Well, I’ve literally observed it (circa 2016 and design has not significantly changed with this in that time) and confirmed my finding with percona.
deadlock was the wrong terminology to use, apologies, I keep writing from my phone as I am travelling at the moment: I meant lock contention, specifically in memory. A deadlock would be a hard stop but what I observed was a bottleneck on memory bandwidth past a certain number of cores (24) with update heavy workloads.
So, appreciate your points but I don't think I am wrong in the core thesis of my statements. x :)
You would not be able to saturate the memory bus if you have a lock contention. Having a lock contention is usually exhibited in under-utilizing your CPU compute and memory bandwidth resources. So, hitting the limit of the available memory bandwidth still sounds like a misrepresentation of the issue you stumbled upon.
deadlock was the wrong terminology to use, apologies, I keep writing from my phone as I am travelling at the moment: I meant lock contention, specifically in memory. A deadlock would be a hard stop but what I observed was a bottleneck on memory bandwidth past a certain number of cores (24) with update heavy workloads.
So, appreciate your points but I don't think I am wrong in the core thesis of my statements. x :)