Multi-Cluster Warehouses: Avoid Paying for Unused Clusters

Multi-Cluster Warehouses: Avoid Paying for Unused Clusters

Auto-scaling is powerful but can be expensive if misconfigured. Understand the min/max cluster settings that prevent cost overruns.

The Problem

Default multi-cluster settings can spin up too many clusters, resulting in unnecessary costs. A misconfigured warehouse can scale to 10 clusters when 3 would suffice.

Configuration Parameters

ALTER WAREHOUSE analytics_wh SET
MIN_CLUSTER_COUNT = 1 -- Start with 1 cluster
MAX_CLUSTER_COUNT = 4 -- Scale up to 4 max
SCALING_POLICY = 'STANDARD' -- or 'ECONOMY'
AUTO_SUSPEND = 60;

Scaling Policies

• STANDARD: Aggressively adds clusters (favors performance)
• ECONOMY: Conservatively adds clusters (favors cost) - waits 6 minutes before adding

Choosing Min Clusters

• 1: Best for variable workloads, lowest cost
• 2+: Use only if you have constant baseline load
• Rule: Set MIN = average concurrent query load

Best Practices

1. Start with MIN=1, MAX=3
2. Use ECONOMY policy unless performance is critical
3. Monitor queue times and scale-out events
4. Set AUTO_SUSPEND aggressively (60-120 seconds)
5. Use separate warehouses for different workload types

Real-World Example

A BI team set MIN=3, MAX=10 with STANDARD policy. Clusters stayed at 3-4 most of the day (over-provisioned) and spiked to 10 during peak hours. By changing to MIN=1, MAX=5 with ECONOMY policy, they reduced average running clusters from 3.5 to 1.8, cutting warehouse costs by 48%.

Key Takeaways

• Start with MIN=1, MAX=3-5
• Use ECONOMY policy unless you need instant scale-out
• Monitor actual concurrency vs configured maximums
• Multi-cluster is for concurrency, not query performance