版本：Nightly

管理 etcd

GreptimeDB 集群默认需要一个 etcd 集群用于元数据存储。让我们使用 Bitnami 的 etcd Helm chart 安装一个 etcd 集群。

先决条件

Kubernetes >= v1.23
kubectl >= v1.18.0
Helm >= v3.0.0

安装

将以下配置保存为文件 etcd.yaml:

global:
  security:
    allowInsecureImages: true

image:
  registry: greptime-registry.cn-hangzhou.cr.aliyuncs.com
  repository: bitnami/etcd
  tag: 3.6.1-debian-12-r3
  
replicaCount: 3

auth:
  rbac:
    create: false
  token:
    enabled: false

persistence:
  storageClass: null
  size: 8Gi

resources:
  limits:
    cpu: '2'
    memory: 8Gi
  requests:
    cpu: '2'
    memory: 8Gi

autoCompactionMode: "revision"
autoCompactionRetention: "100"

extraEnvVars:
  - name: ETCD_QUOTA_BACKEND_BYTES
    value: "8589934592"
  - name: ETCD_ELECTION_TIMEOUT
    value: "2000"
  - name: ETCD_SNAPSHOT_COUNT
    value: "10000"

此配置使用基于 revision 的自动压缩。etcd 每 5 分钟检查一次当前 revision，并压缩至 当前 revision - 100。保留的历史时间窗口取决于写入速率；读取或监听旧 revision 的客户端需要处理历史版本已被压缩的错误。压缩释放的空间可供 etcd 内部复用；要回收磁盘空间，还需要进行碎片整理（defragmentation）。

安装 etcd 集群:

helm upgrade \
  --install etcd oci://greptime-registry.cn-hangzhou.cr.aliyuncs.com/charts/etcd \
  --create-namespace \
  --version 12.0.8 \
  -n etcd-cluster \
  --values etcd.yaml

等待 etcd 集群运行：

kubectl get pod -n etcd-cluster

Expected Output

NAME     READY   STATUS    RESTARTS   AGE
etcd-0   1/1     Running   0          64s
etcd-1   1/1     Running   0          65s
etcd-2   1/1     Running   0          72s

等待 etcd 集群运行完毕，使用以下命令检查 etcd 集群的健康状态：

kubectl -n etcd-cluster \
  exec etcd-0 -- etcdctl \
  --endpoints etcd-0.etcd-headless.etcd-cluster:2379,etcd-1.etcd-headless.etcd-cluster:2379,etcd-2.etcd-headless.etcd-cluster:2379 \
  endpoint status -w table

Expected Output

+----------------------------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+
|                ENDPOINT                |        ID        | VERSION | DB SIZE | IS LEADER | IS LEARNER | RAFT TERM | RAFT INDEX | RAFT APPLIED INDEX | ERRORS |
+----------------------------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+
| etcd-0.etcd-headless.etcd-cluster:2379 | 680910587385ae31 |  3.5.15 |   20 kB |     false |      false |         4 |      73991 |              73991 |        |
| etcd-1.etcd-headless.etcd-cluster:2379 | d6980d56f5e3d817 |  3.5.15 |   20 kB |     false |      false |         4 |      73991 |              73991 |        |
| etcd-2.etcd-headless.etcd-cluster:2379 | 12664fc67659db0a |  3.5.15 |   20 kB |      true |      false |         4 |      73991 |              73991 |        |
+----------------------------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+

备份

在 bitnami etcd chart 中，使用共享存储卷 Network File System (NFS) 存储 etcd 备份数据。通过 Kubernetes 中的 CronJob 进行 etcd 快照备份，并挂载 NFS PersistentVolumeClaim (PVC)，可以将快照传输到 NFS 中。

