主从切换原主库同步失败

[cossbow]

我运行了三个节点，一个主库两个从库，然后运行5个哨兵，一般情况下关闭主库能切换完成，但偶尔发生这种情况：重新再启动原来的主库A的时候，从当前的主库B同步数据时发生错误，具体错误信息没找到，只找到以下日志：
A的INFO日志

I0120 13:38:30.782464 2371 pika.cc:191] Server at: pika.conf
I0120 13:38:30.783274 2371 pika_server.cc:176] Using Networker Interface: ens32
I0120 13:38:30.784133 2371 pika_server.cc:219] host: 172.16.0.81 port: 6483
I0120 13:38:30.784153 2371 pika_server.cc:95] Worker queue limit is 20100
I0120 13:38:30.784390 2371 pika_binlog.cc:111] Binlog: Find the exist file.
I0120 13:38:30.784500 2371 pika_binlog.cc:111] Binlog: Find the exist file.
I0120 13:38:30.784549 2371 pika_binlog.cc:111] Binlog: Find the exist file.
I0120 13:38:30.784627 2371 pika_binlog.cc:111] Binlog: Find the exist file.
I0120 13:38:30.784678 2371 pika_binlog.cc:111] Binlog: Find the exist file.
I0120 13:38:30.784725 2371 pika_binlog.cc:111] Binlog: Find the exist file.
I0120 13:38:30.784775 2371 pika_binlog.cc:111] Binlog: Find the exist file.
I0120 13:38:30.784829 2371 pika_binlog.cc:111] Binlog: Find the exist file.
I0120 13:38:31.831373 2371 pika_partition.cc:87] db0 DB Success
I0120 13:38:31.855065 2371 pika_partition.cc:87] db1 DB Success
I0120 13:38:31.879359 2371 pika_partition.cc:87] db2 DB Success
I0120 13:38:31.921152 2371 pika_partition.cc:87] db3 DB Success
I0120 13:38:31.947860 2371 pika_partition.cc:87] db4 DB Success
I0120 13:38:31.971632 2371 pika_partition.cc:87] db5 DB Success
I0120 13:38:31.996385 2371 pika_partition.cc:87] db6 DB Success
I0120 13:38:32.020203 2371 pika_partition.cc:87] db7 DB Success
I0120 13:38:32.020740 2371 pika_server.cc:273] Pika Server going to start
I0120 13:38:32.020967 2400 pika_repl_client.cc:146] Try Send Meta Sync Request to Master (172.16.0.81:6482)
I0120 13:38:32.022534 2373 pika_server.cc:618] Mark try connect finish
I0120 13:38:32.022550 2373 pika_repl_client_conn.cc:146] Finish to handle meta sync response
I0120 13:38:32.166294 2374 pika_repl_client_conn.cc:258] Partition: db6 TrySync Ok
I0120 13:38:32.166479 2374 pika_repl_client_conn.cc:258] Partition: db4 TrySync Ok
I0120 13:38:32.166548 2373 pika_repl_client_conn.cc:258] Partition: db5 TrySync Ok
I0120 13:38:32.166859 2374 pika_repl_client_conn.cc:258] Partition: db1 TrySync Ok
I0120 13:38:32.166882 2373 pika_repl_client_conn.cc:258] Partition: db3 TrySync Ok
I0120 13:38:32.166898 2373 pika_repl_client_conn.cc:258] Partition: db7 TrySync Ok
I0120 13:38:32.166908 2373 pika_repl_client_conn.cc:258] Partition: db0 TrySync Ok

A的WARNING日志

Log file created at: 2021/01/20 13:38:32
Running on machine: localhost.localdomain
Log line format: [IWEF]mmdd hh:mm:ss.uuuuuu threadid file:line] msg
W0120 13:38:32.166677 2374 pika_repl_client_conn.cc:267] Partition: db2 TrySync Error

B的WARNING日志

W0120 14:37:51.114579 1463 pika_repl_server_conn.cc:274] Slave offset is not a start point of cur log, Slave ip: 172.16.0.81, Slave port: 6483, Partition: db2, cloest start point, filenum: 313, offset: 3796997

请问这个问题怎么手动处理呢？还是只能重建A的数据库（删除重新同步的那种）

[kernelai]

第一次发生主从切换时，B可能会丢失一部分数据，导致A第二次同步时，A与B的binlog点位不一致。这时候，可以强制全同步，在A上执行slaveof B_ip B_port force.从而使A上的数据与B完全一致。

[cossbow]

哦，多谢了；虽然和重新建库一样，不过不用restart和delete，安全一些😁