Skip to main content

Application metrics

An application's Metrics tab shows what it uses over time: CPU, memory, and network in and out, over the last hour, 24 hours or 7 days. Use it to see whether an application is sized right and how it behaves under load. The cluster's Astro Agent reports it; nothing else needs installing. A model also shows latency and tokens: Operate a model.

Before you start​

  • An application AstroPulse deployed, on a cluster with a connected agent.

In the console​

  1. Open the application from Workloads > Applications and select Metrics.

  2. Pick Last Hour, Last 24 Hours or Last 7 Days. Resources charts its CPU, memory and network. An application on GPUs also has Accelerator, with the Accelerator Metrics add-on.

    Open zoom view

    An application's Metrics tab: CPU, memory and network over the selected window

    An application's Metrics tab: CPU, memory and network over the selected windowAn application's Metrics tab: CPU, memory and network over the selected window

You should see​

Figures within a minute or two of the application starting. A window with partial data is drawn dashed, and a window with nothing to measure leaves a gap, never a zero: a flat zero would read as an outage.

Open zoom view

A running application's Metrics tab with CPU usage, memory usage and network I/O over the last hour

A running application's Metrics tab with CPU usage, memory usage and network I/O over the last hourA running application's Metrics tab with CPU usage, memory usage and network I/O over the last hour

Reading the numbers​

CPU and memory are shown against what the application requested. Used well below requested marks it as a candidate for smaller requests; check its peaks and autoscaling before you change them. What its requests cost is on its Cost tab: Cost.

Other ways in​

  • Nova: ask, for example, "is checkout using what it requests?"
  • API: the metrics calls in the API reference.

With astroctl​

Terminal
astroctl application metrics checkout --window 1h

Troubleshooting​

No metrics: the application was deployed outside AstroPulse, or the cluster's agent is not connected (check the cluster's connection pill).

Next​