ringtail / prombench

Prometheus E2E benchmarking tool

Automated Prometheus E2E testing and benchmarking.

It runs with Prow CI on a Google Kubernetes Engine Cluster. It is designed to support adding more k8s providers.

Run tests manually

Create a k8s cluster

---

• Create a new project on Google Cloud.

• Create a Service Account on GKE with role Kubernetes Engine Service Agent & Kubernetes Engine Admin and download the json file.

• Set the following environment variables & deploy cluster.

export PROJECT_ID=<google-cloud project-id>
export CLUSTER_NAME=prombench
export ZONE=us-east1-b
export AUTH_FILE=<path to service-account.json>

./prombench gke cluster create -a $AUTH_FILE -v PROJECT_ID:$PROJECT_ID \
    -v ZONE:$ZONE -v CLUSTER_NAME:$CLUSTER_NAME -f components/prow/cluster.yaml

Deploy Prometheus-Meta & Grafana

This is used for collecting and displaying the test results.

---

• Set the following environment variables

export GCLOUD_SERVICEACCOUNT_CLIENTID=<client-id present in service-account.json>
export GRAFANA_ADMIN_PASSWORD=password

The GCLOUD_SERVICEACCOUNT_CLIENTID is used to grant cluster-admin-rights to the service-account. This is needed to create RBAC roles on GKE.

• Deploy the nginx-ingress-controller which will be used to access Prometheus-Meta & Grafana.

./prombench gke resource apply -a $AUTH_FILE -v PROJECT_ID:$PROJECT_ID -v ZONE:$ZONE \
    -v CLUSTER_NAME:$CLUSTER_NAME -v GCLOUD_SERVICEACCOUNT_CLIENTID:$GCLOUD_SERVICEACCOUNT_CLIENTID \
    -f components/prow/manifests/rbac.yaml -f components/prow/manifests/nginx-controller.yaml

• Export the nginx-ingress-controller IP address.

// Generate auth config so we can use kubectl.
gcloud container clusters get-credentials $CLUSTER_NAME --zone=$ZONE --project=$PROJECT_ID

kubectl get ingress ing -o go-template='{{ range .status.loadBalancer.ingress}}{{.ip}}{{ end }}'

export INGRESS_IP=$(kubectl get ingress ing -o go-template='{{ range .status.loadBalancer.ingress}}{{.ip}}{{ end }}')

• Deploy Prometheus-meta & Grafana.

./prombench gke resource apply -a $AUTH_FILE -v PROJECT_ID:$PROJECT_ID \
    -v ZONE:$ZONE -v CLUSTER_NAME:$CLUSTER_NAME -v INGRESS_IP:$INGRESS_IP \
    -v GRAFANA_ADMIN_PASSWORD:$GRAFANA_ADMIN_PASSWORD -f components/prombench/manifests/results

• The services will be accessible at:
  • Grafana :: http://<INGRESS_IP>/grafana
  • Prometheus :: http://<INGRESS_IP>/prometheus-meta

Start a test

---

• Set the following environment variables.

export RELEASE=<master or any prometheus release(ex: v2.3.0) >
export PR_NUMBER=<PR to benchmark against the selected $RELEASE>

• Create the nodepools for the k8s objects

./prombench gke nodepool create -a ${AUTH_FILE} \
    -v ZONE:${ZONE} -v PROJECT_ID:${PROJECT_ID} -v CLUSTER_NAME:${CLUSTER_NAME} \
    -v PR_NUMBER:${PR_NUMBER} -f components/prombench/nodepools.yaml

• Deploy the k8s objects

./prombench gke resource apply -a ${AUTH_FILE} \
    -v ZONE:${ZONE} -v PROJECT_ID:${PROJECT_ID} -v CLUSTER_NAME:${CLUSTER_NAME} \
    -v PR_NUMBER:${PR_NUMBER} -v RELEASE:${RELEASE} \
    -f components/prombench/manifests/benchmark

Triggered tests by GitHub comments

Create a k8s cluster

---

• Follow the steps mentioned in the Create a k8s cluster in the manual setup.

Setup the GitHub API

---

• Generate a GitHub auth token that will be used to authenticate when sending requests to the GitHub api.

  • Login with the Prombot account and generate a new auth token.
    permissions:public_repo, read:org, write:discussion.

• Set the following environment variables

export HMAC_TOKEN=$(openssl rand -hex 20)
export OAUTH_TOKEN=***Replace with the generated token from github***

• Add a github webhook where to send the events.

  • Content Type: json

  • Send: Issue comments,Pull requests

  • Secret: echo $HMAC_TOKEN

  • Payload URL: http://prombench.prometheus.io/hook

    • Note: The IP DNS record for prombench.prometheus.io will be added once we get it from the ingress deployment.

Deploy Prow

This is used to monitor GitHub comments and starts new tests.

---

• Add all required tokens as k8s secrets.
  • hmac is used when verifying requests from GitHub.
  • oauth is used when sending requests to the GitHub api.
  • gke auth is used when scaling up and down the cluster.

./prombench gke resource apply -a $AUTH_FILE -v ZONE:$ZONE -v CLUSTER_NAME:$CLUSTER_NAME \
    -f components/prow/manifests/secrets.yaml \
    -v HMAC_TOKEN="$(printf $HMAC_TOKEN | base64 -w 0)" \
    -v OAUTH_TOKEN="$(printf $OAUTH_TOKEN | base64 -w 0)" \
    -v GKE_AUTH="$(cat $AUTH_FILE | base64 -w 0)"

• Deploy all internal prow components

Long term plans are to use the prombench cli tool to deploy and manage everything, but at the moment CustomResourceDefinition is WIP in the k8s golang client library. So we use kubectl to deploy CRD.

// Generate auth config so we can use kubectl.
gcloud container clusters get-credentials $CLUSTER_NAME --zone=$ZONE
kubectl apply -f components/prow/manifests/prow_internals_1.yaml

export GITHUB_ORG=prometheus
export GITHUB_REPO=prometheus

./prombench gke resource apply -a $AUTH_FILE -v PROJECT_ID:$PROJECT_ID \
    -v ZONE:$ZONE -v CLUSTER_NAME:$CLUSTER_NAME \
    -v GITHUB_ORG:$GITHUB_ORG -v GITHUB_REPO:$GITHUB_REPO \
    -f components/prow/manifests/prow_internals_2.yaml

Deploy Prometheus-Meta & Grafana

---

• Follow the steps mentioned in the manual setup to deploy prometheus-meta & grafana.

• Set the IP DNS record for prombench.prometheus.io to the nginx-ingress-controller IP address.

• The services will be accessible at:

  • Prow dashboard :: http://prombench.prometheus.io
  • Grafana :: http://prombench.prometheus.io/grafana
  • Prometheus :: http://prombench.prometheus.io/prometheus-meta

Trigger tests via a Github comment.

---

A Prometheus maintainer can comment as follows to benchmark a PR:

• /benchmark (benchmark PR with the master branch.)
• /benchmark master
• /benchmark 2.4.0 (Any release version can be added here. Don't prepend v to the release version here. The benchmark plugin in Prow will prepend it.)

To cancel benchmarking, a mantainer should comment /benchmark cancel.