Documentation Grafana Cloud AI and machine learning Sift Sift analyses Resource contention

Sift analysis - Resource contention

The resource contention analysis reviews resource metrics to determine if any pod in the scope of the investigation is running into resource constraints.

Inputs

• Required:
  • cluster
  • namespace

How it works

Step 1: Find pods with high CPU throttling

Query cAdvisor metrics from Prometheus for any pods are throttled for more than 25% of cfs periods.

Copy

sum by (cluster, namespace, pod, container) (
  rate(container_cpu_cfs_throttled_periods_total{cluster="<cluster>", namespace="<namespace>"}[5m])
) /
sum by (cluster, namespace, pod, container) (
  rate(container_cpu_cfs_periods_total{cluster="<cluster>", namespace="<namespace>"}[5m])
) > 0.25

Step 2: Check for profiling data

Look in Pyroscope to see if there are profiles that can be linked for any pods with high CPU throttling.

What resources does this analysis use?

• Prometheus datasource

How does this analysis determine when a result is interesting?

• If any pods in the scope have a high level of CPU throttling.

What configurations options are available for this analysis?

• This check has no configurable parameters except for the Prometheus datasource.