Add failover tests for storage nodes

The tests are aimed to work on cloud infrastructure (SberCloud VMs)

Signed-off-by: Vladimir Avdeev <v.avdeev@yadro.com>

pytest_tests/helpers/sbercloud_helper.py

@@ -0,0 +1,110 @@
+import json
+from dataclasses import dataclass
+
+import requests
+from yaml import FullLoader
+from yaml import load as yaml_load
+
+
+@dataclass
+class SberCloudCtx:
+    sber_login: str = None
+    sber_password: str = None
+    sber_domain: str = None
+    sber_project_id: str = None
+    sber_iam_url: str = None
+    sber_ecss: list = None
+
+    @staticmethod
+    def from_dict(sbercloud_dict: dict) -> 'SberCloudCtx':
+        return SberCloudCtx(**sbercloud_dict)
+
+    @staticmethod
+    def from_yaml(config: str) -> 'SberCloudCtx':
+        with open(config) as yaml_file:
+            config_from_yaml = yaml_load(yaml_file, Loader=FullLoader)
+        return SberCloudCtx.from_dict(config_from_yaml)
+
+
+class SberCloud:
+    def __init__(self, config: str):
+        self.sbercloud_config = SberCloudCtx().from_yaml(config)
+        self.ecs_url = None
+        self.project_id = None
+        self.token = None
+        self.update_token()
+        self.ecss = self.get_ecss()
+
+    def update_token(self):
+        data = {
+            'auth': {
+                'identity': {
+                    'methods': ['password'],
+                    'password': {
+                        'user': {
+                            'domain': {
+                                'name': self.sbercloud_config.sber_domain
+                            },
+                            'name': self.sbercloud_config.sber_login,
+                            'password': self.sbercloud_config.sber_password
+                        }
+                    }
+                },
+                'scope': {
+                    'project': {
+                        'id': self.sbercloud_config.sber_project_id
+                    }
+                }
+            }
+        }
+        response = requests.post(f'{self.sbercloud_config.sber_iam_url}/v3/auth/tokens', data=json.dumps(data),
+                                 headers={'Content-Type': 'application/json'})
+        self.ecs_url = [catalog['endpoints'][0]['url']
+                        for catalog in response.json()['token']['catalog'] if catalog['type'] == 'ecs'][0]
+        self.project_id = self.ecs_url.split('/')[-1]
+        self.token = response.headers['X-Subject-Token']
+
+    def find_esc_by_ip(self, ip: str, update: bool = False) -> str:
+        if not self.ecss or update:
+            self.ecss = self.get_ecss()
+        ecss = [ecs for ecs in self.ecss if ip in [
+                ecs_ip['addr'] for ecs_ip in [ecs_ip for ecs_ips in ecs['addresses'].values() for ecs_ip in ecs_ips]]]
+        assert len(ecss) == 1
+        return ecss[0]['id']
+
+    def get_ecss(self) -> [dict]:
+        response = requests.get(f'{self.ecs_url}/cloudservers/detail',
+                                headers={'X-Auth-Token': self.token}).json()
+        return response['servers']
+
+    def start_node(self, node_id: str = None, node_ip: str = None):
+        data = {
+            'os-start': {
+                'servers': [
+                    {
+                        'id': node_id or self.find_esc_by_ip(node_ip)
+                    }
+                ]
+            }
+        }
+        response = requests.post(f'{self.ecs_url}/cloudservers/action',
+                                 data=json.dumps(data),
+                                 headers={'Content-Type': 'application/json', 'X-Auth-Token': self.token})
+        assert response.status_code < 300, f'Status:{response.status_code}. Server not started: {response.json()}'
+
+    def stop_node(self, node_id: str = None, node_ip: str = None, hard: bool = False):
+        data = {
+            'os-stop': {
+                'type': 'HARD' if hard else 'SOFT',
+                'servers': [
+                    {
+                        'id': node_id or self.find_esc_by_ip(node_ip)
+                    }
+
+                ]
+            }
+        }
+        response = requests.post(f'{self.ecs_url}/cloudservers/action',
+                                 data=json.dumps(data),
+                                 headers={'Content-Type': 'application/json', 'X-Auth-Token': self.token})
+        assert response.status_code < 300, f'Status:{response.status_code}. Server not stopped: {response.json()}'

