The library for practicing chaos engineering using Vostok.ClusterClient as base http client.

Heavily inspired by Netflix Chaos Monkey and Simmy projects.

This library provide an API to inject the specified additional latency for the http request with specified rate. Using this injection you can observe your system behavior in moments of requests` execution time random increasings. In real world, the reasons for such extra long requests' latencies can be network problems or server's problems.

To configure latency injection use ClusterClientConfiguration's extension methods as follows:

var client = new ClusterClient(new SilentLog(), configuration =>
            {
                configuration.InjectTotalLatency(() => TimeSpan.FromSeconds(1), () => 0.05); //adds extra 1 second latency with 5% probability
            });

Methods parameters are:

latencyProvider - function to get desired Timespan for the injected latency

rateProvider - function to provide desired probability rate of the latency injection. 0.0 - never inject latency, 1.0 - always inject latency for the request.

There are overloads with ILatencyPerformer parameter, so you can implement your own class for the injection decision based on specified rate and impelement your own "sleep" functionality to introduce extra latency for your requests.

There are three ways to configure latency injection using ClusterClient.Chaos:

• Inject total latency
• Inject every call retry latency
• Inject every network call latency

With this configuration the latency is injected at the beginning of the request's pipeline ignoring any retries or different strategies calls.

Usage example:

configuration.InjectTotalLatency(() => TimeSpan.FromSeconds(1), () => 0.05);

Scheme

This configuration injects latency for every request retry. Request retry policy is configured with ClisterClients's RetryPolicy

Usage example:

configuration.InjectLatencyOnEveryRetry(() => TimeSpan.FromSeconds(1), () => 0.05);

Scheme

Using this configuration you can inject latency on every network call, which is implemented by ClusterClient's IRequestSender.SendToReplicaAsync

Usage example:

configuration.InjectLatencyOnEveryNetworkCall(() => TimeSpan.FromSeconds(1), () => 0.05);

Scheme

Along with latency injection, it is possible to inject response transformation into error with provided rate using ClusterClient.Chaos.

var client = new ClusterClient(new SilentLog(), configuration =>
            {
                configuration.InjectCommonServerError(() => 0.05); //inject tranformation of http response into InternalServerError with 5% probability
            });

ClusterClient.Chaos and chaos testing example can be found at this test

Chaos engineering: how to test latency resilience