CVE-2026-94625: vLLM through 0.29.0 Resource Exhaustion via Ownerless Mooncake Transfer Placeholders
vLLM through 0.29.0 contains a resource exhaustion vulnerability in MooncakeConnector where rejected prefill requests create ownerless transfer placeholders that are never reclaimed. Attackers can send rejected requests to exhaust sender task pools, causing valid requests to be delayed by up to 480 seconds while health checks continue returning success.
Affected Software
Event History
Frequently Asked Questions
What access does an attacker need to trigger the issue?
The issue is network-reachable and requires neither privileges nor user interaction. An attacker can trigger it by sending rejected prefill requests.
Which deployments should be investigated?
Investigate vLLM deployments through version 0.29.0 that use the MooncakeConnector component. The affected behavior involves its handling of rejected prefill requests.
How can operators recognize active impact?
Valid requests may be delayed for up to 480 seconds after sender task pools are exhausted. Health checks can continue to report success during the resulting service degradation.