CVE-2026-94627: vLLM through 0.29.0 GPU KV Cache Leak via Mooncake Transfer ID Collision
vLLM Mooncake connector through 0.29.0 fails to properly manage GPU KV cache block ownership when concurrent child requests share a single transfer ID in prefill/decode disaggregated deployments. Attackers can trigger GPU memory exhaustion by submitting completion requests with multiple prompts, causing orphaned KV cache blocks to accumulate until process restart and eventually preventing legitimate requests from executing.
Affected Software
Event History
Frequently Asked Questions
Which deployments are exposed to this issue?
The issue affects vLLM through 0.29.0 when using the Mooncake connector in a prefill/decode disaggregated deployment. The described failure requires concurrent child requests to share one transfer ID.
What does an attacker need to do to cause impact?
An unauthenticated attacker can submit completion requests containing multiple prompts. This can create orphaned GPU KV cache blocks through transfer ID collisions, accumulating GPU memory usage over time.
What is the operational impact?
Accumulated orphaned KV cache blocks can exhaust GPU memory and prevent legitimate requests from executing. Recovery may require restarting the affected process.