Adding and removing capacity.
WHAT MANUAL SCALING IS
Changing the number of replicas.
WHAT AUTOMATIC SCALING DOES
Adjusts that number based on a measurement.
WHAT MEASUREMENTS ARE USED
Processor use, commonly Memory Custom measurements, such as queue depth
WHY QUEUE DEPTH IS FREQUENTLY BETTER
It reflects actual backlog rather than a proxy for it.
WHAT AUTOMATIC SCALING REQUIRES
Resource requests set, so usage can be compared Metrics available to the cluster An application that can run in several instances
WHAT PREVENTS AN APPLICATION SCALING
Local state Sessions held in memory Background work assuming a single instance
WHAT TO DO ABOUT EACH
Move state out Share sessions Make scheduled work safe to run once across instances
WHAT CLUSTER SCALING ADDS
More nodes, when pods cannot be placed.
WHY BOTH ARE NEEDED
More replicas need somewhere to run.
WHAT TO SET
Minimum and maximum replicas.
WHY A MAXIMUM
It bounds cost, and protects downstream systems.
WHAT TO BE CAREFUL WITH
Scaling that overwhelms a database.
WHY
Every new instance opens connections.
WHAT TO TEST
Behaviour while scaling, not only at each size.
WHAT TO WATCH
Whether scaling actually improves anything.