添加以下配置，并将其命名为 etcd-backup.yaml 文件，注意需要将 existingClaim 修改为你的 NFS PVC 名称：

global:
  security:
    allowInsecureImages: true

image:
  registry: greptime-registry.cn-hangzhou.cr.aliyuncs.com
  repository: bitnami/etcd
  tag: 3.6.1-debian-12-r3
  
replicaCount: 3

auth:
  rbac:
    create: false
  token:
    enabled: false

persistence:
  storageClass: null
  size: 8Gi

resources:
  limits:
    cpu: '2'
    memory: 8Gi
  requests:
    cpu: '2'
    memory: 8Gi

autoCompactionMode: "revision"
autoCompactionRetention: "100"

extraEnvVars:
  - name: ETCD_QUOTA_BACKEND_BYTES
    value: "8589934592"
  - name: ETCD_ELECTION_TIMEOUT
    value: "2000"
  - name: ETCD_SNAPSHOT_COUNT
    value: "10000"

# Backup settings
disasterRecovery:
  enabled: true
  cronjob:
    schedule: "*/30 * * * *"
    historyLimit: 2
    snapshotHistoryLimit: 2
  pvc:
    existingClaim: "${YOUR_NFS_PVC_NAME_HERE}"

重新部署 etcd 集群：

helm upgrade \
  --install etcd oci://greptime-registry.cn-hangzhou.cr.aliyuncs.com/charts/etcd \
  --create-namespace \
  --version 12.0.8 \
  -n etcd-cluster \
  --values etcd-backup.yaml

你可以看到 etcd 备份计划任务：

kubectl get cronjob -n etcd-cluster

Expected Output

NAME               SCHEDULE      TIMEZONE   SUSPEND   ACTIVE   LAST SCHEDULE   AGE
etcd-snapshotter   */30 * * * *   <none>    False     0        <none>          36s

kubectl get pod -n etcd-cluster

Expected Output

NAME                              READY   STATUS      RESTARTS   AGE
etcd-0                            1/1     Running     0          35m
etcd-1                            1/1     Running     0          36m
etcd-2                            0/1     Running     0          6m28s
etcd-snapshotter-28936038-tsck8   0/1     Completed   0          4m49s

kubectl logs etcd-snapshotter-28936038-tsck8 -n etcd-cluster

Expected Output

etcd-0.etcd-headless.etcd-cluster.svc.cluster.local:2379 is healthy: successfully committed proposal: took = 2.698457ms
etcd 11:18:07.47 INFO  ==> Snapshotting the keyspace
{"level":"info","ts":"2025-01-06T11:18:07.579095Z","caller":"snapshot/v3_snapshot.go:65","msg":"created temporary db file","path":"/snapshots/db-2025-01-06_11-18.part"}
{"level":"info","ts":"2025-01-06T11:18:07.580335Z","logger":"client","caller":"v3@v3.5.15/maintenance.go:212","msg":"opened snapshot stream; downloading"}
{"level":"info","ts":"2025-01-06T11:18:07.580359Z","caller":"snapshot/v3_snapshot.go:73","msg":"fetching snapshot","endpoint":"etcd-0.etcd-headless.etcd-cluster.svc.cluster.local:2379"}
{"level":"info","ts":"2025-01-06T11:18:07.582124Z","logger":"client","caller":"v3@v3.5.15/maintenance.go:220","msg":"completed snapshot read; closing"}
{"level":"info","ts":"2025-01-06T11:18:07.582688Z","caller":"snapshot/v3_snapshot.go:88","msg":"fetched snapshot","endpoint":"etcd-0.etcd-headless.etcd-cluster.svc.cluster.local:2379","size":"20 kB","took":"now"}
{"level":"info","ts":"2025-01-06T11:18:07.583008Z","caller":"snapshot/v3_snapshot.go:97","msg":"saved","path":"/snapshots/db-2025-01-06_11-18"}
Snapshot saved at /snapshots/db-2025-01-06_11-18

接下来，可以在 NFS 服务器中看到 etcd 备份快照：

ls ${NFS_SERVER_DIRECTORY}

Expected Output

db-2025-01-06_11-18  db-2025-01-06_11-20  db-2025-01-06_11-22

恢复

当您遇到 etcd 数据丢失或损坏（例如，存储在 etcd 中的关键信息被意外删除，或者发生无法恢复的灾难性集群故障）时，您需要执行 etcd 恢复。此外，恢复 etcd 还可用于开发和测试目的。

