Add limesurvey, migrating its data off local-path onto GlusterFS
Validate and Deploy to K3s / validate (push) Failing after 23s
Validate and Deploy to K3s / deploy (push) Skipped

LimeSurvey was never in this repo, and its two PVCs used the k3s local-path
StorageClass: both volumes were pinned to rpi-worker-3's local disk with
persistentVolumeReclaimPolicy: Delete. That put the survey data on a single
SD-backed node with no replication, where deleting the PVC would destroy it.

Both volumes now live on gvolume0 under /mnt/gvolume0/limesurvey/{app,db},
with Retain, matching the pattern used by vaultwarden and tududi. Migrated
2026-09-21 via mysqldump/restore; verified identical afterwards (3 surveys,
742 responses, 2 users, 17 questions, 60 tables).

db-config.yaml is required, not optional: InnoDB's defaults assume a local
POSIX filesystem, and on a GlusterFS FUSE mount native AIO is unavailable and
O_DIRECT unreliable. Without innodb_use_native_aio=0 and
innodb_flush_method=fsync, MariaDB either fails to start or risks corruption.

A nightly dump to /mnt/gvolume0/limesurvey/backups (14 day retention) is
included -- a volume survives losing a node, but a logical dump also survives
corruption, a bad upgrade and mistaken deletes.

