Consumer Autoscaling on Lag¶
Scale consumers from backlog age and processing capacity, while respecting partition parallelism and rebalance cost.
flowchart LR J[Junior: why lag] --> M[Middle: scaling loop] --> S[Senior: flapping failures] --> P[Professional: fleet scale]flowchart LR Kafka --> Lag[Consumer lag] --> KEDA --> Replicas --> Group[Consumer group]| Level | Guide | You are done when | |---|---|---| | Junior | Definition and why | You can explain why CPU misses backlog. | | Middle | How it works | You can turn lag into bounded replicas. | | Senior | Failures and mistakes | You can prevent flapping and unsafe scale-down. | | Professional | Best practices and scale | You can design a stable autoscaling control loop. | Practice rule: Scale to drain work within an SLO, never beyond useful partition parallelism.