The vLLM Mooncake connector through version 0.29.0 fails to properly release GPU KV cache blocks when concurrent child requests share a single transfer ID in prefill/decode disaggregated deployments. Attackers can exploit this by submitting completion requests with multiple prompts, causing orphaned KV cache blocks to accumulate and exhaust GPU memory. This condition persists until the process is restarted, effectively denying service to legitimate requests.

Successful exploitation results in GPU memory exhaustion, causing denial of service by preventing legitimate requests from executing until the affected process is restarted. There is no indication of privilege escalation, data leakage, or code execution from the provided data.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until a fix is available, consider limiting concurrent child requests sharing a single transfer ID or restarting the process periodically to clear orphaned KV cache blocks.