feat: support raft

Chart.yaml

@@ -2,5 +2,5 @@ apiVersion: v2
 name: pgdog-control
 description: PgDog Control
 type: application
-version: 0.2.18
-appVersion: "84c7c56a"
+version: 0.3.0
+appVersion: "f477677f"

templates/_helpers.tpl

@@ -51,6 +51,35 @@ the same cluster don't collide.
 {{- printf "%s-redis" .Release.Name | trunc 63 | trimSuffix "-" }}
 {{- end }}
 
+{{/* Stable DNS service and token Secret for the Raft members. */}}
+{{- define "pgdog-control.raft.fullname" -}}
+{{- printf "%s-raft" .Release.Name | trunc 63 | trimSuffix "-" -}}
+{{- end }}
+
+{{/*
+Reuse the installed token on upgrades. Cache a newly generated token in the
+render context so the Secret, ConfigMap, and pod checksum all agree.
+*/}}
+{{- define "pgdog-control.raft.token" -}}
+{{- if .Values.raft.token -}}
+{{- if not (regexMatch "^[!-~]+$" .Values.raft.token) -}}
+{{- fail "raft.token must contain only non-whitespace printable ASCII characters" -}}
+{{- end -}}
+{{- .Values.raft.token -}}
+{{- else -}}
+{{- if not (hasKey .Values.raft "_generatedToken") -}}
+{{- $existing := lookup "v1" "Secret" .Release.Namespace (include "pgdog-control.raft.fullname" .) | default dict -}}
+{{- $data := $existing.data | default dict -}}
+{{- $token := index $data "token" | default "" | b64dec -}}
+{{- if not $token -}}
+{{- $token = randAlphaNum 64 -}}
+{{- end -}}
+{{- $_ := set .Values.raft "_generatedToken" $token -}}
+{{- end -}}
+{{- .Values.raft._generatedToken -}}
+{{- end -}}
+{{- end }}
+
 {{/*
 Redis URL used by the control plane. redis.url is the public chart setting;
 control.config.redis.url remains supported for backwards compatibility.

templates/configmap.yaml

@@ -112,23 +112,18 @@ data:
     {{- end }}
     {{- end }}
 
-    {{- with $config.leader }}
-
-    [leader]
-    {{- if hasKey . "enabled" }}
-    enabled = {{ .enabled }}
-    {{- end }}
-    {{- with .lease_name }}
-    lease_name = {{ . | quote }}
-    {{- end }}
-    {{- with .lease_duration_secs }}
-    lease_duration_secs = {{ . }}
-    {{- end }}
-    {{- with .renew_interval_secs }}
-    renew_interval_secs = {{ . }}
-    {{- end }}
-    {{- with .release_timeout_secs }}
-    release_timeout_secs = {{ . }}
+    {{- if .Values.raft.enabled }}
+
+    [raft]
+    token = {{ include "pgdog-control.raft.token" . | quote }}
+    storage_path = {{ printf "%s/raft.redb" (trimSuffix "/" .Values.raft.persistence.mountPath) | quote }}
+    cluster_name = {{ .Values.raft.cluster_name | default .Release.Name | quote }}
+    sequence_cache_size = {{ .Values.raft.sequence_cache_size }}
+    {{- range $id := until 3 }}
+
+    [[raft.members]]
+    id = {{ $id }}
+    address = {{ printf "http://%s-%d.%s.%s.svc:%v" (include "pgdog-control.control.fullname" $) $id (include "pgdog-control.raft.fullname" $) $.Release.Namespace $.Values.control.port | quote }}
     {{- end }}
     {{- end }}
 

