cloud Foundation Installer API Reference Guide for the properties, and offers example JSON files for new deployments, convergence, deferred components and the vSphere Foundation services install.

What the schema says (9.1.0.0 specification, SddcSpec) [spec]:

  • Required at the top level: dnsSpec, networkSpecs, sddcId, vcenterSpec.
  • workflowType is one of VCF, VCF_COMPLETE, VCF_EXTEND, VVF, VCF_BOOTSTRAP. For a secondary instance joining a fleet it is VCF_EXTEND.
  • Component blocks: nsxtSpec, vcfOperationsSpec, vcfOperationsCollectorSpec, vcfAutomationSpec, vspClusterSpec (VCF management services), licenseServerSpec, vidbSpec (Identity Broker), fleetLcmSpec, sddcLcmSpec, fleetDepotSpec, datastoreSpec, dvsSpecs, hostSpecs, clusterSpec, securitySpec and others.
  • Flags worth deciding on purpose rather than inheriting from a template: ceipEnabled, skipEsxThumbprintValidation, skipGatewayPingValidation.

Which blocks you need depends on the scenario. The TechDocs table, condensed to the properties it names (values as the page gives them; useExistingDeployment is the property on each component block):

I want to…workflowTypevcfOperationsSpec
useExistingDeployment
vcenterSpec
useExistingDeployment
vspClusterSpec (VCF management services)vidbSpec (Identity Broker)
Deploy a new VCF fleetVCFfalsefalsetruetrue
Deploy a new VCF instanceVCF_EXTENDtruefalsetruetrue / false (mandatory for the primary instance only)
Deploy deferred componentsVCF_COMPLETEtruefalsefalsen/a
Deploy a new vSphere Foundation platformVVFfalsefalsetrue / false (see below)true

The page has more rows (convergence, the VVF services install, existing NSX and VCF Automation); read the table there for those.

vspClusterSpec takes an address pool three ways [documented]: a CIDR (ipv4Pool.cidr), a range (ipv4Pool.ipRange.startIpAddress / endIpAddress), or a list (ipv4Pool.addresses), each with optional excludedAddresses, plus internalClusterCidrIpv4. The services runtime’s internal block is 198.18.0.0/15; “If this subnet is currently in use on your network, you must change the internal CIDR block … to 240.0.0.0/15 or 250.0.0.0/15”. The other internal blocks are in 34-internal-cidrs.md.

Without VCF management services (VVF) the page requires one change on the Installer appliance first (explicit.management.components.deployment=true in /etc/vmware/vcf/domainmanager/application.properties, then restart domainmanager); use the exact commands on the page for your build.

Use the version-pinned spec bundle for field names, not the “latest” page on developer.broadcom.com, which can silently show an older major line. The bundle is a download behind Broadcom’s SDK licence agreement; do not commit it to a repository.


6. Validate before bring-up

In the Installer

After upload, the Installer “validates the information that you provided in the JSON specification and reports any errors or warnings” [documented]:

  • Warnings — address them or Acknowledge to proceed.
  • Errors — resolve them with the remediation text in the message, then “modify and reupload the JSON specification file”. Repeat until clean, then Deploy.

Through the API (before you open the UI) — untested

The 9.1.0.0 specification offers a dry run [spec]. Authenticate, submit the same body the UI would, and poll the result:

# UNTESTED. Windows PowerShell 5.1-safe (no -SkipCertificateCheck): trust the
# Installer's certificate on this machine first, or run it from PowerShell 7
# and add -SkipCertificateCheck to each call.
$installer = 'https://vcf-installer.sfo.example.io'
$cred  = Get-Credential -Message 'VCF Installer admin@local'
$login = @{ username = $cred.UserName
            password = $cred.GetNetworkCredential().Password } | ConvertTo-Json
$token = (Invoke-RestMethod -Method Post -Uri "$installer/v1/tokens" `
          -ContentType 'application/json' -Body $login).accessToken
$h = @{ Authorization = "Bearer $token" }

$spec = Get-Content -Raw -Path .\bringup.json          # keep this file private, section 8
$val  = Invoke-RestMethod -Method Post -Uri "$installer/v1/sddcs/validations" `
          -Headers $h -ContentType 'application/json' -Body $spec
Invoke-RestMethod -Uri "$installer/v1/sddcs/validations/$($val.id)" -Headers $h |
  Select-Object id, executionStatus, resultStatus

What the specification says to expect:

  • A malformed body is refused at submission (the endpoint lists 400 among its responses). That is the schema failing: a wrong or unknown field, a missing required block. Fix the file; nothing was checked about your environment yet.
  • An accepted body returns 202 with a Validation, which you poll at GET /v1/sddcs/validations/{id}. executionStatus is IN_PROGRESS, FAILED, COMPLETED, …; resultStatus is SUCCEEDED, FAILED or WARNING once it has completed. Each entry in validationChecks carries a severity (ERROR, WARNING, INFO) and, for warnings, whether it needs acknowledgement. That is the content failing: DNS, addresses, hosts, credentials — the remediation text names the fix.
  • Related endpoints in the same specification: POST /v1/sddcs/resources-calculation (the infrastructure the spec would need, useful next to 04-sizing.md) and GET /v1/sddcs/{id}/spec (the input spec of an installation that has already started).

For the fake-UUID trick that proves a spec’s shape without real hosts (used for domain and cluster specs), see 21-config-artifacts.md §4. It is not established for the bring-up endpoint; use the Installer’s own validation.


7. Review against the plan

Machine validation proves the file is consistent, not that it is your design. Read the file against the plan:

CheckAgainst
Every FQDN and IP in the file01-network-dns-plan.md: forward and reverse DNS resolve, addresses sit in the planned subnets
NTP and DNS serversThe same plan; every component gets the same time source
Deployment model and sizes04-sizing.md and the sizer result you signed off
Services-runtime pool and internal CIDR30-ops-logs-sizing.md, 34-internal-cidrs.md: room for Day-N workers, no overlap with a routed network
Principal storage type and datastoreThe workbook / intake answer; the datastore exists and is mounted where the type requires it
NSX connectivity and VPC choicesThe management-domain answers in 02-intake.md
Which components are existing vs newThe scenario table in §5
PasswordsPresent for every component the Installer asks for, stored in your secret store, meeting the rules the Installer reports
License Server, Identity Broker, CEIP, skip flagsDeliberate values, written down

If you generated the file (§4), a quick way to find a wrong cell is to diff the JSON against the workbook value by value for the first run; later runs only need the changed rows.


8. Handle the file as a secret

The spec you upload holds every bring-up password, plus FQDNs, addresses and account names: the same sensitivity as a filled planning workbook.

  • Keep it in an encrypted, access-controlled store, never in a public or shared repository, and never in a chat or ticket. Put its folder in .gitignore.
  • The Installer’s download has no passwords (§3), so the downloaded copy is the one to keep as a template; keep the filled copy only as long as you need it.
  • Repeat runs: re-upload the edited file from the Installer homepage rather than retyping the wizard.
  • More on handling captures and templates: 21-config-artifacts.md §5.

9. Sources

ITQ – What's Next. VCF 9.1 deployment-planning and operations helper material. © 2026 Paul van Dieen. hollebollevsan.nl