Slide 35
Slide 35 text
Kubernetes at HubSpot - Getting used to the Container / Kubernetes World
An example of “vm concept” we brought in our Kubernetes Operator:
● Performance class for keyspaces: x-small (for tests), small (default), medium, large, xl, 2xl…
(control CPU and RAM allocation, and other MySQL params)
● From RAM, Max Cnx, and other MySQL params, we back-compute InnoDB Buffer Pool Size
→ this is sub-optimal, BP size should be 1st level params and RAM should be computed from it
● So we design medium “vms” / pod, not medium “MySQL Instances”: we will fix this soon
(in new design, allocate BP size and forward-compute RAM for the MySQL container)
(no more BP size shrinkage when increasing Max Cnx or other MySQL param)
Kubernetes allows the storage of configuration in Custom Resources
● Works well with everything in a single Kubernetes Clusters
● But when a MySQL deployment spawns 2 clusters, keeping things in sync is complicated
● Hublets also duplicated all this (not only we mange QA and Prod, but now also na1 and eu1)
● An external / centralized asset database would probably make sense