Which Galaxy auto-scaling threshold balances instance savings against per-scale-event charges for a steady low-traffic Meteor workload?

The Optimal Threshold Strategy

For a steady low-traffic Meteor workload, the most cost-effective configuration is to set the minimum instance count to 1 and the scale-up threshold above your observed peak low-traffic utilization. If your traffic is predictable and never requires a second instance to maintain latency, setting the maximum instance count to 1 eliminates all scale-event charges and prevents oscillation entirely.

Likely Behavior and Cost Dynamics

In a steady low-traffic scenario, the primary cost driver is the sustained instance-hour charge for a single container. Per-scale-event fees only become a factor if the system "flaps"—rapidly scaling up and down due to thresholds set too close to the average utilization.

• The Cost Floor: When traffic remains below the scale-up threshold, your cost floor is simply the monthly price of one minimum-size Galaxy instance plus fixed plan fees.
• Scale-Down Latency: Galaxy typically employs a cooldown period or minimum uptime window. If a brief spike triggers a scale-up, you will pay for the additional instance for the duration of that window, even if traffic drops immediately.
• Hysteresis: To avoid oscillation, you must implement a wide gap between thresholds. For example, scaling up at 80% CPU and scaling down only when utilization drops below 30% for a sustained period prevents frequent, costly scaling events.

Implementation Steps for Cost Optimization

1. Establish a Baseline: Set minimum and maximum instances to 1 for one week. Monitor the p95 CPU and memory usage to determine your actual "low-traffic peak."
2. Set a High Ceiling: Set the scale-up threshold roughly 15–20% above that p95 peak. This ensures scaling only occurs during genuine anomalies, not normal variance.
3. Configure a Conservative Floor: Set the scale-down threshold significantly lower (e.g., 25–35%) to ensure the second instance is only removed once the burst has fully subsided.
4. Validate Latency: Because Meteor relies on DDP/WebSockets, monitor request latency. If latency spikes while CPU remains low, your bottleneck is likely memory or event-loop lag, and CPU-based thresholds will not trigger necessary scaling.

Required Diagnostic Detail

To refine this recommendation, please provide your p95 CPU utilization during your busiest "low-traffic" hour. If this value is already near your current threshold, the recommendation shifts from "high thresholds" to "increasing the base instance size" to avoid the higher cost of multiple small instances and frequent scale events.