def create_node(self, node_config, tags, count):
conf = copy.deepcopy(node_config)
pod_spec = conf.get("pod", conf)
service_spec = conf.get("service")
ingress_spec = conf.get("ingress")
node_uuid = str(uuid4())
tags[TAG_RAY_CLUSTER_NAME] = self.cluster_name
tags["ray-node-uuid"] = node_uuid
pod_spec["metadata"]["namespace"] = self.namespace
if "labels" in pod_spec["metadata"]:
pod_spec["metadata"]["labels"].update(tags)
else:
pod_spec["metadata"]["labels"] = tags
# Allow Operator-configured service to access the head node.
if tags[TAG_RAY_NODE_KIND] == NODE_KIND_HEAD:
head_selector = head_service_selector(self.cluster_name)
pod_spec["metadata"]["labels"].update(head_selector)
logger.info(
log_prefix + "calling create_namespaced_pod " "(count={}).".format(count)
)
new_nodes = []
for _ in range(count):
pod = core_api().create_namespaced_pod(self.namespace, pod_spec)
new_nodes.append(pod)
new_svcs = []
if service_spec is not None:
logger.info(
log_prefix + "calling create_namespaced_service "
"(count={}).".format(count)
)
for new_node in new_nodes:
metadata = service_spec.get("metadata", {})
metadata["name"] = new_node.metadata.name
service_spec["metadata"] = metadata
service_spec["spec"]["selector"] = {"ray-node-uuid": node_uuid}
svc = core_api().create_namespaced_service(self.namespace, service_spec)
new_svcs.append(svc)
if ingress_spec is not None:
logger.info(
log_prefix + "calling create_namespaced_ingress "
"(count={}).".format(count)
)
for new_svc in new_svcs:
metadata = ingress_spec.get("metadata", {})
metadata["name"] = new_svc.metadata.name
ingress_spec["metadata"] = metadata
ingress_spec = _add_service_name_to_service_port(
ingress_spec, new_svc.metadata.name
)
networking_api().create_namespaced_ingress(self.namespace, ingress_spec)
def terminate_node(self, node_id):
logger.info(log_prefix + "calling delete_namespaced_pod")
try:
core_api().delete_namespaced_pod(node_id, self.namespace)
except ApiException as e:
if e.status == 404:
logger.warning(log_prefix + f"Tried to delete pod {node_id},"
" but the pod was not found (404).")
else:
raise
try:
core_api().delete_namespaced_service(node_id, self.namespace)
except ApiException:
pass
try:
networking_api().delete_namespaced_ingress(
node_id,
self.namespace,
)
except ApiException:
pass