The old local-path PVCs (limesurvey-db-data, limesurvey-uploads) are
deliberately left in place as a rollback and are not managed here.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
jasonwitty
2026-09-20 19:54:29 -07:00
co-authored by Claude Opus 5
parent 097dbeb7ea
commit 91d96eb17a
11 changed files with 387 additions and 0 deletions
+45
View File
@@ -0,0 +1,45 @@
# Added 2026-09-21. The datadir now lives on GlusterFS, which survives losing a
# node -- but a logical dump also survives corruption, a bad upgrade, and
# fat-fingered deletes, which a volume does not. Keeps 14 days.
apiVersion: batch/v1
kind: CronJob
metadata:
name: limesurvey-db-backup
namespace: limesurvey
spec:
schedule: "17 3 * * *"
concurrencyPolicy: Forbid
successfulJobsHistoryLimit: 2
failedJobsHistoryLimit: 3
jobTemplate:
spec:
template:
spec:
restartPolicy: OnFailure
containers:
- name: dump
image: docker.io/library/mariadb:11
env:
- name: MARIADB_ROOT_PASSWORD
valueFrom:
secretKeyRef:
name: limesurvey
key: db-root-password
command:
- sh
- -c
- |
set -e
f=/backups/limesurvey-$(date -u +%Y%m%d-%H%M%S).sql.gz
mariadb-dump -h limesurvey-db -u root -p"$MARIADB_ROOT_PASSWORD" \
--single-transaction --routines --triggers --events \
--databases limesurvey | gzip > "$f"
echo "wrote $f ($(stat -c%s "$f") bytes)"
find /backups -name 'limesurvey-*.sql.gz' -mtime +14 -delete
volumeMounts:
- name: backups
mountPath: /backups
volumes:
- name: backups
persistentVolumeClaim:
claimName: limesurvey-backups
+15
View File
@@ -0,0 +1,15 @@
# InnoDB defaults assume a local POSIX filesystem. The datadir now lives on a
# GlusterFS FUSE mount, where native async I/O is unavailable and O_DIRECT is
# unreliable -- without these, MariaDB either fails to start ("Failed to
# initialize AIO") or risks corruption. Keep these with the gluster PV.
apiVersion: v1
kind: ConfigMap
metadata:
name: limesurvey-db-config
namespace: limesurvey
data:
99-gluster.cnf: |
[mariadb]
innodb_use_native_aio = 0
innodb_flush_method = fsync
innodb_file_per_table = 1
+59
View File
@@ -0,0 +1,59 @@
apiVersion: apps/v1
kind: Deployment
metadata:
name: limesurvey-db
namespace: limesurvey
spec:
replicas: 1
strategy:
type: Recreate
selector:
matchLabels:
app: limesurvey-db
template:
metadata:
labels:
app: limesurvey-db
spec:
containers:
- name: mariadb
image: docker.io/library/mariadb:11
env:
- name: MARIADB_INITDB_SKIP_TZINFO
value: "1"
- name: MARIADB_DATABASE
value: limesurvey
- name: MARIADB_USER
value: limesurvey
- name: MARIADB_PASSWORD
valueFrom:
secretKeyRef:
name: limesurvey
key: db-password
- name: MARIADB_ROOT_PASSWORD
valueFrom:
secretKeyRef:
name: limesurvey
key: db-root-password
ports:
- containerPort: 3306
name: mysql
resources:
requests:
cpu: 100m
memory: 256Mi
limits:
memory: 1Gi
volumeMounts:
- name: data
mountPath: /var/lib/mysql
- name: dbconf
mountPath: /etc/mysql/conf.d/99-gluster.cnf
subPath: 99-gluster.cnf
volumes:
- name: data
persistentVolumeClaim:
claimName: limesurvey-db
- name: dbconf
configMap:
name: limesurvey-db-config
+89
View File
@@ -0,0 +1,89 @@
apiVersion: apps/v1
kind: Deployment
metadata:
name: limesurvey
namespace: limesurvey
spec:
replicas: 1
strategy:
type: Recreate
selector:
matchLabels:
app: limesurvey
template:
metadata:
labels:
app: limesurvey
spec:
# ONE volume mounted twice with subPath. Do not split this into two
# volume entries pointing at the same PVC -- that wedges kubelet's volume
# setup and leaves the pod stuck in Init with no events.
initContainers:
- name: seed-config
image: docker.io/martialblog/limesurvey:6-apache
command:
- sh
- -c
- cp -rn /var/www/html/application/config/. /seed/
volumeMounts:
- name: data
mountPath: /seed
subPath: config
containers:
- name: limesurvey
image: docker.io/martialblog/limesurvey:6-apache
env:
- name: DB_TYPE
value: mysql
- name: DB_HOST
value: limesurvey-db
- name: DB_PORT
value: "3306"
- name: DB_NAME
value: limesurvey
- name: DB_USERNAME
value: limesurvey
- name: DB_PASSWORD
valueFrom:
secretKeyRef:
name: limesurvey
key: db-password
- name: ADMIN_USER
value: admin
- name: ADMIN_NAME
value: Jason
- name: ADMIN_EMAIL
value: admin@wittyoneoff.com
- name: ADMIN_PASSWORD
valueFrom:
secretKeyRef:
name: limesurvey
key: admin-password
- name: PUBLIC_URL
value: https://surveys.skatefight.us
ports:
- containerPort: 8080
name: http
readinessProbe:
httpGet:
path: /
port: 8080
initialDelaySeconds: 20
periodSeconds: 10
resources:
requests:
cpu: 100m
memory: 256Mi
limits:
memory: 1Gi
volumeMounts:
- name: data
mountPath: /var/www/html/upload
subPath: upload
- name: data
mountPath: /var/www/html/application/config
subPath: config
volumes:
- name: data
persistentVolumeClaim:
claimName: limesurvey-app
+28
View File
@@ -0,0 +1,28 @@
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
name: limesurvey
namespace: limesurvey
spec:
ingressClassName: traefik
rules:
- host: surveys.skatefight.us
http:
paths:
- path: /
pathType: Prefix
backend:
service:
name: limesurvey
port:
number: 8080
- host: origin-surveys.wittyoneoff.com
http:
paths:
- path: /
pathType: Prefix
backend:
service:
name: limesurvey
port:
number: 8080
+13
View File
@@ -0,0 +1,13 @@
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
# secret.example.yaml is intentionally NOT listed -- copy it to secret.yaml
# (gitignored), set real values, and apply it once per cluster.
resources:
- namespace.yaml
- storage.yaml
- db-config.yaml
- db-deployment.yaml
- deployment.yaml
- service.yaml
- ingress.yaml
- backup-cronjob.yaml
+4
View File
@@ -0,0 +1,4 @@
apiVersion: v1
kind: Namespace
metadata:
name: limesurvey
+12
View File
@@ -0,0 +1,12 @@
# Copy to secret.yaml (gitignored), set real values, apply once per cluster:
# kubectl apply -f secret.yaml
apiVersion: v1
kind: Secret
metadata:
name: limesurvey
namespace: limesurvey
type: Opaque
stringData:
db-password: CHANGEME
db-root-password: CHANGEME
admin-password: CHANGEME
+27
View File
@@ -0,0 +1,27 @@
apiVersion: v1
kind: Service
metadata:
name: limesurvey
namespace: limesurvey
spec:
type: NodePort
selector:
app: limesurvey
ports:
- port: 8080
targetPort: 8080
# Pinned: the UniFi forward "lime survey" (WAN 30087) targets this.
nodePort: 30087
---
apiVersion: v1
kind: Service
metadata:
name: limesurvey-db
namespace: limesurvey
spec:
type: ClusterIP
selector:
app: limesurvey-db
ports:
- port: 3306
targetPort: 3306
+94
View File
@@ -0,0 +1,94 @@
# LimeSurvey was originally deployed with the k3s `local-path` StorageClass,
# which pinned both volumes to rpi-worker-3's local disk with
# persistentVolumeReclaimPolicy: Delete. That meant the survey data lived on a
# single SD-backed node and would be destroyed by deleting the PVC. Migrated to
# GlusterFS 2026-09-21.
#
# Two separate PVCs are correct here because the app and the database run in
# SEPARATE pods. (The warning in apps/tududi/storage.yaml is about declaring the
# same PVC twice as two volume entries in ONE pod -- that is what wedged an
# earlier limesurvey attempt. The app pod below still uses ONE volume, mounted
# twice with subPath.)
apiVersion: v1
kind: PersistentVolume
metadata:
name: limesurvey-pv
spec:
capacity:
storage: 5Gi
accessModes:
- ReadWriteOnce
persistentVolumeReclaimPolicy: Retain
volumeMode: Filesystem
hostPath:
path: /mnt/gvolume0/limesurvey/app
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: limesurvey-app
namespace: limesurvey
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 5Gi
storageClassName: ""
volumeName: limesurvey-pv
---
apiVersion: v1
kind: PersistentVolume
metadata:
name: limesurvey-db-pv
spec:
capacity:
storage: 5Gi
accessModes:
- ReadWriteOnce
persistentVolumeReclaimPolicy: Retain
volumeMode: Filesystem
hostPath:
path: /mnt/gvolume0/limesurvey/db
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: limesurvey-db
namespace: limesurvey
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 5Gi
storageClassName: ""
volumeName: limesurvey-db-pv
---
apiVersion: v1
kind: PersistentVolume
metadata:
name: limesurvey-backups-pv
spec:
capacity:
storage: 5Gi
accessModes:
- ReadWriteOnce
persistentVolumeReclaimPolicy: Retain
volumeMode: Filesystem
hostPath:
path: /mnt/gvolume0/limesurvey/backups
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: limesurvey-backups
namespace: limesurvey
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 5Gi
storageClassName: ""
volumeName: limesurvey-backups-pv
+1
View File
@@ -10,3 +10,4 @@ resources:
- apps/unified-streaming
- apps/dashwise
- apps/tududi
- apps/limesurvey