Skip to content

Latest commit

 

History

History
64 lines (44 loc) · 2.49 KB

monitoring.md

File metadata and controls

64 lines (44 loc) · 2.49 KB
Raw

Monitoring and Logging

All Unikorn controllers support prometheus metrics, and promtail/loki compatible logs out of the box. Logs in particular are tagged with the instance they relate to and a specific GUID associated with the reconcile request. As such a nice database is in order to make sense of them (unless you can read JSON...)

Performance Metrics

When writing provisioners and other things, it's useful to keep a histogram of run-times so that you can spot any anomalies. This is especially pertinent when OpenStack is in play as it's pretty flaky, and things like load-balancers and ingresses can get stuck very easily. All metric names should be prefixed with unikorn_ for easy identification, then some other context to make things totally obvious.

You will most likely want to install Prometheus and Grafana:

helm repo add prometheus-community https://prometheus-community.github.io/helm-charts
helm install prometheus prometheus-community/kube-prometheus-stack --namespace monitoring --create-namespace

This will pretty much work out of the box when installed with Helm and prometheus support enabled.

Metrics Server

The prometheus stuff has some good metrics collected by default, and the Unikorn binaries give a decent view into heap allocations. You could use the contrainer_cpu_usage_second_total metric in Grafana for example. Using the kubectl top pods command is a quick and easy way to derive utilisation, hence this section. If it's not installed by default on your platform:

kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml

On KinD, this won't work out of the box as the node certificates don't have an IP SAN. To fudge this, add the --kubelet-insecure-tls flag to the server:

kubectl edit -n kube-system deployment/metrics-server

OpenTelemetry Tracing

Unikorn makes use of the OTLP protocol to ship information on spans. This provides a break down of transaction based workloads.

You can install Jaeger's stack with:

helm repo add jaegertracing https://jaegertracing.github.io/helm-charts
helm upgrade jaeger jaegertracing/jaeger --set collector.service.otlp.http.port=4318

Logs

Coming soon!

For now, use kubectl logs -f deployment/unikorn-*. You can boost the verbosity by adding the follwing to the controll manifests:

args:
- -zap-log-level=debug

Additionalal options/details you can get from ./bin/amd64-linux-gnu/unikorn-control-plane-manager --help.