def query_cluster_nodes(ctx_cluster: CtxCluster, exclude_master: bool = True) -> Dict:
"""查询节点数据
包含标签、污点、状态等供前端展示数据
"""
# 获取集群中的节点列表
# NOTE: 现阶段会有两个agent，新版agent上报集群信息到bcs api中，可能会有时延，导致bcs api侧找不到集群信息；处理方式:
# 1. 初始化流程调整，创建集群时，注册一次集群信息
# 2. 应用侧，兼容处理异常
try:
cluster_node_list = Node(ctx_cluster).list(is_format=False)
except Exception as e: # 兼容处理现阶段kube-agent没有注册时，连接不上集群的异常
logger.error("query cluster nodes error, %s", e)
# 查询集群内节点异常，返回空字典
return {}
nodes = {}
for node in cluster_node_list.items:
labels = node.labels
# 现阶段节点页面展示及操作，需要排除master
if exclude_master and node.is_master():
continue
# 使用inner_ip作为key，主要是方便匹配及获取值
nodes[node.inner_ip] = {
"inner_ip": node.inner_ip,
"name": node.name,
"status": node.node_status,
"labels": labels,
"taints": node.taints,
"unschedulable": node.data.spec.unschedulable or False,
}
return nodes
def create_and_delete_master(ctx_cluster):
client = Node(ctx_cluster)
client.update_or_create(
body={
"apiVersion": "v1",
"kind": "Node",
"metadata": {"name": fake_node_name, "labels": {"node-role.kubernetes.io/master": "true"}},
"spec": {},
"status": {
"addresses": [
{"address": fake_inner_ip, "type": "InternalIP"},
],
"conditions": [
{
"lastHeartbeatTime": "2021-10-25T04:13:48Z",
"lastTransitionTime": "2020-10-25T05:24:53Z",
"message": "kubelet is posting ready status",
"reason": "KubeletReady",
"status": "True",
"type": "Ready",
}
],
},
},
name=fake_node_name,
)
yield
client.delete_wait_finished(fake_node_name)
def query_cluster_nodes(ctx_cluster: CtxCluster,
exclude_master: bool = True) -> Dict:
"""查询节点数据
包含标签、污点、状态等供前端展示数据
"""
# 获取集群中的节点列表
node_client = Node(ctx_cluster)
try:
cluster_node_list = node_client.list(is_format=False)
except ApiException:
# 查询集群内节点异常，返回空字典
return {}
nodes = {}
for node in cluster_node_list.items:
labels = node.labels
# 现阶段节点页面展示及操作，需要排除master
if exclude_master and labels.get(
node_constants.K8S_NODE_ROLE_MASTER) == "true":
continue
# 使用inner_ip作为key，主要是方便匹配及获取值
nodes[node.inner_ip] = {
"inner_ip": node.inner_ip,
"name": node.name,
"status": node.node_status,
"labels": labels,
"taints": node.taints,
"unschedulable": node.data.spec.unschedulable or False,
}
return nodes
def query_cluster_nodes(ctx_cluster: CtxCluster) -> Dict[str, Dict]:
# 查询集群下node信息
client = Node(ctx_cluster)
nodes = client.list()
# 根据传入的inner_ip过滤节点信息
data = {}
for node in nodes:
node_data = node.data
# 解析数据用于前端展示
metadata = node_data.get("metadata", {})
labels = metadata.get("labels", {})
# 过滤掉master
if labels.get("node-role.kubernetes.io/master") == "true":
continue
taints = getitems(node_data, ["spec", "taints"], [])
# 组装数据，用于展示
data[node.inner_ip] = {
"inner_ip":
node.inner_ip,
"name":
node.name,
"labels":
labels,
"taints":
taints,
"status":
get_node_status(getitems(node_data, ["status", "conditions"], [])),
"unschedulable":
getitems(node_data, ["spec", "unschedulable"], False),
}
return data
def query_taints(self, request, project_id, cluster_id):
"""查询node的污点"""
params = self.params_validate(slz.QueryNodeListSLZ)
node_client = Node(request.ctx_cluster)
return Response(
node_client.filter_nodes_field_data("taints",
params["node_name_list"]))
def set_taints(self, request, project_id, cluster_id):
"""设置污点"""
params = self.params_validate(slz.NodeTaintListSLZ)
node_client = Node(request.ctx_cluster)
node_client.set_taints_for_multi_nodes(params["node_taint_list"])
