fix(mongodb): 分片集群部署关闭balancer，调整chunk大小；新增mongos对clb的操作；优化mongos自愈拿机器 …

…#9038

dbm-ui/backend/db_services/mongodb/autofix/mongodb_autofix_ticket.py

@@ -18,6 +18,7 @@
 from backend.configuration.constants import DBType
 from backend.configuration.models.dba import DBAdministrator
 from backend.core import notify
+from backend.db_meta.models import Cluster
 from backend.db_services.dbbase.constants import IpSource
 from backend.db_services.redis.autofix.enums import AutofixStatus
 from backend.db_services.redis.autofix.models import RedisAutofixCore
@@ -28,39 +29,75 @@
 logger = logging.getLogger("root")
 
 
-def get_resource_spec(mongos_list: list, mongod_list: list) -> dict:
-    """获取申请机器规格信息，采用故障机与新机器园区相对应"""
+def mongos_get_resource_spec(cluster_id: int, mongos_list: list) -> dict:
+    """获取申请机器规格信息，mongos申请机器，判断如果一半机器在同一园区，则排除该园区，否则任意园区"""
+
+    include_or_exclue = True
+    sub_zone_ids = []
+    # 获取mongos的机器
+    all_mongos = Cluster.objects.get(id=cluster_id).proxyinstance_set.all()
+    # 获取健康mongos机器在园区的比重 {sub_zone_id: number}
+    health_mongos_number = len(all_mongos) - len(mongos_list)  # 所有健康mongos的数量
+    health_mongos_sub_zone = {}  # 健康mongos在每个园区的数量
+    for mongos in all_mongos:
+        if mongos.machine.ip in [host("ip") for host in mongos_list]:
+            continue
+        if health_mongos_sub_zone.get(str(mongos.machine.bk_sub_zone_id)):
+            health_mongos_sub_zone[str(mongos.machine.bk_sub_zone_id)] = 0
+        health_mongos_sub_zone[str(mongos.machine.bk_sub_zone_id)] += 1
+    #  健康mongos在每个园区的机器数量占所有健康mongos的数量的百分比
+    mongos_sub_zone_percent = {
+        sub_zone: num / health_mongos_number for sub_zone, num in health_mongos_sub_zone.items()
+    }
+    for sub_zone, sub_zone_percent in mongos_sub_zone_percent.items():
+        if sub_zone_percent >= 0.5:
+            include_or_exclue = False
+            sub_zone_ids.append(int(sub_zone))
+            break
 
     resource_spec = {}
-    for host in mongos_list + mongod_list:
+    for host in mongos_list:
         resource_spec.update(
             {
                 host["ip"]: {
                     "spec_id": host["spec_id"],
                     "count": 1,
                     "spec_config": host["spec_config"],
-                    "Location_spec": {"city": host["city"], "sub_zone_ids": [host["bk_sub_zone_id"]]},
+                    "location_spec": {
+                        "city": host["city"],
+                        "sub_zone_ids": sub_zone_ids,
+                        "include_or_exclue": include_or_exclue,
+                    },
                 }
             }
         )
     return resource_spec
 
 
 def mongo_create_ticket(cluster: RedisAutofixCore, cluster_ids: list, mongos_list: list, mongod_list: list):
-    """mongodb自愈创建单据"""
+    """mongodb自愈创建单据 以cluster为维度"""
 
     # 获取dba
     mongodb_dba = DBAdministrator.get_biz_db_type_admins(bk_biz_id=cluster.bk_biz_id, db_type=DBType.MongoDB.value)
 
