Automatically exclude liveness/readiness probe endpoints

[roobre]

It is common to add exclusions for liveness/readiness endpoints to the test config, as not doing so may cause probes to fail and the pod to be removed from the service load balancer, or even restarted:

    const fault = {
        // ...
        exclude: '/health,/ready',
    };

However, it might not be immediately obvious to application developers why this needs to be done. As http readiness/liveness probes are defined in the pod spec, it should be possible for the disruptor to check for any http probes present in this spec, and automatically feed them to the proxy.

Proposal

We can make a change in the agent so that, by default, bypasses all requests that include an User-Agent header that starts with kube-probe.

As there can be some cases where we do want for probes to be disrupted, we can add an option to opt-out of this behavior:

{
  includeProbes: true, // Defaults to false.
}

[pablochacin]

However, it might not be immediately obvious to application developers why this needs to be done.

Maybe this should be documented better

[roobre]

An alternative solution to this would be for the disruptor to pass-through requests whose User-Agent starts with kube-probe/.

[pablochacin]

We should consider that in some cases there may be a legitimate case of users wanting to disrupt the probes. Therefore, if we make this the default behavior we should add an option to exclude the probes.