Hello,
We are thinking of changing over from traditional replication to gtid replication.
I would be interested in peoples opinion on gtid replication are there any gotchas we should be aware of?
Thanks for your thoughts.
Hello @Jamie_Downs,
GTID makes things pretty simple as you no longer have to worry about binlog files and positions. I’d say the #1 issue I run into is missing GTIDs when promoting a replica. Let’s say you have a simple S->R setup. You run something seemingly harmless on the R like creating a new user. Your GTID set now contains GTIDs from S and separate GTID from the R itself. Now, S goes down and R becomes the new S (S’). The new S’ continues running the app. S comes back online and tries to connect to S’ as a R (R’). But S’ cannot replicate to R’ because there’s missing GTIDs as those events are no longer in S’ binlogs.
Check out this blog: Errant transactions: Major hurdle for GTID-based failover
Any customer that I configure replication for I use GTIDs as they are slowly becoming the new standard.
Thanks @matthewb ,
This is good to know. Especially:
“Any customer that I configure replication for I use GTIDs as they are slowly becoming the new standard.”
Does anyone have any experiences to share? Thanks.