     # 申请机器规格信息
-    resource_spec = get_resource_spec(mongos_list, mongod_list)
-    if not resource_spec:
-        return
+    resource_spec = {}
+
     # 集群类型
     if mongos_list:
         cluster_type = mongos_list[0]["cluster_type"]
+        # mongos的资源规格
+        mongos_resource_spec = mongos_get_resource_spec(cluster_ids[0], mongos_list)
+        resource_spec.update(mongos_resource_spec)
     if mongod_list:
         cluster_type = mongod_list[0]["cluster_type"]
+        # mongodb的资源规格 TODO
+        # mongod_resource_spec = mongod_get_resource_spec(cluster_ids[0], mongod_list)
+        # mongod_resource_spec.update(mongod_resource_spec)
+
+    if not resource_spec:
+        return
+
     # 单据信息
     details = {
         "ip_source": IpSource.RESOURCE_POOL.value,

dbm-ui/backend/db_services/plugin/nameservice/clb.py

@@ -17,6 +17,7 @@
 from backend.configuration.models import DBAdministrator
 from backend.db_meta import api
 from backend.db_meta.enums import ClusterEntryType
+from backend.db_meta.enums.cluster_type import ClusterType
 from backend.db_meta.models import Cluster, ClusterEntry
 from backend.env import CLB_DOMAIN
 from backend.flow.utils import dns_manage
@@ -132,6 +133,12 @@ def create_lb_and_register_target(cluster_id: int) -> Dict[str, Any]:
     cluster = get_cluster_info(cluster_id=cluster_id)
     immute_domain = cluster["immute_domain"]
 
+    # 获取DB类型
+    if cluster["cluster_type"] == ClusterType.MongoShardedCluster.value:
+        db_type = DBType.MongoDB.value
+    else:
+        db_type = DBType.Redis.value
+
     # 判断clb是否已经存在
     if ClusterEntryType.CLB.value in cluster["clusterentry_set"]:
         message = "clb of cluster:{} has existed".format(immute_domain)
@@ -141,7 +148,7 @@ def create_lb_and_register_target(cluster_id: int) -> Dict[str, Any]:
     bk_biz_id = cluster["bk_biz_id"]
 
     # 通过bk_biz_id获取manager，backupmanager，去除admin
-    users = DBAdministrator().get_biz_db_type_admins(bk_biz_id=bk_biz_id, db_type=DBType.Redis)
+    users = DBAdministrator().get_biz_db_type_admins(bk_biz_id=bk_biz_id, db_type=db_type)
     users = [user for user in users if user != "admin"]
     manager = users[0]
     backupmanager = users[1] if len(users) > 1 else users[0]
@@ -391,7 +398,7 @@ def operate_part_target(cluster_id: int, ips: list, bind: bool) -> dict:
 
     # 获取信息
     cluster = get_cluster_info(cluster_id=cluster_id)
-    clb = cluster["clusterentry_set"]["clb"]
+    clb = cluster["clusterentry_set"]["clb"][0]
     clb_id = clb["clb_id"]
     listener_id = clb["listener_id"]
     clb_region = clb["clb_region"]

dbm-ui/backend/flow/engine/bamboo/scene/mongodb/mongodb_install.py

@@ -219,6 +219,15 @@ def cluster_install_flow(self):
             act_component_code=ExecAddRelationshipOperationComponent.code,
             kwargs=kwargs,
         )
+
+        # cluster初始配置
+        kwargs = self.get_kwargs.get_mongodb_cluster_init_kwargs()
+        pipeline.add_act(
+            act_name=_("MongoDB--cluster初始化设置"),
+            act_component_code=ExecuteDBActuatorJobComponent.code,
+            kwargs=kwargs,
+        )
+
         # 域名写入dns
         kwargs = self.get_kwargs.get_add_domain_to_dns_kwargs(cluster=True)
         pipeline.add_act(

dbm-ui/backend/flow/engine/bamboo/scene/mongodb/sub_task/cluster_monogs_autofix.py

