Feat/split sts

valkey/templates/_helpers.tpl

@@ -196,8 +196,11 @@ Validate sentinel configuration
   {{- if not .Values.replica.enabled }}
     {{- fail "Sentinel mode requires replication to be enabled. Please set replica.enabled=true along with sentinel.enabled=true" }}
   {{- end }}
-  {{- if lt (add (int .Values.replica.replicas) 1) 3 }}
-    {{- fail "Sentinel mode requires at least 3 Valkey pods (replicas: 2) for a stable quorum." }}
+  {{- if lt (int .Values.replica.sentinel.replicas) 3 }}
+    {{- fail "Sentinel mode requires at least 3 pods for a stable quorum." }}
+  {{- end }}
+  {{- if gt (int .Values.replica.sentinel.quorum) (int .Values.replica.sentinel.replicas) }}
+    {{- fail (printf "Sentinel quorum (%d) cannot be greater than sentinels count (%d)." (int .Values.replica.sentinel.quorum) (int .Values.replica.sentinel.replicas)) }}
   {{- end }}
   {{- if and .Values.auth.enabled (not (hasKey .Values.auth.aclUsers .Values.replica.replicationUser)) }}
     {{- fail (printf "Sentinel with auth requires replication user '%s' to be defined in auth.aclUsers" .Values.replica.replicationUser) }}
@@ -219,3 +222,11 @@ Sentinel headless service name
 {{ include "valkey.fullname" . }}-sentinel-headless
 {{- end -}}
 
+{{/*
+Validate haproxy is used in replica mode
+*/}}
+{{- define "valkey.validateHaproxyRequirements" -}}
+{{- if and .Values.haproxy.enabled (or (not .Values.replica.enabled) (not .Values.replica.sentinel.enabled)) }}
+  {{- fail "Haproxy requires replica mode and sentinel to handle incompatible sentinel clients usescases." }}
+{{- end }}
+{{- end -}}

valkey/templates/deploy_valkey.yaml

@@ -3,6 +3,7 @@
 {{- $storage := .Values.dataStorage }}
 {{- $createPVC := and $storage.enabled (not (empty $storage.requestedSize)) (empty $storage.persistentVolumeClaimName) }}
 {{- include "valkey.validateAuthConfig" . }}
+{{- include "valkey.validateHaproxyRequirements" . }}
 apiVersion: apps/v1
 kind: Deployment
 metadata:

valkey/templates/haproxy-configmap.yaml

@@ -19,19 +19,15 @@ data:
       timeout server {{ .Values.haproxy.config.timeout.server }}
       retries 3
 
-    frontend valkey_frontend_write
-      bind *:{{ .Values.haproxy.service.port | default 6379 }}
+    {{- range .Values.haproxy.service.ports }}
+    frontend valkey_frontend_{{ .name }}
+      bind *:{{ .port }}
       mode tcp
       option tcplog
-      default_backend valkey_backend_master
-
-    frontend valkey_frontend_read
-      bind *:{{ .Values.haproxy.service.readPort | default 6380 }}
-      mode tcp
-      option tcplog
-      default_backend valkey_backend_read
+      default_backend valkey_backend_{{ .name }}
+    {{- end }}
 
-    backend valkey_backend_master
+    backend valkey_backend_{{ (index .Values.haproxy.service.ports 0).name }}
       mode tcp
       # no-check: sentinel-watcher solely controls which server is active via
       # the runtime socket (set server addr + enable/disable server).
@@ -41,7 +37,7 @@ data:
       server valkey-{{ $i }} 127.0.0.1:6379 check-send-proxy disabled
       {{- end }}
 
-    backend valkey_backend_read
+    backend valkey_backend_{{ (index .Values.haproxy.service.ports 1).name }}
       mode tcp
       option tcp-check
 