templates/deployment.yaml

@@ -1,31 +1,47 @@
 apiVersion: apps/v1
-kind: Deployment
+kind: {{ ternary "StatefulSet" "Deployment" .Values.raft.enabled }}
 metadata:
   name: {{ include "pgdog-control.control.fullname" . }}
   labels:
     {{- include "pgdog-control.labels" . | nindent 4 }}
 spec:
-  replicas: {{ .Values.control.replicas | default 1 }}
+  replicas: {{ ternary 3 (int (.Values.control.replicas | default 1)) .Values.raft.enabled }}
+  {{- if .Values.raft.enabled }}
+  serviceName: {{ include "pgdog-control.raft.fullname" . }}
+  # Start all members without waiting for an initial quorum to become ready.
+  podManagementPolicy: Parallel
+  updateStrategy:
+    type: RollingUpdate
+  volumeClaimTemplates:
+  - metadata:
+      name: raft
+    spec:
+      accessModes: ["ReadWriteOnce"]
+      {{- if eq .Values.raft.persistence.storageClass "-" }}
+      storageClassName: ""
+      {{- else if .Values.raft.persistence.storageClass }}
+      storageClassName: {{ .Values.raft.persistence.storageClass | quote }}
+      {{- end }}
+      resources:
+        requests:
+          storage: {{ .Values.raft.persistence.size | quote }}
+  {{- end }}
   selector:
     matchLabels:
       {{- include "pgdog-control.selectorLabels" . | nindent 6 }}
-  # Readiness is leader-aware so only the elected control pod receives
-  # Service traffic. New pods normally start as followers, so they do not
-  # become Ready while the old leader still holds the Lease. Allowing the
-  # whole old ReplicaSet to be unavailable lets Kubernetes terminate the
-  # old leader, release the Lease, and promote a new-version pod instead
-  # of waiting forever for a follower to become Ready.
+  {{- if not .Values.raft.enabled }}
   strategy:
     type: RollingUpdate
     rollingUpdate:
       maxSurge: 0
       maxUnavailable: 100%
+  {{- end }}
   template:
     metadata:
       annotations:
         meta.helm.sh/release-name: {{ .Release.Name | quote }}
         meta.helm.sh/release-namespace: {{ .Release.Namespace | quote }}
-        # Roll the deployment when any rendered config / secret content
+        # Roll the workload when any rendered config / secret content
         # changes. Each annotation hashes the *template output*, not
         # values directly, so transformations applied inside the
         # template (defaults, lookups, randAlphaNum on first install)
@@ -46,7 +62,17 @@ spec:
       labels:
         {{- include "pgdog-control.selectorLabels" . | nindent 8 }}
     spec:
-      {{- if gt (int (.Values.control.replicas | default 1)) 1 }}
+      {{- if .Values.raft.enabled }}
+      affinity:
+        podAntiAffinity:
+          preferredDuringSchedulingIgnoredDuringExecution:
+          - weight: 100
+            podAffinityTerm:
+              topologyKey: kubernetes.io/hostname
+              labelSelector:
+                matchLabels:
+                  {{- include "pgdog-control.selectorLabels" . | nindent 18 }}
+      {{- else if gt (int (.Values.control.replicas | default 1)) 1 }}
       topologySpreadConstraints:
       - maxSkew: 1
         topologyKey: kubernetes.io/hostname
@@ -88,6 +114,13 @@ spec:
         env:
         - name: CONTROL_CONFIG
           value: /etc/pgdog-control/control.toml
+        {{- if .Values.raft.enabled }}
+        # control2 extracts the StatefulSet ordinal from the pod name.
+        - name: RAFT_NODE_ID
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.name
+        {{- end }}
         - name: HOME
           value: /var/lib/pgdog-control
         - name: XDG_CACHE_HOME
@@ -168,6 +201,10 @@ spec:
           readOnly: true
         - name: runtime
           mountPath: /var/lib/pgdog-control
+        {{- if .Values.raft.enabled }}
+        - name: raft
+          mountPath: {{ .Values.raft.persistence.mountPath | quote }}
+        {{- end }}
         - name: tmp
           mountPath: /tmp
         ports:

templates/networkpolicy.yaml

@@ -13,6 +13,15 @@ spec:
   - Ingress
   - Egress
   ingress:
+  {{- if .Values.raft.enabled }}
+  - from:
+    - podSelector:
+        matchLabels:
+          {{- include "pgdog-control.selectorLabels" . | nindent 10 }}
+    ports:
+    - protocol: TCP
+      port: {{ .Values.control.port }}
+  {{- end }}
   - from:
     - namespaceSelector:
         matchLabels:
@@ -24,6 +33,15 @@ spec:
   {{- toYaml . | nindent 2 }}
   {{- end }}
   egress:
+  {{- if .Values.raft.enabled }}
+  - to:
+    - podSelector:
+        matchLabels:
+          {{- include "pgdog-control.selectorLabels" . | nindent 10 }}
+    ports:
+    - protocol: TCP
+      port: {{ .Values.control.port }}
+  {{- end }}
   {{- if .Values.redis.enabled }}
   - to:
     - podSelector:

