Initial commit

infrastructure/scylla/cluster-values.yaml

@@ -0,0 +1,39 @@
+# infrastructure/scylla/cluster-values.yaml
+# ScyllaCluster (database)
+
+# Homelab-friendly mode (reduced kernel tuning requirements)
+developerMode: true
+
+# Use your Ceph-backed default StorageClass
+racks:
+  - name: rack-a
+    members: 3
+    storage:
+      storageClassName: rook-ceph-block
+      capacity: 50Gi
+
+    # Requests drive scheduling; keep low for bootstrap.
+    # Limits drive Scylla shard count (cpu limit = shard count).
+    resources:
+      requests:
+        cpu: 10m
+        memory: 512Mi
+      limits:
+        cpu: 2
+        memory: 4Gi
+
+    agentResources:
+      requests:
+        cpu: 100m
+        memory: 128Mi
+
+    placement: {}
+
+alternator:
+  enabled: false
+  insecureEnableHTTP: false
+
+serviceMonitor:
+  create: true
+  labels:
+    release: kube-prometheus-stack

infrastructure/scylla/manager-postsync-fix.sh

@@ -0,0 +1,11 @@
+#!/usr/bin/env bash
+# Post-sync fix for scylla-manager: the chart-created NetworkPolicy
+# blocks cross-node ClusterIP traffic due to a CNI/kube-proxy DNAT
+# interaction. Delete it so the manager API can reach its backend
+# ScyllaCluster via the per-pod service from any node.
+set -euo pipefail
+
+NS=scylla-manager
+
+echo "Removing NetworkPolicy that blocks cross-node ClusterIP traffic..."
+kubectl -n "${NS}" delete networkpolicy scylla-manager-to-scylla-pod --ignore-not-found=true

infrastructure/scylla/manager-values.yaml

@@ -0,0 +1,42 @@
+# infrastructure/scylla/manager-values.yaml
+# Scylla Manager (with its own small Scylla backend)
+
+logLevel: info
+
+resources:
+  requests:
+    cpu: 10m
+    memory: 128Mi
+
+scylla:
+  # Homelab-friendly mode
+  developerMode: true
+
+  # Use Ceph for the manager's backend storage
+  racks:
+    - name: manager-rack
+      members: 1
+      storage:
+        capacity: 10Gi
+        storageClassName: rook-ceph-block
+      resources:
+        # Requests drive Kubernetes scheduling. Keep them extremely low so the
+        # embedded Scylla backend can schedule even when the cluster is
+        # request-saturated.
+        requests:
+          cpu: 1m
+          memory: 1Gi
+        limits:
+          cpu: 1
+          memory: 2Gi
+      agentResources:
+        requests:
+          cpu: 1m
+          memory: 10M
+      placement: {}
+
+
+serviceMonitor:
+  create: true
+  labels:
+    release: kube-prometheus-stack

infrastructure/scylla/manifests/istio-virtualservice.yaml

@@ -0,0 +1,16 @@
+apiVersion: networking.istio.io/v1beta1
+kind: VirtualService
+metadata:
+  name: scylla-manager
+  namespace: scylla-manager
+spec:
+  hosts:
+    - scylla-manager.kube.huskypup.net
+  gateways:
+    - istio-system/edge
+  http:
+    - route:
+        - destination:
+            host: scylla-manager.scylla-manager.svc.cluster.local
+            port:
+              number: 80

infrastructure/scylla/operator-values.yaml

@@ -0,0 +1,18 @@
+# infrastructure/scylla/operator-values.yaml
+# Scylla Operator
+replicas: 2
+logLevel: 2
+
+webhook:
+  # Use cert-manager to generate the self-signed cert
+  createSelfSignedCertificate: true
+
+resources:
+  requests:
+    cpu: 100m
+    memory: 128Mi
+
+webhookServerResources:
+  requests:
+    cpu: 10m
+    memory: 64Mi

infrastructure/scylla/values.yaml

@@ -0,0 +1,51 @@
+# ScyllaCluster (via scylla-operator Helm chart)
+# Bootstrap-friendly defaults:
+# - Uses Rook-Ceph RBD PVCs (rook-ceph-block)
+# - No required node affinity/tolerations
+# - Enables ServiceMonitor for Prometheus
+
+scyllaImage:
+  repository: scylladb/scylla
+  tag: 2025.3.3
+
+agentImage:
+  repository: scylladb/scylla-manager-agent
+  tag: 3.7.0
+
+# Homelab: allow relaxed settings on smaller nodes.
+developerMode: true
+
+# Enable Alternator (DynamoDB API) only if you want it.
+alternator:
+  enabled: false
+  insecureEnableHTTP: false
+
+# Optional: enable cpu pinning once you dedicate nodes.
+cpuset: false
+
+# Datacenter + racks
+# NOTE: This chart creates a ScyllaCluster CR; the operator creates the StatefulSets.
+datacenter: homelab-dc
+racks:
+  - name: rack-a
+    members: 3
+    storage:
+      storageClassName: rook-ceph-block
+      capacity: 50Gi
+    resources:
+      requests:
+        cpu: 1
+        memory: 4Gi
+      limits:
+        cpu: 2
+        memory: 8Gi
+    # No strict scheduling constraints for bootstrap.
+    placement:
+      nodeAffinity: {}
+      tolerations: []
+
+# Prometheus
+serviceMonitor:
+  create: true
+  labels:
+    release: kube-prometheus-stack