@@ -72,7 +68,7 @@ data:
     # Sentinel watcher: polls Sentinel for master changes and updates HAProxy
     # via the runtime socket.
     #
-    # KEY DESIGN: All valkey_backend_master servers are configured as
+    # KEY DESIGN: All valkey_backend_{{ (index .Values.haproxy.service.ports 0).name }} servers are configured as
     # `no-check disabled` in haproxy.cfg. This watcher is the SOLE controller
     # of which server is active. It resolves the master hostname to a Pod IP
     # and uses `set server addr` so HAProxy connects directly to the IP,
@@ -85,7 +81,8 @@ data:
     POLL_INTERVAL="{{ .Values.haproxy.config.checkInterval | default 2 }}"
     VALKEY_PORT="{{ .Values.service.port }}"
     HAPROXY_SOCKET="/var/run/haproxy/admin.sock"
-    BACKEND="valkey_backend_master"
+    BACKEND="valkey_backend_{{ (index .Values.haproxy.service.ports 0).name }}"
+    TOTAL_SENTINELS="{{ .Values.replica.sentinel.replicas }}"
     TOTAL_SERVERS="{{ add (int .Values.replica.replicas) 1 }}"
 
     # Logging to stderr prevents polluting stdout
@@ -148,8 +145,8 @@ data:
 
       # 1. Ask Sentinels who the master is
       i=0
-      while [ "${i}" -lt "${TOTAL_SERVERS}" ]; do
-        S_HOST="{{ include "valkey.fullname" . }}-${i}.{{ include "valkey.headlessServiceName" . }}.{{ .Release.Namespace }}.svc.{{ .Values.clusterDomain }}"
+      while [ "${i}" -lt "${TOTAL_SENTINELS}" ]; do
+        S_HOST="{{ include "valkey.fullname" . }}-sentinel-${i}.{{ include "valkey.sentinel.headlessServiceName" . }}.{{ .Release.Namespace }}.svc.{{ .Values.clusterDomain }}"
 
         RESP=$(sentinel_cmd "${S_HOST}" sentinel get-master-addr-by-name "${MASTER_SET}" 2>/dev/null) || {
           i=$((i + 1))

valkey/templates/haproxy-deployment.yaml

@@ -1,4 +1,5 @@
 {{- if .Values.haproxy.enabled }}
+{{- include "valkey.validateHaproxyRequirements" . }}
 apiVersion: apps/v1
 kind: Deployment
 metadata:
@@ -47,12 +48,11 @@ spec:
           {{- end }}
           imagePullPolicy: {{ .Values.haproxy.image.pullPolicy }}
           ports:
-            - name: valkey-write
-              containerPort: 6379
-              protocol: TCP
-            - name: valkey-read
-              containerPort: 6380
+            {{- range .Values.haproxy.service.ports }}
+            - name: {{ .name }}
+              containerPort: {{ .port }}
               protocol: TCP
+            {{- end }}
           volumeMounts:
             - name: haproxy-config
               mountPath: /usr/local/etc/haproxy/haproxy.cfg

valkey/templates/haproxy-service.yaml

@@ -13,14 +13,19 @@ metadata:
 spec:
   type: {{ .Values.haproxy.service.type }}
   ports:
-    - port: {{ .Values.haproxy.service.port | default 6379 }}
-      targetPort: {{ .Values.haproxy.service.port | default 6379 }}
+    {{- $serviceType := .Values.haproxy.service.type }}
+    {{- range .Values.haproxy.service.ports }}
+    - port: {{ .port }}
+      targetPort: {{ .port }}
       protocol: TCP
-      name: valkey-write
-    - port: {{ .Values.haproxy.service.readPort | default 6380 }}
-      targetPort: {{ .Values.haproxy.service.readPort | default 6380 }}
-      protocol: TCP
-      name: valkey-read
+      name: {{ .name }}
+      {{- if and (eq $serviceType "NodePort") .nodePort }}
+      nodePort: {{ .nodePort }}
+      {{- end }}
+      {{- if .appProtocol }}
+      appProtocol: {{ .appProtocol }}
+      {{- end }}
+    {{- end }}
   selector:
     {{- include "valkey.selectorLabels" . | nindent 4 }}
     app.kubernetes.io/component: haproxy

valkey/templates/init_config.yaml

@@ -11,6 +11,7 @@ data:
 
     # Default config paths
     VALKEY_CONFIG=${VALKEY_CONFIG_PATH:-/data/conf/valkey.conf}
+    BOOTSTRAP_FILE="/data/conf/.bootstrap"
 