@@ -14,8 +14,6 @@
 
 from django.utils.translation import ugettext as _
 
-from backend.db_meta import api
-from backend.db_meta.enums import ClusterEntryType
 from backend.db_meta.enums.cluster_type import ClusterType
 from backend.flow.consts import MongoDBInstanceType, MongoDBManagerUser
 from backend.flow.engine.bamboo.scene.common.builder import SubBuilder
@@ -34,9 +32,10 @@
 )
 from backend.flow.plugins.components.collections.mongodb.mongodb_cmr_4_meta import CMRMongoDBMetaComponent
 from backend.flow.plugins.components.collections.mongodb.send_media import ExecSendMediaOperationComponent
-from backend.flow.plugins.components.collections.name_service.name_service import ExecNameServiceOperationComponent
 from backend.flow.utils.mongodb.mongodb_dataclass import ActKwargs
 
+from .mongos_replace import cluster_clb, mongos_operate_clb
+
 
 def mongos_autofix(root_id: str, ticket_data: Optional[Dict], sub_sub_kwargs: ActKwargs, info: dict) -> SubBuilder:
     """
@@ -109,12 +108,9 @@ def mongos_autofix(root_id: str, ticket_data: Optional[Dict], sub_sub_kwargs: Ac
     sub_sub_get_kwargs.payload["set_id"] = sub_sub_get_kwargs.db_instance["cluster_name"]
 
     # 判断是否有clb
-    clb = False
-    if (
-        ClusterEntryType.CLB.value
-        in api.cluster.nosqlcomm.other.get_cluster_detail(cluster_id=cluster_id)[0]["clusterentry_set"]
-    ):
-        clb = True
+    clb = cluster_clb(cluster_id=cluster_id)
+    creator = sub_sub_get_kwargs.payload.get("created_by")
+
     # dbha已做clb解绑，dns删除
     # # 删除clb中绑定的老ip
     # if clb:
@@ -183,16 +179,12 @@ def mongos_autofix(root_id: str, ticket_data: Optional[Dict], sub_sub_kwargs: Ac
 
     # clb绑定新ip
     if clb:
-        kwargs = {
-            "name_service_operation_type": "clb_register_part_target",
-            "creator": sub_sub_get_kwargs.payload["created_by"],
-            "cluster_id": cluster_id,
-            "ips": ["{}:{}".format(node["ip"], str(sub_sub_get_kwargs.db_instance["port"]))],
-        }
-        sub_sub_pipeline.add_act(
-            act_name=_("MongoDB-clb绑定新ip"),
-            act_component_code=ExecNameServiceOperationComponent.code,
-            kwargs=kwargs,
+        mongos_operate_clb(
+            cluster_id=cluster_id,
+            creator=creator,
+            ips=["{}:{}".format(node["ip"], str(sub_sub_get_kwargs.db_instance["port"]))],
+            bind=True,
+            pipeline=sub_sub_pipeline,
         )
 
     # 添加新的dns

dbm-ui/backend/flow/engine/bamboo/scene/mongodb/sub_task/deinstall.py

@@ -17,6 +17,7 @@
 from backend.db_meta.enums.cluster_type import ClusterType
 from backend.flow.consts import MongoDBInstanceType
 from backend.flow.engine.bamboo.scene.common.builder import SubBuilder
+from backend.flow.engine.bamboo.scene.name_service.name_service import NameServiceFlow
 from backend.flow.plugins.components.collections.mongodb.delete_domain_from_dns import (
     ExecDeleteDomainFromDnsOperationComponent,
 )
@@ -27,6 +28,8 @@
 from backend.flow.plugins.components.collections.mongodb.mongo_shutdown_meta import MongosShutdownMetaComponent
 from backend.flow.utils.mongodb.mongodb_dataclass import ActKwargs
 
+from .mongos_replace import cluster_clb
+
 
 def mongo_deinstall_parallel(sub_get_kwargs: ActKwargs, nodes: list, instance_type: str, force: bool) -> list:
     acts_list = []
@@ -107,6 +110,17 @@ def deinstall(
         )
         if acts_list:
             sub_pipeline.add_parallel_acts(acts_list=acts_list)
+        if not reduce_mongos:
+            # 判断是否有clb
+            clb = cluster_clb(cluster_id=cluster_id)
+            creator = sub_get_kwargs.payload.get("created_by")
+            if clb:
+                kwargs = {
+                    "bk_biz_id": sub_get_kwargs.payload["bk_biz_id"],
+                    "created_by": creator,
+                    "cluster_id": cluster_id,
+                }
+                NameServiceFlow(root_id=root_id, data=kwargs).clb_delete(pipeline=sub_pipeline)
 
