Disclosed weakness · no CVE recorded
vLLM error-message sanitization could stall an API worker
vLLM project · vLLM · validation-error handling
First documented disclosure · Review checkpoint
Summary
An invalid request could consume excessive CPU while its error message was being sanitized.
Severity and exploitation evidence
No CVSS score recorded in this review.
Exploitation not established: The upstream sources document the correction and identify no malicious use.
Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.
How the weakness works
A backtracking regular expression repeatedly scanned long path-like text in a validation error.
Conditions required
A client can send an input that reaches the validation-error sanitizer.
Potential impact
API workers could be delayed and service availability reduced.
Affected versions and configurations
The public report gives no complete release range. The reviewed 0.33.0 source lacks the corrected sanitizer expression.
Fix and mitigation status
0.34.0.
Practical next steps
Upgrade to 0.34.0 or later and retain request limits at the inference service boundary.
Operating lesson
Bound the cost of formatting and sanitizing error responses.
Evidence and limits
The public pull request dates to August 13. Release 0.34.0 lists the fix and its tagged sanitizer contains the corrected expression. A GHSA identifier mentioned in the pull request did not resolve to a public advisory at review, so a CVE or scored severity was not established.
Sources
- vLLM sanitizer issue and merged correction
Primary source ·
- vLLM 0.34.0 release notes
Primary source ·
- vLLM 0.34.0 tagged error sanitizer
Primary source ·