PYTHON-2834 Direct read/write retries to another mongos if possible (#1421)

Author: NoahStapp

pymongo/mongo_client.py

@@ -1277,6 +1277,7 @@ def _select_server(
         server_selector: Callable[[Selection], Selection],
         session: Optional[ClientSession],
         address: Optional[_Address] = None,
+        deprioritized_servers: Optional[list[Server]] = None,
     ) -> Server:
         """Select a server to run an operation on this client.
 
@@ -1300,7 +1301,9 @@ def _select_server(
                 if not server:
                     raise AutoReconnect("server %s:%s no longer available" % address)  # noqa: UP031
             else:
-                server = topology.select_server(server_selector)
+                server = topology.select_server(
+                    server_selector, deprioritized_servers=deprioritized_servers
+                )
             return server
         except PyMongoError as exc:
             # Server selection errors in a transaction are transient.
@@ -2291,6 +2294,7 @@ def __init__(
         )
         self._address = address
         self._server: Server = None  # type: ignore
+        self._deprioritized_servers: list[Server] = []
 
     def run(self) -> T:
         """Runs the supplied func() and attempts a retry
@@ -2359,6 +2363,9 @@ def run(self) -> T:
                     if self._last_error is None:
                         self._last_error = exc
 
+                if self._client.topology_description.topology_type == TOPOLOGY_TYPE.Sharded:
+                    self._deprioritized_servers.append(self._server)
+
     def _is_not_eligible_for_retry(self) -> bool:
         """Checks if the exchange is not eligible for retry"""
         return not self._retryable or (self._is_retrying() and not self._multiple_retries)
@@ -2397,7 +2404,10 @@ def _get_server(self) -> Server:
             Abstraction to connect to server
         """
         return self._client._select_server(
-            self._server_selector, self._session, address=self._address
+            self._server_selector,
+            self._session,
+            address=self._address,
+            deprioritized_servers=self._deprioritized_servers,
         )
 
     def _write(self) -> T:

pymongo/topology.py

@@ -282,8 +282,10 @@ def _select_server(
         selector: Callable[[Selection], Selection],
         server_selection_timeout: Optional[float] = None,
         address: Optional[_Address] = None,
+        deprioritized_servers: Optional[list[Server]] = None,
     ) -> Server:
         servers = self.select_servers(selector, server_selection_timeout, address)
+        servers = _filter_servers(servers, deprioritized_servers)
         if len(servers) == 1:
             return servers[0]
         server1, server2 = random.sample(servers, 2)
@@ -297,9 +299,12 @@ def select_server(
         selector: Callable[[Selection], Selection],
         server_selection_timeout: Optional[float] = None,
         address: Optional[_Address] = None,
+        deprioritized_servers: Optional[list[Server]] = None,
     ) -> Server:
         """Like select_servers, but choose a random server if several match."""
-        server = self._select_server(selector, server_selection_timeout, address)
+        server = self._select_server(
+            selector, server_selection_timeout, address, deprioritized_servers
+        )
         if _csot.get_timeout():
             _csot.set_rtt(server.description.min_round_trip_time)
         return server
@@ -931,3 +936,16 @@ def _is_stale_server_description(current_sd: ServerDescription, new_sd: ServerDe
     if current_tv["processId"] != new_tv["processId"]:
         return False
     return current_tv["counter"] > new_tv["counter"]
+
+
+def _filter_servers(
+    candidates: list[Server], deprioritized_servers: Optional[list[Server]] = None
+) -> list[Server]:
+    """Filter out deprioritized servers from a list of server candidates."""
+    if not deprioritized_servers:
+        return candidates
+
+    filtered = [server for server in candidates if server not in deprioritized_servers]
+
+    # If not possible to pick a prioritized server, return the original list
+    return filtered or candidates

test/test_retryable_reads.py

@@ -20,6 +20,9 @@
 import sys
 import threading
 
+from bson import SON
+from pymongo.errors import AutoReconnect
+
 sys.path[0:0] = [""]
 
 from test import (
@@ -31,9 +34,12 @@
 )
 from test.utils import (
     CMAPListener,
+    EventListener,
     OvertCommandListener,
     SpecTestCreator,
+    rs_client,
     rs_or_single_client,
+    set_fail_point,
 )
 from test.utils_spec_runner import SpecRunner
 
@@ -221,5 +227,48 @@ def test_pool_paused_error_is_retryable(self):
         self.assertEqual(1, len(failed), msg)
 
