universe/k8s/zoo-mainnet/zood-mv-statefulset.yaml

# Zoo mainnet validators — Zoo EVM chainID 200200 on Lux primary networkID 1.
#
# THIS FILE EXISTS BECAUSE THE STATEFULSET DID NOT. zood-mv ran for months with
# no manifest anywhere in any repo: it was created by hand and every subsequent
# change was a `kubectl patch`, so the cluster was the only record of what Zoo
# mainnet actually was. Generated from the live object and reconciled — keep it
# that way, and change Zoo mainnet by editing this file and applying it.
#
# THE TWO IMAGES BELOW ARE ONE VALUE. init.sh does
#     cp /luxd/build/plugins/* /data/plugins/
# and the node runs --plugin-dir=/data/plugins, so the INIT container's image —
# not the node's — decides which EVM plugin binary gets loaded, on every restart.
# When init sat at v1.36.2 under a v1.36.55 node, every C-Chain init died with
#     VM initialization failed error="zap decode initialize response: unexpected EOF"
# while the plugin itself logged "ZAP VM initialized successfully": the plugin was
# healthy, the CLIENT could not decode a reply carrying a field (quasarExport) that
# the older plugin never emitted. 200200 had no C-Chain at all until they matched.
#
# LIVENESS MUST NOT DEPEND ON THE C-CHAIN. Liveness KILLS. It was an
# `eth_blockNumber` exec probe, so a node that had not finished bootstrapping was
# SIGTERMed at 300s + 10*30s = 600s, and a node killed every 600s never finishes
# bootstrapping — the probe suppressed the thing that would have cleared it.
# "Is the process wedged?" (liveness) and "should this node serve RPC?"
# (readiness, which only de-registers from the Service) are different questions.
apiVersion: apps/v1
kind: StatefulSet
metadata:
  labels:
    app: zood-mv
  name: zood-mv
spec:
  persistentVolumeClaimRetentionPolicy:
    whenDeleted: Retain
    whenScaled: Retain
  podManagementPolicy: Parallel
  replicas: 5
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      app: zood-mv
  serviceName: zood-mv
  template:
    metadata:
      labels:
        app: zood-mv
    spec:
      affinity:
        podAntiAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - podAffinityTerm:
              labelSelector:
                matchLabels:
                  app: zood-mv
              topologyKey: kubernetes.io/hostname
            weight: 100
      containers:
      - command:
        - /bin/sh
        - /scripts/startup.sh
        image: ghcr.io/luxfi/node:v1.36.181@sha256:3ff4841e76b407e774b59ea33cdb691bc0b5b6dfdebc0e5ea71bd1119d059cad
        imagePullPolicy: IfNotPresent
        livenessProbe:
          failureThreshold: 6
          httpGet:
            path: /v1/health/liveness
            port: 9630
            scheme: HTTP
          initialDelaySeconds: 120
          periodSeconds: 30
          successThreshold: 1
          timeoutSeconds: 10
        name: zood
        # GOMEMLIMIT is the SOFT ceiling that makes Go GC harder as it nears the
        # cgroup limit instead of sailing into a SIGKILL. Its absence is not a
        # restart-noise problem: an OOM landing mid-write can leave a node holding
        # a divergent CONSENSUS-CERTIFIED chain, which it then GOSSIPS — that is
        # what took hanzo-mainnet down for hours on 2026-08-06 and needed a
        # node-by-node state transplant to clear. Held below the 3Gi hard limit.
        env:
        - name: GOMEMLIMIT
          value: 2GiB
        ports:
        - containerPort: 9630
          name: http
          protocol: TCP
        - containerPort: 9631
          name: staking
          protocol: TCP
        readinessProbe:
          failureThreshold: 6
          httpGet:
            path: /v1/health/liveness
            port: 9630
            scheme: HTTP
          periodSeconds: 10
          successThreshold: 1
          timeoutSeconds: 5
        resources:
          limits:
            cpu: '2'
            memory: 3Gi
          requests:
            cpu: 300m
            memory: 900Mi
        startupProbe:
          failureThreshold: 60
          initialDelaySeconds: 10
          periodSeconds: 10
          successThreshold: 1
          tcpSocket:
            port: 9630
          timeoutSeconds: 1
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /data
          name: data
        - mountPath: /scripts
          name: scripts
        - mountPath: /genesis
          name: genesis
          readOnly: true
      dnsPolicy: ClusterFirst
      imagePullSecrets:
      - name: ghcr-zooai
      - name: ghcr-secret
      - name: hanzo
      initContainers:
      - command:
        - /bin/sh
        - /scripts/init.sh
        image: ghcr.io/luxfi/node:v1.36.181@sha256:3ff4841e76b407e774b59ea33cdb691bc0b5b6dfdebc0e5ea71bd1119d059cad
        imagePullPolicy: IfNotPresent
        name: setup
        resources: {}
        securityContext:
          runAsGroup: 0
          runAsUser: 0
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /data
          name: data
        - mountPath: /scripts
          name: scripts
        - mountPath: /staking-keys
          name: staking
          readOnly: true
      restartPolicy: Always
      schedulerName: default-scheduler
      securityContext:
        fsGroup: 1000
        runAsGroup: 1000
        runAsUser: 1000
      terminationGracePeriodSeconds: 30
      volumes:
      - configMap:
          defaultMode: 493
          name: zood-mv-startup
        name: scripts
      - configMap:
          defaultMode: 420
          name: zood-mv-genesis
        name: genesis
      - name: staking
        secret:
          defaultMode: 256
          secretName: zood-mv-staking
  updateStrategy:
    type: RollingUpdate
  volumeClaimTemplates:
  - apiVersion: v1
    kind: PersistentVolumeClaim
    metadata:
      name: data
    spec:
      accessModes:
      - ReadWriteOnce
      resources:
        requests:
          storage: 40Gi
      storageClassName: do-block-storage
      volumeMode: Filesystem