replication and binlog pointer position saving

[Eugene Zheganin]

Hello,

I'm using Mysql replication sets for I don't really remember how long (more than 10 years for now), and one thing really bothers me: it seems like something has changed between 5.7 and 8.0 in the binlog position syncing. I'll explain on examples: imagine the replica crashses due to the power outage or server reset/OS crash. Or you're taking the zfs snapshot without stopping the server and then cloning it/rolling back. In the 5.7 releases this was safe: after restart the replica was continuing from the plase where it was shut down. In 8.0 releases this is 99% chance of getting huge amount of duplicates on primary keys or rows that cannot be found for updates: and this basically tells me that the master_exec_pos was not saved, and it actually points back, to the place that was already processed.

So, is there something I've missed, may be I should take some measures bu setting some of the replication variables so the binlog position for a slave thread will be more of forcefully synchronized ?