恢复前需要停止向 etcd 集群写入数据（停止 GreptimeDB Metasrv 对 etcd 的写入），并创建最新的快照文件用于恢复。

添加以下配置文件，命名为 etcd-restore.yaml。注意，existingClaim 是你的 NFS PVC 的名字，snapshotFilename 为 etcd 快照文件名：

global:
  security:
    allowInsecureImages: true

image:
  registry: greptime-registry.cn-hangzhou.cr.aliyuncs.com
  repository: bitnami/etcd
  tag: 3.6.1-debian-12-r3
  
replicaCount: 3

auth:
  rbac:
    create: false
  token:
    enabled: false

persistence:
  storageClass: null
  size: 8Gi

resources:
  limits:
    cpu: '2'
    memory: 8Gi
  requests:
    cpu: '2'
    memory: 8Gi

autoCompactionMode: "revision"
autoCompactionRetention: "100"

extraEnvVars:
  - name: ETCD_QUOTA_BACKEND_BYTES
    value: "8589934592"
  - name: ETCD_ELECTION_TIMEOUT
    value: "2000"
  - name: ETCD_SNAPSHOT_COUNT
    value: "10000"

# Restore settings
startFromSnapshot:
  enabled: true
  existingClaim: "${YOUR_NFS_PVC_NAME_HERE}"
  snapshotFilename: "${YOUR_ETCD_SNAPSHOT_FILE_NAME}"

部署 etcd 恢复集群：

helm upgrade \
  --install etcd-recover oci://greptime-registry.cn-hangzhou.cr.aliyuncs.com/charts/etcd \
  --create-namespace \
  --version 12.0.8 \
  -n etcd-cluster \
  --values etcd-restore.yaml

等待 etcd 恢复集群运行：

kubectl get pod -n etcd-cluster -l app.kubernetes.io/instance=etcd-recover

Expected Output

NAME             READY   STATUS    RESTARTS   AGE
etcd-recover-0   1/1     Running   0          91s
etcd-recover-1   1/1     Running   0          91s
etcd-recover-2   1/1     Running   0          91s

NOTE

chart 版本之间的配置结构已发生变化:

旧版本: meta.etcdEndpoints
新版本: meta.backendStorage.etcd.endpoints

请参考 chart 仓库中配置 values.yaml 以获取最新的结构。

接着，将 Metasrv 的 backendStorage.etcd.endpoints 改成新的 etcd recover 集群，本例中为 "etcd-recover.etcd-cluster.svc.cluster.local:2379"：

apiVersion: greptime.io/v1alpha1
kind: GreptimeDBCluster
metadata:
  name: greptimedb
spec:
  # 其他配置
  meta:
    backendStorage:
      etcd:
        endpoints:
          - "etcd-recover.etcd-cluster.svc.cluster.local:2379"

然后重启 GreptimeDB Metasrv 完成 etcd 恢复。

Monitoring

安装 Prometheus Operator (例如: kube-prometheus-stack)。
安装 podmonitor CRD。

将以下配置保存为文件 etcd-monitoring.yaml:

global:
  security:
    allowInsecureImages: true

image:
  registry: greptime-registry.cn-hangzhou.cr.aliyuncs.com
  repository: bitnami/etcd
  tag: 3.6.1-debian-12-r3
  
replicaCount: 3

auth:
  rbac:
    create: false
  token:
    enabled: false

persistence:
  storageClass: null
  size: 8Gi

resources:
  limits:
    cpu: '2'
    memory: 8Gi
  requests:
    cpu: '2'
    memory: 8Gi

autoCompactionMode: "revision"
autoCompactionRetention: "100"

extraEnvVars:
  - name: ETCD_QUOTA_BACKEND_BYTES
    value: "8589934592"
  - name: ETCD_ELECTION_TIMEOUT
    value: "2000"
  - name: ETCD_SNAPSHOT_COUNT
    value: "10000"

# Monitoring settings
metrics:
  enabled: true
  podMonitor:
    enabled: true
    namespace: etcd-cluster
    interval: 10s
    scrapeTimeout: 10s
    additionalLabels:
      release: prometheus

