Because of a rare network problem, my 3-node testbed of innodb cluster got into a strange condition. The replicationLag of two RO nodes were both over 7 hours, and the status of both RO nodes were still ONLINE. There was no error in mysqld.log but some warnings only.
(I don't think it is a right condition, but this is not the issue I would discuss here.)
The actual problem was MySQL router still read data from RO nodes even most data were out-of-data already.
I thought the expected result should be -- when the replicationLag is over a specified threshold, the un-synced nodes should not be used any more.
Is there any system variables could prevent this?