triplewy / art

ART: Automated Resilience Testing for microservices

ART: Automated Resilience Testing for microservices

ART is a framework for automated resilience testing on microservices deployed on top of Istio. Our purpose is to provide meaningful insights of application behavior during failure and improve ease-of-use of microservice fault-tolerance testing. ART accomplishes these goals by identifying critical application components, automatically conducting fault-injection experiments, and presenting the results to the user.

Full Paper

Usage

Build

Run make at the top level directory to build the client binary.

Dependencies:

• python3/pip3
• go 1.14

If make is successful, you can run the client binary with ./bin/client. By default, the client library will connect to a test cluster we already have running so all tracing and fault injection functionality should be already set up.

Shell commands:

• experiment - Runs a full experiment as described from the paper
• continue - Continues an experiment in case of temporary failures
• kiali - Uses kiali instead of sub-graph mining to conduct an experiment

The sections below are for deploying ART on your own application cluster. You can disregard them if you are simply demo-ing ART with our test cluster.

Fault Service

ART requires the fault service be deployed in the application cluster. This allows for the ART client to apply fault injections on the cluster without having the necessary Kuberentes certificates. We have provided yaml files in kubernetes-manifests/ to help you deploy the fault service.

Client

The client is an interactive shell, which you can start with the following command:

$ client --jaeger <jaeger-addr> --fault <fault-addr> --kiali <kiali-addr>

Flags:

• jaeger: Address of jaeger component where you can query traces from
• fault: Address of fault service deployed in the cluster
• kiali: Address of kiali