Server:
Vmware virtual machineOperating system: Centos7.4
CPU: 2 * 8=16 cores
Mem: 32 g
Disk: 1 t
Mysql version: 5.7.25

Description:
MGR single main mode, each node is configured with the weight, in decreasing order of 1, 2, 3, using keepalived generate VIP, also configuration and MGR consistent weight,
Business through the VIP access the database, and one from the library, data replication by VIP from MGR master node, the program end there are a lot of time task, in which two tasks were once every ten minutes to extract data inserted into a library, a 15000, another task is once every half an hour, 25000 pieces of data at a time, written to the library, each task done transaction control, every submitted article 100, the current database innodb status shows not young quantity is more,
Recent problems turned out of the cluster nodes, network error nodes errorlog of deportation, but monitoring, see only network quality is poorer, sometimes there will be a delay of more than 30 ms, did not appear network interruption,
Errolog:
The 2019-08-16 T23:20:07. 669964 + 08:00 0 [Warning] Plugin group_replication reported: 'Member with address sha11xxdb03:3306 has become unreachable.'
The 2019-08-16 T23:20:07. 687859 + 08:00 0 [Warning] Plugin group_replication reported: 'Member with address sha11xxdb01:3306 has become unreachable.'
The 2019-08-16 T23:20:07. 687926 + 08:00 0 [ERROR] Plugin group_replication reported: 'This server is not able to reach a majority of members in the group. This server will now block all updates. The server will remain blocked for the next 300 seconds. Unless the contact with the majority is restored, after This time, the member will ERROR out and leave the group. It is possible to use group_replication_force_members to force a new group membership.'
T23:2019-08-16 20:08. 399913 + 08:00 0 [ERROR] Plugin group_replication reported: 'Member was expelled from the group, due to the network failures, changing member status to the ERROR.
'
Do to transaction limit MGR, each transaction cannot exceed 10 m, and do to io_capacity and io_capacity_max limit, are 300 and 600 respectively, innodb_lru_scan_depth=256,

Ever encountered this problem, please friends, hope to help give directions, or have idea also please guide, is the problem for a long time
In addition, when the VIP from primary drift to the secondary, the library does not automatically switch to the new continue to copy the data on the master node, note: synchronization via the VIP from the library