Couchbase 在生产中的高可用性：XDCR、Kubernetes Operator 和多区域部署

#!/bin/bash
# couchbase-backup.sh — Automated backup to S3-compatible storage

set -euo pipefail

CLUSTER_HOST="cb-production-srv.couchbase.svc.cluster.local"
BACKUP_DIR="/backup/couchbase"
REPO_NAME="prod-$(date +%Y%m%d)"
S3_BUCKET="s3://couchbase-backups/production"
RETENTION_DAYS=14

log() { echo "[$(date +'%Y-%m-%d %H:%M:%S')] $*"; }

log "Starting Couchbase backup for cluster: $CLUSTER_HOST"

if [ ! -d "$BACKUP_DIR/$REPO_NAME" ]; then
  log "Configuring new backup repository: $REPO_NAME"
  cbbackupmgr config \
    --archive "$BACKUP_DIR" \
    --repo "$REPO_NAME" \
    --include-data production-data \
    --include-data user-profiles
fi

log "Running incremental backup..."
cbbackupmgr backup \
  --archive "$BACKUP_DIR" \
  --repo "$REPO_NAME" \
  --cluster "couchbase://$CLUSTER_HOST" \
  --username "$CB_USERNAME" \
  --password "$CB_PASSWORD" \
  --threads 4 \
  --no-progress-bar

BACKUP_SIZE=$(du -sh "$BACKUP_DIR/$REPO_NAME" | cut -f1)
log "Backup complete. Size: $BACKUP_SIZE"

log "Syncing to S3: $S3_BUCKET/$REPO_NAME"
aws s3 sync "$BACKUP_DIR/$REPO_NAME" "$S3_BUCKET/$REPO_NAME" \
  --storage-class STANDARD_IA \
  --sse aws:kms

log "Cleaning up backups older than $RETENTION_DAYS days..."
find "$BACKUP_DIR" -maxdepth 1 -name "prod-*" -mtime +"$RETENTION_DAYS" -exec rm -rf {} \;

log "Backup pipeline complete."

Couchbase备份 CRD（运营商管理）

自治操作员提供用于自动备份管理的 CRD：

apiVersion: couchbase.com/v2
kind: CouchbaseBackup
metadata:
  name: cb-daily-backup
  namespace: couchbase
spec:
  strategy: full_incremental
  full:
    schedule: "0 2 * * 0"   # Full backup every Sunday at 2 AM
  incremental:
    schedule: "0 2 * * 1-6" # Incremental Mon-Sat at 2 AM
  successfulJobsHistoryLimit: 5
  failedJobsHistoryLimit: 3
  backOffLimit: 3
  logRetention: 168h
  size: 100Gi
  s3bucket: s3://couchbase-backups/production
---
apiVersion: couchbase.com/v2
kind: CouchbaseBackupRestore
metadata:
  name: cb-restore-pitr
  namespace: couchbase
spec:
  backup: cb-daily-backup
  repo: "20260412"
  start:
    int: 1
  end:
    int: 5
  backOffLimit: 3

N1QL 查询性能调整

N1QL（JSON 的 SQL++）是 Couchbase 的查询语言。调整 N1QL 性能需要了解查询规划器、索引设计和服务器端优化。

索引策略：GSI 和 FTS

-- Global Secondary Index (GSI) for common query patterns

-- Composite index for user lookups
CREATE INDEX idx_users_email_status
  ON `user-profiles`(email, status)
  WHERE type = 'user'
  WITH {"num_replica": 1, "defer_build": false};

-- Covering index (includes all queried fields to avoid fetch)
CREATE INDEX idx_orders_covering
  ON `production-data`(customer_id, order_date, total_amount, status)
  WHERE type = 'order'
  WITH {"num_replica": 1};

-- Array index for nested documents
CREATE INDEX idx_order_items
  ON `production-data`(DISTINCT ARRAY item.product_id FOR item IN items END)
  WHERE type = 'order'
  WITH {"num_replica": 1};

-- Partial index for active records only
CREATE INDEX idx_active_sessions
  ON `production-data`(user_id, created_at)
  WHERE type = 'session' AND status = 'active'
  WITH {"num_replica": 1};

-- Adaptive index for dynamic query patterns
CREATE INDEX idx_adaptive_products
  ON `production-data`(DISTINCT PAIRS(self))
  WHERE type = 'product'
  WITH {"num_replica": 1};

