mysql - Galera集群节点不触发wsrep_notify_cmd和wsrep_sst_method

Question

我确实使用 3 个 docker 容器 设置了具有 3 个节点的 Galera 集群。有一个要求，当数据从donor节点同步到其他节点时，基于同步时的wsrep_notify_cmd触发器或wsrep_sst_method触发器节点的数据也需要填充到该节点对应的Redis队列中。问题是这 2 个触发器仅在我启动集群时调用。有日志说当一个节点加入集群时调用了这 2 个触发器。但是，当我尝试在一个节点中修改架构或执行 CUD 操作时，触发器不会在其他节点上触发。我不知道是我配置不正确，还是这些触发器的工作方式不对。

下面是我用来使集群正常工作的文件

• docker-compose.yml

  version: '3'
  services:
    node1:
        build: ./galera/
        image: galera_mariadb:latest
        container_name: "galera_cluster_node1"
        hostname: node1
        ports:
          - 13306:3306
        networks:
          - galera_cluster
        volumes:
          - ./galera/conf.d/galera.cnf:/etc/mysql/conf.d/galera.cnf
          - /var/data/galera/mysql/node1:/var/lib/mysql/
          # ./galera/scripts contains the bash script which is executed by wsrep_notify_cmd trigger
          - ./galera/scripts/:/etc/mysql/scripts/
        environment:
          - MYSQL_ROOT_PASSWORD=123
          - REPLICATION_PASSWORD=123
          - MYSQL_DATABASE=test_db
          - MYSQL_USER=maria
          - MYSQL_PASSWORD=123
          - GALERA=On
          - NODE_NAME=node1
          - CLUSTER_NAME=maria_cluster
          - CLUSTER_ADDRESS=gcomm://
        command: --wsrep-new-cluster
  
    node2:
        image: galera_mariadb:latest
        container_name: "galera_cluster_node2"
        hostname: node2
        links:
          - node1
        ports:
          - 23306:3306
        networks:
          - galera_cluster
        volumes:
          - ./galera/conf.d/galera.cnf:/etc/mysql/conf.d/galera.cnf
          - /var/data/galera/mysql/node2:/var/lib/mysql/
          - ./galera/scripts/:/etc/mysql/scripts/
        environment:
          - REPLICATION_PASSWORD=123
          - GALERA=On
          - NODE_NAME=node2
          - CLUSTER_NAME=maria_cluster
          - CLUSTER_ADDRESS=gcomm://node1
  
    node3:
        image: galera_mariadb:latest
        container_name: "galera_cluster_node3"
        hostname: node3
        links:
          - node1
        ports:
          - 33306:3306
        networks:
          - galera_cluster
        volumes:
          - ./galera/conf.d/galera.cnf:/etc/mysql/conf.d/galera.cnf
          - /var/data/galera/mysql/node3:/var/lib/mysql/
          - ./galera/scripts/:/etc/mysql/scripts/
        environment:
          - REPLICATION_PASSWORD=123
          - GALERA=On
          - NODE_NAME=node3
          - CLUSTER_NAME=maria_cluster
          - CLUSTER_ADDRESS=gcomm://node1
  
  networks:
    galera_cluster:
        driver: bridge
  

• 用于搭建3个galera集群节点的Dockerfile

  # Galera Cluster Dockerfile
  FROM hauptmedia/mariadb:10.1
  RUN apt-get update \
      && apt-get -y install \
          vim \
          python \
          redis-tools
  
  # remove the default galera.cnf in the original image
  RUN rm -rf /etc/mysql/conf.d/galera.cnf
  # add the custom galera.cnf
  COPY ./conf.d/galera.cnf /etc/mysql/conf.d/galera.cnf
  # grant access and execution right
  RUN chmod 755 /etc/mysql/conf.d/galera.cnf
  

• galera.cnf文件

  [galera]
  wsrep_on=ON
  
  # wsrep only supports binlog_format='ROW' and storage-engine=innodb
  binlog_format=row
  default_storage_engine=InnoDB
  
  # to avoid issues with 'bulk mode inserts' using autoinc
  innodb_autoinc_lock_mode=2
  
  bind-address=0.0.0.0
  
  # relax flushing logs when running in galera mode
  innodb_flush_log_at_trx_commit=0
  sync_binlog=0
  
  # Query Cache is supported since version 10.0.14 with wsrep
  query_cache_size=8000000
  query_cache_type=1
  
  wsrep_provider=/usr/lib/galera/libgalera_smm.so
  # use the built-in method to manage State Snapshot Transfers
  # we can customize this script to perform a specific logic
  wsrep_sst_method=xtrabackup-v2
  # This bash is volumed from the host which is used to populate synchronized data to the Redis queue
  wsrep_notify_cmd=/etc/mysql/scripts/wsrep_notify.sh
  
  # force transaction level to be read commited
  #transaction-isolation = READ-COMMITTED
  

• wsrep_notify.sh

  #!/bin/sh -eu
  
  wsrep_log()
  {
      # echo everything to stderr so that it gets into common error log
      # deliberately made to look different from the rest of the log
      local readonly tst="$(date +%Y%m%d\ %H:%M:%S.%N | cut -b -21)"
      echo "WSREP_SST: $* ($tst)" >&2
  }
  
  wsrep_log_info()
  {
      wsrep_log "[INFO] $*"
  }
  
  STATUS=""
  CLUSTER_UUID=""
  PRIMARY=""
  MEMBERS=""
  INDEX=""
  
  while [ $# -gt 0 ]
  do
     case $1 in
        --status)
           STATUS=$2
           shift
           ;;
        --uuid)
           CLUSTER_UUID=$2
           shift
           ;;
        --primary)
           PRIMARY=$2
           shift
           ;;
        --index)
           INDEX=$2
           shift
           ;;
        --members)
           MEMBERS=$2
           shift
           ;;
           esac
           shift
     done
  
  wsrep_log_info "--status $STATUS --uuid $CLUSTER_UUID --primary $PRIMARY --members $MEMBERS --index $INDEX"
  

• 这里是3个节点的日志文件

节点 1:

https://drive.google.com/file/d/0B2q2F62RQxVjbkRaQlFrV2NyYnc/view?usp=sharing

节点2:

https://drive.google.com/file/d/0B2q2F62RQxVjX3hYZHBpQ2FRV0U/view?usp=sharing

节点 3:

https://drive.google.com/file/d/0B2q2F62RQxVjelZHQTN3ZDRNZ0k/view?usp=sharing

我一直在用谷歌搜索这个问题，但没有成功。我希望任何有 Galera Cluster 设置经验的人都可以帮助我解决这个问题。或者有另一种方法来解决需求请告诉我。非常感谢

Answer 1

wsrep_notify_cmd

Defines the command the node runs whenever cluster membership or the state of the node changes.

因此，如果节点将下面列表中描述的状态更改为任何其他状态，脚本将在节点上启动:

The possible statuses are:

Undefined The node has just started up and is not connected to any Primary Component.

Joiner The node is connected to a primary component and now is receiving state snapshot.

Donor The node is connected to primary component and now is sending state snapshot.

Joined The node has a complete state and now is catching up with the cluster.

Synced The node has synchronized itself with the cluster.

Error( if available>) The node is in an error state.

您会看到脚本在节点启动和更改其状态时发出通知。当数据刚刚在 galera 集群节点之间同步时，它不会通知。