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

import logging
import os
import random
from time import sleep

import allure
import pytest
from frostfs_testlib.healthcheck.interfaces import Healthcheck
from frostfs_testlib.resources.wellknown_acl import PUBLIC_ACL
from frostfs_testlib.steps.cli.container import create_container
from frostfs_testlib.steps.cli.object import (
    get_object,
    get_object_nodes,
    neo_go_query_height,
    put_object,
    put_object_to_random_node,
)
from frostfs_testlib.steps.storage_object import delete_objects
from frostfs_testlib.storage.cluster import ClusterNode
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 Interfaces, StorageObjectInfo
from frostfs_testlib.testing.cluster_test_base import ClusterTestBase
from frostfs_testlib.testing.parallel import parallel
from frostfs_testlib.utils.failover_utils import wait_object_replication
from frostfs_testlib.utils.file_utils import generate_file, get_file_hash

logger = logging.getLogger("NeoLogger")
STORAGE_NODE_COMMUNICATION_PORT = "8080"
STORAGE_NODE_COMMUNICATION_PORT_TLS = "8082"
PORTS_TO_BLOCK = [STORAGE_NODE_COMMUNICATION_PORT, STORAGE_NODE_COMMUNICATION_PORT_TLS]
blocked_nodes: list[ClusterNode] = []
file_wait_delete = []

OBJECT_ATTRIBUTES = [
    None,
    {"key1": 1, "key2": "abc", "common_key": "common_value"},
    {"key1": 2, "common_key": "common_value"},
]


@pytest.mark.failover
@pytest.mark.failover_network
class TestFailoverNetwork(ClusterTestBase):
    @pytest.fixture(autouse=True)
    @allure.title("Restore network")
    def restore_network(self, healthcheck: Healthcheck, cluster_state_controller: ClusterStateController):
        yield
        with allure.step(f"Count blocked nodes {len(blocked_nodes)}"):
            not_empty = len(blocked_nodes) != 0
        for node in list(blocked_nodes):
            with allure.step(f"Restore network for {node}"):
                cluster_state_controller.restore_traffic(mode="ports", node=node)
            blocked_nodes.remove(node)
        if not_empty:
            parallel(healthcheck.storage_healthcheck, self.cluster.cluster_nodes)

    @pytest.fixture()
    @allure.title("Restore drop traffic to system")
    def restore_down_interfaces(self, cluster_state_controller: ClusterStateController):
        yield
        cluster_state_controller.restore_interfaces()

    @pytest.fixture()
    @allure.title("Delete file after test")
    def delete_file_after_test(self) -> None:
        yield
        for path in file_wait_delete:
            os.remove(path)
        file_wait_delete.clear()

    @pytest.fixture()
    def storage_objects(
        self,
        simple_object_size: ObjectSize,
        default_wallet: str,
    ) -> list[StorageObjectInfo]:

        file_path = generate_file(simple_object_size.value)
        file_hash = get_file_hash(file_path)

        with allure.step("Create container"):
            placement_rule = "REP 1 CBF 1"
            cid = create_container(
                wallet=default_wallet,
                shell=self.shell,
                endpoint=self.cluster.default_rpc_endpoint,
                rule=placement_rule,
                await_mode=True,
            )

        storage_objects = []

        with allure.step("Put object"):
            for attribute in OBJECT_ATTRIBUTES:
                oid = put_object_to_random_node(
                    wallet=default_wallet,
                    path=file_path,
                    cid=cid,
                    shell=self.shell,
                    cluster=self.cluster,
                )

            storage_object = StorageObjectInfo(cid=cid, oid=oid)
            storage_object.size = simple_object_size.value
            storage_object.wallet_file_path = default_wallet
            storage_object.file_path = file_path
            storage_object.file_hash = file_hash
            storage_object.attributes = attribute

            storage_objects.append(storage_object)

        yield storage_objects

        delete_objects(storage_objects, self.shell, self.cluster)