-- Check index status
SELECT name, state, num_docs_indexed, num_docs_pending
FROM system:indexes
WHERE keyspace_id = 'production-data';

-- Analyze query execution plan
EXPLAIN SELECT u.name, u.email, COUNT(o.id) AS order_count
FROM `user-profiles` u
JOIN `production-data` o ON o.customer_id = u.id
WHERE u.status = 'active' AND o.type = 'order'
GROUP BY u.name, u.email
ORDER BY order_count DESC
LIMIT 100;

-- Use ADVISE to get index recommendations
ADVISE SELECT * FROM `production-data`
WHERE type = 'order'
  AND customer_id = 'cust-12345'
  AND order_date BETWEEN '2026-01-01' AND '2026-04-12'
ORDER BY order_date DESC;

查询优化技巧

-- Use PREPARE for frequently executed queries (cached plan)
PREPARE get_user_orders AS
SELECT o.id, o.order_date, o.total_amount, o.status
FROM `production-data` o
WHERE o.type = 'order'
  AND o.customer_id = $customer_id
ORDER BY o.order_date DESC
LIMIT $page_size OFFSET $page_offset;

-- Execute prepared statement
EXECUTE get_user_orders
USING {"customer_id": "cust-12345", "page_size": 20, "page_offset": 0};

-- Use META().id for direct key-value lookups (fastest path)
SELECT META().id, *
FROM `production-data`
USE KEYS ["order::2026-001", "order::2026-002", "order::2026-003"];

-- Correlated subquery with USE KEYS for joins
SELECT u.name,
  (SELECT o.id, o.total_amount
   FROM `production-data` o
   USE KEYS u.order_ids
   WHERE o.status = 'completed') AS completed_orders
FROM `user-profiles` u
WHERE META(u).id = 'user::12345';

-- Use INFER to understand document schema
INFER `production-data` WITH {"sample_size": 10000, "similarity_metric": 0.6};

内存管理和存储桶配置

Couchbase 的内存优先架构意味着 RAM 分配直接影响性能。每个服务都有自己的内存配额，桶共享数据服务配额。适当的大小可以防止缓存驱逐而降低延迟。

# Configure cluster-level memory quotas
/opt/couchbase/bin/couchbase-cli setting-cluster \
  --cluster localhost:8091 \
  --username Administrator \
  --password password \
  --cluster-ramsize 8192 \
  --cluster-index-ramsize 4096 \
  --cluster-fts-ramsize 2048 \
  --cluster-eventing-ramsize 2048 \
  --cluster-analytics-ramsize 4096

# Memory allocation guidelines:
# Data Service:      60% of available node RAM
# Index Service:     20% of available node RAM
# Search Service:    10% of available node RAM
# OS/overhead:       10% reserved

# Bucket memory sizing formula:
# Required RAM = (avg_doc_size * num_docs * 2.5) / num_data_nodes
# The 2.5 multiplier accounts for:
#   - Metadata overhead (~56 bytes per document)
#   - Internal fragmentation
#   - Replica copies in memory

# Create an optimized production bucket
curl -X POST http://localhost:8091/pools/default/buckets \
  -u Administrator:password \
  -d name=production-data \
  -d ramQuota=4096 \
  -d bucketType=couchbase \
  -d replicaNumber=2 \
  -d threadsNumber=8 \
  -d evictionPolicy=valueOnly \
  -d compressionMode=active \
  -d maxTTL=0 \
  -d conflictResolutionType=lww \
  -d flushEnabled=0 \
  -d durabilityMinLevel=majorityAndPersistActive

# Eviction policies:
# valueOnly  - Evicts document values but keeps metadata in RAM
#              Best for workloads where key access patterns are predictable
# fullEviction - Evicts both values and metadata
#                Best for very large datasets that exceed available RAM
# noEviction - (Ephemeral buckets only) Rejects writes when RAM is full
#              Best for caching use cases

TLS 加密和 RBAC

在生产中保护 Couchbase 需要对传输中的数据进行加密 (TLS)、基于角色的细粒度访问控制 (RBAC) 和审核日志记录。

# Enable TLS for all Couchbase services
/opt/couchbase/bin/couchbase-cli ssl-manage \
  --cluster localhost:8091 \
  --username Administrator \
  --password password \
  --set-node-certificate

