multirunners

var Command = cli.Command{
	Name:      "multirunners",
	ShortName: "mrunners",
	Usage:     "packages runner as job and deploy runners into kubernetes",
	Subcommands: []cli.Command{
		runCommand,
		waitCommand,
		resultCommand,
	},
}

Command represents multirunners sub-command.

Subcommand multirunners is to deploy multiple runners as kubernetes jobs. Since one runner could run out of networking bandwidth on one host, the multirunners deploys runners on different hosts and reduces the impact of limited networking resource.

Command line interface:

kperf mrunners run --help

Options:

--kubeconfig   PATH     (default: empty_string, use token if it's empty)
--namespace    STRING   (default: empty_string, required)
--runner-image STRING   (default: empty_string, required)
--runners      []STRING (default: empty, required)
--wait         BOOLEAN  (default: false)

Details:

The --runners format is defined by URI.

- file:///abs_path?numbers=10 - configmap:///namespace/name?numbers=2 - ...

The schema:://PATH is used to get runner's configuration. It can be local path or stored as configmap in target kubernetes cluster. The query part is to define what the job looks like. Currently, that command just requires the number of pods in that job. At the beginning, we just need to file://. The number of runners defines the number of jobs.

All the jobs are referenced by one configmap (ownerReference). The configmap name will be output to stdout. The name is progress tracker ID. By default, there is only one progress tracker ID in one namespace.

kperf mrunners wait --help

Args:

0: namespace (STRING)

Options:

--kubeconfig   PATH     (default: empty_string, use token if it's empty)

Wait it to wait until jobs finish.

kperf mrunners result --help

Args:

0: namespace (STRING)

Options:

--kubeconfig   PATH     (default: empty_string, use token if it's empty)

Result retrieves the result for jobs. If jobs is still running, that command will fail.