Associate Cloud Engineer free question 4 / 10 · Domain 2: Planning and implementing a cloud solution
An application in a MIG takes about five minutes from VM start until it can serve traffic, and traffic rises at about the same time every day. You want new VMs to be ready to serve by the time the load arrives. According to the autoscaler page, which configuration fits?
- Create a schedule that starts at the time the load rises and keep the default initialization period of 60 seconds
- Turn on predictive autoscaling and keep the default initialization period of 60 seconds
- Turn on predictive autoscaling and set the initialization period to 300 seconds
- Create a schedule that starts at the time the load rises and set the initialization period to 300 seconds
Show the answer and explanation
Answer: C
Why this is correct
Predictive autoscaling forecasts future load from historical data and scales the MIG out ahead of the forecast, so new instances are ready when the load arrives. The page says it works best when an application takes a long time to initialize (for example, an initialization period longer than two minutes) and when load changes in a predictable daily or weekly cycle. With an initialization period of 300 seconds, predictive autoscaling creates VMs five minutes before the forecast load.
Why the other options are wrong
"Create a schedule that starts at the time the load rises and set the initialization period to 300 seconds" … The page on scaling based on schedules says a schedule starts creating VMs at its start time, and the VMs are not ready at that exact time, so you should start the schedule earlier by the time it takes to start and initialize (for example, 10 minutes before you need capacity if creation to serving takes 10 minutes). If the start time matches the time the load rises, VMs begin to be created then and need about five more minutes to serve. Creating VMs ahead of the need by the initialization period is what predictive mode does; with a schedule you move the start time itself earlier.
"Turn on predictive autoscaling and keep the default initialization period of 60 seconds" … The page says that in predictive mode VMs are created ahead of time by the initialization period (300 seconds means five minutes before the forecast load). The default initialization period is 60 seconds, so VMs would be created only one minute ahead, and an application that needs about five minutes would not be ready when the load arrives.
"Create a schedule that starts at the time the load rises and keep the default initialization period of 60 seconds" … A schedule begins creating VMs at its start time, so starting it when the load rises means VMs are created at the same time as the load. The default 60-second initialization period also leaves the window in which scale-out decisions ignore initializing VMs shorter than the real start time. In both respects, VMs are not ready to serve when the load arrives.
How the initialization period is handled
A schedule-based approach can also prepare VMs before the load if you move its start time earlier by the initialization time. The two scheduled options in this question fall short because both start at the time the load rises.
The initialization period used to be called the cooldown period. For scale-out decisions the autoscaler ignores the utilization of VMs that are still initializing, and for scale-in decisions it averages over the group including initializing VMs. Real initialization time varies with many factors, so the page recommends testing how long it actually takes.
Autohealing is the mechanism by which a MIG recreates VMs that fail a health check, and it works independently of autoscaling. A scale-in controller keeps the group from shrinking too fast when another peak comes right after load drops. Neither prepares VMs ahead of the load.
Go to the Associate Cloud Engineer practice app (10 free questions; purchase unlocks all questions)