Add write-cache loss test

Signed-off-by: Andrey Berezin <a.berezin@yadro.com>

pytest_tests/testsuites/failovers/conftest.py

@@ -0,0 +1,29 @@
+import random
+from datetime import datetime
+
+import allure
+import pytest
+from frostfs_testlib.storage.cluster import Cluster, ClusterNode
+from frostfs_testlib.storage.controllers import ShardsWatcher
+
+
+@pytest.fixture()
+@allure.title("Select random node for testing")
+def node_under_test(cluster: Cluster) -> ClusterNode:
+    selected_node = random.choice(cluster.cluster_nodes)
+    allure.attach(f"{selected_node}", "Selected node", allure.attachment_type.TEXT)
+    return selected_node
+
+
+@pytest.fixture()
+@allure.title("Provide Shards watcher")
+def shards_watcher(node_under_test: ClusterNode) -> ShardsWatcher:
+    watcher = ShardsWatcher(node_under_test)
+    return watcher
+
+
+@pytest.fixture()
+@allure.title("Test start time")
+def test_start_time() -> datetime:
+    start_time = datetime.utcnow()
+    return start_time

pytest_tests/testsuites/failovers/test_failover_storage.py

@@ -1,4 +1,5 @@
 import logging
+from datetime import datetime
 from time import sleep
 
 import allure
@@ -9,7 +10,11 @@ from frostfs_testlib.resources.common import MORPH_BLOCK_TIME
 from frostfs_testlib.resources.wellknown_acl import PUBLIC_ACL
 from frostfs_testlib.s3 import AwsCliClient, Boto3ClientWrapper, S3ClientWrapper, VersioningStatus
 from frostfs_testlib.shell import CommandOptions, Shell
-from frostfs_testlib.steps.cli.container import create_container
+from frostfs_testlib.steps.cli.container import (
+    StorageContainer,
+    StorageContainerInfo,
+    create_container,
+)
 from frostfs_testlib.steps.cli.object import get_object, put_object_to_random_node
 from frostfs_testlib.steps.node_management import (
     check_node_in_map,
@@ -20,8 +25,10 @@ from frostfs_testlib.steps.node_management import (
     wait_for_node_to_be_ready,
 )
 from frostfs_testlib.steps.s3 import s3_helper
-from frostfs_testlib.storage.cluster import Cluster, StorageNode
-from frostfs_testlib.storage.controllers import ClusterStateController
+from frostfs_testlib.storage.cluster import Cluster, ClusterNode, StorageNode
+from frostfs_testlib.storage.controllers import ClusterStateController, ShardsWatcher
+from frostfs_testlib.storage.dataclasses.storage_object_info import StorageObjectInfo
+from frostfs_testlib.storage.dataclasses.wallet import WalletInfo
 from frostfs_testlib.testing.cluster_test_base import ClusterTestBase
 from frostfs_testlib.testing.test_control import expect_not_raises
 from frostfs_testlib.utils import datetime_utils
@@ -580,3 +587,90 @@ class TestStorageDataLoss(ClusterTestBase):
         with allure.step("Delete bucket"):
             with expect_not_raises():
                 s3_client.delete_bucket(bucket)
+
+    @allure.title(
+        "Write cache loss on one node should not affect shards and should not produce errors in log"
+    )
+    @pytest.mark.write_cache_loss
+    def test_write_cache_loss_on_one_node(
+        self,
+        node_under_test: ClusterNode,
+        simple_object_size: int,
+        cluster_state_controller: ClusterStateController,
+        shards_watcher: ShardsWatcher,
+        default_wallet: str,
+        test_start_time: datetime,
+        after_run_return_all_stopped_services: str,
+    ):
+        exception_messages = []
+        allure.dynamic.description(after_run_return_all_stopped_services)
+
+        with allure.step(f"Create container on node {node_under_test}"):
+            locode = node_under_test.storage_node.get_un_locode()
+            placement_rule = f"""REP 1 IN X
+                                 CBF 1 
+                                 SELECT 1 FROM C AS X 
+                                 FILTER 'UN-LOCODE' EQ '{locode}' AS C"""
+            cid = create_container(
+                default_wallet,
+                self.shell,
+                node_under_test.storage_node.get_rpc_endpoint(),
+                rule=placement_rule,
+            )
+            container = StorageContainer(
+                StorageContainerInfo(cid, WalletInfo(default_wallet)),
+                self.shell,
+                cluster_state_controller.cluster,
+            )
+
+        with allure.step(f"Put couple objects to container on node {node_under_test}"):
+            storage_objects: list[StorageObjectInfo] = []
+            for _ in range(5):
+                storage_object = container.generate_object(
+                    simple_object_size, endpoint=node_under_test.storage_node.get_rpc_endpoint()
+                )
+                storage_objects.append(storage_object)
+
+        with allure.step("Take shards snapshot"):
+            shards_watcher.take_shards_snapshot()
+
+        with allure.step(f"Stop storage service on node {node_under_test}"):
+            cluster_state_controller.stop_storage_service(node_under_test)
+
+        with allure.step(f"Delete write cache from node {node_under_test}"):
+            node_under_test.storage_node.delete_write_cache()
+
+        with allure.step(f"Start storage service on node {node_under_test}"):
+            cluster_state_controller.start_storage_service(node_under_test)
+
+        with allure.step("Objects should be available"):
+            for storage_object in storage_objects:
+                get_object(
+                    storage_object.wallet_file_path,
+                    container.get_id(),
+                    storage_object.oid,
+                    self.shell,
+                    node_under_test.storage_node.get_rpc_endpoint(),
+                )
+
+        with allure.step("No shards should have new errors"):
+            shards_watcher.take_shards_snapshot()
+            shards_with_errors = shards_watcher.get_shards_with_new_errors()
+            if shards_with_errors:
+                exception_messages.append(f"Shards have new errors: {shards_with_errors}")
+
+        with allure.step("No shards should have degraded status"):
+            snapshot = shards_watcher.get_shards_snapshot()
+            for shard in snapshot:
+                status = snapshot[shard]["mode"]
+                if status != "read-write":
+                    exception_messages.append(f"Shard {shard} changed status to {status}")
+
+        with allure.step("No related errors should be in log"):
+            if node_under_test.host.is_message_in_logs(
+                message_regex=r"\Wno such file or directory\W", since=test_start_time
+            ):
+                exception_messages.append(f"Node {node_under_test} have shard errors in logs")
+
+        with allure.step("Pass test if no errors found"):
+            assert not exception_messages, "\n".join(exception_messages)