feature: 后台组件服务支持配置超时时间(closed TencentBlueKing#1696)

apps/backend/components/collections/agent_new/configure_policy.py

@@ -11,7 +11,7 @@
 
 from django.utils.translation import ugettext_lazy as _
 
-from apps.backend.api.constants import POLLING_INTERVAL, POLLING_TIMEOUT
+from apps.backend.api.constants import POLLING_INTERVAL
 from apps.backend.components.collections.agent_new.base import AgentBaseService
 from apps.node_man import models
 from apps.node_man.handlers.security_group import get_security_group_factory
@@ -56,7 +56,7 @@ def _schedule(self, data, parent_data, callback_data=None):
             self.finish_schedule()
             return True
 
-        elif polling_time + POLLING_INTERVAL > POLLING_TIMEOUT / 2:
+        elif polling_time + POLLING_INTERVAL > self.service_polling_timeout / 2:
             self.move_insts_to_failed(subscription_instance_ids, _("配置到Gse和Nginx的策略失败请联系节点管理维护人员"))
             self.finish_schedule()
             return False

apps/backend/components/collections/agent_new/get_agent_status.py

@@ -13,7 +13,7 @@
 
 from django.utils.translation import ugettext_lazy as _
 
-from apps.backend.api.constants import POLLING_INTERVAL, POLLING_TIMEOUT
+from apps.backend.api.constants import POLLING_INTERVAL
 from apps.node_man import constants, models
 from pipeline.core.flow import Service, StaticIntervalGenerator
 
@@ -141,7 +141,7 @@ def _schedule(self, data, parent_data, callback_data=None):
             return
 
         polling_time = data.get_one_of_outputs("polling_time")
-        if polling_time + POLLING_INTERVAL > POLLING_TIMEOUT:
+        if polling_time + POLLING_INTERVAL > self.service_polling_timeout:
             sub_inst_ids = [host_id__sub_inst_id_map[host_id] for host_id in host_ids_need_to_query]
             self.move_insts_to_failed(sub_inst_ids=sub_inst_ids, log_content=_("查询 GSE 超时"))
             self.finish_schedule()

apps/backend/components/collections/agent_new/install.py

@@ -26,7 +26,7 @@
 
 from apps.backend.agent import solution_maker
 from apps.backend.agent.tools import InstallationTools
-from apps.backend.api.constants import POLLING_INTERVAL, POLLING_TIMEOUT
+from apps.backend.api.constants import POLLING_INTERVAL
 from apps.backend.constants import (
     REDIS_AGENT_CONF_KEY_TPL,
     REDIS_INSTALL_CALLBACK_KEY_TPL,
@@ -365,10 +365,11 @@ def _execute(self, data, parent_data, common_data: base.AgentCommonData):
             # 缓存 Agent 配置
             pipeline.mset(cache_key__config_content_map)
             # 设置过期时间
+            polling_timeout = self.service_polling_timeout
             for cache_key in cache_key__config_content_map:
                 # 根据调度超时时间预估一个过期时间
                 # 由于此时还未执行「命令下发」动作，随机增量过期时长，避免缓存雪崩
-                pipeline.expire(cache_key, POLLING_TIMEOUT + random.randint(POLLING_TIMEOUT, 2 * POLLING_TIMEOUT))
+                pipeline.expire(cache_key, polling_timeout + random.randint(polling_timeout, 2 * polling_timeout))
             pipeline.execute()
 
         remote_conn_helpers_gby_result_type = self.bulk_check_ssh(remote_conn_helpers=lan_windows_sub_inst)
@@ -866,7 +867,7 @@ def _schedule(self, data, parent_data, callback_data=None):
             return True
 
         polling_time = data.get_one_of_outputs("polling_time")
-        if polling_time + POLLING_INTERVAL > POLLING_TIMEOUT:
+        if polling_time + POLLING_INTERVAL > self.service_polling_timeout:
             self.move_insts_to_failed(left_scheduling_sub_inst_ids, _("安装超时"))
             self.finish_schedule()
         data.outputs.polling_time = polling_time + POLLING_INTERVAL

apps/backend/components/collections/agent_new/push_host_identifier.py

@@ -12,7 +12,7 @@
 
 from django.utils.translation import ugettext_lazy as _
 
-from apps.backend.api.constants import POLLING_INTERVAL, POLLING_TIMEOUT
+from apps.backend.api.constants import POLLING_INTERVAL
 from apps.core.concurrent import controller
 from apps.utils import concurrent
 from common.api import CCApi
@@ -104,7 +104,7 @@ def _schedule(self, data, parent_data, callback_data=None):
             self.finish_schedule()
             return
 
