创建session修改参数报错

[dakeleblack]

Graphscope默认参数创建session时出现内存不够的异常，修改参数后报如下错，貌似不支持参数修改？

graphscope.set_option(k8s_coordinator_mem="2Gi")
s=graphscope.session()

Initializing graphscope session with parameters: {'addr': None, 'cluster_type': 'k8s', 'num_workers': 2, 'preemptive': True, 'k8s_namespace': None, 'k8s_service_type': 'NodePort', 'k8s_gs_image': 'registry.cn-hongkong.aliyuncs.com/graphscope/graphscope:0.4.0', 'k8s_etcd_image': 'quay.io/coreos/etcd:v3.4.13', 'k8s_image_pull_policy': 'IfNotPresent', 'k8s_image_pull_secrets': [], 'k8s_gie_graph_manager_image': 'registry.cn-hongkong.aliyuncs.com/graphscope/maxgraph_standalone_manager:0.4.0', 'k8s_zookeeper_image': 'zookeeper:3.4.14', 'k8s_coordinator_cpu': 1.0, 'k8s_coordinator_mem': '2Gi', 'k8s_etcd_num_pods': 3, 'k8s_etcd_cpu': 0.5, 'k8s_etcd_mem': '128Mi', 'k8s_zookeeper_cpu': 0.5, 'k8s_zookeeper_mem': '256Mi', 'k8s_gie_graph_manager_cpu': 1.0, 'k8s_gie_graph_manager_mem': '4Gi', 'k8s_vineyard_daemonset': '', 'k8s_vineyard_cpu': 0.5, 'k8s_vineyard_mem': '512Mi', 'vineyard_shared_mem': '4Gi', 'k8s_engine_cpu': 0.5, 'k8s_engine_mem': '4Gi', 'k8s_mars_worker_cpu': 0.5, 'k8s_mars_worker_mem': '4Gi', 'k8s_mars_scheduler_cpu': 0.5, 'k8s_mars_scheduler_mem': '2Gi', 'with_mars': False, 'k8s_volumes': {}, 'k8s_waiting_for_delete': False, 'timeout_seconds': 600, 'dangling_timeout_seconds': 600, 'k8s_client_config': {}}
2021-05-24 10:59:15,535 [INFO][cluster:446]: Launching coordinator...

ApiException Traceback (most recent call last)
/opt/conda/lib/python3.8/site-packages/graphscope/deploy/kubernetes/cluster.py in start(self)
659
--> 660 self._create_services()
661 time.sleep(1)

/opt/conda/lib/python3.8/site-packages/graphscope/deploy/kubernetes/cluster.py in _create_services(self)
583 def _create_services(self):
--> 584 self._create_coordinator()
585

/opt/conda/lib/python3.8/site-packages/graphscope/deploy/kubernetes/cluster.py in _create_coordinator(self)
489 targets.append(
--> 490 self._app_api.create_namespaced_deployment(
491 self._namespace, coordinator_builder.build()

/opt/conda/lib/python3.8/site-packages/kubernetes/client/api/apps_v1_api.py in create_namespaced_deployment(self, namespace, body, **kwargs)
341 kwargs['_return_http_data_only'] = True
--> 342 return self.create_namespaced_deployment_with_http_info(namespace, body, **kwargs) # noqa: E501
343

/opt/conda/lib/python3.8/site-packages/kubernetes/client/api/apps_v1_api.py in create_namespaced_deployment_with_http_info(self, namespace, body, **kwargs)
436
--> 437 return self.api_client.call_api(
438 '/apis/apps/v1/namespaces/{namespace}/deployments', 'POST',

/opt/conda/lib/python3.8/site-packages/kubernetes/client/api_client.py in call_api(self, resource_path, method, path_params, query_params, header_params, body, post_params, files, response_type, auth_settings, async_req, _return_http_data_only, collection_formats, _preload_content, _request_timeout, _host)
347 if not async_req:
--> 348 return self.__call_api(resource_path, method,
349 path_params, query_params, header_params,

/opt/conda/lib/python3.8/site-packages/kubernetes/client/api_client.py in __call_api(self, resource_path, method, path_params, query_params, header_params, body, post_params, files, response_type, auth_settings, _return_http_data_only, collection_formats, _preload_content, _request_timeout, _host)
179 # perform request and return response
--> 180 response_data = self.request(
181 method, url, query_params=query_params, headers=header_params,

/opt/conda/lib/python3.8/site-packages/kubernetes/client/api_client.py in request(self, method, url, query_params, headers, post_params, body, _preload_content, _request_timeout)
390 elif method == "POST":
--> 391 return self.rest_client.POST(url,
392 query_params=query_params,