    @allure.title("Block Storage node traffic")
    def test_block_storage_node_traffic(
        self,
        default_wallet: str,
        require_multiple_hosts,
        simple_object_size: ObjectSize,
        cluster_state_controller: ClusterStateController,
    ):
        """
        Block storage nodes traffic using iptables and wait for replication for objects.
        """
        wallet = default_wallet
        placement_rule = "REP 2 IN X CBF 2 SELECT 2 FROM * AS X"
        wakeup_node_timeout = 10  # timeout to let nodes detect that traffic has blocked
        nodes_to_block_count = 2

        source_file_path = generate_file(simple_object_size.value)
        cid = create_container(
            wallet,
            shell=self.shell,
            endpoint=self.cluster.default_rpc_endpoint,
            rule=placement_rule,
            basic_acl=PUBLIC_ACL,
        )
        oid = put_object_to_random_node(wallet, source_file_path, cid, shell=self.shell, cluster=self.cluster)

        nodes = wait_object_replication(cid, oid, 2, shell=self.shell, nodes=self.cluster.storage_nodes)

        logger.info(f"Nodes are {nodes}")
        nodes_to_block = nodes
        if nodes_to_block_count > len(nodes):
            # TODO: the intent of this logic is not clear, need to revisit
            nodes_to_block = random.choices(nodes, k=2)

        excluded_nodes = []
        for node in nodes_to_block:
            with allure.step(f"Block incoming traffic at node {node} on port {PORTS_TO_BLOCK}"):
                block_node = [
                    cluster_node for cluster_node in self.cluster.cluster_nodes if cluster_node.storage_node == node
                ]
                blocked_nodes.append(*block_node)
                excluded_nodes.append(node)
                cluster_state_controller.drop_traffic(
                    mode="ports",
                    node=node,
                    wakeup_timeout=wakeup_node_timeout,
                    ports=PORTS_TO_BLOCK,
                )

            with allure.step(f"Check object is not stored on node {node}"):
                new_nodes = wait_object_replication(
                    cid,
                    oid,
                    2,
                    shell=self.shell,
                    nodes=list(set(self.cluster.storage_nodes) - set(excluded_nodes)),
                )
                assert node not in new_nodes

            with allure.step("Check object data is not corrupted"):
                got_file_path = get_object(wallet, cid, oid, endpoint=new_nodes[0].get_rpc_endpoint(), shell=self.shell)
                assert get_file_hash(source_file_path) == get_file_hash(got_file_path)

        for node in nodes_to_block:
            with allure.step(f"Unblock incoming traffic at host {node} on port {PORTS_TO_BLOCK}"):
                cluster_state_controller.restore_traffic(mode="ports", node=node)
                block_node = [
                    cluster_node for cluster_node in self.cluster.cluster_nodes if cluster_node.storage_node == node
                ]
                blocked_nodes.remove(*block_node)
                sleep(wakeup_node_timeout)

        with allure.step("Check object data is not corrupted"):
            new_nodes = wait_object_replication(cid, oid, 2, shell=self.shell, nodes=self.cluster.storage_nodes)

            got_file_path = get_object(wallet, cid, oid, shell=self.shell, endpoint=new_nodes[0].get_rpc_endpoint())
            assert get_file_hash(source_file_path) == get_file_hash(got_file_path)

    @pytest.mark.interfaces
    @allure.title("Block DATA interface node")
    def test_block_data_interface(
        self,
        cluster_state_controller: ClusterStateController,
        default_wallet: str,
        restore_down_interfaces: None,
        delete_file_after_test: None,
        storage_objects: list[StorageObjectInfo],
    ):
        storage_object = storage_objects[0]

        with allure.step("Search nodes with object"):
            nodes_with_object = get_object_nodes(
                cluster=self.cluster,
                wallet=default_wallet,
                cid=storage_object.cid,
                oid=storage_object.oid,
                shell=self.shell,
                endpoint=self.cluster.default_rpc_endpoint,
            )

        with allure.step("Get data interface to node"):
            config_interfaces = list(nodes_with_object[0].host.config.interfaces.keys())
            with allure.step(f"Get data in {config_interfaces}"):
                data_interfaces = [interface for interface in config_interfaces if "data" in interface]

        with allure.step("Block data interfaces for node"):
            for interface in data_interfaces:
                cluster_state_controller.down_interface(nodes=nodes_with_object, interface=interface)

        with allure.step("Tick epoch and wait 2 block"):
            nodes_without_an_object = list(set(self.cluster.cluster_nodes) - set(nodes_with_object))
            self.tick_epochs(1, alive_node=nodes_without_an_object[0].storage_node, wait_block=2)

        with allure.step("Get object for target nodes to data interfaces, expect false"):
            with pytest.raises(RuntimeError, match="return code: 1"):
                get_object(
                    wallet=default_wallet,
                    cid=storage_object.cid,
                    oid=storage_object.oid,
                    shell=self.shell,
                    endpoint=nodes_with_object[0].storage_node.get_rpc_endpoint(),
                )

        with allure.step(f"Get object others nodes, expect true"):
            input_file = get_object(
                wallet=default_wallet,
                cid=storage_object.cid,
                oid=storage_object.oid,
                shell=self.shell,
                endpoint=nodes_without_an_object[0].storage_node.get_rpc_endpoint(),
            )
            file_wait_delete.append(input_file)

        with allure.step("Restore interface and tick 1 epoch, wait 2 block"):
            cluster_state_controller.restore_interfaces()
            self.tick_epochs(1, alive_node=nodes_without_an_object[0].storage_node, wait_block=2)

