vlambda博客
学习文章列表

Loki日志系统分布式部署实践之 Cassandra

 系列文章

1. 说明

Loki 支持文件系统、对象存储、NoSQL,因为对象存储大多都要使用公有云,所以暂时使用 Cassandra 作为存储,目前的实现里它支持 index 和 chunk

2. 基本知识

Cassandra 是一个开源的、分布式、无中心节点、弹性可扩展、高可用、容错、一致性协调、面向列的 NoSQL 数据库

2.1 架构层级

Cluster - Data center(s) - Rack(s) - Server(s) - Node (more accurately, a vnode)
  • Node(节点):一个运行 cassandra 的实例
  • Rack(机架):一组 nodes 的集合
  • DataCenter(数据中心):一组 racks 的集合
  • Cluster(集群):映射到拥有一个完整令牌圆环所有节点的集合

2.2 Cassandra 一致性

Apache Cassandra[1]依赖于 Amazon Dynamo 分布式存储键值系统的多种技术。Dynamo 系统中的每个节点都有三个主要组件:

  1. 请求对分区数据集进行协调,客户端连接到某一节点发起读写请求时,该节点充当客户端应用与集群中拥有相应数据节点间的桥梁,称为协调者,以根据集群配置确定环(ring)中的哪个节点应当获取这个请求
  2. 环成员(Ring membership)和故障检测
  3. 本地持久性(存储)引擎

复制策略

注意:所有生产部署都应使用 NetworkTopologyStrategy,而 SimpleStrategy 复制策略仅对尚不知道集群数据中心布局的集群进行测试有用

  • NetworkTopologyStrategy
  • SimpleStrategy

一致性级别

  • ONE 只有单个副本必须响应
  • TWO 两个副本必须响应
  • THREE 三个副本必须响应
  • QUORUM 多数副本( n / 2 + 1)必须响应
  • ALL 所有副本都必须响应
  • LOCAL_QUORUM 本地数据中心(协调器所在的任何数据中心)中的大多数副本都必须响应
  • EACH_QUORUM 每个数据中心中的大多数副本必须响应
  • LOCAL_ONE 只有单个副本必须响应。在多数据中心群集中,这也保证读取请求不会发送到远程数据中心中的副本
  • ANY 单个副本可以响应,或者协调器可以存储提示。如果存储了提示,则协调器稍后将尝试重播提示并将突变传递给副本。仅写操作接受此一致性级别。

复制因子和一致性级别

Cassandra 提供了可调节的一致性,允许我们选定需要的一致性水平与可用性水平,在二者间找到平衡点。因为客户端可以控制在更新到达多少个副本之前,必须阻塞系统。这是通过设置副本因子(replication factor)来调节与之相对的一致性级别。通过副本因子(replication factor),你可以决定准备牺牲多少性能来换取一致性。副本因子是你要求更新在集群中传播到的节点数(注意,更新包括所有增加、删除和更新操作)。

客户端每次操作还必须设置一个一致性级别(consistency level)参数,这个参数决定了多少个副本写入成功才可以认定写操作是成功的,或者读取过程中读到多少个副本正确就可以认定是读成功的。如果需要的话,你可以设定一致性级别和副本因子相等,从而达到一个较高的一致性水平,不过这样就必须付出同步阻塞操作的代价,只有所有节点都被更新完成才能成功返回一次更新。而实际上,Cassandra 一般都不会这么来用。而如果一个客户端设置一致性级别低于副本因子的话,即使有节点宕机了,仍然可以写成功。

3.安装

3.1 下载包

# wget -O cassandra-operator-v7.1.0.tar.gz https://github.com/instaclustr/cassandra-operator/archive/v7.1.0.tar.gz
# tar -zxf cassandra-operator-v7.1.0.tar.gz
# cd cassandra-operator-7.1.0/
  • 部署Cassandra CRD [2]
# kubectl apply -f deploy/crds.yaml
customresourcedefinition.apiextensions.k8s.io/cassandrabackups.cassandraoperator.instaclustr.com created
customresourcedefinition.apiextensions.k8s.io/cassandraclusters.cassandraoperator.instaclustr.com created
customresourcedefinition.apiextensions.k8s.io/cassandradatacenters.cassandraoperator.instaclustr.com created

