组复制旨在确保数据库服务持续可用，即使由于计划内维护或意外问题，形成组的一些服务器当前无法参与其中。只要剩余成员占组的多数，他们就可以选举出新的主服务器，并继续作为一个组运行。然而，如果复制组的每个成员都离开了组，并且通过STOP GROUP_REPLICATION语句或系统关闭在每个成员上停止了组复制，该组现在仅在理论上存在，作为成员上的配置。在这种情况下，要重新创建组，必须通过引导启动，就像首次启动一样。Group Replication is designed to ensure that the database service is continuously available, even if some of the servers that form the group are currently unable to participate in it due to planned maintenance or unplanned issues. As long as the remaining members are a majority of the group they can elect a new primary and continue to function as a group. However, if every member of a replication group leaves the group, and Group Replication is stopped on every member by a STOP GROUP_REPLICATION statement or system shutdown, the group now only exists in theory, as a configuration on the members. In that situation, to re-create the group, it must be started by bootstrapping as if it was being started for the first time.

第一次为一个组引导（bootstrapping）与第二次或随后进行引导的区别在于，在后者情况下，一个已经关闭的组的成员之间可能拥有不同的事务集，这取决于它们被停止或失败的顺序。如果一个成员拥有的事务在其他组成员中不存在，该成员就无法加入该组。对于组复制（Group Replication）来说，这包括已经提交并应用的事务，这些事务在 gtid_executed GTID 集中，以及已认证但尚未应用的事务，这些事务在 group_replication_applier 通道中。事务被提交的具体时间点取决于为组设置的事务一致性级别（参见 第 20.5.3 节，“事务一致性保证”）。然而，组复制的组成员从不移除已经认证的事务，这表示成员有意提交该事务。The difference between bootstrapping a group for the first time and doing it for the second or subsequent times is that in the latter situation, the members of a group that was shut down might have different transaction sets from each other, depending on the order in which they were stopped or failed. A member cannot join a group if it has transactions that are not present on the other group members. For Group Replication, this includes both transactions that have been committed and applied, which are in the gtid_executed GTID set, and transactions that have been certified but not yet applied, which are in the group_replication_applier channel. The exact point at which a transaction is committed depends on the transaction consistency level that is set for the group (see Section 20.5.3, “Transaction Consistency Guarantees”). However, a Group Replication group member never removes a transaction that has been certified, which is a declaration of the member’s intent to commit the transaction.

因此，复制组必须从最新的成员开始重启，也就是执行和认证事务最多的成员。事务较少的成员随后可以加入，并通过分布式恢复赶上缺失的事务。不能假设组中最后已知的主成员就是组中最新的成员，因为比主成员稍后关闭的成员可能有更多的事务。因此，必须重启每个成员以检查事务，比较所有事务集，并确定最新的成员。然后可以使用该成员来引导组。The replication group must therefore be restarted beginning with the most up to date member, that is, the member that has the most transactions executed and certified. The members with fewer transactions can then join and catch up with the transactions they are missing through distributed recovery. It is not correct to assume that the last known primary member of the group is the most up to date member of the group, because a member that was shut down later than the primary might have more transactions. You must therefore restart each member to check the transactions, compare all the transaction sets, and identify the most up to date member. This member can then be used to bootstrap the group.

在每个成员关闭后，请按照此程序安全地重启复制组。Follow this procedure to restart a replication group safely after every member shuts down.