/opt/conda/lib/python3.8/site-packages/kubernetes/client/rest.py in POST(self, url, headers, query_params, post_params, body, _preload_content, _request_timeout)
273 body=None, _preload_content=True, _request_timeout=None):
--> 274 return self.request("POST", url,
275 headers=headers,

/opt/conda/lib/python3.8/site-packages/kubernetes/client/rest.py in request(self, method, url, query_params, headers, body, post_params, _preload_content, _request_timeout)
232 if not 200 <= r.status <= 299:
--> 233 raise ApiException(http_resp=r)
234

ApiException: (422)
Reason: Unprocessable Entity
HTTP response headers: HTTPHeaderDict({'Audit-Id': 'c17a08d3-9fb5-4377-ba5f-dec41471fa5a', 'Cache-Control': 'no-cache, private', 'Content-Type': 'application/json', 'Date': 'Mon, 24 May 2021 10:59:16 GMT', 'Content-Length': '540'})
HTTP response body: {"kind":"Status","apiVersion":"v1","metadata":{},"status":"Failure","message":"Deployment.apps "coordinator-zgixua" is invalid: spec.template.spec.containers[0].resources.requests: Invalid value: "4Gi": must be less than or equal to memory limit","reason":"Invalid","details":{"name":"coordinator-zgixua","group":"apps","kind":"Deployment","causes":[{"reason":"FieldValueInvalid","message":"Invalid value: "4Gi": must be less than or equal to memory limit","field":"spec.template.spec.containers[0].resources.requests"}]},"code":422}

The above exception was the direct cause of the following exception:

K8sError Traceback (most recent call last)
in
----> 1 s=graphscope.session()

/opt/conda/lib/python3.8/site-packages/graphscope/client/utils.py in wrapper(*args, **kwargs)
154 func.defaults = tuple(new_defaults)
155
--> 156 return_value = func(*args, **kwargs)
157
158 # Restore original defaults.

/opt/conda/lib/python3.8/site-packages/graphscope/client/session.py in init(self, config, cluster_type, addr, num_workers, preemptive, k8s_namespace, k8s_service_type, k8s_gs_image, k8s_etcd_image, k8s_gie_graph_manager_image, k8s_zookeeper_image, k8s_image_pull_policy, k8s_image_pull_secrets, k8s_coordinator_cpu, k8s_coordinator_mem, k8s_etcd_num_pods, k8s_etcd_cpu, k8s_etcd_mem, k8s_zookeeper_cpu, k8s_zookeeper_mem, k8s_gie_graph_manager_cpu, k8s_gie_graph_manager_mem, k8s_vineyard_daemonset, k8s_vineyard_cpu, k8s_vineyard_mem, vineyard_shared_mem, k8s_engine_cpu, k8s_engine_mem, k8s_mars_worker_cpu, k8s_mars_worker_mem, k8s_mars_scheduler_cpu, k8s_mars_scheduler_mem, k8s_volumes, k8s_waiting_for_delete, timeout_seconds, dangling_timeout_seconds, with_mars, **kw)
489 # create and connect session
490 with CaptureKeyboardInterrupt(self.close):
--> 491 self._connect()
492
493 self._disconnected = False

/opt/conda/lib/python3.8/site-packages/graphscope/client/session.py in _connect(self)
826 # launching graphscope service
827 if self._launcher is not None:
--> 828 self._launcher.start()
829 self._coordinator_endpoint = self._launcher.coordinator_endpoint
830

/opt/conda/lib/python3.8/site-packages/graphscope/deploy/kubernetes/cluster.py in start(self)
667 self._dump_coordinator_failed_status()
668 self.stop()
--> 669 raise K8sError(
670 "Error when launching Coordinator on kubernetes cluster"
671 ) from e

K8sError: 'Error when launching Coordinator on kubernetes cluster'

[lidongze0629]

Hi, @bitao666, thank you for posting this question in the discussions, and welcome to the GraphScope discussions.

From your log, the exception resources.requests: Invalid value: "4Gi": must be less than or equal to memory limit'. occurs when resource limits less than requests.

In your current version of GraphScope, the coordinator memory requests defaults to '4Gi'. So your parameters k8s_coordinator_mem="2Gi" cannot pass the API check. You can solve this problem by passing preemptive=False together.

graphscope.set_option(k8s_coordinator_mem="2Gi", preemptive=False)
s=graphscope.session()

From GraphScope Doc

preemptive (bool, optional) – If True, GraphScope instance will treat resource params (e.g. k8s_coordinator_cpu) as limits and provide the minimum available value as requests, but this will make pod has a Burstable QOS, which can be preempted by other pods with high QOS. Otherwise, it will set both requests and limits with the same value.

[dakeleblack]

it works, thanks for your answer