3.2 部署 Cassandra Operator

# vi deploy/bundle.yaml
# 对里面的资源都加上 namespace: grafana,因为我这里需要和 loki 配合使用。podsecuritypolicy 是集群级别的,可以不用加
      containers:
        - name: cassandra-operator
          #image: "gcr.io/cassandra-operator/cassandra-operator:latest"
          image: "ops-harbor.hupu.io/k8s/cassandra-operator:v7.1.0"

# kubectl apply -f deploy/bundle.yaml
serviceaccount/cassandra created
role.rbac.authorization.k8s.io/cassandra created
rolebinding.rbac.authorization.k8s.io/cassandra created
podsecuritypolicy.policy/cassandra created
serviceaccount/cassandra-performance created
role.rbac.authorization.k8s.io/cassandra-performance created
rolebinding.rbac.authorization.k8s.io/cassandra-performance created
podsecuritypolicy.policy/cassandra-performance created
configmap/cassandra-operator-default-config created
deployment.apps/cassandra-operator created
podsecuritypolicy.policy/cassandra-operator created
rolebinding.rbac.authorization.k8s.io/cassandra-operator created
role.rbac.authorization.k8s.io/cassandra-operator created
serviceaccount/cassandra-operator created

# kubectl  get pod -n grafana -l name=cassandra-operator 
NAME                                  READY   STATUS    RESTARTS   AGE
cassandra-operator-6f685694c5-l7m27   1/1     Running   0          40s

查看日志:

# kubectl logs -f -n grafana $(kubectl  get pod -n grafana -l name=cassandra-operator -o name)

3.3 部署 Cassandra Cluster

Cassandra Cluster参考1[3]

Cassandra Cluster参考2[4]

Cassandra operator 支持将自定义 ConfigMap 通过 ConfigMapVolumeSource 装载到 cassandra 容器中:

注意:所有 Cassandra 和 JVM 配置都存在于容器内的 /etc/Cassandra 下

$ ls -l /etc/cassandra/
total 48
-rw-r--r-- 1 cassandra cassandra   19 Nov  9 13:29 cassandra-env.sh
drwxr-xr-x 2 cassandra cassandra 4096 Nov 28 05:39 cassandra-env.sh.d
-rw-r--r-- 1 cassandra cassandra   19 Nov  9 13:29 cassandra-exporter.conf
-rw-r--r-- 1 cassandra cassandra   70 Nov 28 05:39 cassandra-rackdc.properties
drwxr-xr-x 2 cassandra cassandra 4096 Nov 28 05:39 cassandra.yaml.d
-rw-r--r-- 1 cassandra cassandra   82 Nov  9 13:29 jvm-jmx.options
-rw-r--r-- 1 cassandra cassandra  143 Nov  9 13:29 jvm-operator.options
-rw-r--r-- 1 cassandra cassandra  600 Nov  9 13:29 jvm.options
drwxr-xr-x 2 cassandra cassandra 4096 Nov 28 05:39 jvm.options.d
-rw-r--r-- 1 cassandra cassandra 1239 Nov  9 13:29 logback-tools.xml
-rw-r--r-- 1 cassandra cassandra  538 Nov  9 13:29 logback.xml
drwxr-xr-x 2 cassandra cassandra 4096 Nov  9 13:31 logback.xml.d

$ ls -l /etc/cassandra/cassandra-env.sh.d
total 4
-rw-r--r-- 1 cassandra cassandra 130 Nov 28 05:39 001-cassandra-exporter.sh