    @pytest.mark.interfaces
    @allure.title("Block INTERNAL interface node")
    def test_block_internal_interface(
        self,
        cluster_state_controller: ClusterStateController,
        default_wallet: str,
        restore_down_interfaces: None,
        delete_file_after_test: None,
        storage_objects: list[StorageObjectInfo],
        simple_object_size: ObjectSize,
    ):
        storage_object = storage_objects[0]

        with allure.step("Search nodes with object"):
            nodes_with_object = get_object_nodes(
                cluster=self.cluster,
                wallet=default_wallet,
                cid=storage_object.cid,
                oid=storage_object.oid,
                shell=self.shell,
                endpoint=self.cluster.default_rpc_endpoint,
            )

        with allure.step("Get internal interface to node"):
            config_interfaces = list(nodes_with_object[0].host.config.interfaces.keys())
            with allure.step(f"Get internal in {config_interfaces}"):
                internal_interfaces = [interface for interface in config_interfaces if "internal" in interface]

        with allure.step("Block internal interfaces for node"):
            for interface in internal_interfaces:
                cluster_state_controller.down_interface(nodes=nodes_with_object, interface=interface)

        with allure.step("Tick epoch and wait 2 block"):
            nodes_without_an_object = list(set(self.cluster.cluster_nodes) - set(nodes_with_object))
            self.tick_epochs(1, alive_node=nodes_without_an_object[0].storage_node, wait_block=2)

        with allure.step("Get object others node, expect false"):
            with pytest.raises(RuntimeError, match="return code: 1"):
                get_object(
                    wallet=default_wallet,
                    cid=storage_object.cid,
                    oid=storage_object.oid,
                    shell=self.shell,
                    endpoint=nodes_without_an_object[0].storage_node.get_rpc_endpoint(),
                )

        with allure.step("Put object, others node, expect false"):
            with pytest.raises(RuntimeError, match="return code: 1"):
                put_object(
                    wallet=default_wallet,
                    path=storage_object.file_path,
                    cid=storage_object.cid,
                    shell=self.shell,
                    endpoint=nodes_without_an_object[0].storage_node.get_rpc_endpoint(),
                )

        with allure.step(f"Get object nodes with object, expect true"):
            input_file = get_object(
                wallet=default_wallet,
                cid=storage_object.cid,
                oid=storage_object.oid,
                shell=self.shell,
                endpoint=nodes_with_object[0].storage_node.get_rpc_endpoint(),
            )
            file_wait_delete.append(input_file)

        with allure.step(f"Put object nodes with object, expect true"):
            temp_file_path = generate_file(simple_object_size.value)
            _ = put_object(
                wallet=default_wallet,
                path=temp_file_path,
                cid=storage_object.cid,
                shell=self.shell,
                endpoint=nodes_with_object[0].storage_node.get_rpc_endpoint(),
            )
            file_wait_delete.append(temp_file_path)
        with allure.step("Restore interface and tick 1 epoch, wait 2 block"):
            cluster_state_controller.restore_interfaces()
            self.tick_epochs(1, alive_node=nodes_without_an_object[0].storage_node, wait_block=2)

    @pytest.mark.interfaces
    @pytest.mark.failover_baremetal
    @pytest.mark.parametrize(
        "block_interface, other_interface",
        [(Interfaces.DATA_O, Interfaces.DATA_1), (Interfaces.DATA_1, Interfaces.DATA_O)],
    )
    @allure.title("Down data interfaces to all nodes(interface={block_interface})")
    def test_down_data_interface(
        self,
        require_multiple_interfaces,
        cluster_state_controller: ClusterStateController,
        default_wallet: str,
        simple_object_size: ObjectSize,
        delete_file_after_test: None,
        restore_down_interfaces: None,
        block_interface: Interfaces,
        other_interface: Interfaces,
    ):
        cluster_nodes = self.cluster.cluster_nodes
        with allure.step(f"Block {block_interface.value} interfaces"):
            cluster_state_controller.down_interface(cluster_nodes, block_interface.value)

