8000 Cross-host dedup of Broadcom HCL queries · Issue #15 · dominikzorgnotti/esx-hcl-check · GitHub
[go: up one dir, main page]

Skip to content

Cross-host dedup of Broadcom HCL queries #15

Closed

Description

@dominikzorgnotti

Summary

queryBroadcomAPI runs per device, per host. Identical hardware across a cluster (e.g. the same NIC/HBA/CPU/server model on 50 hosts) issues up to 50 identical network calls to Broadcom — a real rate-limit and wall-clock cost at scale. The per-host pciMap only dedups within a host.

Approach

Memoize the Broadcom baseline lookup across hosts within a single run, keyed by the query inputs (programId, filters, keywords, targetRelease). The API result is a pure function of those inputs, so this is exact caching, not a heuristic.

  • Add a queryCache (map + mutex; the mutex readies it for the concurrency work in the next item) that wraps queryBroadcomAPI
  • Route all call sites (server, cpu, io PCI, vsan disk + its filterless retry) through the cache
  • Do not cache CertError — a transient failure on the first host must not poison every identical device in the run; let later occurrences retry

Firmware/driver certification is unaffected: it comes from the local vSAN offline DB (evaluateVsanPCI/evaluateVsanDisk), which is per-host and involves no network call. So no per-host values are lost — this sidesteps the roadmap's stated "reapply per-host firmware/driver" risk.

Acceptance criteria

  • Identical hardware across N hosts triggers one Broadcom baseline call, not N
  • Output is unchanged vs. today (same verdicts per host)
  • A transient CertError is not cached
  • Non-breaking; no schema change

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions

      0