GPU VulnDB

Database/Container, Kubernetes & orchestration

Envoy: large requests crash the TCP connection pool via a null buffer watermark callback

CVSS 6.6CVE-2025-62409Container, Kubernetes & orchestrationcurated

Impact

A crash in the Envoy proxy process takes down whatever traffic it fronts. On GPU fleets Envoy is commonly the ingress and the service-mesh sidecar in front of inference endpoints, so a dereference of a null buffer watermark callback during connection teardown drops in-flight inference requests and, for sidecar deployments, can be triggered per pod by whoever can send a large request or elicit a large response. The flaw needs no authentication and reaches both TCP proxy and mixed HTTP/1-HTTP/2 ALPN configurations. Availability only - there is no indication of memory disclosure or code execution.

Who can reach it

Anyone who can reach an Envoy listener, unauthenticated, and send a large request or cause a large upstream response while the connection is closing.

What to do

Upgrade to Envoy 1.36.1, 1.35.5, 1.34.9 or 1.33.10. Rolling restart of the proxy fleet; for sidecar mesh deployments this means a rollout that restarts pods, which on GPU workloads should be sequenced with job boundaries rather than forced mid-training.

References

Related entries

All Container, Kubernetes & orchestration entries

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.