Metrics with Prometheus

Metrics in Fission

Fission exposes metrics in the Prometheus standard, which can be readily scraped and used using a Prometheus server and visualized using Grafana. The metrics help monitor the state of the Functions as well as the Fission components.

Prometheus

Prometheus is a monitoring and alerting tool. It uses a multi-dimensional data model with time series data identified by metric name and key/value pairs.

Fission exposes metrics which are pulled and operated by Prometheus at regular intervals.

Grafana

Grafana is a visualization tool which can query, visualize, alert on and understand metrics. It supports Prometheus as it’s data source.

Setting up

There are different ways to install Prometheus. It can be installed and run in and outside containers. Since Fission itself runs in Kubernetes, we’ll use the Prometheus Operator which is a way of installing Prometheus as Kubernetes Custom Resource.

Prerequisite

Install Prometheus and Grafana

We’ll install Prometheus and Grafana in a namespace named monitoring.

To create the namespace, run the following command in a terminal:

$ expose METRICS_NAMESPACE=monitoring
$ kubectl create namespace $METRICS_NAMESPACE

Install Prometheus and Grafana with the release name fission-metrics.

$ helm repo add prometheus-community https://prometheus-community.github.io/helm-charts
$ helm repo add stable https://kubernetes-charts.storage.googleapis.com/
$ helm repo update
$ helm install fission-metrics --namespace monitoring prometheus-community/kube-prometheus-stack \
  --set kubelet.serviceMonitor.https=true \
  --set prometheus.prometheusSpec.serviceMonitorSelectorNilUsesHelmValues=false \
  --set prometheus.prometheusSpec.podMonitorSelectorNilUsesHelmValues=false \
  --set prometheus.prometheusSpec.ruleSelectorNilUsesHelmValues=false

This will install Prometheus and Grafana in the monitoring namespace. Along with the Prometheus server, it’ll also install other components viz. node-exporter, kube-state-metrics and pushgateway.

Adding ServiceMonitor for Fission

By default, this installation will not discover the metrics exposed by Fission. To be able to scrape those metrics, we can install ServiceMonitor, which is a Kubernetes custom resource used to by the Prometheus Operator to add new targets.

Create the manifest:

$ cat <<EOF > servicemonitors.yaml
apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: executor-service-app
spec:
  selector:
    matchLabels:
      svc: executor
  endpoints:
  - targetPort: 8080
---
apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: router-service-app
spec:
  selector:
    matchLabels:
      svc: router
  endpoints:
  - targetPort: 8080
EOF

Install the ServiceMonitors.

$ kubectl apply -f servicemonitors.yaml --namespace fission

This will install the ServiceMonitors in the fission namespace, the same where Fission is installed.

Accessing Grafana UI

The installation creates a Service named fission-metrics-grafana. To access this, you can use Kubernetes port forwarding

$ kubectl --namespace monitoring port-forward svc/fission-metrics-grafana 3000:80

The Grafana can be now accessed on http://localhost:3000.

This installation also adds Prometheus as a data source for Grafana automatically. You can verify and update this in the Data Sources section of the UI.

Metrics Queries

Once Prometheus is configured, we can now run queries in Grafana over Fission metrics. Individual queries can be run under Explore section.

Fission exposes a set of metrics. For example to query the total number of function calls, run

fission_function_calls_total

Calls for a specific function can be queried using

fission_function_calls_total{name="foo"}

To track the duration of a specific function

fission_function_duration_seconds{name="hello"}

There are other more Fission metrics, which can be found under Metrics in the Explore screen.

Fission Dashboard

With Grafana, visuals dashboards can be created to monitor multiple metrics in an organized way.

One such dashboard can be found here. This dashboard shows log metrics from all the major components of Fission.

Once imported, the dashboard will look similar to below image.

Prometheus Fission Functions dashboard

There will be more dashboards added to the same location over time.

Last modified October 27, 2020: Markdown lint reformat (#221) (8fc8f1c)