When should you scale your message queue infrastructure?

When should you scale your message queue infrastructure?

• A. Throughput approaches 800k messages per second or consumer lag is consistently growing.
• B. Latency is consistently under 1 millisecond.
• C. The system currently handles 2 million messages per second without issues.
• D. The number of topics has doubled in the last hour.

Answer: (A)

The signal to scale is when demand starts to outpace what your current setup can handle, or when the system begins to accumulate a backlog. In practice, that means watching two concrete indicators: throughput approaching the system’s capacity ceiling and consumer lag that keeps growing. If you’re hitting around a high threshold (for example, around 800k messages per second) or if consumers can’t keep up and lag persists, it’s time to scale—adding more partitions, brokers, or increasing parallelism to restore balance. Low latency (latency consistently under 1 millisecond) is a sign of good performance, but it isn’t by itself a reason to scale. You can maintain low latency even as you grow, but you’ll need to scale when throughput climbs enough that lag starts to rise or when your capacity limit is reached. Conversely, handling 2 million messages per second without issues suggests current capacity is sufficient for now, but it doesn’t guarantee you’ll stay in the clear if demand grows; scaling decisions should be driven by observed load and backlog, not a snapshot of current throughput alone. An increasing number of topics doesn’t directly indicate saturation or backlog; it can complicate routing and management without signaling a need to scale.

The signal to scale is when demand starts to outpace what your current setup can handle, or when the system begins to accumulate a backlog. In practice, that means watching two concrete indicators: throughput approaching the system’s capacity ceiling and consumer lag that keeps growing. If you’re hitting around a high threshold (for example, around 800k messages per second) or if consumers can’t keep up and lag persists, it’s time to scale—adding more partitions, brokers, or increasing parallelism to restore balance.

Low latency (latency consistently under 1 millisecond) is a sign of good performance, but it isn’t by itself a reason to scale. You can maintain low latency even as you grow, but you’ll need to scale when throughput climbs enough that lag starts to rise or when your capacity limit is reached. Conversely, handling 2 million messages per second without issues suggests current capacity is sufficient for now, but it doesn’t guarantee you’ll stay in the clear if demand grows; scaling decisions should be driven by observed load and backlog, not a snapshot of current throughput alone. An increasing number of topics doesn’t directly indicate saturation or backlog; it can complicate routing and management without signaling a need to scale.