-        if polling_time + POLLING_INTERVAL > POLLING_TIMEOUT:
+        if polling_time + POLLING_INTERVAL > self.service_polling_timeout:
             self.move_insts_to_failed(sub_inst_ids=pending_push_instance_ids, log_content=_("推送主机身份超时"))
             self.finish_schedule()
             return

apps/backend/components/collections/base.py

@@ -12,6 +12,7 @@
 import os
 import traceback
 import typing
+
 from dataclasses import dataclass
 from typing import (
     Any,
@@ -33,6 +34,7 @@
 from django.utils.translation import ugettext as _
 
 from apps.adapters.api.gse import GseApiBaseHelper, get_gse_api_helper
+from apps.backend.api.constants import POLLING_TIMEOUT
 from apps.backend.subscription import errors
 from apps.core.files.storage import get_storage
 from apps.exceptions import parse_exception
@@ -218,6 +220,19 @@ def setup_pagent_file_content(self) -> bytes:
                 return fh.read().encode()
 
 
+class PollingTimeoutMixin:
+    @property
+    def service_polling_timeout(self) -> int:
+        service_name = self.__class__.__name__
+        all_service_polling_timeout: dict = models.GlobalSettings.get_config(
+            key=models.GlobalSettings.KeyEnum.BACKEND_SERVICE_POLLING_TIMEOUT.value,
+            default={},
+        )
+
+        service_polling_timeout = all_service_polling_timeout.get(service_name, POLLING_TIMEOUT)
+        return service_polling_timeout
+
+
 @dataclass
 class CommonData:
     """
@@ -237,7 +252,7 @@ class CommonData:
     subscription_instance_ids: Set[int]
 
 
-class BaseService(Service, LogMixin, DBHelperMixin):
+class BaseService(Service, LogMixin, DBHelperMixin, PollingTimeoutMixin):
 
     # 失败订阅实例ID - 失败原因 映射关系
     failed_subscription_instance_id_reason_map: Optional[Dict[int, Any]] = None

apps/backend/components/collections/job.py

@@ -21,7 +21,7 @@
 from django.conf import settings
 from django.utils.translation import ugettext_lazy as _
 
-from apps.backend.api.constants import POLLING_INTERVAL, POLLING_TIMEOUT
+from apps.backend.api.constants import POLLING_INTERVAL
 from apps.backend.api.job import process_parms
 from apps.backend.components.collections.base import BaseService, CommonData
 from apps.core.files.storage import get_storage
@@ -340,7 +340,7 @@ def _schedule(self, data, parent_data, callback_data=None):
         ).exists()
         if is_finished:
             self.finish_schedule()
-        elif polling_time + POLLING_INTERVAL > POLLING_TIMEOUT:
+        elif polling_time + POLLING_INTERVAL > self.service_polling_timeout:
             # 由于JOB的超时机制可能会失效，因此这里节点管理自己需要有超时机制进行兜底
             pending_job_sub_maps = models.JobSubscriptionInstanceMap.objects.filter(
                 node_id=self.id, status=constants.BkJobStatus.PENDING

apps/backend/components/collections/plugin.py

@@ -27,7 +27,6 @@
 from apps.backend.api.constants import (
     GSE_RUNNING_TASK_CODE,
     POLLING_INTERVAL,
-    POLLING_TIMEOUT,
     SUFFIX_MAP,
     GseDataErrCode,
 )
@@ -1249,7 +1248,7 @@ def handle_error_code(
         if error_code == GseDataErrCode.RUNNING:
             # 只要有运行中的任务，则认为未完成，标记 is_finished
             is_finished = False
-            if polling_time + POLLING_INTERVAL > POLLING_TIMEOUT:
+            if polling_time + POLLING_INTERVAL > self.service_polling_timeout:
                 self.move_insts_to_failed([subscription_instance.id], _("GSE任务轮询超时"))
         elif success_conditions:
             # 状态码非 SUCCESS 的，但满足成功的特殊条件，认为是成功的，无需做任何处理

