Been running Traceloop on our production K8s clusters for about 8 months now. Mainly for OpenTelemetry trace-based cost allocation and to track some internal SLAs.
The good:
* The OpenTelemetry pipeline it sets up is solid. No complaints there.
* The `traceloop` CLI is useful for ad-hoc queries when you need to trace a specific workflow's cost or performance.
* The automated span attributes for cost (like `gcp.project.id`) save a ton of manual tagging.
The bad:
* The UI feels slow, especially for complex queries over a 7-day window. We often fall back to CLI or direct TraceQL.
* The "AI-powered" root cause stuff is noise. We turned it off. It just creates alert fatigue with vague correlations.
* Pricing got opaque after the initial pilot. Be prepared to negotiate based on span volume.
Our config is pretty minimal, just the auto-instrumentation and some custom business attributes:
```yaml
traceloop:
autodetect:
kubernetes: true
clouds: [gcp]
attributes:
- business_unit: "team-platform"
- cost_center: "cc-2024"
```
Biggest pitfall: It's not a replacement for your existing APM (we use Tempo). Think of it as a specialized layer on top. If your tracing isn't already mature, you'll get little value.
So, who else is in production? What's your actual retention rate on the insights? We find we only act on about 20% of what it surfaces.
Benchmarks > marketing.