     # 删除dns
     kwargs = sub_get_kwargs.get_delete_domain_kwargs()

dbm-ui/backend/flow/engine/bamboo/scene/mongodb/sub_task/increase_mongos.py

@@ -24,6 +24,7 @@
 from backend.flow.utils.mongodb.mongodb_dataclass import ActKwargs
 
 from .mongos_install import mongos_install
+from .mongos_replace import cluster_clb, mongos_operate_clb
 
 
 def increase_mongos(root_id: str, ticket_data: Optional[Dict], sub_kwargs: ActKwargs, info: dict) -> SubBuilder:
@@ -66,8 +67,9 @@ def increase_mongos(root_id: str, ticket_data: Optional[Dict], sub_kwargs: ActKw
     )
 
     # 获取信息
-    sub_get_kwargs.get_cluster_info_deinstall(cluster_id=info["cluster_id"])
-    sub_get_kwargs.payload["cluster_id"] = info["cluster_id"]
+    cluster_id = info["cluster_id"]
+    sub_get_kwargs.get_cluster_info_deinstall(cluster_id=cluster_id)
+    sub_get_kwargs.payload["cluster_id"] = cluster_id
     sub_get_kwargs.cluster_type = sub_get_kwargs.payload["cluster_type"]
     sub_get_kwargs.mongos_info = info
     sub_get_kwargs.mongos_info["nodes"] = info["mongos"]
@@ -85,6 +87,9 @@ def increase_mongos(root_id: str, ticket_data: Optional[Dict], sub_kwargs: ActKw
     sub_get_kwargs.payload["mongos"]["nodes"] = info["mongos"]
     sub_get_kwargs.payload["mongos"]["domain"] = sub_get_kwargs.payload["mongos_nodes"][0]["domain"]
     sub_get_kwargs.payload["mongos"]["port"] = sub_get_kwargs.payload["mongos_nodes"][0]["port"]
+    # 判断是否有clb
+    clb = cluster_clb(cluster_id=cluster_id)
+    creator = sub_get_kwargs.payload["created_by"]
 
     # 进行mongos安装——子流程
     sub_sub_pipeline = mongos_install(
@@ -95,6 +100,16 @@ def increase_mongos(root_id: str, ticket_data: Optional[Dict], sub_kwargs: ActKw
     )
     sub_pipeline.add_sub_pipeline(sub_sub_pipeline)
 
+    # 实例绑定clb
+    if clb:
+        mongos_operate_clb(
+            cluster_id=cluster_id,
+            creator=creator,
+            ips=["{}:{}".format(node["ip"], str(sub_get_kwargs.mongos_info["port"])) for node in info["mongos"]],
+            bind=True,
+            pipeline=sub_pipeline,
+        )
+
     # dns新增实例
     kwargs = sub_get_kwargs.get_add_domain_to_dns_kwargs(cluster=True)
     sub_pipeline.add_act(

