This has been going on in my mind for years and I'd like to get some thoughts.
We've running SBR on masters forever but RBR on slaves. Our candidate masters run SBR.
The motivation for this was to be able to at some point migrate to RBR everywhere, but this has been proven to be unlikely in the last few (more than 5 years), due to various reasons, starting with not being able to fully fully ensure our data is consistent everywhere, especially on those tables that didn't have PK for a long time (in some case till recently).
By mistake, some years ago we ran s5 with RBR everywhere for a lots of months and never happened, but we were always a bit nervous about it.
The fact that we run RBR on the slaves but not on the candidate master, it makes us very limited in our choice of candidate masters in case a master failure.
I'd like to propose we migrate sX to SBR everywhere (not on backup sources, sanitarium master, sanitarium and clouddb*).
x1 should remain RBR as it's been proven for years it is all good there.
@Ladsgroup @jcrespo @FCeratto-WMF thoughts? Either way, to move forward or to keep the current status are welcome.
Progress
- s1
- eqiad
- codfw
- s2
- eqiad
- codfw
- s3
- eqiad
- codfw
- s4
- eqiad
- codfw
- s5
- eqiad
- codfw
- s6
- eqiad
- codfw
- s7
- eqiad
- codfw
- s8
- eqiad
- codfw
- x3
- eqiad
- codfw