Improve wait_actor_pool_recovered

mars/oscar/backends/pool.py

@@ -864,15 +864,11 @@ async def handle_control_command(self,
                     self.sub_processes[message.address],
                     timeout=timeout,
                     force=force)
-                if self._auto_recover:
-                    self._recover_events[message.address] = asyncio.Event()
                 processor.result = ResultMessage(message.message_id, True,
                                                  protocol=message.protocol)
             elif message.control_message_type == ControlMessageType.wait_pool_recovered:
-                # check the aliveness of sub pool first, in case monitor task haven't found it.
-                if not await self.is_sub_pool_alive(self.sub_processes[message.address]):
-                    if self._auto_recover and message.address not in self._recover_events:
-                        self._recover_events[message.address] = asyncio.Event()
+                if self._auto_recover and message.address not in self._recover_events:
+                    self._recover_events[message.address] = asyncio.Event()
 
                 event = self._recover_events.get(message.address, None)
                 if event is not None:
@@ -1019,6 +1015,10 @@ async def is_sub_pool_alive(self, process: SubProcessHandle):
         bool
         """
 
+    @abstractmethod
+    def recover_sub_pool(self, address):
+        """Recover a sub actor pool"""
+
     def process_sub_pool_lost(self, address: str):
         if self._auto_recover in (False, 'process'):
             # process down, when not auto_recover
@@ -1030,18 +1030,18 @@ async def monitor_sub_pools(self):
             while not self._stopped.is_set():
                 for address in self.sub_processes:
                     process = self.sub_processes[address]
+                    recover_events_discovered = (address in self._recover_events)
                     if not await self.is_sub_pool_alive(process):  # pragma: no cover
                         if self._on_process_down is not None:
                             self._on_process_down(self, address)
                         self.process_sub_pool_lost(address)
                         if self._auto_recover:
-                            if address not in self._recover_events:
-                                self._recover_events[address] = asyncio.Event()
                             await self.recover_sub_pool(address)
                             if self._on_process_recover is not None:
                                 self._on_process_recover(self, address)
-                            event = self._recover_events.pop(address)
-                            event.set()
+                    if recover_events_discovered:
+                        event = self._recover_events.pop(address)
+                        event.set()
 
                 # check every half second
                 await asyncio.sleep(.5)

mars/oscar/backends/ray/pool.py

@@ -29,6 +29,7 @@
 from ..message import CreateActorMessage
 from ..pool import AbstractActorPool, MainActorPoolBase, SubActorPoolBase, create_actor_pool, _register_message_handler
 from ..router import Router
+from ... import ServerClosed
 from ....serialization.ray import register_ray_serializers
 from ....utils import lazy_import
 
@@ -174,6 +175,8 @@ def _set_ray_server(self, actor_pool: AbstractActorPool):
     async def __on_ray_recv__(self, channel_id: ChannelID, message):
         """Method for communication based on ray actors"""
         try:
+            if self._ray_server is None:
+                raise ServerClosed(f'Remote server {channel_id.dest_address} closed')
             return await self._ray_server.__on_ray_recv__(channel_id, message)
         except Exception:  # pragma: no cover
             return RayChannelException(*sys.exc_info())

mars/oscar/backends/ray/tests/test_ray_pool.py

@@ -11,17 +11,33 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import asyncio
+import os
 
 import pytest
 
+import mars.oscar as mo
+from mars.oscar.errors import ServerClosed
+from mars.oscar.backends.allocate_strategy import ProcessIndex, MainPool
+from mars.oscar.backends.ray.pool import RayMainPool, RayMainActorPool, create_actor_pool, PoolStatus
+from mars.oscar.backends.ray.utils import process_placement_to_address
+from mars.oscar.context import get_context
 from mars.tests.core import require_ray
-from .....utils import lazy_import
-from ..pool import RayMainPool, RayMainActorPool, create_actor_pool
-from ..utils import process_placement_to_address
+from mars.utils import lazy_import
 
 ray = lazy_import('ray')
 
