Native Gateway API fault injection with Envoy Gateway
Feature request
Chaos Mesh HTTPChaos relies on pod-level traffic interception, which can conflict with service meshes and gateway-managed traffic. #5041 addresses this for Istio by using its native fault-injection API.
Envoy Gateway also supports native delay and abort faults through BackendTrafficPolicy, attached to HTTPRoute or GRPCRoute. A Chaos Mesh integration could manage those policies for the lifetime of an experiment and restore the original state during recovery.
Proposed MVP
- Target one
HTTPRouteorGRPCRoute - Inject a fixed delay or HTTP/gRPC abort
- Configure the affected request percentage
- Reject conflicting ownership
- Restore or remove the managed policy on recovery
- Support workflows and schedules
This would avoid CNI/tproxy dependencies for gateway traffic and would also cover inference stacks built on Gateway API, including KServe and Envoy AI Gateway.
The Gateway API itself does not define fault injection, so the first implementation would be Envoy Gateway-specific. Before working on the API, I would like feedback on whether this should be named GatewayChaos, EnvoyGatewayChaos, or modeled as another provider-specific integration.
Source: chaos-mesh/chaos-mesh