Repository navigation
Description
Summary
Investigation into a "true full offline mode" (local HCL for all device classes, not just vSAN storage). Two candidate data sources were examined against what the tool actually needs. Conclusion: neither enables offline hardware HCL. Filing so the findings are visible and the research isn't repeated.
Background
Today -offline verifies only storage controllers / vSAN SSD-NVMe against the local vSAN offline DB (vsan-offline-hcl.json). Everything else (system chassis, CPU, NICs, other I/O, GPUs) is live-API-only and becomes SKIPPED offline. Full offline mode needs a local, authoritative source for those classes, keyed by the identifiers the tool collects (PCI VID/DID/SVID/SSID, CPUID, model).
Candidate 1 — VCF offline-depot VmwareCompatibilityData.json
From a VCF offline depot at …/PROD/metadata/Compatibility/v2/VmwareCompatibilityData.json (basic auth; would be a user-supplied depot URL + credentials). ~28 MB.
It is a product interoperability matrix, not a hardware HCL. Self-identifies as bundleType.name = "vcf-interop-bundle", type = "productsBundle", params.include = "interop,upgrade".
- Structure:
products[] → releases[] → { interops[], upgrades[] }. 11 top-level products (ESX, vCenter, NSX, VCF Operations, HCX, SDDC Manager, …), 768 releases, ~94.5k interop edges, ~7.1k upgrade edges. - The entire file contains only six array keys:
products,releases,interops,upgrades,footnotes, anddlGroups(always empty). Nowhere for hardware data to live. - Exhaustive search for
vid/did/svid/ssid/cpuid/gpu/driver/firmware/model/pciid/… → zero occurrences. No IO devices, CPUID, GPUs, server models, or SSD entries.
Verdict: closes none of the offline gaps (io/server/cpu/ssd/gpu). Useful for a different future feature: it's an authoritative version ↔ gaBuildNumber ↔ releaseApiVersion map and would enable an ESXi ↔ vCenter ↔ NSX interop / upgrade-path check — fully offline-able from this file, but that's interop, not HCL.
Candidate 2 — "IO Devices" guide CSV export
A manual CSV export of the I/O Devices compatibility guide (~1.2 MB, ~7.7k device rows).
It's a lossy projection that drops the join key. Columns: Brand Name, Model, Device Type, Supported Releases, _ (blank/Restricted).
- No VID/DID/SVID/SSID, no driver, no firmware. No key to join the tool's collected PCI IDs against, and driver/firmware certification is simply absent (the tool produces both
hw_certifiedanddriver_certified). - 261 groups of rows are byte-for-byte identical across every exported column (e.g.
Intel … PRO/1000 PF Dual Port×4) — distinct HCL entries differing only in the dropped PCI-ID/driver columns. - Also not valid RFC 4180: a model name contains an unescaped inner quote (
2.5") that crashes a strict CSV parser (needsLazyQuotes).
Verdict: supports at most "some device with this model string is listed on ESXi X" — never a PCI-ID-keyed, driver-aware verdict. The needed data lives on the guide's per-productId detail pages (which carry PCI IDs + per-release driver/firmware), not the search-results export.
Possible paths forward (none built)
- Check the guide's export for a column chooser / "export all fields" including VID/DID. Cheapest decisive test. If present, a PCI-ID-keyed offline DB from CSV becomes viable.
- Harvest the per-
productIddetail pages into a bundled DB — real scraping, with maintenance + terms-of-use questions. The only path seen that yields driver/firmware-aware offline data for io/gpu. - GPUs are already close to offline-able: the GPU guide (
program=sptg) is only ~63 rows and is fetched whole (as of v0.5.3). Persisting that fetch would give an offline GPU DB almost for free — a plausible first step.
Absent one of these, full offline HCL is not achievable from the files examined, and offline mode stays vSAN-DB-only by design.
Filed as a research/findings record, not a bug. No code change proposed here.
Activity