图数据库选型与集群运维：从 Causal Cluster 到分布式对比

# docker-compose：3 个 CORE + 2 个 READ_REPLICA
services:
  core1:
    image: neo4j:5.15-enterprise
    environment:
      NEO4J_AUTH: neo4j/password
      NEO4J_initial_mode: CORE
      NEO4J_initial_dbms_default__database: neo4j
      NEO4J_dbms_discovery_advertised__address: core1:5000
      NEO4J_dbms_cluster_advertised__address: core1:6000
      NEO4J_dbms_cluster_advertised__membership__address: core1:6000
      NEO4J_server_routing_advertised__address: core1:7687
    ports: ["7474:7474", "7687:7687"]
  core2:
    image: neo4j:5.15-enterprise
    environment:
      NEO4J_AUTH: neo4j/password
      NEO4J_initial_mode: CORE
      # ... 同上，地址改为 core2
  read1:
    image: neo4j:5.15-enterprise
    environment:
      NEO4J_AUTH: neo4j/password
      NEO4J_initial_mode: READ_REPLICA
      # 加入 core1 引导的集群
      NEO4J_initial_discovery_members: core1:5000

# 查看集群状态
neo4j-admin server status
# 或通过 API
curl -s -u neo4j:password http://localhost:7474/api/cluster/status | python3 -m json.tool

# 驱动使用路由地址（bolt+routing），自动感知集群拓扑
from neo4j import GraphDatabase

driver = GraphDatabase.driver(
    "bolt+routing://core1:7687",
    auth=("neo4j", "password")
)

with driver.session() as session:
    # 读请求 → READ_REPLICA 或 CORE（自动选择）
    res = session.execute_read(lambda tx: tx.run("MATCH (p:Person) RETURN count(p)").single()[0])
    # 写请求 → 强制路由到 CORE 领导者
    session.execute_write(lambda tx: tx.run("CREATE (:Person {name:'Alice'})"))

from neo4j import GraphDatabase, READ_ACCESS, WRITE_ACCESS
from neo4j.workload import Workload

# 默认：近期写入优先（匹配 causal）
# 可用配置覆盖：
driver.execute_query(
    "MATCH (p:Person) RETURN count(p)",
    database_="neo4j",
    routing_="r",            # r=read, w=write
)

# 在会话间传递 bookmark 保持因果
with driver.session() as s1:
    s1.run("CREATE (:Person {name:'Alice'})")
    bookmark = s1.last_bookmark()

with driver.session(bookmarks=[bookmark]) as s2:
    # s2 的读必然能看到 Alice
    cnt = s2.run("MATCH (p:Person {name:'Alice'}) RETURN count(p)").single()[0]
    assert cnt == 1

# NebulaGraph 集群规模建议
metad: 3 节点（Raft）
storaged: 3-30 节点（按数据量，分区可再平衡）
graphd: 2-20 节点（无状态，LB 分发）

# 停止数据库后导出（离线一致性最可靠）
neo4j-admin database dump --to-file=/backup/neo4j.dump

# 恢复
neo4j-admin database load --from-file=/backup/neo4j.dump --database=neo4j

# 在线备份（无需停机，保证因果一致）
neo4j-admin database backup \
  --from=bolt://core1:7687 \
  --to=/backup/ \
  --database=neo4j

# 备份结果校验
neo4j-admin database check --database=neo4j

# 1. 定期全量备份（每日）
neo4j-admin database backup --from=bolt://core1:7687 --to=/backup/full/

# 2. 连续归档事务日志（开启）
# neo4j.conf
server.logs.rotation.size=100m
server.db.tx_log.preallocate=true

# 3. 恢复到指定时间点
neo4j-admin database restore \
  --from=/backup/full/ \
  --database=neo4j \
  --restore-to=neo4j-restored

# prometheus.yml 抓取 Neo4j
scrape_configs:
  - job_name: 'neo4j'
    static_configs:
      - targets: ['core1:9100', 'core2:9100', 'core3:9100']

# 常用 PromQL 示例
# 页缓存命中率
neo4j_page_cache_hits / (neo4j_page_cache_hits + neo4j_page_cache_faults)
# 事务平均耗时
rate(neo4j_transaction_last_committed_time_seconds_total[5m])
# 复制滞后
neo4j_cluster_read_replica_lag

# alertmanager 规则片段
groups:
  - name: neo4j.rules
    rules:
      - alert: Neo4jCoreDown
        expr: up{job="neo4j"} == 0
        for: 2m
        severity: critical
      - alert: Neo4jPageCacheMissHigh
        expr: neo4j_page_cache_hit_rate < 0.9
        for: 10m
        severity: warning
      - alert: Neo4jTxQueueTooLong
        expr: neo4j_transaction_queue_size > 100
        for: 5m
        severity: warning

# 查询日志：慢查询定位
dbms.logs.query.enabled=true
dbms.logs.query.threshold=200ms

# 审计日志（企业版）：敏感操作追踪
dbms.security.audit.enabled=true