Kubernetes
Provision AnsibleForms from a ConfigMap and Secrets
Mount the seed from a ConfigMap and its secrets from a Secret:
apiVersion: apps/v1
kind: Deployment
metadata:
name: ansibleforms
spec:
# AF_ROLE unset : the web app and the worker in one pod. See the note below.
replicas: 1
strategy:
type: Recreate
selector:
matchLabels: { app: ansibleforms }
template:
metadata:
labels: { app: ansibleforms }
spec:
terminationGracePeriodSeconds: 120
containers:
- name: ansibleforms
image: ghcr.io/ansibleforms/ansibleforms:7
env:
- name: CONFIG_SEED_PATH
value: /seed/seed.yaml
# a ConfigMap edited in git is remounted here within about a minute and
# applied from there, so changing the seed does not roll the pod
- name: CONFIG_SEED_RELOAD_SECONDS
value: "60"
# the environment is declared here, so the settings pages must not
# write persistent/.env behind this manifest's back
- name: ALLOW_ENV_EDIT
value: "0"
envFrom:
- secretRef:
name: ansibleforms-seed-secrets # SEED_AWX_TOKEN, SEED_LDAP_PW, SEED_CHAT_API_KEY, ...
- secretRef:
name: ansibleforms-secrets # DB_PASSWORD, ENCRYPTION_SECRET, ...
volumeMounts:
- name: seed
mountPath: /seed
readOnly: true
- name: persistent
mountPath: /app/dist/persistent
livenessProbe:
# answers as long as the process runs, without asking the database
httpGet: { path: /api/v2/live, port: 8000 }
readinessProbe:
# 503 until the database answers and the schema is complete, and while stopping
httpGet: { path: /api/v2/ready, port: 8000 }
volumes:
- name: seed
configMap:
name: ansibleforms-seed
- name: persistent
persistentVolumeClaim:
claimName: ansibleforms-persistent
One pod, with
AF_ROLEunset. This pod runs the web app and the worker in one process. The schedules, the nightly backup and the other background work run only in the process that holds the worker lock in the database, the designer lock is a database table, and the playbooks run on an RTE, not in this pod. Keepreplicas: 1with theRecreatestrategy all the same: on an upgrade the new pod waits for the worker to patch the schema, and the old pod holds the worker lock until it stops. To run several web nodes, split them intoAF_ROLE=appnodes and a worker, as Scaling describes.
ENCRYPTION_SECRET must be set before any credential is entered. Once credentials come from the seed, the database is effectively a cache, so rotating it means re-applying the seed rather than re-entering everything by hand.
The two probes need no login. Under a BASE_URL their paths start with it, and Monitoring describes what each one checks.