Backups: pg_dump → S3 (MEGA S4); drop local PVC

templates/backup-cronjob.yaml

@@ -1,10 +1,16 @@
 {{- if .Values.backups.enabled -}}
-# Daily dump job. Same image as the Postgres pod, so pg_dump is
+# Daily dump → S3.
-# version-matched. Output goes to the dedicated backup PVC; the same
-# job script prunes older dumps to honor `backups.retain`.
 #
-# Tower's "Backup Now" feature creates a one-off Job from this same
+# Architecture: pg_dump in postgres:alpine, pipe through gzip, then
-# template at request time — see backend/cmd/api/backups.go.
+# `aws s3 cp -` to push the stream straight to MEGA S4. We use a
+# single multi-stage shell command (no init container) so the dump
+# never lands on the customer-server's local disk — the instance
+# data and the backup destination are deliberately separated.
+#
+# AWS credentials come from a K8s Secret (default `s3-backup-creds`)
+# provisioned out-of-band by Tower's bootstrap. Endpoint + bucket +
+# prefix are committed in this file's values; only the access/secret
+# pair lives in the Secret.
 apiVersion: batch/v1
 kind: CronJob
 metadata:
@@ -32,8 +38,11 @@ spec:
         spec:
           restartPolicy: Never
           containers:
-            - name: pgdump
+            - name: pgdump-s3
-              image: "{{ .Values.postgres.image }}:{{ .Values.postgres.tag }}"
+              # Image carries both pg_dump (postgresql-client) and
+              # aws-cli. We build it from alpine + apk install on
+              # first run; for now bitnami's prebuilt covers both.
+              image: bitnami/postgresql:16
               imagePullPolicy: IfNotPresent
               env:
                 - name: PGHOST
@@ -53,29 +62,55 @@ spec:
                     secretKeyRef:
                       name: {{ include "instance.fullname" . }}-pg
                       key: POSTGRES_DB
+                - name: AWS_ACCESS_KEY_ID
+                  valueFrom:
+                    secretKeyRef:
+                      name: {{ .Values.backups.credentialsSecret }}
+                      key: AWS_ACCESS_KEY_ID
+                - name: AWS_SECRET_ACCESS_KEY
+                  valueFrom:
+                    secretKeyRef:
+                      name: {{ .Values.backups.credentialsSecret }}
+                      key: AWS_SECRET_ACCESS_KEY
+                - name: S3_ENDPOINT
+                  value: {{ .Values.backups.s3.endpoint | quote }}
+                - name: AWS_DEFAULT_REGION
+                  value: {{ .Values.backups.s3.region | quote }}
+                - name: S3_BUCKET
+                  value: {{ .Values.backups.s3.bucket | quote }}
+                - name: S3_PREFIX
+                  value: {{ .Values.instance.code | quote }}
                 - name: RETAIN
                   value: {{ .Values.backups.retain | quote }}
               command:
-                - /bin/sh
+                - /bin/bash
                 - -c
                 - |
-                  set -e
+                  set -euo pipefail
                   TS=$(date -u +%Y%m%dT%H%M%SZ)
-                  OUT=/backups/${TS}.sql.gz
+                  KEY="${S3_PREFIX}/${TS}.sql.gz"
-                  mkdir -p /backups
+                  echo ">>> dumping to s3://${S3_BUCKET}/${KEY}"
-                  echo ">>> pg_dump → $OUT"
+                  # Install aws-cli on first run. bitnami/postgresql is
+                  # debian-based so apt is available and fast.
+                  if ! command -v aws >/dev/null 2>&1; then
+                    apt-get update -qq && apt-get install -y -qq awscli >/dev/null
+                  fi
                   pg_dump --format=plain --clean --if-exists --no-owner --no-acl \
-                    | gzip -9 > "$OUT"
+                    | gzip -9 \
-                  echo ">>> wrote $(du -h "$OUT" | cut -f1)"
+                    | aws --endpoint-url "$S3_ENDPOINT" s3 cp - "s3://${S3_BUCKET}/${KEY}"
-                  # Rotate: keep only the newest $RETAIN dumps.
+                  echo ">>> uploaded"
-                  cd /backups
+                  echo ">>> rotating: keep last $RETAIN under ${S3_PREFIX}/"
-                  ls -1t *.sql.gz 2>/dev/null \
+                  # List, sort newest-first, drop the top N, delete the rest.
-                    | awk -v n=$RETAIN 'NR > n' \
+                  aws --endpoint-url "$S3_ENDPOINT" s3api list-objects-v2 \
-                    | xargs -r rm -v
+                    --bucket "$S3_BUCKET" --prefix "${S3_PREFIX}/" \
-                  ls -lh /backups
+                    --query 'Contents[].Key' --output text 2>/dev/null \
-              volumeMounts:
+                    | tr '\t' '\n' | sort -r | tail -n +$((RETAIN + 1)) \
-                - name: backups
+                    | while read OLDKEY; do
-                  mountPath: /backups
+                        [ -n "$OLDKEY" ] || continue
+                        echo ">>> deleting old: $OLDKEY"
+                        aws --endpoint-url "$S3_ENDPOINT" s3 rm "s3://${S3_BUCKET}/${OLDKEY}"
+                      done
+                  echo ">>> done"
               resources:
                 requests:
                   cpu: 100m
@@ -83,8 +118,4 @@ spec:
                 limits:
                   cpu: "1"
                   memory: 1Gi
-          volumes:
-            - name: backups
-              persistentVolumeClaim:
-                claimName: {{ include "instance.fullname" . }}-backups
 {{- end }}

templates/backup-pvc.yaml

@@ -1,14 +0,0 @@
-{{- if .Values.backups.enabled -}}
-apiVersion: v1
-kind: PersistentVolumeClaim
-metadata:
-  name: {{ include "instance.fullname" . }}-backups
-  labels:
-    {{- include "instance.labels" . | nindent 4 }}
-    odoosky.io/role: backups
-spec:
-  accessModes: [ReadWriteOnce]
-  resources:
-    requests:
-      storage: {{ .Values.backups.storage | quote }}
-{{- end }}

values.yaml

@@ -78,11 +78,24 @@ backups:
   # Cron schedule for the automatic backup job. Default 03:00 UTC
   # daily — quiet hour for most timezones, non-business in EU/US/AS.
   schedule: "0 3 * * *"
-  # PVC size for retained dumps. Holds ~7 days of dumps for a small
+  # How many dumps to retain in S3. The backup job prunes older
-  # instance; scale up via overlay if the instance has a large DB.
+  # objects matching the instance's prefix on every successful run.
-  storage: 10Gi
-  # How many dumps to retain. Older ones are pruned by the same Job.
   retain: 7
+  # S3-compatible destination. The endpoint + region + bucket are
+  # NON-secret and live in this committed values.yaml; the AWS
+  # credentials live in a K8s Secret named by `credentialsSecret`,
+  # provisioned out-of-band by Tower's bootstrap script (which reads
+  # from OpenBao). The chart never sees access/secret keys directly.
+  s3:
+    endpoint: https://s3.eu-central-1.s4.mega.io
+    region: eu-central-1
+    bucket: odoosky-v3-backups
+    # Per-instance S3 key prefix. Each instance writes under its own
+    # code/ subdirectory inside the shared bucket.
+    prefix: "{{ .Values.instance.code }}"
+  # Name of the K8s Secret holding AWS_ACCESS_KEY_ID +
+  # AWS_SECRET_ACCESS_KEY. Mounted via envFrom on the backup Job.
+  credentialsSecret: s3-backup-creds
 
 ingress:
   # Traefik entrypoint name (set on the Traefik install in the