It turns out there are two bugs causing two failures in this ticket. This one is about the second failure.
When multiple slaves connects to the master at the same time, the socket buffers were not correctly created, causing them sharing buffers and resulting in data loss in the slaves.
One of the slaves
1. connect slave (add node)
2. connect another slave (add node)
In this order
3. (master + 2 slaves)
Data loss in slave, no clear evidence.
It’s hard as there is not much evidences.
The socket buffer wasn’t created correctly for the 2nd and later slaves. The buffers were all messed up...