How do you stop one user's runaway agent from consuming all capacity?
Why A is the answer
Per-user rate and budget limits
Limits protect shared capacity from loops.
In detail
- Per-user limits protect shared capacity.
- Limits protect other users.
Use case
An illustrative scenario, not a real incident.
One looping agent would have consumed all capacity, but a budget limit stops it.
See it in different ways
sequenceDiagram participant U as Users participant L as Rate and budget limits participant A as Agent replicas participant S as External state participant M as GPU inference U->>L: Requests L->>A: Within limits A->>S: Reads and saves state A->>M: Model calls, in parallel if independent
flowchart TD
Q{"Is the user within budget?"}
Q -->|"Yes"| A0["Continue"]
Q -->|"No"| A1["Stop and report"]Key point: Per-user limits protect capacity.
Why the other options are wrong
- B. A larger context window for that user
- That lets it use more.
- C. A longer idle timeout on the load balancer
- That keeps it running longer.
- D. A higher temperature on its model calls
- Temperature is unrelated.