[Action] Load generation / k6 evalution and discussion

[matthewi]

Description

Continuing the discussion from #182 (comment) and this week's wg meeting.

K6 provides a way generate and manage a distributed network load for testing applications. It does not provide a mechanism for running lower level stress tests.

High level questions:

• Is K6 the right tool for network loads for the project?
• Do we want to use a separate tool for managing lower level stress tests for load generation?
• Do we need both cases to be handled by 1 tool? Does such a tool exist?

K6 open questions:

• What method do we use to operate the runners?
  • the SaaS cloud is easy but long term we want to limit external clouds and leverage community resources when possible.
  • k6-operator: this was recommended as the simplest way to get started.
    • Con: since the load is coming from within the cluster it will mean we can't run end to end network tests with projects such as Istio.
    • We should prevent the runners from scheduling on the node(s) used by Falco (or other project(s) being tested)
• technical requirements for the load test itself - number of VUs, duration, failure/success thresholds, etc. (This generally applies to any network load generator)

K6 Resources:

• Sample application and test scripts
• Prometheus stats output
• k6-operator

...

Milestones

...

Input

...

Outcome

...

Todos

...

[leonardpahlke]

@nikimanoledaki @AntonioDiTuri added the issue/needs-triage label.

[nikimanoledaki]

We will not be using k6 as part of the benchmarking in WG Green Reviews after all so we can close this issue and re-open it if this changes. @matthewi it would be great to have your input on the new proposals being drafted regarding this! cncf-tags/green-reviews-tooling#83