Files
fit/k8s.yaml
scott 81d795955d
All checks were successful
Build and Deploy / build-and-deploy (push) Successful in 56s
Let cert-manager issue the certificate
The ingress pointed at scottyah-tls, a secret hand-copied between
namespaces and absent from this one, so Traefik fell back to a
self-signed certificate. That is invisible behind Cloudflare until the
SSL mode is set to Full (strict), which then fails with a 526.

Issue into the namespace over DNS-01 instead. Nothing to copy, and the
renewal is no longer a thing anyone has to remember.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-01 22:30:47 -07:00

164 lines
4.2 KiB
YAML

---
apiVersion: v1
kind: Namespace
metadata:
name: fit
---
# The whole log is one JSON file, so the "database" is this volume. It is
# ReadWriteOnce, which is why the deployment below replaces rather than rolls.
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: fit-data-pvc
namespace: fit
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 1Gi
---
apiVersion: v1
kind: Service
metadata:
name: fit-svc
namespace: fit
spec:
selector:
app: fit
ports:
- port: 80
targetPort: 8080
protocol: TCP
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: fit-dep
namespace: fit
spec:
replicas: 1
# Two pods would both hold the file in memory and overwrite each other's
# writes, and the volume only attaches to one node anyway. Recreate makes the
# old pod let go before the new one starts.
strategy:
type: Recreate
selector:
matchLabels:
app: fit
template:
metadata:
labels:
app: fit
spec:
imagePullSecrets:
- name: harborcred
securityContext:
runAsNonRoot: true
runAsUser: 65534
runAsGroup: 65534
# Without fsGroup the volume comes up owned by root and nobody cannot
# write the file it exists to hold.
fsGroup: 65534
containers:
- name: fit
image: harbor.scottyah.com/scottyah/fit:latest
imagePullPolicy: Always
ports:
- containerPort: 8080
env:
- name: ADDR
value: ":8080"
- name: BASE_URL
value: "https://fit.scottyah.com"
- name: DATA_PATH
value: "/data/fit.json"
# Every date in the app is a local date, so this decides when the
# day rolls over.
- name: TZ_NAME
value: "America/Los_Angeles"
- name: GOMEMLIMIT
value: "40MiB"
volumeMounts:
- name: data
mountPath: /data
resources:
requests:
memory: "16Mi"
cpu: "10m"
limits:
memory: "64Mi"
cpu: "200m"
livenessProbe:
httpGet:
path: /healthz
port: 8080
initialDelaySeconds: 5
periodSeconds: 20
readinessProbe:
httpGet:
path: /healthz
port: 8080
initialDelaySeconds: 2
periodSeconds: 10
securityContext:
allowPrivilegeEscalation: false
# /data is the only thing the process writes.
readOnlyRootFilesystem: true
capabilities:
drop: ["ALL"]
volumes:
- name: data
persistentVolumeClaim:
claimName: fit-data-pvc
---
# The app has no login of its own; this is the only thing between the log and
# the internet, so the Service must not be exposed any other way.
apiVersion: traefik.io/v1alpha1
kind: Middleware
metadata:
name: fit-basic-auth
namespace: fit
spec:
basicAuth:
secret: fit-basic-auth
---
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
name: fit-ingress
namespace: fit
annotations:
traefik.ingress.kubernetes.io/router.entrypoints: websecure
traefik.ingress.kubernetes.io/router.tls: "true"
traefik.ingress.kubernetes.io/router.middlewares: fit-fit-basic-auth@kubernetescrd
# cert-manager watches the tls block below and issues into this namespace,
# over DNS-01 at Cloudflare. A TLS secret cannot be shared across
# namespaces, and hand-copying one leaves a certificate nothing renews.
cert-manager.io/cluster-issuer: letsencrypt-prod-cloudflare
spec:
ingressClassName: traefik
tls:
- hosts:
- fit.scottyah.com
# Created and renewed by cert-manager. It does not exist until the
# first issue succeeds, and the site does not serve a valid cert
# until then.
secretName: fit-tls
rules:
- host: fit.scottyah.com
http:
paths:
- path: /
pathType: Prefix
backend:
service:
name: fit-svc
port:
number: 80