dbm-ui/backend/flow/engine/bamboo/scene/mongodb/sub_task/mongos_replace.py

@@ -32,6 +32,41 @@
 from backend.flow.plugins.components.collections.mongodb.exec_actuator_job import ExecuteDBActuatorJobComponent
 from backend.flow.plugins.components.collections.name_service.name_service import ExecNameServiceOperationComponent
 from backend.flow.utils.mongodb.mongodb_dataclass import ActKwargs
+from backend.flow.utils.name_service.name_service_dataclass import TransDataKwargs
+
+
+def cluster_clb(cluster_id: int) -> bool:
+    """判断分片集群是否有clb"""
+
+    if (
+        ClusterEntryType.CLB.value
+        in api.cluster.nosqlcomm.other.get_cluster_detail(cluster_id=cluster_id)[0]["clusterentry_set"]
+    ):
+        return True
+    return False
+
+
+def mongos_operate_clb(cluster_id: int, creator: str, ips: list, bind: bool, pipeline: SubBuilder):
+    """mongos操作clb"""
+
+    if bind:
+        name_service_operation_type = "clb_register_part_target"
+        act_name = _("MongoDB-clb绑定新ip")
+    else:
+        name_service_operation_type = "clb_deregister_part_target"
+        act_name = _("MongoDB-clb解绑老ip")
+    kwargs = {
+        "set_trans_data_dataclass": TransDataKwargs.__name__,
+        "name_service_operation_type": name_service_operation_type,
+        "creator": creator,
+        "cluster_id": cluster_id,
+        "ips": ips,
+    }
+    pipeline.add_act(
+        act_name=_(act_name),
+        act_component_code=ExecNameServiceOperationComponent.code,
+        kwargs=kwargs,
+    )
 
 
 def mongos_replace(root_id: str, ticket_data: Optional[Dict], sub_sub_kwargs: ActKwargs, info: dict) -> SubBuilder:
@@ -92,25 +127,17 @@ def mongos_replace(root_id: str, ticket_data: Optional[Dict], sub_sub_kwargs: Ac
     sub_sub_get_kwargs.payload["bk_cloud_id"] = info["bk_cloud_id"]
     sub_sub_get_kwargs.payload["set_id"] = sub_sub_get_kwargs.db_instance["cluster_name"]
     # 判断是否有clb
-    clb = False
-    if (
-        ClusterEntryType.CLB.value
-        in api.cluster.nosqlcomm.other.get_cluster_detail(cluster_id=cluster_id)[0]["clusterentry_set"]
-    ):
-        clb = True
+    clb = cluster_clb(cluster_id=cluster_id)
+    creator = sub_sub_get_kwargs.payload["created_by"]
 
     # clb解绑老ip
     if clb:
-        kwargs = {
-            "name_service_operation_type": "clb_deregister_part_target",
-            "creator": sub_sub_get_kwargs.payload["creator"],
-            "cluster_id": cluster_id,
-            "ips": ["{}:{}".format(info["ip"], str(sub_sub_get_kwargs.db_instance["port"]))],
-        }
-        sub_sub_pipeline.add_act(
-            act_name=_("MongoDB-clb解绑替换ip"),
-            act_component_code=ExecNameServiceOperationComponent.code,
-            kwargs=kwargs,
+        mongos_operate_clb(
+            cluster_id=cluster_id,
+            creator=creator,
+            ips=["{}:{}".format(info["ip"], str(sub_sub_get_kwargs.db_instance["port"]))],
+            bind=False,
+            pipeline=sub_sub_pipeline,
         )
 
     # 删除老的dns
@@ -131,16 +158,12 @@ def mongos_replace(root_id: str, ticket_data: Optional[Dict], sub_sub_kwargs: Ac
 
     # clb绑定新ip
     if clb:
-        kwargs = {
-            "name_service_operation_type": "clb_register_part_target",
-            "creator": sub_sub_get_kwargs.payload["creator"],
-            "cluster_id": cluster_id,
-            "ips": ["{}:{}".format(node["ip"], str(sub_sub_get_kwargs.db_instance["port"]))],
-        }
-        sub_sub_pipeline.add_act(
-            act_name=_("MongoDB-clb绑定新ip"),
-            act_component_code=ExecNameServiceOperationComponent.code,
-            kwargs=kwargs,
+        mongos_operate_clb(
+            cluster_id=cluster_id,
+            creator=creator,
+            ips=["{}:{}".format(node["ip"], str(sub_sub_get_kwargs.db_instance["port"]))],
+            bind=True,
+            pipeline=sub_sub_pipeline,
         )
 
     # 添加新的dns