$ ls -l /etc/cassandra/cassandra.yaml.d
total 16
-rw-r--r-- 1 cassandra cassandra 187 Nov  9 13:29 001-directories.yaml
-rw-r--r-- 1 cassandra cassandra 404 Nov 28 05:39 001-operator-overrides.yaml
-rw-r--r-- 1 cassandra cassandra  59 Nov 28 05:39 004-broadcast_rpc_address.yaml
-rw-r--r-- 1 cassandra cassandra  29 Nov 28 05:39 cassandra-config.yaml

$ ls -l /etc/cassandra/jvm.options.d
total 4
-rw-r--r-- 1 cassandra cassandra 416 Nov 28 05:39 001-jvm-memory-gc.options

/$ ls -l /etc/cassandra/logback.xml.d
total 0

3.4 自定义 Cassandra 配置

# vi cassandra-config.yaml
# 空闲连接超时,默认为禁用
#native_transport_idle_timeout_in_ms: 60000
# 协调器应等待读取操作完成的时间
read_request_timeout_in_ms: 30000
# 协调器应等待 seq 或 index 扫描完成的时间
range_request_timeout_in_ms: 30000
# 协调器应等待多长时间才能完成写操作
write_request_timeout_in_ms: 30000
# 协调器应等待多长时间才能完成计数器写入
counter_write_request_timeout_in_ms: 30000
# 协调器应继续重试与同一行的其他提议相抵触的 CAS 操作的时间
cas_contention_timeout_in_ms: 30000
# 协调器应等待多长时间才能完成截断操作(这可能会更长,因为除非禁用了auto_snapshot,否则我们需要先刷新,以便可以在删除数据之前进行快照。)
truncate_request_timeout_in_ms: 60000
# 其他杂项操作的默认超时
request_timeout_in_ms: 30000
# 记录慢速查询的时间
slow_query_log_timeout_in_ms: 5000
# 默认情况下,此选项已被注释掉,启用节点之间的操作超时信息交换,以准确测量请求超时。如果禁用,副本将假定协调员将请求立即转发给他们,这意味着在过载情况下,我们将浪费大量额外的时间来处理已超时的请求。
# 警告:通常假定用户在其集群上设置了 NTP,并且时钟适度同步,因为这是最后写入获胜者的总体正确性的要求。
#cross_node_timeout: true
# 默认情况下,此选项已被注释掉。
#internode_application_send_queue_capacity_in_bytes: 4194304
# 用于稳定块缓存和缓冲池的最大内存。其中的 32MB 保留用于缓冲池,其余部分用作保存未压缩的稳定块的缓存。默认为堆的 1/4 或 512MB 中的较小者。该池是堆外分配的,因此除了分配给堆的内存之外。缓存还有堆内存开销,每个块大约 128 字节(如果使用默认的 64k 块大小,则为保留大小的 0.2%)。仅在需要时才分配内存。
file_cache_size_in_mb: 2048 

# kubectl create configmap cassandra-new --from-file=cassandra-config.yaml -n grafana
  • 方法一:
# vi cassandra-operator-7.1.0/examples/example-datacenter.yaml 
spec:
  userConfigMapVolumeSource:
    # the name of the ConfigMap
    name: cassandra-new
    # ConfigMap keys -> file paths (relative to /etc/cassandra)
    items:
      - key: cassandra-config.yaml
        path: cassandra.yaml.d/cassandra-config.yaml
  • 方法二:
# kubectl edit CassandraDataCenter -n grafana cassandra-dc1 
spec:
  userConfigMapVolumeSource:
    # the name of the ConfigMap
    name: cassandra-new
    # ConfigMap keys -> file paths (relative to /etc/cassandra)
    items:
      - key: cassandra-config.yaml
        path: cassandra.yaml.d/cassandra-config.yaml

注意:也可以自定义 JVM 选项,通过在 jvm.options.d/gc.options 下指定一个选项文件来提供      默认的 JVM 选项:

# kubectl get cm -n grafana cassandra-cassandra-dc1-dc1-operator-config -o yaml
data:
  jvm_options_d_001_jvm_memory_gc_options: |
    -Xms1073741824
    -Xmx1073741824
    -Xmn4194304
    -XX:+UseParNewGC
    -XX:+UseConcMarkSweepGC
    -XX:+CMSParallelRemarkEnabled
    -XX:SurvivorRatio=8
    -XX:MaxTenuringThreshold=1
    -XX:CMSInitiatingOccupancyFraction=75
    -XX:+UseCMSInitiatingOccupancyOnly
    -XX:CMSWaitDuration=10000
    -XX:+CMSParallelInitialMarkEnabled
    -XX:+CMSEdenChunksRecordAlways
    -XX:+CMSClassUnloadingEnabled
    -XX:+HeapDumpOnOutOfMemoryError
    -XX:+CrashOnOutOfMemoryError

注意:这会全局覆盖默认的,而不是额外添加

# vi gc.options
-XX:+UseG1GC
-XX:ParallelGCThreads=8
-XX:MaxGCPauseMillis=200
#-Xms8g
#-Xmx8g
#-Xmn4g
-XX:+UseContainerSupport
#-XX:InitialRAMPercentage=15.0
#-XX:MinRAMPercentage=15.0
#-XX:MaxRAMPercentage=75.0

# kubectl delete configmap cassandra-new -n grafana
# kubectl create configmap cassandra-new --from-file=gc.options --from-file=cassandra-config.yaml -n grafana

# vi cassandra-operator-7.1.0/examples/example-datacenter.yaml 
spec:
  userConfigMapVolumeSource:
    # the name of the ConfigMap
    name: cassandra-new
    # ConfigMap keys -> file paths (relative to /etc/cassandra)
    items:
      - key: cassandra-config.yaml
        path: cassandra.yaml.d/cassandra-config.yaml
      - key: gc.options
        path: jvm.options.d/gc.options
修改 CassandraDataCenter 示例:

# vi examples/example-datacenter.yaml 
apiVersion: cassandraoperator.instaclustr.com/v1alpha1
kind: CassandraDataCenter
metadata:
  name: cassandra-dc1
  namespace: grafana
  labels:
    app: cassandra
datacenter: dc1
cluster: cassandra-dc1
spec:
  initImage: ops-harbor.hupu.io/base/alpine:v3.10
  # 通过 cassandra-operator-metrics ServiceMonitor 来实现,镜像中内置了 cassandra-exporter
  prometheusSupport: true
  optimizeKernelParams: true
  serviceAccountName: cassandra-performance
  nodes: 3
  racks:
  - name: rack1
    # 实际效果为 nodeSelector
    labels:
      failure-domain.beta.kubernetes.io/zone: cn-hangzhou-g
    # 容忍
    tolerations:
    - key: "app"
      operator: "Equal"
      value: "cassandra"
      effect: "NoSchedule"
    # 亲和性
    affinity: 
      # Pod 反亲和
      podAntiAffinity:
        # Pod 硬反亲和
        requiredDuringSchedulingIgnoredDuringExecution:
        - labelSelector:
            matchExpressions:
            - key: app
              operator: In
              values:
              - cassandra
          topologyKey: "kubernetes.io/hostname"
        # Pod 软反亲和
        #preferredDuringSchedulingIgnoredDuringExecution:
        #- podAffinityTerm:
        #    labelSelector:
        #      matchExpressions:
        #      - key: app
        #        operator: In
        #        values:
        #        - cassandra
        #    topologyKey: kubernetes.io/hostname
        #  weight: 100
      # 节点亲和性
      #nodeAffinity:
      #  requiredDuringSchedulingIgnoredDuringExecution:
      #    nodeSelectorTerms:
      #    - matchExpressions:
      #      - key: system
      #        operator: NotIn
      #        values:
      #        - management
      #      - key: app
      #        operator: In
      #        values:
      #        - cassandra
      #  #preferredDuringSchedulingIgnoredDuringExecution:
      #  #- weight: 60
      #  #  preference:
      #  #    matchExpressions:
      #  #    - {key: zone, operator: In, values: ["shanghai2", "shanghai3", "shanghai4"]}
      #  #- weight: 40
      #  #  preference:
      #  #    matchFields:
      #  #    - {key: ssd, operator: Exists, values: ["sanxing", "dongzhi"]}