# Enforce minimum TLS version
/opt/couchbase/bin/couchbase-cli setting-security \
  --cluster localhost:8091 \
  --username Administrator \
  --password password \
  --set \
  --tls-min-version tlsv1.2 \
  --tls-honor-cipher-order 1 \
  --hsts-max-age 31536000 \
  --hsts-preload-enabled 1

# Create application-specific RBAC users
/opt/couchbase/bin/couchbase-cli user-manage \
  --cluster localhost:8091 \
  --username Administrator \
  --password password \
  --set \
  --rbac-username app-service \
  --rbac-password "$(openssl rand -base64 32)" \
  --rbac-name "Application Service Account" \
  --roles 'data_reader[production-data],data_writer[production-data],query_select[production-data],query_insert[production-data],query_update[production-data],query_delete[production-data]' \
  --auth-domain local

# Create a read-only analytics user
/opt/couchbase/bin/couchbase-cli user-manage \
  --cluster localhost:8091 \
  --username Administrator \
  --password password \
  --set \
  --rbac-username analytics-reader \
  --rbac-password "$(openssl rand -base64 32)" \
  --roles 'data_reader[production-data],query_select[production-data],analytics_reader[production-data]' \
  --auth-domain local

# Enable audit logging
/opt/couchbase/bin/couchbase-cli setting-audit \
  --cluster localhost:8091 \
  --username Administrator \
  --password password \
  --set \
  --audit-enabled 1 \
  --audit-log-path /opt/couchbase/var/lib/couchbase/logs \
  --audit-log-rotate-interval 86400 \
  --audit-log-rotate-size 20971520

Kubernetes TLS 与证书管理器

# Certificate for Couchbase server TLS
apiVersion: cert-manager.io/v1
kind: Certificate
metadata:
  name: couchbase-server-tls
  namespace: couchbase
spec:
  secretName: couchbase-server-tls
  duration: 8760h   # 1 year
  renewBefore: 720h  # 30 days before expiry
  privateKey:
    algorithm: RSA
    size: 4096
  usages:
  - server auth
  - client auth
  dnsNames:
  - "*.cb-production.couchbase.svc.cluster.local"
  - "*.cb-production.couchbase.svc"
  - "cb-production-srv.couchbase.svc.cluster.local"
  - "localhost"
  issuerRef:
    name: couchbase-ca-issuer
    kind: ClusterIssuer

使用 Prometheus 导出器进行监控

Couchbase 通过其 REST API 公开丰富的指标。couchbase-exporter将这些转换为 Prometheus 格式以进行全面监控。

# Deploy Couchbase Prometheus Exporter
apiVersion: apps/v1
kind: Deployment
metadata:
  name: couchbase-exporter
  namespace: couchbase
spec:
  replicas: 1
  selector:
    matchLabels:
      app: couchbase-exporter
  template:
    metadata:
      labels:
        app: couchbase-exporter
      annotations:
        prometheus.io/scrape: "true"
        prometheus.io/port: "9091"
    spec:
      containers:
      - name: exporter
        image: couchbase/exporter:1.0.9
        args:
        - --couchbase-address=cb-production-srv.couchbase.svc.cluster.local
        - --couchbase-port=8091
        - --couchbase-username=$(CB_USERNAME)
        - --couchbase-password=$(CB_PASSWORD)
        - --server-address=0.0.0.0:9091
        - --per-node-refresh=5
        env:
        - name: CB_USERNAME
          valueFrom:
            secretKeyRef:
              name: cb-admin-credentials
              key: username
        - name: CB_PASSWORD
          valueFrom:
            secretKeyRef:
              name: cb-admin-credentials
              key: password
        ports:
        - containerPort: 9091
          name: metrics
        resources:
          requests:
            cpu: 100m
            memory: 128Mi
          limits:
            cpu: 500m
            memory: 256Mi
---
apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: couchbase-monitor
  namespace: couchbase
spec:
  selector:
    matchLabels:
      app: couchbase-exporter
  endpoints:
  - port: metrics
    interval: 15s
    path: /metrics