        with allure.step("Tick 1 epoch and wait 2 block for sync all nodes"):
            self.tick_epochs(1, alive_node=cluster_nodes[0].storage_node, wait_block=2)

        with allure.step("Create container"):
            cid = create_container(
                wallet=default_wallet,
                shell=self.shell,
                endpoint=f"{cluster_nodes[0].get_data_interface(other_interface.value)[0]}:8080",
                rule="REP 4 CBF 1",
            )

        with allure.step("Put object"):
            file_path = generate_file(simple_object_size.value)
            file_wait_delete.append(file_path)

            oid = put_object(
                wallet=default_wallet,
                path=file_path,
                cid=cid,
                shell=self.shell,
                endpoint=f"{cluster_nodes[0].get_data_interface(other_interface.value)[0]}:8080",
            )

        with allure.step("Get object"):
            file_get_path = get_object(
                wallet=default_wallet,
                cid=cid,
                oid=oid,
                shell=self.shell,
                endpoint=f"{cluster_nodes[0].get_data_interface(other_interface.value)[0]}:8080",
            )
            file_wait_delete.append(file_get_path)

        with allure.step("Restore interfaces all nodes"):
            cluster_state_controller.restore_interfaces()
            self.tick_epochs(1, alive_node=cluster_nodes[0].storage_node, wait_block=2)

    @pytest.mark.interfaces
    @pytest.mark.failover_baremetal
    @pytest.mark.parametrize("interface", [Interfaces.INTERNAL_0, Interfaces.INTERNAL_1])
    @allure.title("Down internal interfaces to all nodes(interface={interface})")
    def test_down_internal_interface(
        self,
        require_multiple_interfaces,
        cluster_state_controller: ClusterStateController,
        default_wallet: str,
        simple_object_size: ObjectSize,
        delete_file_after_test: None,
        restore_down_interfaces: None,
        interface: Interfaces,
    ):
        cluster_nodes = self.cluster.cluster_nodes
        latest_block = {}

        with allure.step("Get block all nodes"):
            for cluster_node in cluster_nodes:
                latest_block[cluster_node] = neo_go_query_height(
                    shell=cluster_node.host.get_shell(), endpoint=cluster_node.morph_chain.get_http_endpoint()
                )

        with allure.step(f"Block {interface} interfaces"):
            cluster_state_controller.down_interface(cluster_nodes, interface.value)

        with allure.step("Tick 1 epoch and wait 2 block for sync all nodes"):
            self.tick_epochs(1, alive_node=cluster_nodes[0].storage_node, wait_block=2)

        with allure.step("Create container"):
            cid = create_container(
                wallet=default_wallet,
                shell=self.shell,
                endpoint=self.cluster.default_rpc_endpoint,
                rule="REP 4 CBF 1",
            )

        with allure.step(f"Put object, after down {interface}"):
            file_path = generate_file(simple_object_size.value)
            file_wait_delete.append(file_path)

            oid = put_object(
                wallet=default_wallet,
                path=file_path,
                cid=cid,
                shell=self.shell,
                endpoint=self.cluster.default_rpc_endpoint,
            )

        with allure.step("Get object"):
            file_get_path = get_object(
                wallet=default_wallet,
                cid=cid,
                oid=oid,
                shell=self.shell,
                endpoint=self.cluster.default_rpc_endpoint,
            )
            file_wait_delete.append(file_get_path)

        now_block = {}

        with allure.step("Get actual block"):
            for cluster_node in cluster_nodes:
                now_block[cluster_node] = neo_go_query_height(
                    shell=cluster_node.host.get_shell(), endpoint=cluster_node.morph_chain.get_http_endpoint()
                )
        with allure.step(f"Compare block"):
            for cluster_node, items in now_block.items():
                with allure.step(
                    f"Node - {cluster_node.host_ip}, old block - {latest_block[cluster_node]['Latest block']}, "
                    f"now block - {now_block[cluster_node]['Latest block']}"
                ):
                    assert latest_block[cluster_node]["Latest block"] < now_block[cluster_node]["Latest block"]

        with allure.step("Restore interfaces all nodes"):
            cluster_state_controller.restore_interfaces()
            self.tick_epochs(1, alive_node=self.cluster.cluster_nodes[0].storage_node, wait_block=2)