The admission control system in CockroachDB prioritizes work during different types of node overload to ensure cluster stability and minimize the impact on critical operations. When individual nodes detect excessive workload, the system queues work into different priority levels, ensuring that critical operations continue uninterrupted while lower-priority work is throttled. The system actively manages three types of overload: CPU, Storage, and Replication, each using indicators such as runnable goroutines per CPU or storage health to determine when overload occurs. By dynamically queuing work based on priority and limiting resource availability, the admission control system minimizes the risk of overload while preserving normal throughput and latency for foreground workload. This system provides temporary protection during node overload, ensuring critical cluster operations continue while other work is properly paced, and minimizing the impact of lower-priority work such as backups and changefeeds when a node is under heavy load.