 
+class TestActor(mo.Actor):
+    async def kill(self, address, uid):
+        actor_ref = await mo.actor_ref(address, uid)
+        task = asyncio.create_task(actor_ref.crash())
+        return await task
+
+    async def crash(self):
+        os._exit(0)
+
+
 @require_ray
 @pytest.mark.asyncio
 async def test_main_pool(ray_start_regular):
@@ -71,3 +87,96 @@ async def test_shutdown_sub_pool(ray_start_regular):
         await sub_pool_handle1.actor_pool.remote('health_check')
     with pytest.raises(AttributeError, match='NoneType'):
         await sub_pool_handle2.actor_pool.remote('health_check')
+
+
+@require_ray
+@pytest.mark.asyncio
+async def test_server_closed(ray_start_regular):
+    pg_name, n_process = 'ray_cluster', 1
+    pg = ray.util.placement_group(name=pg_name, bundles=[{'CPU': n_process}])
+    ray.get(pg.ready())
+    address = process_placement_to_address(pg_name, 0, process_index=0)
+    # start the actor pool
+    actor_handle = await mo.create_actor_pool(address, n_process=n_process)
+    await actor_handle.actor_pool.remote('start')
+
+    ctx = get_context()
+    actor_main = await ctx.create_actor(
+        TestActor, address=address, uid='Test-main',
+        allocate_strategy=ProcessIndex(0))
+
+    actor_sub = await ctx.create_actor(
+        TestActor, address=address, uid='Test-sub',
+        allocate_strategy=ProcessIndex(1))
+
+    # test calling from ray driver to ray actor
+    task = asyncio.create_task(actor_sub.crash())
+
+    with pytest.raises(ServerClosed):
+        # process already died,
+        # ServerClosed will be raised
+        await task
+
+    # wait for recover of sub pool
+    await ctx.wait_actor_pool_recovered(actor_sub.address, address)
+
+    # test calling from ray actor to ray actor
+    task = asyncio.create_task(actor_main.kill(actor_sub.address, 'Test-sub'))
+
+    with pytest.raises(ServerClosed):
+        await task
+
+
+@require_ray
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    'auto_recover',
+    [False, True, 'actor', 'process']
+)
+async def test_auto_recover(ray_start_regular, auto_recover):
+    pg_name, n_process = 'ray_cluster', 1
+    pg = ray.util.placement_group(name=pg_name, bundles=[{'CPU': n_process}])
+    assert pg.wait(timeout_seconds=20)
+    address = process_placement_to_address(pg_name, 0, process_index=0)
+    actor_handle = await mo.create_actor_pool(address, n_process=n_process, auto_recover=auto_recover)
+    await actor_handle.actor_pool.remote('start')
+
+    ctx = get_context()
+
+    # wait for recover of main pool always returned immediately
+    await ctx.wait_actor_pool_recovered(address, address)
+
+    # create actor on main
+    actor_ref = await ctx.create_actor(
+        TestActor, address=address,
+        allocate_strategy=MainPool())
+
+    with pytest.raises(ValueError):
+        # cannot kill actors on main pool
+        await mo.kill_actor(actor_ref)
+
+    # create actor
+    actor_ref = await ctx.create_actor(
+        TestActor, address=address,
+        allocate_strategy=ProcessIndex(1))
+    # kill_actor will cause kill corresponding process
+    await ctx.kill_actor(actor_ref)
+
+    if auto_recover:
+        await ctx.wait_actor_pool_recovered(actor_ref.address, address)
+        sub_pool_address = process_placement_to_address(pg_name, 0, process_index=1)
+        sub_pool_handle = ray.get_actor(sub_pool_address)
+        assert await sub_pool_handle.actor_pool.remote('health_check') == PoolStatus.HEALTHY
+
+        expect_has_actor = True if auto_recover in ['actor', True] else False
+        assert await ctx.has_actor(actor_ref) is expect_has_actor
+    else:
+        with pytest.raises((ServerClosed, ConnectionError)):
+            await ctx.has_actor(actor_ref)
+
+    if 'COV_CORE_SOURCE' in os.environ:
+        for addr in [process_placement_to_address(pg_name, 0, process_index=i) for i in range(2)]:
+            # must save the local reference until this is fixed:
+            # https://github.com/ray-project/ray/issues/7815
+            ray_actor = ray.get_actor(addr)
+            ray.get(ray_actor.cleanup.remote())