Monitor Endpoints

Track custom metrics from your deployment endpoints by printing JSON metadata. Valohai automatically collects and visualizes these metrics.

Collect metrics

Print JSON with vh_metadata to log metrics from your endpoint:

import json

print(
    json.dumps(
        {
            "vh_metadata": {
                "accuracy": 0.9247,
                "best_guess": "dog",
                "confidence": 0.87,
            },
        },
    ),
)

Valohai parses this output and makes it available for visualization.

View metrics

Access deployment metrics from your deployment version:

1. Open the Deployment tab

2. Select your deployment

3. Click on a version

4. View metrics in the Monitoring tab

Available visualizations:

• Time series charts (metrics over time)

• Histograms (distribution of values)

• Request logs (individual predictions)

Common metrics to track

Model performance:

print(
    json.dumps(
        {
            "vh_metadata": {
                "prediction_confidence": 0.94,
                "processing_time_ms": 23,
            },
        },
    ),
)

Business metrics:

print(
    json.dumps(
        {
            "vh_metadata": {
                "fraud_detected": True,
                "transaction_amount": 1250.00,
            },
        },
    ),
)

System metrics:

import time

start = time.time()
# ... inference code ...
duration = time.time() - start

print(
    json.dumps(
        {
            "vh_metadata": {
                "inference_latency_ms": int(duration * 1000),
            },
        },
    ),
)

View endpoint logs

Access full logs for debugging or detailed inspection:

1. Navigate to your deployment version

2. Select an endpoint

3. Click Log tab

4. Filter by time range or search for specific events

---

Next: Learn how to troubleshoot failing endpoints.