Skip to content
CSA Loom — the Microsoft Fabric experience for Azure tenants where Fabric isn't yet available: lakehouses, warehouses, notebooks, semantic models, Activator rules, Data Agents, across Commercial, GCC, GCC-High, and DoD IL5

Performance & benchmarks admin page

Surface: /admin/performance BFF: apps/fiab-console/app/api/admin/performance/{run,cache-stats,copilot-slo}/route.ts Store: Cosmos perf-benchmarks (PK /runId) — trend history retained across rolls

The Performance & benchmarks page runs the repeatable PSR-1 performance suite on demand and trends the results against the published Microsoft Fabric bars, so an operator can prove — with real numbers, not claims — that Loom meets or beats the Fabric-grade floor.

Each run measures p50 / p95 / p99 plus cold-vs-warm latency for the Loom-critical paths: Spark session attach, warehouse / ADX query, dashboard tile time-to-interactive, a Copilot turn (first token + full turn), and page TTI. All backends are Azure-native and the run executes against the live deployment — no synthetic timings.

What you can do

  • Run the suite — POST /api/admin/performance/run executes the probes and writes one perf-benchmarks row per (runId, metric); the trend chart plots successive runs.
  • Compare to Fabric — each metric is annotated with the published Fabric bar so regressions and wins are obvious.
  • Cache stats — /api/admin/performance/cache-stats surfaces the query / result-cache hit-rate feeding the warehouse & ADX paths.
  • Copilot SLO — /api/admin/performance/copilot-slo shows the Copilot turn latency budget (first-token + full-turn) and whether the current window is in budget.

Backend

Control Backend
Benchmark run lib/perf/* probes against the live Azure backends
Trend history Cosmos perf-benchmarks (PK /runId, no TTL)
CI regression gate .github/workflows/perf-gate.yml + perf-budgets.json
Optional export LoomPerf_CL Log Analytics DCR (honest-gated; off until LOOM_PERF_DCR_* set)

RBAC & honest gates

Runs as the Console UAMI with Monitoring Reader. When a probed backend isn't provisioned the metric shows an honest gate rather than a zero; the optional Log Analytics export stays a silent no-op until its DCR env vars are present.