1. Dashboard Rule#

Folder: Database
Dashboard: Valkey - Overview

Rows:
    Overview:
        engine CPU、host CPU、memory、connections、eviction、hit rate、replication lag

    Drilldown:
        network in/out、replication lag detail

    Capacity:
        memory usage trend、host freeable memory、connection trend、network trend、eviction trend

    Debug:
        swap usage 这类宿主机压力信号，不放第一屏；排查 host 压力时查看

2. Overview#

Panel Metrics Why
Engine CPU EngineCPUUtilization Redis / Valkey 主线程压力
Host CPU CPUUtilization host 层资源压力
Memory usage DatabaseMemoryUsagePercentage 判断缓存空间是否接近打满
Host freeable memory FreeableMemory 辅助看宿主机内存缓冲空间，不等于可用缓存内存
Evictions per period Evictions 当前 CloudWatch / YACE 周期里发生了多少 eviction
Hit rate CacheHits, CacheMisses cache 是否真正有效
Replication lag ReplicationLag replica / failover 风险
Connections CurrConnections, NewConnections client 连接压力

3. Drilldown#

Panel Metrics Why
Network in / out NetworkBytesIn, NetworkBytesOut 看流量方向和波动
Replication lag detail ReplicationLag 作为 overview 之外的第二视角看复制延迟

4. Capacity#

Panel Metrics Why
Memory usage trend DatabaseMemoryUsagePercentage 评估是否需要扩容或调整 TTL
Host freeable memory FreeableMemory 观察 host 是否还有可回收内存，但不作为缓存剩余空间的直接判断
Eviction trend Evictions 1h rolling sum 看 eviction 是偶发还是持续存在
Connection trend CurrConnections, NewConnections 评估 client pool 和 maxclients
Network trend NetworkBytesIn, NetworkBytesOut 评估 node type 和流量增长
Hit rate trend hit rate from hits / misses 判断 cache 投入是否有效

5. Debug#

Panel Metrics Why
Swap SwapUsage swap 会显著影响 latency

6. YACE Metrics#

如果 Valkey dashboard 使用 aws_elasticache_* metrics，需要 YACE 采集 AWS/ElastiCache。Valkey / Redis OSS 内部 keyspace、command 细节仍然建议用 redis_exporter 补充。

YACE config:

apiVersion: v1alpha1
sts-region: ap-northeast-1
discovery:
  jobs:
    - type: AWS/ElastiCache
      regions:
        - ap-northeast-1
      customTags:
        - key: environment
          value: uat
      period: 300
      length: 600
      nilToZero: true
      metrics:
        - name: EngineCPUUtilization
          statistics: [Average]
        - name: CPUUtilization
          statistics: [Average]
        - name: DatabaseMemoryUsagePercentage
          statistics: [Average]
        - name: FreeableMemory
          statistics: [Average]
        - name: Evictions
          statistics: [Sum]
        - name: CacheHits
          statistics: [Sum]
        - name: CacheMisses
          statistics: [Sum]
        - name: ReplicationLag
          statistics: [Maximum]
        - name: CurrConnections
          statistics: [Average]
        - name: NewConnections
          statistics: [Sum]
        - name: NetworkBytesIn
          statistics: [Sum]
        - name: NetworkBytesOut
          statistics: [Sum]
        - name: SwapUsage
          statistics: [Average]

Exporter naming note:

当前 Valkey dashboard 按实际 YACE metric name 对齐，例如:
    aws_elasticache_engine_cpuutilization_average
    aws_elasticache_cpuutilization_average

FreeableMemory note:

FreeableMemory 是 host-level freeable memory，单位 bytes。
它不等于 Valkey 可用缓存内存。

判断缓存是否接近打满，优先看:
    DatabaseMemoryUsagePercentage
    Evictions
    SwapUsage

SwapUsage note:

SwapUsage 是 host-level 指标，不是 engine-level 指标。
它有值，不一定代表 CPU 或缓存使用率异常。

更需要关注的是:
    SwapUsage 持续升高
    或同时伴随 latency / memory pressure 变差

Evictions note:

Evictions 是因为触发 maxmemory 限制，被 Valkey / Redis OSS 淘汰掉的 key 数量。

Evictions = 0:
    很多时候是正常的，通常说明当前还没有因为内存压力开始淘汰 key。

Evictions > 0:
    说明新写入已经开始逼近容量边界，engine 需要腾空间。

它不是“缓存有没有在工作”的指标。
判断容量是否真的紧张，应该结合:
    DatabaseMemoryUsagePercentage
    Evictions
    CacheMisses
    SwapUsage
一起看。

确认 metric name:

{__name__=~"aws_elasticache_.*"}

7. Do Not Put In Overview#

不要放:
    每个 command 的全量延迟
    keyspace 详细分布
    slowlog 文本内容

原因:
    ElastiCache overview 重点是 CPU、memory、eviction、hit rate、replication、connections
Backward ElastiCache YACE Dashboard Forward