  #racks:
  #  - name: "west1-b"
  #    labels:
  #      failure-domain.beta.kubernetes.io/zone: europe-west1-b
  #  - name: "west1-c"
  #    labels:
  #      failure-domain.beta.kubernetes.io/zone: europe-west1-c
  #  - name: "west1-a"
  #    labels:
  #      failure-domain.beta.kubernetes.io/zone: europe-west1-a
  #cassandraImage: "gcr.io/cassandra-operator/cassandra-3.11.6:latest"
  cassandraImage: "ops-harbor.hupu.io/k8s/cassandra-3.11.9:latest"
  #sidecarImage: "gcr.io/cassandra-operator/instaclustr-icarus:latest"
  sidecarImage: "ops-harbor.hupu.io/k8s/instaclustr-icarus:latest"
  imagePullPolicy: Always
  imagePullSecrets:
    - name: regcred
  # 不生效,官方已经注释掉了该字段:https://github.com/liwang0513/cassandra-operator/commit/b4f8b596013e5cbeaf222957b5aaa9b52a91efd7
  podManagementPolicy: Parallel
  # 用于存放 loki 日志时,发现经常超过 2GB 导致 OOM,平均内存在 4GB 左右,CPU 1.5 核心左右
  readinessProbe:
    exec:
      command:
      - /usr/bin/cql-readiness-probe
    failureThreshold: 3
    initialDelaySeconds: 5
    periodSeconds: 10
    successThreshold: 1
    timeoutSeconds: 5
  resources:
    limits:
      memory: 32Gi
      cpu: "16"
    requests:
      memory: 4Gi
      cpu: "2"
  sidecarResources:
    limits:
      memory: 512Mi
    requests:
      memory: 512Mi
  dataVolumeClaimSpec:
    storageClassName: alicloud-disk-efficiency-cn-hangzhou-g
    accessModes:
      - ReadWriteOnce
    resources:
      requests:
        storage: 2048Gi 

  # 将自定义 ConfigMap 通过 ConfigMapVolumeSource 装载到 cassandra 容器中达到修改 Cassandra.yaml 参数目的
  userConfigMapVolumeSource:
    # the name of the ConfigMap
    name: cassandra-new
    type: array
    # ConfigMap keys -> file paths (relative to /etc/cassandra)
    items:
      - key: cassandra-config.yaml
        path: cassandra.yaml.d/cassandra-config.yaml
      - key: gc.options
        path: jvm.options.d/gc.options

#  userSecretVolumeSource:
#    secretName: test-cassandra-dc-ssl
#
#  sidecarSecretVolumeSource:
#    secretName: test-cassandra-dc-ssl-sidecar

  cassandraAuth:
    authenticator: PasswordAuthenticator
    authorizer: CassandraAuthorizer
    roleManager: CassandraRoleManager

#  operatorLabels:
#    prometheusService:
#      cassandratestdclabel: testdc
#    nodesService:
#      mynodesservicelabel: labelvalue1
#    statefulSet:
#      mystatefullabel: labelvalue2
#    podTemplate:
#      mypodlabel: label1
#      myanotherpod: label2
#
#  operatorAnnotations:
#    prometheusService:
#      p1 : pv1
#    nodesService:
#      n1: nv1
#      n2: nv2
#    statefulSet:
#      s1: sv1
#      s2: sv2
#    podTemplate:
#      pt1: ptv1
#      pt2: ptv2

