bzm-vitals-dashboard

bzm-vitals-dashboard

Generator, not viewer: point it at a BlazeMeter master (Report) id and get one self-contained static HTML file with the data baked in. This package currently implements issue #2 — the fetch/extract/cache layer the later views build on.

The fetch/cache script

scripts/fetch-cache.ts is the only thing that talks to the live BlazeMeter API. It walks the API chain for a master id and writes every Engine's artifact to a local cache, extracted and namespaced by sessionId. The test suite reads that cache and never touches the network; the script is run by hand and fails loudly.

BLAZEMETER_API_KEY=./api-key.json \
  npx tsx packages/dashboard/scripts/fetch-cache.ts <masterId> [<masterId> ...]

It prints, per Engine: sessionId, locationId, status, and either artifact present — N files or NO ARTIFACT (an Engine that terminated ungracefully emits no zip — a normal outcome, not an error).

What it writes

.artifact-cache/{masterId}/manifest.json     session list + per-Engine outcome
.artifact-cache/{masterId}/{sessionId}.zip   raw fetched bytes
.artifact-cache/{masterId}/{sessionId}/...    extracted, namespaced by sessionId

Cross-Engine attachment basenames are identical by construction (basename is sha1 of the fixed absolute path Taurus passes to every Engine), so extracting two Engines into one directory would silently destroy half the Samples. Namespacing by sessionId is what prevents that.

Re-running is cheap

A second run against a cached master does zero network calls: the manifest is trusted and extraction re-runs from the cached raw zip. The raw bytes are cached alongside the extraction on purpose, so an extraction-logic change never forces a refetch. Delete .artifact-cache/{masterId}/ to force a fresh fetch.

Credentials — BlazeMeter's own convention

export BLAZEMETER_API_KEY=./api-key.json        # {"id": "...", "secret": "..."}

# CI — a file is awkward, so discrete vars are accepted:
BLAZEMETER_API_KEY_ID / BLAZEMETER_API_KEY_SECRET

The file form wins when both are present. Credentials authenticate only the two listing calls; the artifact download URL is pre-signed and needs none.

api-key.json is gitignored and must stay so — a stray git add -A must never commit a live key.

The cache is gitignored and re-fetchable

.artifact-cache/ is gitignored. Artifacts are a re-fetchable resource, not a committed corpus — archiving is a storage tier, not an expiry, so a Report's zips stay reachable long after the run. Anyone with account access can repopulate the cache. The one committed exception is the stripped probe collision fixture in test/fixtures/probe/ (see its README).