apps/backend/tests/components/collections/agent_new/base.py

@@ -17,8 +17,10 @@
 from typing import Any, Dict, List, Optional
 
 import mock
+from django.test import TestCase
 
-from apps.backend.api.constants import POLLING_INTERVAL
+from apps.backend.api.constants import POLLING_INTERVAL, POLLING_TIMEOUT
+from apps.backend.components.collections.base import PollingTimeoutMixin
 from apps.mock_data import api_mkd
 from apps.mock_data import utils as mock_data_utils
 from apps.node_man import constants, models
@@ -31,6 +33,23 @@
 from . import utils
 
 
+class PollingTimeoutMixinTestCase(TestCase):
+    def setUp(self):
+        self.mixin = PollingTimeoutMixin()
+
+    def test_default_polling_time(self):
+        expected_timeout = POLLING_TIMEOUT
+        timeout = self.mixin.service_polling_timeout
+        self.assertEqual(timeout, expected_timeout)
+
+    def test_component_polling_time(self):
+        expected_timeout = 20
+        component_polling_timeout = {"PollingTimeoutMixin": expected_timeout}
+        with mock.patch('apps.node_man.models.GlobalSettings.get_config', return_value=component_polling_timeout):
+            timeout = self.mixin.service_polling_timeout
+        self.assertEqual(timeout, expected_timeout)
+
+
 class JobBaseTestCase(utils.AgentServiceBaseTestCase, ABC):
     """用于JobV3BaseService子类的测试基类"""
 
@@ -241,7 +260,8 @@ def structure_mock_data(cls):
     def setUp(self) -> None:
         super().setUp()
         mock.patch(
-            "apps.backend.components.collections.job.POLLING_TIMEOUT", (self.POLLING_COUNT - 1) * POLLING_INTERVAL
+            "apps.backend.components.collections.base.PollingTimeoutMixin.service_polling_timeout",
+            (self.POLLING_COUNT - 1) * POLLING_INTERVAL
         ).start()
 
     def fetch_schedule_assertion(self) -> List[ScheduleAssertion]:

apps/backend/tests/components/collections/agent_new/test_get_agent_status.py

@@ -230,7 +230,7 @@ def adjust_test_data_in_db(cls):
     def setUp(self) -> None:
         super().setUp()
         mock.patch(
-            "apps.backend.components.collections.agent_new.get_agent_status.POLLING_TIMEOUT",
+            "apps.backend.components.collections.base.PollingTimeoutMixin.service_polling_timeout",
             2 * POLLING_INTERVAL - 1,
         ).start()
 

apps/backend/tests/components/collections/agent_new/test_push_host_identifier.py

@@ -193,7 +193,7 @@ def get_default_case_name(cls) -> str:
     def setUp(self) -> None:
         super().setUp()
         mock.patch(
-            "apps.backend.api.constants.POLLING_TIMEOUT",
+            "apps.backend.components.collections.base.PollingTimeoutMixin.service_polling_timeout",
             2 * POLLING_INTERVAL - 1,
         ).start()
 

apps/node_man/models.py

@@ -146,6 +146,8 @@ class KeyEnum(Enum):
         ENABLE_AP_VERSION_MUTEX = "ENABLE_AP_VERSION_MUTEX"
         # 记录所有业务ID,用于同步新业务到灰度列表对比使用
         ALL_BIZ_IDS = "ALL_BIZ_IDS"
+        # 后台组件服务超时时间
+        BACKEND_SERVICE_POLLING_TIMEOUT = "BACKEND_SERVICE_POLLING_TIMEOUT"
 
     key = models.CharField(_("键"), max_length=255, db_index=True, primary_key=True)
     v_json = JSONField(_("值"))