  # Needed to run on AKS
  fsGroup: 999

# kubectl apply -f examples/example-datacenter.yaml    
cassandradatacenter.cassandraoperator.instaclustr.com/cassandra-dc1 created

# kubectl get pod -n grafana -l cassandra-operator.instaclustr.com/cluster=cassandra-dc1
NAME                                  READY   STATUS    RESTARTS   AGE
cassandra-cassandra-dc1-dc1-rack1-0   2/2     Running   0          16m
cassandra-cassandra-dc1-dc1-rack1-1   2/2     Running   0          14m
cassandra-cassandra-dc1-dc1-rack1-2   2/2     Running   0          12m

# kubectl get svc -n grafana |grep cassandra
cassandra-cassandra-dc1-dc1-nodes        ClusterIP   None           <none>        9042/TCP,7199/TCP   176m
cassandra-cassandra-dc1-dc1-prometheus   ClusterIP   None           <none>        9500/TCP            176m
cassandra-cassandra-dc1-dc1-seeds        ClusterIP   None           <none>        7000/TCP            176m
cassandra-operator-metrics               ClusterIP   172.21.5.56    <none>        8383/TCP,8686/TCP   4h5m

3.5 验证集群健康

# kubectl exec cassandra-cassandra-dc1-dc1-rack1-0 -c cassandra -n grafana -- nodetool status
Datacenter: dc1
===============
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
--  Address        Load       Tokens       Owns (effective)  Host ID                               Rack
UN  10.41.180.90   85.66 KiB  256          67.9%             0c59d821-e2ad-4040-98e5-363f7029f3ca  rack1
UN  10.41.182.254  81.05 KiB  256          66.3%             3fb8f3f7-2305-4829-99bc-bc62ce30af56  rack1
UN  10.41.190.194  94.99 KiB  256          65.7%             4f70b0e2-6edb-4b51-8472-fe497fa22b88  rack1

3.6 测试查看下数据

# kubectl exec cassandra-cassandra-dc1-dc1-rack1-0 -c cassandra -n grafana -- cqlsh -e "SELECT now() FROM system.local;" cassandra-cassandra-dc1-dc1-nodes -ucassandra -pcassandra

 system.now()
--------------------------------------
 c05823f0-2e08-11eb-a1bc-4ddca19f6dc1

(1 rows)

3.7 cassandra 部分操作

  • 查看用户下信息
# describe cluster;
Cluster: cassandra-dc1
Partitioner: Murmur3Partitioner
  • 查看所有 keyspace
# describe keyspaces;
system_schema  system_auth  system  loki  system_distributed  system_traces
  • 查看 keyspace 内容
# describe keyspace loki;
  • 创建 keyspace
# CREATE KEYSPACE loki WITH REPLICATION = { 'class' : 'SimpleStrategy', 'replication_factor' : 1 };
  • Replication Factor : 复制因数

一条新数据应该被复制到多少个节点。常用奇数。比如我们项目组设置的 replication_factor=3    Replica placement strategy : 复制策略。默认的是 SimpleStrategy. 如果是单机架、单数据中心的模式,保持使用 SimpleStrtegy 即可    或对于多数据中心策略:

# CREATE KEYSPACE loki WITH REPLICATION = { 'class' : 'NetworkTopologyStrategy', 'dc1' : 1 };
  • 删除 keyspace
# DROP KEYSPACE loki;
  • 查看详情:
# SELECT * FROM system_schema.keyspaces;
 keyspace_name      | durable_writes | replication
--------------------+----------------+-------------------------------------------------------------------------------------
        system_auth |           True | {'class''org.apache.cassandra.locator.SimpleStrategy''replication_factor''1'}
      system_schema |           True |                             {'class''org.apache.cassandra.locator.LocalStrategy'}
 system_distributed |           True | {'class''org.apache.cassandra.locator.SimpleStrategy''replication_factor''3'}
             system |           True |                             {'class''org.apache.cassandra.locator.LocalStrategy'}
               loki |           True | {'class''org.apache.cassandra.locator.SimpleStrategy''replication_factor''3'}
      system_traces |           True | {'class''org.apache.cassandra.locator.SimpleStrategy''replication_factor''2'}

