Sharing Finite Resources
A warehouse serves many queries from dashboards, reports, and loads at once. Workload management decides which query runs now, how much memory and CPU it gets, and what waits. Without it a single giant scan can starve every fast interactive query.
Common Controls
- Queues: route queries into groups, such as a fast lane for dashboards and a heavy lane for batch reports.
- Concurrency limits: cap how many run at once so each gets enough memory.
- Priorities: let urgent queries jump ahead of background jobs.
- Timeouts: kill runaway queries before they hog resources.
Why It Matters
Good workload management trades raw throughput for predictable latency on the queries that users feel. It protects interactive work from large analytical jobs sharing the same cluster.
Key idea
Workload management uses queues, concurrency limits, priorities, and timeouts to keep fast interactive queries responsive even when heavy analytical jobs share the same warehouse.