# 获取节点名称，用于审计
node_names = get_nodes_repr([n["node_name"] for n in params["node_taint_list"]])
request.audit_ctx.update_fields(resource=node_names, extra=params, description=_("节点设置污点"))
return Response()
def _get_cluster_masters(self) -> List[Dict]:
"""查询集群中的master ip和name"""
node_client = Node(self.ctx_cluster)
# NOTE: 返回节点出现异常，直接报错
cluster_nodes = node_client.list(is_format=False)
# 过滤 master 信息
masters = []
for node in cluster_nodes.items:
if not node.is_master():
return
masters.append({"inner_ip": node.inner_ip, "host_name": node.name})
return masters
def set_labels(ctx_cluster: CtxCluster, label_list: List):
"""节点设置标签
ctx_cluster: 集群模型数据
taint_list: 节点的污点内容，格式: [{"node_name": "demo", "labels": {"key1": "value1", "key2": "value2"}]
"""
client = Node(ctx_cluster)
# 下发的body格式: {"metadata": {"labels": {"demo": "demo"}}}
tasks = [
functools.partial(client.patch, {"metadata": {
"labels": l["labels"]
}}, l["node_name"]) for l in label_list
]
# 当有操作失败的，抛出异常
async_run(tasks)
def set_taints(ctx_cluster: CtxCluster, taint_list: List):
"""节点设置污点，因为可能有多个节点分别调用接口完成打污点，使用asyncio处理，减少耗时
ctx_cluster: 集群模型数据
taint_list: 节点的污点内容，格式: [{"node_name": "demo", "taints": [{"key": xxx, "value": xxx, "effect": xxx}]]
"""
client = Node(ctx_cluster)
# 下发的body格式: {"spec": {"taints": [{"key": xxx, "value": xxx, "effect": xxx}]}}
tasks = [
functools.partial(client.patch, {"spec": {
"taints": t["taints"]
}}, t["node_name"]) for t in taint_list
]
# 当有操作失败的，抛出异常
async_run(tasks)
def set_schedule_status(self, request, project_id, cluster_id):
"""设置节点调度状态
通过传递状态, 设置节点的调度状态
"""
params = self.params_validate(slz.NodeStatusSLZ)
# NOTE: 如果状态为REMOVABLE，则期望的是停止调度状态, 以便于进一步操作
unschedulable = True if params["status"] == node_status.REMOVABLE else False
client = Node(request.ctx_cluster)
client.set_nodes_schedule_status(unschedulable, params["node_name_list"])
request.audit_ctx.update_fields(
resource=get_nodes_repr(params["node_name_list"]),
extra=params,
description=_("节点停止调度") if unschedulable else _("节点允许调度"),
)
return Response()
def get_cluster_nodes(access_token, project_id, cluster_id):
"""获取集群下的node信息
NOTE: 节点数据通过集群中获取，避免数据不一致
"""
ctx_cluster = CtxCluster.create(
id=cluster_id,
project_id=project_id,
token=access_token,
)
try:
cluster_nodes = Node(ctx_cluster).list(is_format=False)
except Exception as e:
logger.error("查询集群内节点数据异常, %s", e)
return []
return [{
"inner_ip": node.inner_ip,
"status": node.node_status
} for node in cluster_nodes.items]
def detail(self) -> Dict[str, Any]:
# 通过集群获取
client = Node(self.ctx_cluster)
node_data = client.get(self.name).get("data", {})
# 过滤需要的信息: 包含节点名称、IP、版本、OS、运行时、镜像、标签、污点、pods
return self._detail(node_data)
def client(self, ctx_cluster):
return Node(ctx_cluster)
def __init__(self, ctx_cluster: CtxCluster):
self.node_client = Node(ctx_cluster)
def set_taints(self, request, project_id, cluster_id):
"""设置污点"""
params = self.params_validate(slz.NodeTaintListSLZ)
node_client = Node(request.ctx_cluster)
node_client.set_taints_for_multi_nodes(params["node_taint_list"])
return Response()
def set_labels(self, request, project_id, cluster_id):
"""设置节点标签"""
params = self.params_validate(slz.NodeLabelListSLZ)
node_client = Node(request.ctx_cluster)
node_client.set_labels_for_multi_nodes(params["node_label_list"])
return Response()