用于监控

• cb_bucket_ops_per_sec— 每个存储桶每秒的总操作数。确定正常吞吐量的基线并对异常情况发出警报。
• cb_bucket_mem_used_bytes— 存储桶内存使用情况。当接近 RAM 配额时发出警报，以防止驱逐。
• cb_bucket_cache_miss_ratio— 未命中缓存且需要磁盘获取的请求比率。应保持在 2% 以下以获得最佳性能。
• cb_bucket_disk_queue_items— 磁盘写入队列深度。队列不断增长表明磁盘 I/O 无法跟上写入吞吐量。
• cb_xdcr_changes_left— 等待 XDCR 复制的突变数量。表示跨区域复制滞后。
• cb_xdcr_docs_writing— 每秒通过 XDCR 复制的文档。
• cb_node_cpu_utilization_percent— 每个节点 CPU 使用情况。 Couchbase 是 CPU 的压缩和索引密集型产品。
• cb_bucket_vbucket_active_num— 每个节点的活动 vBucket 数量。数据节点之间应该大致均匀。
• cb_index_num_docs_pending— 文档等待索引更新。表示索引构建滞后。
• cb_n1ql_requests_per_sec— N1QL 查询吞吐量。结合平均延迟，可以识别查询性能问题。

Prometheus 警报规则

apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  name: couchbase-alerts
  namespace: couchbase
spec:
  groups:
  - name: couchbase.rules
    rules:
    - alert: CouchbaseNodeDown
      expr: cb_node_healthy == 0
      for: 1m
      labels:
        severity: critical
      annotations:
        summary: "Couchbase node {{ $labels.node }} is unhealthy"
    - alert: CouchbaseHighCacheMissRate
      expr: cb_bucket_cache_miss_ratio > 0.05
      for: 5m
      labels:
        severity: warning
      annotations:
        summary: "Cache miss rate {{ $value | humanizePercentage }} on bucket {{ $labels.bucket }}"
    - alert: CouchbaseXDCRLag
      expr: cb_xdcr_changes_left > 10000
      for: 10m
      labels:
        severity: warning
      annotations:
        summary: "XDCR replication lag: {{ $value }} pending mutations"
    - alert: CouchbaseDiskQueueGrowing
      expr: rate(cb_bucket_disk_queue_items[5m]) > 100
      for: 10m
      labels:
        severity: warning
      annotations:
        summary: "Disk queue growing on bucket {{ $labels.bucket }}"
    - alert: CouchbaseMemoryPressure
      expr: cb_bucket_mem_used_bytes / cb_bucket_mem_quota_bytes > 0.9
      for: 5m
      labels:
        severity: critical
      annotations:
        summary: "Memory usage at {{ $value | humanizePercentage }} for bucket {{ $labels.bucket }}"

SDK 连接字符串配置

Couchbase SDK 具有拓扑感知能力 - 它们维护内部集群映射并将操作直接路由到正确的节点。正确的 SDK 配置对于 HA 至关重要，可确保快速故障转移检测和针对瞬态错误的自动重试。

// Node.js SDK configuration for HA
const couchbase = require('couchbase');

const clusterConnStr = 'couchbases://cb-node1.example.com,cb-node2.example.com,cb-node3.example.com';

const cluster = await couchbase.connect(clusterConnStr, {
  username: process.env.CB_USERNAME,
  password: process.env.CB_PASSWORD,
  timeouts: {
    kvTimeout: 2500,           // Key-value operation timeout (ms)
    kvDurableTimeout: 10000,   // Durable write timeout
    queryTimeout: 75000,       // N1QL query timeout
    searchTimeout: 75000,      // FTS search timeout
    analyticsTimeout: 75000,   // Analytics query timeout
    connectTimeout: 10000,     // Initial connection timeout
    managementTimeout: 75000   // Management API timeout
  },
  security: {
    trustStorePath: '/etc/couchbase/ca.pem'
  },
  transactions: {
    durabilityLevel: couchbase.DurabilityLevel.MajorityAndPersistToActive,
    timeout: 15000
  }
});

const bucket = cluster.bucket('production-data');
const collection = bucket.defaultCollection();

// Durable write with observe-based durability
await collection.upsert('order::2026-001', orderDocument, {
  durabilityLevel: couchbase.DurabilityLevel.MajorityAndPersistToActive,
  timeout: 10000
});

// Read with replica fallback for HA
try {
  const result = await collection.get('user::12345');
} catch (err) {
  if (err instanceof couchbase.errors.TimeoutError) {
    const replicaResult = await collection.getAnyReplica('user::12345');
  }
}