 
+class TestRetryableReads(IntegrationTest):
+    @client_context.require_multiple_mongoses
+    @client_context.require_failCommand_fail_point
+    def test_retryable_reads_in_sharded_cluster_multiple_available(self):
+        fail_command = {
+            "configureFailPoint": "failCommand",
+            "mode": {"times": 1},
+            "data": {
+                "failCommands": ["find"],
+                "closeConnection": True,
+                "appName": "retryableReadTest",
+            },
+        }
+
+        mongos_clients = []
+
+        for mongos in client_context.mongos_seeds().split(","):
+            client = rs_or_single_client(mongos)
+            set_fail_point(client, fail_command)
+            self.addCleanup(client.close)
+            mongos_clients.append(client)
+
+        listener = OvertCommandListener()
+        client = rs_or_single_client(
+            client_context.mongos_seeds(),
+            appName="retryableReadTest",
+            event_listeners=[listener],
+            retryReads=True,
+        )
+
+        with self.fail_point(fail_command):
+            with self.assertRaises(AutoReconnect):
+                client.t.t.find_one({})
+
+        # Disable failpoints on each mongos
+        for client in mongos_clients:
+            fail_command["mode"] = "off"
+            set_fail_point(client, fail_command)
+
+        self.assertEqual(len(listener.failed_events), 2)
+        self.assertEqual(len(listener.succeeded_events), 0)
+
+
 if __name__ == "__main__":
     unittest.main()

test/test_retryable_writes.py

@@ -31,6 +31,7 @@
     OvertCommandListener,
     SpecTestCreator,
     rs_or_single_client,
+    set_fail_point,
 )
 from test.utils_spec_runner import SpecRunner
 from test.version import Version
@@ -40,6 +41,7 @@
 from bson.raw_bson import RawBSONDocument
 from bson.son import SON
 from pymongo.errors import (
+    AutoReconnect,
     ConnectionFailure,
     OperationFailure,
     ServerSelectionTimeoutError,
@@ -469,6 +471,46 @@ def test_batch_splitting_retry_fails(self):
             self.assertEqual(final_txn, expected_txn)
         self.assertEqual(coll.find_one(projection={"_id": True}), {"_id": 1})
 
+    @client_context.require_multiple_mongoses
+    @client_context.require_failCommand_fail_point
+    def test_retryable_writes_in_sharded_cluster_multiple_available(self):
+        fail_command = {
+            "configureFailPoint": "failCommand",
+            "mode": {"times": 1},
+            "data": {
+                "failCommands": ["insert"],
+                "closeConnection": True,
+                "appName": "retryableWriteTest",
+            },
+        }
+
+        mongos_clients = []
+
+        for mongos in client_context.mongos_seeds().split(","):
+            client = rs_or_single_client(mongos)
+            set_fail_point(client, fail_command)
+            self.addCleanup(client.close)
+            mongos_clients.append(client)
+
+        listener = OvertCommandListener()
+        client = rs_or_single_client(
+            client_context.mongos_seeds(),
+            appName="retryableWriteTest",
+            event_listeners=[listener],
+            retryWrites=True,
+        )
+
+        with self.assertRaises(AutoReconnect):
+            client.t.t.insert_one({"x": 1})
+
+        # Disable failpoints on each mongos
+        for client in mongos_clients:
+            fail_command["mode"] = "off"
+            set_fail_point(client, fail_command)
+
+        self.assertEqual(len(listener.failed_events), 2)
+        self.assertEqual(len(listener.succeeded_events), 0)
+
 
 class TestWriteConcernError(IntegrationTest):
     RUN_ON_LOAD_BALANCER = True

test/test_topology.py

@@ -30,11 +30,12 @@
 from pymongo.monitor import Monitor
 from pymongo.pool import PoolOptions
 from pymongo.read_preferences import ReadPreference, Secondary
+from pymongo.server import Server
 from pymongo.server_description import ServerDescription
 from pymongo.server_selectors import any_server_selector, writable_server_selector
 from pymongo.server_type import SERVER_TYPE
 from pymongo.settings import TopologySettings
-from pymongo.topology import Topology, _ErrorContext
+from pymongo.topology import Topology, _ErrorContext, _filter_servers
 from pymongo.topology_description import TOPOLOGY_TYPE
 
 
@@ -685,6 +686,23 @@ def test_unexpected_load_balancer(self):
         self.assertNotIn(("a", 27017), t.description.server_descriptions())
         self.assertEqual(t.description.topology_type_name, "Unknown")
 
+    def test_filtered_server_selection(self):
+        s1 = Server(ServerDescription(("localhost", 27017)), pool=object(), monitor=object())  # type: ignore[arg-type]
+        s2 = Server(ServerDescription(("localhost2", 27017)), pool=object(), monitor=object())  # type: ignore[arg-type]
+        servers = [s1, s2]
+
+        result = _filter_servers(servers, deprioritized_servers=[s2])
+        self.assertEqual(result, [s1])
+
+        result = _filter_servers(servers, deprioritized_servers=[s1, s2])
+        self.assertEqual(result, servers)
+
+        result = _filter_servers(servers, deprioritized_servers=[])
+        self.assertEqual(result, servers)
+
+        result = _filter_servers(servers)
+        self.assertEqual(result, servers)
+
 
 def wait_for_primary(topology):
     """Wait for a Topology to discover a writable server.

test/utils.py

@@ -1153,3 +1153,9 @@ def prepare_spec_arguments(spec, arguments, opname, entity_map, with_txn_callbac
                 raise AssertionError(f"Unsupported cursorType: {cursor_type}")
         else:
             arguments[c2s] = arguments.pop(arg_name)
+
+
+def set_fail_point(client, command_args):
+    cmd = SON([("configureFailPoint", "failCommand")])
+    cmd.update(command_args)
+    client.admin.command(cmd)