Ensure agents improvements

changelogs/unreleased/ensure-agents-improvements.yml

@@ -0,0 +1,10 @@
+description: "Made various improvements to the AutostartedAgent._ensure_agents method"
+sections:
+  bugfix: "Fixed a race condition where autostarted agents might become unresponsive for 30s when restarted"
+issue-nr: 7612
+change-type: patch
+destination-branches:
+  - master
+  - iso7
+  - iso6
+

src/inmanta/agent/agent.py

@@ -977,16 +977,18 @@ async def _init_agent_map(self) -> None:
             self.agent_map = dict(cfg.agent_map.get())
 
     async def _init_endpoint_names(self) -> None:
-        if self.hostname is not None:
-            await self.add_end_point_name(self.hostname)
-        else:
-            # load agent names from the config file
-            agent_names = cfg.agent_names.get()
-            if agent_names is not None:
-                for name in agent_names:
-                    if "$" in name:
-                        name = name.replace("$node-name", self.node_name)
-                    await self.add_end_point_name(name)
+        assert self.agent_map is not None
+        endpoints: Iterable[str] = (
+            [self.hostname]
+            if self.hostname is not None
+            else (
+                self.agent_map.keys()
+                if cfg.use_autostart_agent_map.get()
+                else (name if "$" not in name else name.replace("$node-name", self.node_name) for name in cfg.agent_names.get())
+            )
+        )
+        for endpoint in endpoints:
+            await self.add_end_point_name(endpoint)
 
     async def stop(self) -> None:
         await super().stop()
@@ -1069,6 +1071,13 @@ async def update_agent_map(self, agent_map: dict[str, str]) -> None:
             await self._update_agent_map(agent_map)
 
     async def _update_agent_map(self, agent_map: dict[str, str]) -> None:
+        if "internal" not in agent_map:
+            LOGGER.warning(
+                "Agent received an update_agent_map() trigger without internal agent in the agent_map %s",
+                agent_map,
+            )
+            agent_map = {"internal": "local:", **agent_map}
+
         async with self._instances_lock:
             self.agent_map = agent_map
             # Add missing agents

src/inmanta/data/__init__.py

@@ -29,7 +29,7 @@
 import warnings
 from abc import ABC, abstractmethod
 from collections import abc, defaultdict
-from collections.abc import Awaitable, Callable, Iterable, Sequence
+from collections.abc import Awaitable, Callable, Iterable, Sequence, Set
 from configparser import RawConfigParser
 from contextlib import AbstractAsyncContextManager
 from itertools import chain
@@ -1219,7 +1219,7 @@ def get_connection(
         """
         if connection is not None:
             return util.nullcontext(connection)
-        # Make pypi happy
+        # Make mypy happy
         assert cls._connection_pool is not None
         return cls._connection_pool.acquire()
 
@@ -3344,10 +3344,12 @@ def get_valid_field_names(cls) -> list[str]:
         return super().get_valid_field_names() + ["process_name", "status"]
 
     @classmethod
-    async def get_statuses(cls, env_id: uuid.UUID, agent_names: set[str]) -> dict[str, Optional[AgentStatus]]:
+    async def get_statuses(
+        cls, env_id: uuid.UUID, agent_names: Set[str], *, connection: Optional[asyncpg.connection.Connection] = None
+    ) -> dict[str, Optional[AgentStatus]]:
         result: dict[str, Optional[AgentStatus]] = {}
         for agent_name in agent_names:
-            agent = await cls.get_one(environment=env_id, name=agent_name)
+            agent = await cls.get_one(environment=env_id, name=agent_name, connection=connection)
             if agent:
                 result[agent_name] = agent.get_status()
             else: