1   TL;DR¶

Follow #com-efd at LSSTC Slack for updates.

2   Why are we doing this?¶

The EFD is a solution based on Kafka and InfluxDB for recording telemetry, commands, and events for LSST. It was prototyped and tested with the simulator for the M1M3 subsystem. The next logical step is to deploy and test the EFD with real hardware.

The ability to deploy the EFD at different environments quickly and reproduce these deployments is crucial. To solve this problem we’ve adopted Kubernetes as our deployment platform, Helm as the Kubernetes package manager and Argo CD as the continuous delivery tool for Kubernetes.

In this technote, we demonstrate that we can deploy the EFD on a single machine with Docker and k3s (“Kubes”) while the final on-premise deployment platform is not ready.

3   Provisioning a k3s (“kubes”) cluster¶

The first step is to provision our Kubernetes cluster. K3s is a lightweight Kubernetes that runs in a container.

sudo systemctl enable docker

export K3S_PORT=6443
export K3S_URL=http://localhost:${K3S_PORT}
export K3S_TOKEN=$(date | base64)
export HOST_PATH=/data # change depending on your host
export CONTAINER_PATH=/opt/local-path-provisioner
sudo docker run  -d --restart always --tmpfs /run --tmpfs /var/run --volume ${HOST_PATH}:${CONTAINER_PATH} -e K3S_URL=${K3S_URL} -e K3S_TOKEN=${K3S_TOKEN} --privileged --network host --name master docker.io/rancher/k3s:v0.5.0-rc1 server --https-listen-port ${K3S_PORT} --no-deploy traefik

sudo docker cp master:/etc/rancher/k3s/k3s.yaml k3s.yaml
export KUBECONFIG=$(pwd)/k3s.yaml
kubectl cluster-info

Kubernetes master is running at https://localhost:6443
CoreDNS is running at https://localhost:6443/api/v1/namespaces/kube-system/services/kube-dns:dns/proxy

To further debug and diagnose cluster problems, use 'kubectl cluster-info dump'.

kubectl apply -f https://raw.githubusercontent.com/rancher/local-path-provisioner/master/deploy/local-path-storage.yaml

kubectl get pods --all-namespaces
NAMESPACE            NAME                                      READY   STATUS    RESTARTS   AGE
kube-system          coredns-695688789-r9gkt                   1/1     Running   0          5m
local-path-storage   local-path-provisioner-5d4b898474-vz2np   1/1     Running   0          4s

sudo docker cp master:/var/lib/rancher/k3s/server/node-token node-token

export SERVER_URL=https://<master external IP>:${K3S_PORT}
export NODE_TOKEN=$(cat node-token)
export WORKER=kube-0
export HOST_PATH=/data # change depending on your host
export CONTAINER_PATH=/opt/local-path-provisioner
sudo docker run -d --tmpfs /run --tmpfs /var/run -v ${HOST_PATH}:${CONTAINER_PATH} -e K3S_URL=${SERVER_URL} -e K3S_TOKEN=${NODE_TOKEN} --privileged --name ${WORKER} rancher/k3s:v0.5.0-rc1

4   Install Argo CD¶

With Argo CD we keep the EFD deployment configuration on GitHub. This way we can use GitHub to control changes in the EFD deployments and easily bootstrap new EFD deployments.

kubectl create namespace argocd
kubectl apply -n argocd -f https://raw.githubusercontent.com/argoproj/argo-cd/stable/manifests/install.yaml

This creates a new namespace, argocd, where Argo CD services and application resources will live.

Follow these instructions to install Argo CD CLI.

Additional Argo CD configuration includes Single Sign On (SSO) and the Role Based Access Control (RBAC).

In particular, this is the RBAC configuration we added to the ArgoCD ConfigMap argocd-rbac-cm to let members of the EFD ops team in the lsst-sqre GitHub organization to synchronize the EFD configuration, while granting read-only permission for other members.

data:
  policy.csv: |
    p, lsst-sqre:EFD ops, applications, sync, default/*, allow

    g, lsst-sqre:EFD ops, role:admin
    policy.default: role:readonly

5   Deploy the EFD¶

Once Argo CD is installed we can deploy the EFD.

Argo CD manages the deployment of the EFD on multiple environments. The possible environments are summit, tucson-teststand, ncsa-teststand, ldf, and gke.

For example, the following bootstraps an EFD deployment using the configuration for the summit environment:

kubectl port-forward svc/argocd-server -n argocd 8080:443

argocd login localhot:8080
argocd app create efd --dest-namespace argocd --dest-server https://kubernetes.default.svc --repo https://github.com/lsst-sqre/argocd-efd.git --path apps/efd --helm-set env=summit
argocd app sync efd

The secrets used by the EFD are stored on LSST’s Vault Service but you need to create at least one secret manually with the VAULT_TOKEN:

export VAULT_TOKEN=<vault token>
export VAULT_TOKEN_LEASE_DURATION=<vault token lease duration>

kubectl create secret generic vault-secrets-operator --from-literal=VAULT_TOKEN=$VAULT_TOKEN --from-literal=VAULT_TOKEN_LEASE_DURATION=$VAULT_TOKEN_LEASE_DURATION --namespace vault-secrets-operator

Service names for the apps follow the convention <app>-<environment>-efd.lsst.codes, for example, chronograf-summit-efd.lsst.codes

In particular, the broker URL for the Summit EFD is kafka-0-summit-efd.lsst.codes:30190.

6   Testing the EFD¶

The EFD deployment can be tested using kafkacat a command line utility implemented with librdkafka the Apache Kafka C/C++ client library.

Run in producer mode (-P) to produce messages for a test topic:

Run in metadata listing mode (-L) to retrieve metadata from the cluster:

kafkacat -L -b <kafka broker url>

The -d option enables librdkafka debugging. For instance, -d broker can be used to debug connection issues with the cluster:

kafkacat -L -b <kafka broker url>  -d broker

Run in consumer mode (-C) to consume topics from the cluster:

kafkacat -C -b <kafka broker url> -t <topic name>

7   Restarting the EFD manually¶

k3s is configured to automatically start after a system reboot (--restart-always flag). In case you need to start the k3s master manually, first check its status:

sudo docker ps -a

If k3s master status is Exited start with the following command:

sudo docker start master

After a few minutes, all Kubernetes pods should be running again:

export KUBECONFIG=$(pwd)/k3s.yaml
kubectl cluster-info
kubectl get pods --all-namespaces

8   Inspecting logs, restarting pods, etc.¶

You can inspect logs, restart pods and doing other operations including synchronization of the deployment using the Argo CD UI or the CLI.

In the case of the EFD, we have most often needed to restart the kafka connector pod confluent-cp-kafka-connect in the cp-helm-charts namespace.

9   Accessing EFD data¶

Use the Chronograf interface for time-series visualization and dashboarding.

In this notebook we show how to extract data from the EFD using aioinflux, a Python client for InfluxDB, and proceed with data analysis using Pandas dataframes.