Scaling Policy Types

Target tracking scaling is the simplest and most commonly recommended policy type: you specify a target value for a chosen metric (e.g. 'keep average CPU at 50%'), and AWS's algorithm automatically calculates and adjusts capacity to hit that target, similar in spirit to a thermostat maintaining a room temperature — you don't manually decide how many instances to add, you just declare the goal.

Step scaling gives more manual control: you define specific thresholds and exactly how many instances to add or remove at each threshold (e.g. 'if CPU > 70%, add 2 instances; if CPU > 90%, add 5 instances'). This is more work to tune correctly but allows more aggressive or nuanced responses than target tracking's single-target model, useful when you understand your traffic pattern precisely.

Scheduled scaling proactively sets capacity ahead of a known time-based pattern (e.g. 'scale to 10 instances every weekday at 8:55am before the morning traffic surge, scale back to 2 at 8pm') — useful for predictable daily/weekly cycles where you'd rather pre-warm capacity than wait for reactive scaling to catch up. Scheduled and target-tracking/step policies are commonly layered together: scheduled scaling sets a sensible baseline ahead of known patterns, while a target-tracking policy still handles unexpected additional spikes on top of that baseline.