From 5ec844417aa2e8c7f73b6a6ff356d23feaacd9eb Mon Sep 17 00:00:00 2001 From: "a.berezin" Date: Thu, 27 Jun 2024 12:31:02 +0300 Subject: [PATCH] [#259] Optimize failover tests by paralleling long steps Signed-off-by: a.berezin --- .../failovers/test_failover_server.py | 158 ++++++++---------- 1 file changed, 67 insertions(+), 91 deletions(-) diff --git a/pytest_tests/testsuites/failovers/test_failover_server.py b/pytest_tests/testsuites/failovers/test_failover_server.py index 39f30d5e..6b03fca3 100644 --- a/pytest_tests/testsuites/failovers/test_failover_server.py +++ b/pytest_tests/testsuites/failovers/test_failover_server.py @@ -1,5 +1,5 @@ +import itertools import logging -import os.path import random import allure @@ -17,7 +17,6 @@ from frostfs_testlib.storage.dataclasses.wallet import WalletInfo from frostfs_testlib.testing.cluster_test_base import ClusterTestBase from frostfs_testlib.testing.parallel import parallel from frostfs_testlib.testing.test_control import wait_for_success -from frostfs_testlib.utils.failover_utils import wait_object_replication from frostfs_testlib.utils.file_utils import get_file_hash from pytest import FixtureRequest @@ -35,7 +34,7 @@ class TestFailoverServer(ClusterTestBase): def wait_node_in_map(self, *args, **kwargs): check_node_in_map(*args, **kwargs) - @reporter.step("Create {count_containers} containers and {count_files} objects") + @allure.title("[Test] Create containers") @pytest.fixture def containers( self, @@ -45,22 +44,24 @@ class TestFailoverServer(ClusterTestBase): placement_rule = "REP 2 CBF 2 SELECT 2 FROM *" - containers = [] + containers_count = request.param + results = parallel( + [create_container for _ in range(containers_count)], + wallet=default_wallet, + shell=self.shell, + endpoint=self.cluster.default_rpc_endpoint, + rule=placement_rule, + basic_acl=PUBLIC_ACL, + ) - for _ in range(request.param): - cont_id = create_container( - default_wallet, - shell=self.shell, - endpoint=self.cluster.default_rpc_endpoint, - rule=placement_rule, - basic_acl=PUBLIC_ACL, - ) - storage_cont_info = StorageContainerInfo(cont_id, default_wallet) - containers.append(StorageContainer(storage_cont_info, self.shell, self.cluster)) + containers = [ + StorageContainer(StorageContainerInfo(result.result(), default_wallet), self.shell, self.cluster) + for result in results + ] return containers - @reporter.step("Creation container") + @allure.title("[Test] Create container") @pytest.fixture() def container(self, default_wallet: WalletInfo) -> StorageContainer: select = len(self.cluster.cluster_nodes) @@ -75,7 +76,7 @@ class TestFailoverServer(ClusterTestBase): storage_cont_info = StorageContainerInfo(cont_id, default_wallet) return StorageContainer(storage_cont_info, self.shell, self.cluster) - @reporter.step("Create object and delete after test") + @allure.title("[Class] Create objects") @pytest.fixture(scope="class") def storage_objects( self, @@ -83,69 +84,48 @@ class TestFailoverServer(ClusterTestBase): containers: list[StorageContainer], simple_object_size: ObjectSize, complex_object_size: ObjectSize, - ) -> StorageObjectInfo: - count_object = request.param - object_sizes = [simple_object_size, complex_object_size] - object_list: list[StorageObjectInfo] = [] - for cont in containers: - for _ in range(count_object): - object_list.append(cont.generate_object(size=random.choice(object_sizes).value)) - - for storage_object in object_list: - os.remove(storage_object.file_path) - - yield object_list - - @reporter.step("Upload object with nodes and compare") - def get_corrupted_objects_list( - self, nodes: list[StorageNode], storage_objects: list[StorageObjectInfo] ) -> list[StorageObjectInfo]: - corrupted_objects = [] - errors_get = [] - for node in nodes: - for storage_object in storage_objects: - try: - got_file_path = get_object( - storage_object.wallet, - storage_object.cid, - storage_object.oid, - endpoint=node.get_rpc_endpoint(), - shell=self.shell, - timeout="60s", - ) - if storage_object.file_hash != get_file_hash(got_file_path): - corrupted_objects.append(storage_object) - os.remove(got_file_path) - except RuntimeError: - errors_get.append(storage_object.oid) + object_count = request.param + sizes_samples = [simple_object_size, complex_object_size] + samples_count = len(sizes_samples) + assert object_count >= samples_count, f"Object count is too low, must be >= {samples_count}" - assert len(errors_get) == 0, f"Get failed - {errors_get}" - return corrupted_objects + sizes_weights = [2, 1] + sizes = sizes_samples + random.choices(sizes_samples, weights=sizes_weights, k=object_count - samples_count) - def check_objects_replication( - self, storage_objects: list[StorageObjectInfo], storage_nodes: list[StorageNode] - ) -> None: - for storage_object in storage_objects: - wait_object_replication( - storage_object.cid, - storage_object.oid, - 2, - shell=self.shell, - nodes=storage_nodes, - sleep_interval=45, - attempts=60, - ) + results = parallel( + [container.generate_object for _ in sizes for container in containers], + size=itertools.cycle([size.value for size in sizes]), + ) + return [result.result() for result in results] + + @allure.title("[Test] Create objects and get nodes with object") @pytest.fixture() def object_and_nodes( self, simple_object_size: ObjectSize, container: StorageContainer ) -> tuple[StorageObjectInfo, list[ClusterNode]]: object_info = container.generate_object(simple_object_size.value) - object_nodes = get_object_nodes( - cluster=self.cluster, cid=object_info.cid, oid=object_info.oid, alive_node=self.cluster.cluster_nodes[0] - ) + object_nodes = get_object_nodes(self.cluster, object_info.cid, object_info.oid, self.cluster.cluster_nodes[0]) return object_info, object_nodes + def _verify_object(self, storage_object: StorageObjectInfo, node: StorageNode): + with reporter.step(f"Verify object {storage_object.oid} from node {node}"): + file_path = get_object( + storage_object.wallet, + storage_object.cid, + storage_object.oid, + endpoint=node.get_rpc_endpoint(), + shell=self.shell, + timeout="60s", + ) + + assert storage_object.file_hash == get_file_hash(file_path) + + @reporter.step("Verify objects") + def verify_objects(self, nodes: list[StorageNode], storage_objects: list[StorageObjectInfo]) -> None: + parallel(self._verify_object, storage_objects * len(nodes), node=itertools.cycle(nodes)) + @allure.title("Full shutdown node") @pytest.mark.parametrize("containers, storage_objects", [(5, 10)], indirect=True) def test_complete_node_shutdown( @@ -154,22 +134,21 @@ class TestFailoverServer(ClusterTestBase): node_under_test: ClusterNode, cluster_state_controller: ClusterStateController, ): - with reporter.step(f"Remove {node_under_test} from the list of nodes"): + with reporter.step(f"Remove one node from the list of nodes"): alive_nodes = list(set(self.cluster.cluster_nodes) - {node_under_test}) storage_nodes = [cluster.storage_node for cluster in alive_nodes] - with reporter.step("Tick epoch and wait for 2 blocks"): - self.tick_epochs(1, storage_nodes[0], wait_block=2) + with reporter.step("Tick 2 epochs and wait for 2 blocks"): + self.tick_epochs(2, storage_nodes[0], wait_block=2) with reporter.step(f"Stop node"): - cluster_state_controller.stop_node_host(node=node_under_test, mode="hard") + cluster_state_controller.stop_node_host(node_under_test, "hard") with reporter.step("Verify that there are no corrupted objects"): - corrupted_objects_list = self.get_corrupted_objects_list(storage_nodes, storage_objects) - assert not corrupted_objects_list + self.verify_objects(storage_nodes, storage_objects) - with reporter.step(f"check {node_under_test.storage_node} in