     LOGFILE="/data/init.log"
     DATA_DIR="/data/conf"
@@ -175,7 +176,10 @@ data:
     # 1. Determine Initial Topology or Restored State
     REPLICA_TARGET=""
     if [ "${SENTINEL_MODE:-false}" = "true" ]; then
-      if [ -n "$SAVED_REPLICA_LINE" ]; then
+      if [ -f "$BOOTSTRAP_FILE" ] && [ -z "$SAVED_REPLICA_LINE" ] ; then
+        log "Sentinel mode: Restoring master state from previous run"
+        REPLICA_TARGET=""
+      elif [ -n "$SAVED_REPLICA_LINE" ]; then
         # CRITICAL: Restore topology written by Sentinel (CONFIG REWRITE)
         log "Sentinel Mode: Restoring replication state from previous run: $SAVED_REPLICA_LINE"
         REPLICA_TARGET="$SAVED_REPLICA_LINE"
@@ -269,3 +273,4 @@ data:
       log "Appending files in /extravalkeyconfigs/"
       cat /extravalkeyconfigs/* >>"$VALKEY_CONFIG"
     fi
+    touch $BOOTSTRAP_FILE

valkey/templates/sentinel-configmap.yaml

@@ -19,17 +19,46 @@ data:
     FAILOVER_TIMEOUT="{{ .Values.replica.sentinel.failoverTimeout }}"
     PARALLEL_SYNCS="{{ .Values.replica.sentinel.parallelSyncs }}"
 
-    # Master is always pod-0 of the valkey statefulset initially
-    MASTER_HOST="{{ include "valkey.fullname" . }}-0.{{ include "valkey.headlessServiceName" . }}.{{ .Release.Namespace }}.svc.{{ .Values.clusterDomain }}"
-    MASTER_PORT="{{ .Values.service.port }}"
-
-    SENTINEL_CONF="/data/sentinel.conf"
-
     # Logging function (outputs to stderr to avoid polluting stdout command substitutions)
     log() {
       echo "$(date) $1" >&2
     }
 
+    has_master() {
+      for i in $(seq 0 "{{ (int .Values.replica.replicas) }}"); do
+        valkey-cli -h {{ include "valkey.fullname" . }}-$i.{{ include "valkey.headlessServiceName" . }}.{{ .Release.Namespace }}.svc.{{ .Values.clusterDomain }} INFO replication 2>/dev/null \
+          | grep -q "^role:master" && return 0
+        done
+      return 1
+    }
+
+    # Wait until a master is UP
+    until has_master; do
+      log "Master not ready"
+      sleep 5
+    done
+
+    # Discover who is the current master
+    MASTER_HOST=""
+    for i in $(seq 0 "{{ (int .Values.replica.replicas) }}"); do
+        HOST="{{ include "valkey.fullname" . }}-$i.{{ include "valkey.headlessServiceName" . }}.{{ .Release.Namespace }}.svc.{{ .Values.clusterDomain }}"
+
+        ROLE=$(redis-cli -h "$HOST" INFO replication 2>/dev/null \
+          | awk -F: '/^role:/ {print $2}' \
+          | tr -d '\r')
+
+        if [ "$ROLE" = "master" ]; then
+            MASTER_HOST="$HOST"
+            break
+        fi
+
+        i=$((i + 1))
+    done
+    MASTER_PORT="{{ .Values.service.port }}"
+    log "Found existing master $MASTER_HOST"
+
+    SENTINEL_CONF="/sentinel-data/sentinel.conf"
+
     log "Initializing Sentinel configuration..."
 
     {{- if .Values.auth.enabled }}
@@ -78,7 +107,7 @@ data:
     # Announce IP for discovery in Kubernetes
     sentinel resolve-hostnames yes
     sentinel announce-hostnames yes
-    sentinel announce-ip ${POD_NAME}.{{ include "valkey.headlessServiceName" . }}.{{ .Release.Namespace }}.svc.{{ .Values.clusterDomain }}
+    sentinel announce-ip ${POD_NAME}.{{ include "valkey.sentinel.headlessServiceName" . }}.{{ .Release.Namespace }}.svc.{{ .Values.clusterDomain }}
     sentinel announce-port ${SENTINEL_PORT}
 
     # Monitor the master