(6 rows)
  • 修改复制策略
# ALTER KEYSPACE loki WITH REPLICATION = {'class' : 'NetworkTopologyStrategy', 'replication_factor': '3'};
  • 使用 keyspace

# use loki;
  • 查示所有表
# describe tables; 
# desc tables; 
  • 查看表结构
# describe columnfamaliy abc;
# desc table stocks
  • 创建表
# create table abc ( id int primary key, name varchar, age int );
  • 表删除
#drop table user;

4. 错误处理

4.1 错误 1

# kubectl logs -f cassandra-cassandra-dc1-dc1-rack1-0 -c cassandra
INFO  [main] Server.java:159 Starting listening for CQL clients on /0.0.0.0:9042 (unencrypted)...
INFO  [main] CassandraDaemon.java:564 Not starting RPC server as requested. Use JMX (StorageService->startRPCServer()) or nodetool (enablethrift) to start it
INFO  [main] CassandraDaemon.java:650 Startup complete
WARN  [OptionalTasks:1] CassandraRoleManager.java:377 CassandraRoleManager skipped default role setup: some nodes were not ready
INFO  [OptionalTasks:1] CassandraRoleManager.java:416 Setup task failed with error, rescheduling
WARN  [OptionalTasks:1] CassandraRoleManager.java:377 CassandraRoleManager skipped default role setup: some nodes were not ready
INFO  [OptionalTasks:1] CassandraRoleManager.java:416 Setup task failed with error, rescheduling

解决:当重新部署的时候无法建立成员关系,因为 statefulset 只能单个创建,而不能并行创建导致

# kubectl get pod -n grafana
NAME                                  READY   STATUS             RESTARTS   AGE
cassandra-cassandra-dc1-dc1-rack1-0   1/2     Running            0          6m12s
cassandra-operator-6f685694c5-l7m27   1/1     Running            0          4d7h

# kubectl get pvc
NAME                                              STATUS   VOLUME                                      CAPACITY   ACCESS MODES   STORAGECLASS                             AGE
data-volume-cassandra-cassandra-dc1-dc1-rack1-0   Bound    disk-29d3cfdd-dc5a-457e-bae1-6b72dcc34c37   2Ti        RWO            alicloud-disk-efficiency-cn-hangzhou-g   4h12m
data-volume-cassandra-cassandra-dc1-dc1-rack1-1   Bound    disk-9a8621f6-3f8b-428e-b69d-72cde007c7cf   2Ti        RWO            alicloud-disk-efficiency-cn-hangzhou-g   4h6m
data-volume-cassandra-cassandra-dc1-dc1-rack1-2   Bound    disk-1971e0c4-fdf5-4adf-85fa-c1e9e53b7658   2Ti        RWO            alicloud-disk-efficiency-cn-hangzhou-g   4h5m
data-volume-cassandra-cassandra-dc1-dc1-rack1-3   Bound    disk-5be7e523-a3cc-4b32-9149-6a3ab5e44ed2   2Ti        RWO            alicloud-disk-efficiency-cn-hangzhou-g   4h3m
data-volume-cassandra-cassandra-dc1-dc1-rack1-4   Bound    disk-4a4d235b-871f-45ff-be57-c4ed7c9b4ad2   2Ti        RWO            alicloud-disk-efficiency-cn-hangzhou-g   4h2m
data-volume-cassandra-cassandra-dc1-dc1-rack1-5   Bound    disk-b9c45b99-f169-413b-b8dc-65b97d205264   2Ti        RWO            alicloud-disk-efficiency-cn-hangzhou-g   4h
data-volume-cassandra-cassandra-dc1-dc1-rack1-6   Bound    disk-c2bf3596-a986-4099-b746-316ddaf36c8f   2Ti        RWO            alicloud-disk-efficiency-cn-hangzhou-g   72m
data-volume-cassandra-cassandra-dc1-dc1-rack1-7   Bound    disk-89fae7ec-9f5a-4b2f-9191-631f66ac71b8   2Ti        RWO            alicloud-disk-efficiency-cn-hangzhou-g   57m

