frostfs-testcases/pytest_tests/testsuites/failovers/test_failover_server.py

import logging
import os.path
import random

import allure
import pytest
from frostfs_testlib import reporter
from frostfs_testlib.resources.wellknown_acl import PUBLIC_ACL
from frostfs_testlib.steps.cli.container import StorageContainer, StorageContainerInfo, create_container
from frostfs_testlib.steps.cli.object import get_object, get_object_nodes, put_object
from frostfs_testlib.steps.node_management import check_node_in_map, check_node_not_in_map
from frostfs_testlib.storage.cluster import ClusterNode, StorageNode
from frostfs_testlib.storage.controllers import ClusterStateController
from frostfs_testlib.storage.dataclasses.object_size import ObjectSize
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.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

logger = logging.getLogger("NeoLogger")


@pytest.mark.failover
@pytest.mark.failover_server
class TestFailoverServer(ClusterTestBase):
    @wait_for_success(max_wait_time=120, interval=1)
    def wait_node_not_in_map(self, *args, **kwargs):
        check_node_not_in_map(*args, **kwargs)

    @wait_for_success(max_wait_time=120, interval=1)
    def wait_node_in_map(self, *args, **kwargs):
        check_node_in_map(*args, **kwargs)

    @reporter.step("Create {count_containers} containers and {count_files} objects")
    @pytest.fixture
    def containers(
        self,
        request: FixtureRequest,
        default_wallet: WalletInfo,
    ) -> list[StorageContainer]:

        placement_rule = "REP 2 CBF 2 SELECT 2 FROM *"

        containers = []

        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))

        return containers

    @reporter.step("Creation container")
    @pytest.fixture()
    def container(self, default_wallet: WalletInfo) -> StorageContainer:
        select = len(self.cluster.cluster_nodes)
        placement_rule = f"REP {select - 1} CBF 1 SELECT {select} FROM *"
        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)
        return StorageContainer(storage_cont_info, self.shell, self.cluster)

    @reporter.step("Create object and delete after test")
    @pytest.fixture(scope="class")
    def storage_objects(
        self,
        request: FixtureRequest,
        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("Select random node to stop and start it after test")
    @pytest.fixture
    def node_to_stop(
        self, node_under_test: ClusterNode, cluster_state_controller: ClusterStateController
    ) -> ClusterNode:
        yield node_under_test
        with reporter.step(f"start {node_under_test.storage_node}"):
            cluster_state_controller.start_stopped_hosts()

    @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)

        assert len(errors_get) == 0, f"Get failed - {errors_get}"
        return corrupted_objects

    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,
            )

    @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]
        )
        return object_info, object_nodes

    @pytest.fixture()
    def up_stop_nodes(self, cluster_state_controller: ClusterStateController):
        yield
        cluster_state_controller.start_stopped_hosts()

    @allure.title("Full shutdown node")
    @pytest.mark.parametrize("containers, storage_objects", [(5, 10)], indirect=True)
    def test_complete_node_shutdown(
        self,
        storage_objects: list[StorageObjectInfo],
        node_to_stop: ClusterNode,
        cluster_state_controller: ClusterStateController,
    ):
        with reporter.step(f"Remove {node_to_stop} from the list of nodes"):
            alive_nodes = list(set(self.cluster.cluster_nodes) - {node_to_stop})

        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(f"Stop node"):
            cluster_state_controller.stop_node_host(node=node_to_stop, mode="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

        with reporter.step(f"check {node_to_stop.storage_node} in map"):
            self.wait_node_in_map(node_to_stop.storage_node, self.shell, alive_node=storage_nodes[0])

        count_tick_epoch = int(alive_nodes[0].ir_node.get_netmap_cleaner_threshold()) + 4

        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_to_stop} in not map"):
            self.wait_node_not_in_map(node_to_stop.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

    @allure.title("Temporarily disable a node")
    @pytest.mark.parametrize("containers, storage_objects", [(5, 10)], indirect=True)
    def test_temporarily_disable_a_node(
        self,
        storage_objects: list[StorageObjectInfo],
        node_to_stop: ClusterNode,
        cluster_state_controller: ClusterStateController,
    ):
        with reporter.step(f"Remove {node_to_stop} from the list of nodes"):
            storage_nodes = list(set(self.cluster.storage_nodes) - {node_to_stop.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(f"Stop node"):
            cluster_state_controller.stop_node_host(node_to_stop, "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

        with reporter.step(f"Check {node_to_stop} in map"):
            self.wait_node_in_map(node_to_stop.storage_node, self.shell, alive_node=storage_nodes[0])

        cluster_state_controller.start_node_host(node_to_stop)

        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

    @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(
        self,
        object_and_nodes: tuple[StorageObjectInfo, list[ClusterNode]],
        default_wallet: WalletInfo,
        cluster_state_controller: ClusterStateController,
        simple_file: str,
        up_stop_nodes: None,
    ):
        object_info, object_nodes = object_and_nodes
        endpoint_without_object = list(set(self.cluster.cluster_nodes) - set(object_nodes))[
            0
        ].storage_node.get_rpc_endpoint()
        endpoint_with_object = object_nodes[0].storage_node.get_rpc_endpoint()

        with reporter.step("Stop all nodes with object except first one"):
            parallel(cluster_state_controller.stop_node_host, object_nodes[1:], mode="hard")

        with reporter.step(f"Get object from node without object"):
            get_object(default_wallet, object_info.cid, object_info.oid, self.shell, endpoint_without_object)

        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 pytest.raises(RuntimeError):
                put_object(default_wallet, simple_file, object_info.cid, self.shell, endpoint_with_object)

    @allure.title("Not enough nodes in the container with policy - 'REP 2 CBF 2 SELECT 4 FROM *'")
    def test_not_enough_nodes_in_container_rep_2(
        self,
        default_wallet: WalletInfo,
        cluster_state_controller: ClusterStateController,
        simple_file: str,
        up_stop_nodes: None,
    ):
        with reporter.step("Create container with full network map"):
            node_count = len(self.cluster.cluster_nodes)
            placement_rule = f"REP {node_count - 2} IN X CBF 2 SELECT {node_count} FROM * AS X"
            cid = create_container(
                default_wallet,
                self.shell,
                self.cluster.default_rpc_endpoint,
                rule=placement_rule,
                basic_acl=PUBLIC_ACL,
            )

        with reporter.step("Put object"):
            oid = put_object(default_wallet, simple_file, cid, self.shell, self.cluster.default_rpc_endpoint)

        with reporter.step("Search nodes with object"):
            object_nodes = get_object_nodes(self.cluster, cid, oid, self.cluster.cluster_nodes[0])

        with reporter.step("Choose node to stop"):
            node_to_stop = random.choice(object_nodes)
            alive_node_with_object = random.choice(list(set(object_nodes) - {node_to_stop}))
            alive_endpoint_with_object = alive_node_with_object.storage_node.get_rpc_endpoint()

        with reporter.step("Stop random node with object"):
            cluster_state_controller.stop_node_host(node_to_stop, "hard")

        with reporter.step("Put object to alive node with object"):
            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)

        with reporter.step("Create container on alive node"):
            create_container(
                default_wallet,
                self.shell,
                alive_endpoint_with_object,
                rule=placement_rule,
                basic_acl=PUBLIC_ACL,
            )