# Java SDK configuration for HA
import com.couchbase.client.java.*;
import com.couchbase.client.java.env.*;
import java.time.Duration;

ClusterEnvironment env = ClusterEnvironment.builder()
    .timeoutConfig(TimeoutConfig.builder()
        .kvTimeout(Duration.ofMillis(2500))
        .kvDurableTimeout(Duration.ofSeconds(10))
        .queryTimeout(Duration.ofSeconds(75))
        .connectTimeout(Duration.ofSeconds(10))
        .build())
    .ioConfig(IoConfig.builder()
        .numKvConnections(4)
        .enableMutationTokens(true)
        .enableDnsSrv(true)
        .build())
    .securityConfig(SecurityConfig.builder()
        .enableTls(true)
        .trustCertificate(Paths.get("/etc/couchbase/ca.pem"))
        .build())
    .build();

Cluster cluster = Cluster.connect(
    "couchbases://cb-node1.example.com,cb-node2.example.com",
    ClusterOptions.clusterOptions("username", "password")
        .environment(env)
);

Kubernetes 用于 SDK 连接的 DNS 服务

# When using the Autonomous Operator, connect via the headless service:
# couchbase://cb-production-srv.couchbase.svc.cluster.local
#
# The operator creates these services:
# cb-production-srv      - Headless service for SDK auto-discovery
# cb-production-ui       - Web Console (port 8091/18091)
# cb-production-cloud    - External connectivity (NodePort/LoadBalancer)
#
# For external SDK access (outside Kubernetes), use:
# - NodePort with explicit node addresses
# - LoadBalancer with MetalLB (bare metal)
# - Ingress with TCP passthrough for port 11210 (SDK) and 11207 (SDK TLS)

Couchbase 用于边缘部署的移动和同步网关

Couchbase Mobile 将 Couchbase 生态系统扩展到边缘设备和移动应用。Couchbase Lite嵌入在iOS、Android和IoT设备上运行，而同步网关充当Couchbase Lite和Couchbase服务器之间的同步中间件。

// Sync Gateway configuration for production
{
  "interface": ":4984",
  "adminInterface": "127.0.0.1:4985",
  "logging": {
    "console": {
      "log_level": "info",
      "log_keys": ["HTTP", "Sync", "Auth", "Changes"]
    }
  },
  "databases": {
    "mobile-app": {
      "server": "couchbases://cb-production-srv.couchbase.svc.cluster.local",
      "bucket": "production-data",
      "username": "sync-gateway",
      "password": "${SG_PASSWORD}",
      "enable_shared_bucket_access": true,
      "import_docs": true,
      "num_index_replicas": 1,
      "delta_sync": {
        "enabled": true,
        "rev_max_age_seconds": 86400
      },
      "cache": {
        "channel_cache": {
          "max_number": 50000,
          "compact_high_watermark_pct": 80,
          "compact_low_watermark_pct": 60
        },
        "rev_cache": {
          "size": 5000,
          "shard_count": 16
        }
      },
      "users": {
        "GUEST": {"disabled": true}
      },
      "sync": "function(doc, oldDoc) { if (doc.type === 'user-data') { channel(doc.channels); requireAccess(doc.channels); } else { channel('public'); } }"
    }
  }
}

# Deploy Sync Gateway on Kubernetes
apiVersion: apps/v1
kind: Deployment
metadata:
  name: sync-gateway
  namespace: couchbase
spec:
  replicas: 3
  selector:
    matchLabels:
      app: sync-gateway
  template:
    metadata:
      labels:
        app: sync-gateway
    spec:
      containers:
      - name: sync-gateway
        image: couchbase/sync-gateway:3.1.4-enterprise
        args: ["/etc/sync-gateway/config.json"]
        ports:
        - containerPort: 4984
          name: public
        - containerPort: 4985
          name: admin
        resources:
          requests:
            cpu: "2"
            memory: 4Gi
          limits:
            cpu: "4"
            memory: 8Gi
        volumeMounts:
        - name: config
          mountPath: /etc/sync-gateway
      volumes:
      - name: config
        configMap:
          name: sync-gateway-config

容量规划和规模调整

正确的容量规划对于 Couchbase 性能和成本优化至关重要。 下表提供了基于工作负载层的规模调整指南：

尺码公式

# Data Service RAM sizing
# Required RAM per node = (num_documents * (doc_metadata_size + avg_value_size)) / num_data_nodes * (1 + num_replicas)
# doc_metadata_size = 56 bytes (fixed overhead per document)
# Include 25% headroom for fragmentation and growth

