Skip to content

Section 4 — Monitor in Real Time

Monitoring isn’t an afterthought — it’s built into every part of the SkyPortal workflow.

Observability Dashboards

As soon as a job runs, you get:

  • Training metrics: loss, accuracy, MAE, MSE
  • System metrics: GPU utilization, CPU load, memory, I/O
  • Budget insights: cost per run, warnings on overspend

Log Streams

All log output — stdout, stderr, system events — streams in real time.

Alerts

SkyPortal doesn't send metric alerts. Ask SARA what changed instead: it compares recent metrics with the previous 15 minutes. See Monitoring for what is collected from hosts and clusters.

Advanced Views

  • Multi-host overviews across clouds
  • Historical comparisons between runs
  • Experiment tracking tied to parameters, datasets, and outcomes

This consolidated observability eliminates the need for external dashboards or tools.