def _launch_katib_experiment(request, katib_experiment, namespace):
"""Launch Katib experiment."""
try:
katibutils.create_experiment(katib_experiment, namespace)
except ApiException as e:
request.log.exception("Failed to launch Katib experiment")
raise RPCUnhandledError(message="Failed to launch Katib experiment",
details=str(e),
trans_id=request.trans_id)
def _get_k8s_co_client(trans_id):
try:
kubernetes.config.load_incluster_config()
except Exception: # Not in a notebook server
try:
kubernetes.config.load_kube_config()
except Exception:
raise RPCUnhandledError(details="Could not load Kubernetes config",
trans_id=trans_id)
return kubernetes.client.CustomObjectsApi()
def get_experiment(request, experiment, namespace):
"""Get a Katib Experiment.
This RPC is used by the labextension when polling for the state of a
running Experiment.
Args:
request: RPC request object
experiment: Name of the Katib experiment
namespace: Namespace of the experiment
Returns (dict): a dict describing the status of the running experiment
"""
k8s_co_client = k8sutils.get_co_client()
co_group = "kubeflow.org"
co_version = "v1alpha3"
co_plural = "experiments"
try:
exp = k8s_co_client.get_namespaced_custom_object(
co_group, co_version, namespace, co_plural, experiment)
except ApiException as e:
request.log.exception("Failed to get Katib experiment")
raise RPCUnhandledError(message="Failed to get Katib experiment",
details=str(e),
trans_id=request.trans_id)
ret = _construct_experiment_return_base(exp, namespace)
if exp.get("status") is None:
return ret
status, reason, message = _get_experiment_status(exp["status"])
ret.update({
"status": status,
"reason": reason,
"message": message,
"trials": exp["status"].get("trials", 0),
"trialsFailed": exp["status"].get("trialsFailed", 0),
"trialsRunning": exp["status"].get("trialsRunning", 0),
"trialsSucceeded": exp["status"].get("trialsSucceeded", 0),
"currentOptimalTrial": exp["status"].get("currentOptimalTrial")
})
return ret
def _launch_katib_experiment(request, katib_experiment, namespace):
"""Launch Katib experiment."""
k8s_co_client = _get_k8s_co_client(request.trans_id)
co_group = "kubeflow.org"
co_version = "v1alpha3"
co_plural = "experiments"
request.log.debug("Launching Katib Experiment '%s'...",
katib_experiment["metadata"]["name"])
try:
k8s_co_client.create_namespaced_custom_object(co_group, co_version,
namespace, co_plural,
katib_experiment)
except ApiException as e:
request.log.exception("Failed to launch Katib experiment")
raise RPCUnhandledError(message="Failed to launch Katib experiment",
details=str(e), trans_id=request.trans_id)
request.log.info("Successfully launched Katib Experiment")
def get_experiment(request, experiment, namespace):
"""Get a Katib Experiment.
This RPC is used by the labextension when polling for the state of a
running Experiment.
Args:
request: RPC request object
experiment: Name of the Katib experiment
namespace: Namespace of the experiment
Returns (dict): a dict describing the status of the running experiment
"""
k8s_co_client = _get_k8s_co_client(request.trans_id)
co_group = "kubeflow.org"
co_version = "v1alpha3"
co_plural = "experiments"
try:
exp = k8s_co_client.get_namespaced_custom_object(
co_group, co_version, namespace, co_plural, experiment)
except ApiException as e:
request.log.exception("Failed to get Katib experiment")
raise RPCUnhandledError(message="Failed to get Katib experiment",
details=str(e),
trans_id=request.trans_id)
return {
"name": experiment,
"namespace": namespace,
"status": _get_experiment_status(exp["status"]),
"trials": exp["status"].get("trials", 0),
"trialsFailed": exp["status"].get("trialsFailed", 0),
"trialsRunning": exp["status"].get("trialsRunning", 0),
"trialsSucceeded": exp["status"].get("trialsSucceeded", 0),
"maxTrialCount": exp["spec"]["maxTrialCount"],
"currentOptimalTrial": exp["status"].get("currentOptimalTrial")
}