We deploy clusters in very heterogeneous environments, not always on a HW under our control and often combined with various public cloud provider IaaS. After a cluster is spawned on such an infrastructure, how can we run basic sanity checks before declaring the setup truly successful and handing it over for consumption?
There are several criterias of a successful setup that can be validated by deploying a special workload. For example:
- cluster/namespace capacity (check by deploying pods with defined
.specs.containers.requests.{cpu,memory}configuration) - network connectivity, (internal, external) incl.
NetworkPolicysetup (check by running client/server connections between pods as per the needs) - DNS resolution tests
The workload requirements would be described in form of an input for a central test controller. Based on these inputs, the controller would run a series of tests, then collect and present the results.