I) An upgrade which hasn’t gone as planned
One of our customer needed help after trying to upgrade their
slave from Percona Server 5.6 to Percona Server 5.7. (here 5.7.17
Indeed, from time to time, the replication lag was linearly growing, which could be quite annoying…
(good guess, 5.7 is the red dashed line!)
Their use case is quite interesting :
- a lot of Queries Per Second (read & write, nearly 10k queries/s on this slave)
- a lot of tables (about 600k spread across different databases)
- a lot of writes
- 24 CPUs, and 386G of RAM
- A big buffer pool (270G)
- not that fast SSD disks.
Because we have a high QPS but also a lot of writes, we need to
fine tune the innodb settings.