Seconds_Behind_Master gives impossible values

After upgrade from 10.6.17 to 10.6.18, our monitoring started to give alerts multiple times a day from all replicas becuase Seconds_Behind_Master > 3600.

It looks to me that value is no more seconds.

Tested by checking value every two seconds. value jumps from 0 to 40726, might increase a bit for checks, and then reset back to 0.

It is completey impossible for this value to jump from 0 to 40726 seconds in two seconds.
Clearly a regression.

It looks like time increases at the same time on all replicas, replication is set to replicate from galera cluster.