Warehouse

• Multi-cluster WH: auto-scale or maximized.
  • Auto-scale factors: memory capacity and degree of concurrency
  • scaling_policy: standard: minimize queuing, or economy: keep clusters fully loaded
• Each server within a VWH, supports 8 threads, can load 8 files in parallel
• economy auto scaling policy spins up a new cluster if the new cluster can be kept busy for at least 6 minutes
• when spinning down, check is made every minute to see if the load can be redistributed to other clusters
  • for economy, check is made 6 consecutive times, for standard check is made twice
• An open connection doesn't prevent WH from auto-suspending
• Snowpark optimized warehouse offer 16x memory per node
  • charged at 1.5 times
  • provides 16x memory / node
  • BP: minimize concurrency level to ensure max memory available to snowpark workload alter warehouse ... set max_concurrency_level = 1
• QAS offloads large scan that result in relatively low amount of data

Auto-scaling Policies

policy	standard	economy
start	queued query or no spare capacity	keep WH busy for at least 6 min
scale up	multiple if max_cluster_count > 10	one at a time
scale down	multiple if max_cluster_count > 10	multiple if max_cluster_count > 10
shutdown	after 2-3 checks per min	after 5-6 checks per min