部署并监控 etcd:

helm upgrade \
  --install etcd oci://greptime-registry.cn-hangzhou.cr.aliyuncs.com/charts/etcd \
  --create-namespace \
  --version 12.0.8 \
  -n etcd-cluster \
  --values etcd-monitoring.yaml

Grafana dashboard

使用 ETCD Cluster Overview dashboard (ID: 15308) 来监控 etcd 的指标。

登录你的 Grafana。
导航至 Dashboards -> New -> Import。
输入 Dashboard ID: 15308, 选择数据源并加载图表。

ETCD Cluster Overview dashboard

⚠️ Defrag - Critical Warning

Defrag 是一项高风险操作，可能会严重影响您的 ETCD 集群及其依赖系统（例如 GreptimeDB）：

执行期间会阻止所有读/写操作（集群将不可用）。
高 I/O 使用率可能导致客户端应用程序超时。
如果碎片整理耗时过长，可能会触发领导者选举。
如果资源分配不当，可能会导致 OOM 终止。
如果在过程中中断，可能会损坏数据。

ETCD 使用多版本并发控制 (MVCC) 机制，用于存储多个版本的 KV。随着时间的推移，随着数据的更新和删除，后端数据库可能会变得碎片化，从而增加存储空间并降低性能。Defrag 是指压缩这些存储空间以回收空间并提高性能的过程。

在 etcd-defrag.yaml 文件中添加以下与 defrag 相关的配置：

global:
  security:
    allowInsecureImages: true

image:
  registry: greptime-registry.cn-hangzhou.cr.aliyuncs.com
  repository: bitnami/etcd
  tag: 3.6.1-debian-12-r3
  
replicaCount: 3

auth:
  rbac:
    create: false
  token:
    enabled: false

persistence:
  storageClass: null
  size: 8Gi

resources:
  limits:
    cpu: '2'
    memory: 8Gi
  requests:
    cpu: '2'
    memory: 8Gi

autoCompactionMode: "revision"
autoCompactionRetention: "100"

extraEnvVars:
  - name: ETCD_QUOTA_BACKEND_BYTES
    value: "8589934592"
  - name: ETCD_ELECTION_TIMEOUT
    value: "2000"
  - name: ETCD_SNAPSHOT_COUNT
    value: "10000"

# Defragmentation settings
defrag:
  enabled: true
  cronjob:
    schedule: "0 3 * * *"  # Daily at 3:00 AM
    suspend: false
    successfulJobsHistoryLimit: 1
    failedJobsHistoryLimit: 1

部署 etcd 集群并开启 defrag 功能:

helm upgrade \
  --install etcd oci://greptime-registry.cn-hangzhou.cr.aliyuncs.com/charts/etcd \
  --create-namespace \
  --version 12.0.8 \
  -n etcd-cluster \
  --values etcd-defrag.yaml

你可以看到 etcd defrag 定时任务:

kubectl get cronjob -n etcd-cluster

Expected Output

NAME          SCHEDULE      TIMEZONE   SUSPEND   ACTIVE   LAST SCHEDULE   AGE
etcd-defrag   0 3 * * *     <none>     False     0        <none>          34s

kubectl get pod -n etcd-cluster

Expected Output

NAME                         READY   STATUS      RESTARTS   AGE
etcd-0                       1/1     Running     0          4m30s
etcd-1                       1/1     Running     0          4m29s
etcd-2                       1/1     Running     0          4m29s
etcd-defrag-29128518-sstbf   0/1     Completed   0          90s

kubectl logs etcd-defrag-29128518-sstbf -n etcd-cluster

Expected Output

Finished defragmenting etcd member[http://etcd-0.etcd-headless.etcd-cluster.svc.cluster.local:2379]
Finished defragmenting etcd member[http://etcd-1.etcd-headless.etcd-cluster.svc.cluster.local:2379]
Finished defragmenting etcd member[http://etcd-2.etcd-headless.etcd-cluster.svc.cluster.local:2379]

管理 etcd

先决条件​

安装​

备份​

恢复​

Monitoring​

Grafana dashboard​

⚠️ Defrag - Critical Warning​

先决条件

安装

备份

恢复

Monitoring

Grafana dashboard

⚠️ Defrag - Critical Warning