How to run a daemon AND another command in a kubernetes pod

Question

I am trying to setup an automated influxdb instance using kubernetes, nothing too exotic I thought, but : I need to run the "influx setup ...." command for the initial setup, and this needs to be ran after the daemon is started (obviously).

I tried multiple syntax inside the deploy yaml, here is one attempt:

    spec:
      containers:
        - name: influxdb-test
          image: influxdb:2.0.7
          command:
            - sh
            - "-c"
            - |
              /bin/bash <<'EOF'
              influxd&
              influx setup --org ORG --bucket dummy --username USER --password USER --force
              EOF

But this gives me the following error :

Error: failed to determine if instance has been configured: Get "http://localhost:8086/api/v2/setup": dial tcp 127.0.0.1:8086: connect: connection refused
See 'influx setup -h' for help 

Well I do not need a solution for this specific error, but more a solution about how can I have a daemon running (influxd) and run some command after this daemon (influx setup) and then keep the container up.

I first thought that adding the simple "influx setup" command would be enough, but any command given inside a yaml deploy file overwrite the container initial command (influxd in this case) so I tried to reproduce the whole stuff manually : start the daemon, then run the setup then I am done, but no luck.

Of course I am a k8s newbie.

I am looking at "init container" as this sounds close to fixing my issue ....

Answer 1

I finally found some way to achieve this kind of task: Creating a container only for a single and ephemeral job can be achieved through the .... Job API :)

See this doc.

So I now have a deployment for my long living task (influxdb) + a Job definition for my on-time job (influxdb initial setup). Here is an example (I thought that could be useful for someone) :

apiVersion: apps/v1
kind: Deployment
metadata:
  name: influxdb
  namespace: influxdb
spec:
  selector:
    matchLabels:
      app: influxdb
  replicas: 1
  template:
    metadata:
      labels:
        app: influxdb

    spec:
      volumes:
        - name: influxdb-data
          persistentVolumeClaim:
            claimName: nfs-pvc-influx

      containers:
        - name: influxdb-daemon
          image: influxdb:2.0.7
          volumeMounts:
            - mountPath: /home/influxdb/
              name: influxdb-data
          env: 
            - name: INFLUXD_CONFIG_PATH
              value: /etc/influxdb2/
          ports:       
          - containerPort: 8086
            name: pbe
            protocol: TCP

And the job :

apiVersion: batch/v1
kind: Job
metadata:
  name: influxdb-init
  namespace: influxdb
spec:
  template:
    spec:
      restartPolicy: Never
      containers:
        - name: influxdb-init
          image: influxdb:2.0.7
          command:
            - sh
            - "-c"
            - |
              sleep 15
              influx setup --host http://influxdb-nodeport.influxdb.svc.cluster.local:8086 --org organistation1 --bucket dummy --username user --password user --token token --force && exit
  backoffLimit: 0

Answer 2

Since influxd is running background it will have a job id of 1 (check with the command 'jobs') So I think all you need to do is bring back influxd to the foregroud after the setup with 'fg 1'

  command:
        - sh
        - "-c"
        - |
          /bin/bash <<'EOF'
          influxd&
          sleep 15
          influx setup --org org1 --bucket dummy --username user --password passwd --force
          fg 1
          EOF

Answer 3

I think your current approach is probably the correct one. You are just missing a sleep to allow influxd time to actually get running. See Serve Laurijssen anwser.

You also could instead run some kind of process manager to handle this like s6-overlay but I wouldn't really use it for this example but it is handy in some other situations.

I don't think init container would work for you because it runs before the main container and is supposed to be a short lived command.

Container Lifecycle Hooks would probably also work here but its just more complexity then needed. To use PostStart hook you would need to curl http://localhost:8086/api/v2/setup in a loop until you got a good reply and then run your command.

Answer 4

Another option is to have multiple entrypoints in a Dockerfile. Example with setting up an nginx reverse proxy.

First create the script that does the configuring.

envsubst '${NODE_NAME}' < /etc/nginx/conf.d/default.conf.template > /etc/nginx/conf.d/default.conf

exec "$@"

Add the script to the container in the Dockerfile.

FROM nginx
ADD docker-entrypoint.sh /
ENTRYPOINT ["/docker-entrypoint.sh"]
CMD ["nginx", "-g", "daemon off;"]

Set script as ENTRYPOINT, which will create the config inside the container. Afterwards nginx is started with desired config.

Maybe it does not work for influx but it's a clean option for a "config and run" container.