TraceLoop

Metrics

Numbers that know what normal looks like.

Infrastructure saturation, application golden signals and custom business KPIs — ingested at 1.2M points per second, baselined per metric, and alerted on only when the deviation actually matters. Threshold fatigue ends here.

Golden Signals

The four numbers, live from the checkout-platform workspace

Fictional data, real interface. Every card drills into full-resolution history with deploy markers, related traces and the logs from the same window.

requests / sec

8,412

▲ 4.2%

p99 latency

142ms

▼ 8ms

error rate

0.12%

stable

cluster cpu

46.8%

▲ 2.1%

throughput by hour

requests · last 24h · aggregated 1h

0006121824

memory by node pool

GB used · 96 GB pool budget

HEALTHY
pool-general61%
pool-compute74%
pool-memory52%

Autoscaler forecast: pool-compute scales 8 → 10 nodes at 19:30 UTC peak.

Anomaly Detection

Detectors that learned your seasonality

Every metric gets a per-series baseline across daily and weekly cycles. Detectors fire on scored deviations — and attach the correlated deploy, trace and incident automatically.

detector board

seasonal baselines · checkout-platform · scoring every 60s

3 ACTIVE
  • FIRING

    checkout-svc · p99 latency

    2.4σ above seasonal baseline for 12m · linked to deploy v2.41.3

    score 0.91
  • WATCHING

    postgres-01 · deadlock rate

    New pattern since 14:20 UTC · correlated with INC-4417

    score 0.82
  • WATCHING

    redis-cache · evictions/sec

    2.1σ above baseline · memory pressure on shard 2

    score 0.64
  • NORMAL

    api-gateway · rps

    Within expected evening peak envelope

    score 0.18

Capabilities

A metrics engine built for real production, not demos

High-resolution, long retention

10-second resolution for 48 hours, 1-minute for 90 days, pre-aggregated years. Zoom from a quarter view to a single spike without losing fidelity.

Seasonal anomaly detection

Per-metric baselines learn your daily and weekly seasonality, then flag deviations with an anomaly score — no thresholds to babysit, no static noise.

Cardinality without fear

Label sets that would melt a time-series database are handled by columnar storage with adaptive rollups. Tag by user tier, region and build SHA freely.

Cost-aware ingest

See exactly which emitters drive your bill, down to the metric name. Pre-aggregate at the edge and drop what nobody queries.

Your metrics already know the answer

Give them a platform that listens. Baselines, detectors and cost-aware ingest — live in an afternoon.