Skip to content

📊 Observability Dashboard: The Nerve Center 🧠

The FlowyML Dashboard is a professional-grade, interactive command center designed to give you X-ray vision into your ML infrastructure. Whether you are tracking a single hyperparameter sweep or managing thousands of production pipelines, this is where you monitor the "heartbeat" of your system.

🚀 Real-Time HUD

The "Heads-Up Display" (HUD) auto-refreshes every 30 seconds. No manual reloads. See runs as they happen.

🌙 Premium Dark Mode

Designed for high-contrast visibility. Our Slate Night theme reduces eye strain during long debugging sessions.

📈 Live Visualizations

Interactive Gantt charts, Lineage Trees, and Drift Overlays are generated instantly from the metadata store.

📊 Core KPI Cards

At the top of the dashboard, you'll find the Nerve Center Metrics. These are your high-level health indicators:

KPI Technical Meaning Metric Optimization
Total Runs Volume of pipeline executions in the current cycle. Use for capacity planning.
Success Rate The ratio of success status vs. failed. Critical: Investigate if < 95%.
Avg Duration Mean execution time from start to finalize. Optimize if trending upwards.
Cache Efficiency Percentage of steps skipped due to content_hash matches. Goal: > 40% for faster iteration.

🕵️ Lineage Explorer (Artifact Tracking)

Because FlowyML is Artifact-Centric, the dashboard allows you to click on any data asset (Dataset, Model, Metric) and see its "DNA":

  1. Immutability Checksum: The SHA-256 content hash of the data.
  2. Parentage: A visual graph showing exactly which raw data and which code version produced this specific model.
  3. Metadata Inspection: View Pydantic-validated hyperparameters, training metrics, and environment tags.

⏱️ Execution Timeline (Gantt View)

Every pipeline run features a high-fidelity Waterfall Timeline.

  • Parallel Streams: See which steps ran concurrently.
  • Critical Path Analysis: FlowyML highlights the longest-running chain of dependencies so you know exactly where to optimize.
  • Cold-Start Latency: See precisely how long container pull and environment setup took vs. actual code execution.

🤖 GenAI & LLM Traces

The traces tab provides a dedicated view for LLM observability:

  • Waterfall Traces: Full nesting of chains, retrievers, and LLM calls.
  • Token Economy: Individual token counts (Input/Output/Total) and calculated $USD cost per call.
  • Prompt Benchmarking: Compare side-by-side completions for different prompt versions.

🔌 Programmatic Integration

You can extract all dashboard data into your own BI tools (Grafana, Datadog) or CI/CD pipelines.

1
2
3
4
5
6
7
import requests

# Fetch the current health of the 'retraining-v2' project
resp = requests.get("http://localhost:8080/api/metrics/observability/orchestrator")
health = resp.json()

print(f"Project Health: {health['success_rate']:.0%}")

💡 Pro Tips

Performance Monitoring

If your Success Rate is high but Avg Duration is increasing, use the Gantt View to check for "Storage Bottlenecks" (slow artifact downloads).

Lineage as Documentation

Never write a "Model Card" manually again. Export the Lineage Tree directly from the UI as a PDF to document your model's provenance for compliance audits.