参考官方文档,好像注释了 podManagementPolicy: Parallel 部分功能:

@@ -242,7 +241,7 @@ private V1beta2StatefulSet generateStatefulSet(DataCenterKey dataCenterKey, V1Co
                )
                .spec(new V1beta2StatefulSetSpec()
                        .serviceName("cassandra")
                        .podManagementPolicy("Parallel")
                        //.podManagementPolicy("Parallel")
                        .replicas(dataCenter.getSpec().getReplicas().intValue())
                        .selector(new V1LabelSelector().putMatchLabelsItem("cassandra-datacenter", dataCenterKey.name))
                        .template(new V1PodTemplateSpec()

目前只能临时解决:删除 PVC 重新创建 cassandradatacenter    目前也提交了 issue:https://github.com/instaclustr/cassandra-operator/issues/397

4.2 错误 2:

# kubectl exec cassandra-cassandra-dc1-dc1-rack1-0 -c cassandra -n grafana -- cqlsh -e "ALTER KEYSPACE loki WITH REPLICATION = {'class' : 'NetworkTopologyStrategy', 'replication_factor': '3'};" cassandra-cassandra-dc1-dc1-nodes -ucassandra -pcassandra
<stdin>:1:ConfigurationException: replication_factor is an option for SimpleStrategy, not NetworkTopologyStrategy

解决:replication_factor 不兼容 NetworkTopologyStrategy 策略,NetworkTopologyStrategy 允许您为每个 DC 定义不同的复制。正确的查询应为:

# CREATE KEYSPACE NTSkeyspace WITH REPLICATION = { 'class' : 'NetworkTopologyStrategy', 'datacenter1' : 1 };

无法指定 initContainer 镜像问题:

Events:
  Type     Reason          Age                From                                Message
  ----     ------          ----               ----                                -------
  Normal   Scheduled       <unknown>          default-scheduler                   Successfully assigned grafana/cassandra-cassandra-dc1-dc1-rack1-0 to cn-hangzhou.10.41.128.145
  Normal   Pulling         45s                kubelet, cn-hangzhou.10.41.128.145  Pulling image "busybox:latest"

# egrep -r busybox cassandra-operator-7.1.0/
cassandra-operator-7.1.0/pkg/controller/cassandradatacenter/statefulset.go:     var image = "busybox:latest"

解决:

cassandra-operator#379[5]

added initImage field into spec for init container, up to now, it was always busybox:latest. It defaults to this image if that field is empty.

# vi cassandra-operator-7.1.0/examples/example-datacenter.yaml
spec:
  initImage: ops-harbor.hupu.io/base/alpine:v3.10

参考资料

[1]

Apache Cassandra:https://cassandra.apache.org/doc/latest/architecture/dynamo.htm

[2]

部署Cassandra CRD:https://github.com/instaclustr/cassandra-operator/wiki/Installation-and-deployment

[3]

Cassandra Cluster参考1:https://github.com/instaclustr/cassandra-operator/wiki/Custom-configuration

[4]

Cassandra Cluster参考2:https://cassandra.apache.org/doc/latest/configuration/index.html

[5]

cassandra-operator#379:https://github.com/instaclustr/cassandra-operator/issues/379


Loki日志系统分布式部署实践之 Cassandra


  
    
    
  
Loki日志系统分布式部署实践之 Cassandra




  
    
    
  


  
    
    
  
点个赞+在看,少个 bug   👇