# Example: 100M documents, 1KB avg size, 3 data nodes, 1 replica
# RAM = (100,000,000 * (56 + 1024)) / 3 * 2 = ~72 GB per node
# With 25% headroom: ~90 GB per node

# Index Service RAM sizing (memory-optimized)
# RAM = total_index_size * 3 (for build/merge overhead)
# Use system:indexes to check current index sizes

# Disk sizing
# Disk = (num_documents * avg_doc_size * (1 + num_replicas)) * 3 (compaction headroom)
# Use SSD/NVMe with provisioned IOPS for predictable performance

灾难恢复和故障转移程序

全面的灾难恢复计划可确保基础设施故障超出自动故障转移范围时的业务连续性。

单节点故障（自动）

# Auto-failover handles single node failures automatically.
# Verify failover occurred:
/opt/couchbase/bin/couchbase-cli server-list \
  --cluster localhost:8091 \
  --username Administrator \
  --password password

# After replacing the failed node, add and rebalance:
/opt/couchbase/bin/couchbase-cli server-add \
  --cluster localhost:8091 \
  --username Administrator \
  --password password \
  --server-add new-node.example.com:8091 \
  --server-add-username Administrator \
  --server-add-password password \
  --services data,index

/opt/couchbase/bin/couchbase-cli rebalance \
  --cluster localhost:8091 \
  --username Administrator \
  --password password

完全集群故障（手动）

# Scenario: Primary region (US-EAST) completely lost

# Step 1: Verify XDCR target cluster (EU-WEST) has latest data
# Check XDCR replication status before failure
curl -s http://eu-west-node:8091/pools/default/remoteClusters \
  -u Administrator:password | jq .

# Step 2: Pause XDCR replications pointing to the failed cluster
/opt/couchbase/bin/couchbase-cli xdcr-replicate \
  --cluster cb-eu-west.example.com:8091 \
  --username Administrator \
  --password password \
  --pause \
  --xdcr-replicator <replication-id>

# Step 3: Update application connection strings to EU-WEST cluster
# (via DNS update, service mesh, or environment variable change)

# Step 4: Scale up EU-WEST cluster if needed to handle full production load
# In Kubernetes, update the CouchbaseCluster CRD:
kubectl patch couchbasecluster cb-eu-west -n couchbase --type merge \
  -p '{"spec":{"servers":[{"name":"data-zone-a","size":4}]}}'

# Step 5: After US-EAST cluster is restored, re-establish XDCR
# and perform a full resync from EU-WEST back to US-EAST

平稳的故障转移和恢复

# Graceful failover (for maintenance, drains data before removal)
/opt/couchbase/bin/couchbase-cli failover \
  --cluster localhost:8091 \
  --username Administrator \
  --password password \
  --server-failover node-to-remove.example.com:8091

# Recovery (re-add the node after maintenance)
/opt/couchbase/bin/couchbase-cli recovery \
  --cluster localhost:8091 \
  --username Administrator \
  --password password \
  --server-recovery node-to-recover.example.com:8091 \
  --recovery-type delta

# Delta recovery re-synchronizes only the changed data,
# which is much faster than full recovery.
# Full recovery rebuilds the node from scratch.

# Rebalance to complete the recovery
/opt/couchbase/bin/couchbase-cli rebalance \
  --cluster localhost:8091 \
  --username Administrator \
  --password password

Helm 完整生产部署的价值

下面是一个全面的 Helm 值文件，用于在生产环境中使用 Autonomous Operator 部署 Couchbase：

# helm-values-production.yaml
couchbase-operator:
  operator:
    image:
      repository: couchbase/operator
      tag: 2.7.1
    resources:
      requests:
        cpu: 500m
        memory: 512Mi
      limits:
        cpu: "1"
        memory: 1Gi
  admissionController:
    enabled: true
    resources:
      requests:
        cpu: 100m
        memory: 128Mi