map"): + with reporter.step(f"Check node still in map"): self.wait_node_in_map(node_under_test.storage_node, self.shell, alive_node=storage_nodes[0]) count_tick_epoch = int(alive_nodes[0].ir_node.get_netmap_cleaner_threshold()) + 4 @@ -177,12 +156,11 @@ class TestFailoverServer(ClusterTestBase): with reporter.step(f"Tick {count_tick_epoch} epochs and wait for 2 blocks"): self.tick_epochs(count_tick_epoch, storage_nodes[0], wait_block=2) - with reporter.step(f"Check {node_under_test} in not map"): + with reporter.step(f"Check node in not map after {count_tick_epoch} epochs"): self.wait_node_not_in_map(node_under_test.storage_node, self.shell, alive_node=storage_nodes[0]) - with reporter.step(f"Verify that there are no corrupted objects after {count_tick_epoch} epoch"): - corrupted_objects_list = self.get_corrupted_objects_list(storage_nodes, storage_objects) - assert not corrupted_objects_list + with reporter.step(f"Verify that there are no corrupted objects after {count_tick_epoch} epochs"): + self.verify_objects(storage_nodes, storage_objects) @allure.title("Temporarily disable a node") @pytest.mark.parametrize("containers, storage_objects", [(5, 10)], indirect=True) @@ -192,27 +170,26 @@ class TestFailoverServer(ClusterTestBase): node_under_test: ClusterNode, cluster_state_controller: ClusterStateController, ): - with reporter.step(f"Remove {node_under_test} from the list of nodes"): + with reporter.step(f"Remove one node from the list"): storage_nodes = list(set(self.cluster.storage_nodes) - {node_under_test.storage_node}) - with reporter.step("Tick epoch and wait for 2 blocks"): - self.tick_epochs(1, storage_nodes[0], wait_block=2) + with reporter.step("Tick 2 epochs and wait for 2 blocks"): + self.tick_epochs(2, storage_nodes[0], wait_block=2) with reporter.step(f"Stop node"): cluster_state_controller.stop_node_host(node_under_test, "hard") with reporter.step("Verify that there are no corrupted objects"): - corrupted_objects_list = self.get_corrupted_objects_list(storage_nodes, storage_objects) - assert not corrupted_objects_list + self.verify_objects(storage_nodes, storage_objects) - with reporter.step(f"Check {node_under_test} in map"): + with reporter.step(f"Check node still in map"): self.wait_node_in_map(node_under_test.storage_node, self.shell, alive_node=storage_nodes[0]) - cluster_state_controller.start_node_host(node_under_test) + with reporter.step(f"Start node"): + cluster_state_controller.start_node_host(node_under_test) with reporter.step("Verify that there are no corrupted objects"): - corrupted_objects_list = self.get_corrupted_objects_list(storage_nodes, storage_objects) - assert not corrupted_objects_list + self.verify_objects(storage_nodes, storage_objects) @allure.title("Not enough nodes in the container with policy - 'REP 3 CBF 1 SELECT 4 FROM *'") def test_not_enough_nodes_in_container_rep_3( @@ -237,7 +214,7 @@ class TestFailoverServer(ClusterTestBase): with reporter.step(f"Get object from node with object"): get_object(default_wallet, object_info.cid, object_info.oid, self.shell, endpoint_with_object) - with reporter.step(f"Put operation to node with object, expect error"): + with reporter.step(f"[Negative] Put operation to node with object"): with pytest.raises(RuntimeError): put_object(default_wallet, simple_file, object_info.cid, self.shell, endpoint_with_object) @@ -277,8 +254,7 @@ class TestFailoverServer(ClusterTestBase): oid_2 = put_object(default_wallet, simple_file, cid, self.shell, alive_endpoint_with_object) with reporter.step("Get object from alive node with object"): - get_file = get_object(default_wallet, cid, oid_2, self.shell, alive_endpoint_with_object) - os.remove(get_file) + get_object(default_wallet, cid, oid_2, self.shell, alive_endpoint_with_object) with reporter.step("Create container on alive node"): create_container(