Orchestration: autoscaling
The instance group and its bounds, threshold and target-tracking policies, the observation window and the cooldown, replacing an unhealthy instance, and the conditions for it to work: a stateless service and a load balancer in front.
- 26 cards
- 5 topics
- Free
The outline, topic by topic
Each topic is weighted the way the exam weighs it.
- Concepts
- Questions
- Sort and link
- The numbers
- Scenarios
A few cards from the deck
Autoscaling group: what is it?
Show the answer
A set of identical instances whose number varies on its own between a minimum and a maximum, following a policy, behind a load balancer.
The minimum guarantees a service always present, the maximum caps the bill, and the desired count, between the two, is what the policy asks for at each moment.
What is the cooldown of an instance group for?
- Giving the previous adjustment time to act
- Cooling the processors of overloaded instances
- Waiting for traffic to return to zero before removing an instance
- Capping the total number of instances the group can create
Show the answer
Giving the previous adjustment time to act
Added instances only relieve the load once started. Without a delay, the group would still see high load and add too many. The cap is the maximum.
An instance group has a … count, a … count and a desired count that always stays between the two.
Show the answer
An instance group has a minimum count, a maximum count and a desired count that always stays between the two.
Even if the metric asks for 20 instances, the group stops at the maximum: it must therefore be set with the budget and quotas in mind.
Horizontal scaling: what is it?
Show the answer
Increasing capacity by adding identical machines, rather than by making a single machine more powerful.
It escapes the ceiling of a single machine and survives failures better, but requires the service to work on several machines at once. The opposite, vertical, changes an instance's size and often requires a restart.