CPU, memory and spherelet count

Reading · 7 min · Module 3, lesson 1 of 438 min left in this module

Module 3 · MetricsLesson 1 of 4

Goal: Read a service's Metrics tab, and compare CPU and memory with the ceiling of its shape.

Key idea

The Metrics tab shows three things over time: how much CPU and memory your service uses, and how many spherelets run it. A number only means something next to its ceiling, and the ceiling is your shape.

Where it is

Open a service and choose Metrics. Web services, private services and background workers have it; cron jobs don't.

At the top, a row of three numbers: CPU, Memory and Spherelets. Each shows the latest value and how much it changed over the window. Below them, the CPU chart runs full width, with Memory and Spherelets side by side underneath.

The time range picker sets the window. It opens on Last 1 hour; the shortest is Last 5 minutes and the longest Last 90 days.

What each chart measures

  • CPU, in vCPU. Each point is an average over a short stretch, so a spike of a few seconds looks lower than it was.
  • Memory, in MB or GB: what the app holds right now. A line that climbs and never comes down is worth a look.
  • Spherelets: how many are running. Flat at your count, except during an update, a restart or autoscaling.

Compare with the ceiling

Each spherelet is guaranteed a share of CPU, can use up to its shape's vCPU when spare CPU is available, and never more. Memory is a fixed ceiling:

ShapeCPU ceilingGuaranteed CPUMemory ceiling
Flex 512 MB0.5 vCPU0.125 vCPU512 MB
Standard 11 vCPU0.5 vCPU2 GB
Standard 22 vCPU1 vCPU4 GB

So 0.4 vCPU is nearly full on Flex 512 MB and light on Standard 2. Find your service's shape in Settings, under Spherelets, before you read anything into the charts.

Sum, Avg and Max

With more than one spherelet, the CPU and Memory charts combine them. The toggle on each chart picks how, and the chart's title shows the choice, such as CPU · sum or Memory · max:

  • Sum (the default): the total across spherelets.
  • Avg: a typical spherelet.
  • Max: the busiest one.

Compare Max with one spherelet's ceiling. A sum of 0.6 vCPU across two Flex spherelets could be two at 0.3, or one pinned at 0.5 while the other barely works. Only Max tells you which.

With one spherelet all three read the same. The toggle is there for when you scale out (Module 7).

Check yourself

Your service runs on 3 Flex spherelets and CPU on Sum shows 1.2 vCPU. Is any spherelet at its ceiling?
You deployed five minutes ago and the Metrics tab shows nothing new since you opened it. Why?