Rather than rely on the orchestration system to configure read-threads and write-threads, I think it's possible that we can auto-tune the threads based on innodb_purge_threads, which is set to the number of vCPUs on an aurora system.
We may need to improve throttling slightly, since it currently is just based on commit latency. We should also have throttling based on current threads* exceeding the number of vCPUs.
- = Not from threads_running, but from performance_schema and excluding any threads stuck on the redo log, which are blocked and likely not consuming CPU.
Rather than rely on the orchestration system to configure
read-threadsandwrite-threads, I think it's possible that we can auto-tune the threads based oninnodb_purge_threads, which is set to the number of vCPUs on an aurora system.We may need to improve throttling slightly, since it currently is just based on commit latency. We should also have throttling based on current threads* exceeding the number of vCPUs.