cluster:
  image: couchbase/server:7.6.1-enterprise
  antiAffinity: true
  autoFailoverTimeout: 30s
  autoFailoverMaxCount: 3
  autoFailoverOnDataDiskIssues: true
  autoFailoverServerGroup: true
  security:
    adminSecret: cb-admin-credentials
  networking:
    tls:
      static:
        serverSecret: couchbase-server-tls
        operatorSecret: couchbase-operator-tls
    exposeAdminConsole: true
    adminConsoleServiceType: NodePort
  buckets:
    managed: true
  servers:
    data:
      size: 4
      services:
      - data
      - index
      serverGroups:
      - zone-a
      - zone-b
      resources:
        requests:
          cpu: "4"
          memory: 16Gi
        limits:
          cpu: "8"
          memory: 20Gi
      volumeMounts:
        default: couchbase-data
        data: couchbase-data
        index: couchbase-index
    query:
      size: 2
      services:
      - query
      - search
      resources:
        requests:
          cpu: "4"
          memory: 8Gi
        limits:
          cpu: "8"
          memory: 12Gi
      volumeMounts:
        default: couchbase-default
    analytics:
      size: 2
      services:
      - analytics
      - eventing
      serverGroups:
      - zone-c
      resources:
        requests:
          cpu: "8"
          memory: 32Gi
        limits:
          cpu: "16"
          memory: 40Gi
      volumeMounts:
        default: couchbase-analytics
        analytics:
        - couchbase-analytics
  serverGroups:
  - zone-a
  - zone-b
  - zone-c
  volumeClaimTemplates:
  - metadata:
      name: couchbase-data
    spec:
      storageClassName: ebs-gp3-couchbase
      resources:
        requests:
          storage: 100Gi
  - metadata:
      name: couchbase-index
    spec:
      storageClassName: ebs-gp3-couchbase
      resources:
        requests:
          storage: 50Gi
  - metadata:
      name: couchbase-default
    spec:
      storageClassName: ebs-gp3-couchbase
      resources:
        requests:
          storage: 20Gi
  - metadata:
      name: couchbase-analytics
    spec:
      storageClassName: ebs-gp3-couchbase
      resources:
        requests:
          storage: 200Gi

结论

Couchbase 服务器的架构围绕基于 vBucket 的分片、内存优先数据访问和集成多模型服务构建，为高可用性生产部署提供了独特的强大基础。集群内复制与自动故障转移的结合可确保透明地处理单节点故障，而 XDCR 将这种弹性扩展到全球应用程序的跨地理区域。

Kubernetes 的 Couchbase 自主操作员将复杂的手动操作转变为声明式、自我修复部署。服务器组提供机架/区域感知，操作员在扩展事件期间管理重新平衡操作，集成备份 CRD 自动执行灾难恢复准备。

本指南的要点：

• 利用多维扩展— 将数据、索引、查询、搜索、分析和事件服务分离到专用节点池上，以实现独立扩展和资源隔离。
• 配置 XDCR 以实现多区域弹性— 具有基于时间戳的冲突解决功能的双向 XDCR 支持跨 AWS、Azure 和 GCP 的主动-主动部署。始终确保 NTP 同步。
• 使用服务器组进行区域感知— 将服务器组映射到可用区域或机架，以保证活动 vBucket 和副本 vBucket 位于不同的故障域中。
• 仔细调整内存大小— Couchbase 的性能与 RAM 中容纳的工作集大小直接相关。使用大小调整公式并监控缓存未命中率。
• 实施全面监控— 从第一天起就部署 Prometheus 导出器。 XDCR 复制延迟、缓存未命中率、磁盘队列深度和节点运行状况是您的关键信号。
• 使用 cbbackupmgr 自动备份— 将完整备份和增量备份与云快照相结合。定期测试恢复过程。
• 通过 TLS 和 RBAC 实现安全— 启用节点到节点和客户端到节点 TLS 加密。为每个应用程序服务帐户使用细粒度的 RBAC 角色。
• 为 HA配置 SDK — 使用多个引导节点、配置适当的超时、实施副本读取作为回退，并利用关键数据的持久写入。
• 灾难恢复规划— 记录并演练单节点、多节点和完整集群故障场景的故障转移过程。 XDCR 备用集群应随时准备好升级。

有了这个全面的基础，您就可以在任何基础设施的高可用性生产环境中部署和操作 Couchbase 服务器——从 AWS、Azure 和 GCP 上的托管 Kubernetes 到 Rancher 管理的裸机 k3s 集群。 Couchbase 的本机分布式架构与 Kubernetes 编排相结合，提供了一个满足现代全球分布式应用程序需求的数据库平台。