Database/AI/ML frameworks & serving
vLLM: one request with oversized min_tokens kills the EngineCore input thread while /health stays 200
Impact
vLLM before 0.31.0 skips the min_tokens <= max_tokens check when max_tokens is null on /inference/v1/generate, so an uncaught VLLMValidationError later kills the EngineCore input thread. Every subsequent generation request hangs, and the operational sting is that /health keeps returning 200 - load balancers and Kubernetes probes go on routing traffic to a replica that will never answer, so the outage looks like a mystery latency cliff rather than a dead pod. Recovery requires a manual restart; the GPUs stay allocated and idle in the meantime.
Who can reach it
A single request to /inference/v1/generate. The advisory describes unauthenticated remote attackers; the CVSS v4 vector records PR:L, so treat any client that can reach the endpoint as sufficient.
What to do
Upgrade vLLM to 0.31.0 or later and restart the serving processes. If you cannot upgrade now, reject requests with a null max_tokens or an out-of-range min_tokens at the gateway, and stop trusting /health alone - add a liveness probe that actually completes a short generation, so a wedged replica gets cycled instead of being fed traffic.
References
Related entries
- vLLM: caller-controlled EXIF ImageID and cache_salt collide cache keys across requestsCVE-2026-114054 · vLLM (multimodal media cache and prefix KV cache key derivation)High
- Flexible GraphRAG: Cypher injection through ingested document content on the FalkorDB backendCVE-2026-114510 · Flexible GraphRAG FalkorDBAdapter (_escape / add_graph_documents Cypher assembly)High
- vLLM: audio decompression bomb via input_audio or audio_url exhausts serving node memoryCVE-2026-116175 · vLLM (AudioMediaIO load_audio, missing max_duration_s limit)High
- vLLM: out-of-vocabulary stop_token_ids crash the engine and write past the V2 runner logits bufferCVE-2026-116181 · vLLM (OpenAI-compatible API, stop_token_ids validation)High
- NVIDIA TensorRT-LLM: server-side request forgery lets a local user make the serving process issue arbitrary requestsCVE-2026-24269 · NVIDIA TensorRT-LLM (server-side request forgery in the serving component)High
- vLLM (`MediaConnector`): SSRF, recurrence of CVE-2025-6242CVE-2026-24779 · vLLM (`MediaConnector`)High
This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.