templates/raft-secret.yaml

@@ -0,0 +1,11 @@
+{{- if .Values.raft.enabled }}
+apiVersion: v1
+kind: Secret
+metadata:
+  name: {{ include "pgdog-control.raft.fullname" . }}
+  labels:
+    {{- include "pgdog-control.labels" . | nindent 4 }}
+type: Opaque
+data:
+  token: {{ include "pgdog-control.raft.token" . | b64enc | quote }}
+{{- end }}

templates/raft-service.yaml

@@ -0,0 +1,19 @@
+{{- if .Values.raft.enabled }}
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ include "pgdog-control.raft.fullname" . }}
+  labels:
+    {{- include "pgdog-control.labels" . | nindent 4 }}
+spec:
+  clusterIP: None
+  # Peers must be discoverable while the cluster is starting or recovering.
+  publishNotReadyAddresses: true
+  ports:
+  - port: {{ .Values.control.port }}
+    targetPort: http
+    protocol: TCP
+    name: http
+  selector:
+    {{- include "pgdog-control.selectorLabels" . | nindent 4 }}
+{{- end }}

templates/rbac.yaml

@@ -54,35 +54,6 @@ subjects:
   - kind: ServiceAccount
     name: {{ include "pgdog-control.control.serviceAccountName" . }}
     namespace: {{ .Release.Namespace }}
----
-# Namespace-scoped access for control-plane leader election.
-apiVersion: rbac.authorization.k8s.io/v1
-kind: Role
-metadata:
-  name: {{ include "pgdog-control.control.clusterFullname" . }}-leader
-  namespace: {{ .Release.Namespace }}
-  labels:
-    {{- include "pgdog-control.labels" . | nindent 4 }}
-rules:
-  - apiGroups: ["coordination.k8s.io"]
-    resources: ["leases"]
-    verbs: ["get", "list", "watch", "create", "update", "patch", "delete"]
----
-apiVersion: rbac.authorization.k8s.io/v1
-kind: RoleBinding
-metadata:
-  name: {{ include "pgdog-control.control.clusterFullname" . }}-leader
-  namespace: {{ .Release.Namespace }}
-  labels:
-    {{- include "pgdog-control.labels" . | nindent 4 }}
-roleRef:
-  apiGroup: rbac.authorization.k8s.io
-  kind: Role
-  name: {{ include "pgdog-control.control.clusterFullname" . }}-leader
-subjects:
-  - kind: ServiceAccount
-    name: {{ include "pgdog-control.control.serviceAccountName" . }}
-    namespace: {{ .Release.Namespace }}
 {{- range $ns := .Values.control.rbac.writeNamespaces }}
 ---
 # Namespace-scoped write access for workloads the control plane manages.

test/values-full.yaml

@@ -37,12 +37,6 @@ control:
       autoreload: "immediately"
     autoscaling:
       pool_size: true
-    leader:
-      enabled: true
-      lease_name: pgdog-control-leader
-      lease_duration_secs: 20
-      renew_interval_secs: 7
-      release_timeout_secs: 4
     helm:
       chart: pgdog
       repo: pgdogdev

test/values-raft.yaml

@@ -0,0 +1,2 @@
+raft:
+  enabled: true

values.yaml

@@ -11,6 +11,21 @@ nodeSelector: {}
 # tolerations allows pods to be scheduled on nodes with matching taints
 tolerations: []
 
+raft:
+  # Run three control pods with stable identities and durable Raft storage.
+  # Prefer separate machines, but allow pods to share a node when needed.
+  enabled: false
+  # Shared peer token. Empty generates a token retained in a Kubernetes Secret.
+  token: ""
+  # Empty defaults to the Helm release name.
+  cluster_name: ""
+  sequence_cache_size: 1000
+  persistence:
+    size: 1Gi
+    # Empty uses the cluster's default StorageClass; "-" disables dynamic provisioning.
+    storageClass: ""
+    mountPath: /var/lib/pgdog-control/raft
+
 control:
   port: 8080
   aws:
@@ -115,12 +130,6 @@ control:
       # autoreload: off
     autoscaling: {}
       # pool_size: false
-    leader: {}
-      # enabled: true
-      # lease_name: ""              # Empty means derive from Helm release.
-      # lease_duration_secs: 15
-      # renew_interval_secs: 5
-      # release_timeout_secs: 3
     helm: {}
       # chart: pgdog
       # repo: pgdogdev