pytest_tests/testsuites/conftest.py

@@ -49,6 +49,7 @@ def init_wallet_with_address():
 @allure.title('Prepare tmp directory')
 def prepare_tmp_dir():
     full_path = f'{os.getcwd()}/{ASSETS_DIR}'
+    shutil.rmtree(full_path, ignore_errors=True)
     os.mkdir(full_path)
     yield
     shutil.rmtree(full_path)

pytest_tests/testsuites/failovers/test_failover_storage.py

@@ -0,0 +1,104 @@
+import logging
+import os
+from time import sleep
+
+import allure
+import pytest
+from common import STORAGE_NODE_PWD, STORAGE_NODE_USER
+from python_keywords.container import create_container
+from python_keywords.neofs_verbs import get_object, put_object
+from python_keywords.utility_keywords import get_file_hash
+from sbercloud_helper import SberCloud
+from ssh_helper import HostClient, HostIsNotAvailable
+from storage_policy import get_nodes_with_object
+from wellknown_acl import PUBLIC_ACL
+
+logger = logging.getLogger('NeoLogger')
+stopped_hosts = []
+
+
+@pytest.fixture(scope='session')
+def sbercloud_client():
+    yield SberCloud(f'{os.getcwd()}/configuration/sbercloud.yaml')
+
+
+@pytest.fixture(scope='session', autouse=True)
+def return_all_storage_nodes_fixture(sbercloud_client):
+    yield
+    return_all_storage_nodes(sbercloud_client)
+
+
+@allure.title('Hard reboot host via magic SysRq option')
+def panic_reboot_host(ip: str = None):
+    ssh = HostClient(ip=ip, login=STORAGE_NODE_USER, password=STORAGE_NODE_PWD)
+    ssh.create_connection(attempts=1)
+    ssh.exec('echo 1 > /proc/sys/kernel/sysrq')
+    with pytest.raises(HostIsNotAvailable):
+        ssh.exec('echo b > /proc/sysrq-trigger', timeout=1)
+
+
+def return_all_storage_nodes(sbercloud_client: SberCloud):
+    for host in stopped_hosts:
+        sbercloud_client.start_node(node_ip=host.split(':')[-2])
+        stopped_hosts.remove(host)
+
+
+def wait_object_replication(wallet, cid, oid, expected_copies: int) -> [str]:
+    sleep_interval, attempts = 10, 12
+    nodes = []
+    for __attempt in range(attempts):
+        nodes = get_nodes_with_object(wallet, cid, oid)
+        if len(nodes) == expected_copies:
+            return nodes
+        sleep(sleep_interval)
+    raise AssertionError(f'Expected {expected_copies} copies of object, but found {len(nodes)} ')
+
+
+@allure.title('Lost and return nodes')
+@pytest.mark.parametrize('hard_reboot', [True, False])
+def test_lost_storage_node(prepare_wallet_and_deposit, generate_file, sbercloud_client: SberCloud, hard_reboot: bool):
+    wallet, _ = prepare_wallet_and_deposit
+    placement_rule = 'REP 2 IN X CBF 2 SELECT 2 FROM * AS X'
+    cid = create_container(wallet, rule=placement_rule, basic_acl=PUBLIC_ACL)
+    oid = put_object(wallet, generate_file, cid)
+    nodes = wait_object_replication(wallet, cid, oid, 2)
+
+    new_nodes = []
+    for node in nodes:
+        with allure.step(f'Stop storage node {node}'):
+            sbercloud_client.stop_node(node_ip=node.split(':')[-2], hard=hard_reboot)
+        new_nodes = wait_object_replication(wallet, cid, oid, 2)
+
+    assert not [node for node in nodes if node in new_nodes]
+    got_file_path = get_object(wallet, cid, oid)
+    assert get_file_hash(generate_file) == get_file_hash(got_file_path)
+
+    with allure.step(f'Return storage nodes'):
+        return_all_storage_nodes(sbercloud_client)
+
+    wait_object_replication(wallet, cid, oid, 2)
+
+    got_file_path = get_object(wallet, cid, oid)
+    assert get_file_hash(generate_file) == get_file_hash(got_file_path)
+
+
+@allure.title('Panic storage node(s)')
+@pytest.mark.parametrize('sequence', [True, False])
+def test_panic_storage_node(prepare_wallet_and_deposit, generate_file, sequence: bool):
+    wallet, _ = prepare_wallet_and_deposit
+    placement_rule = 'REP 2 IN X CBF 2 SELECT 2 FROM * AS X'
+    cid = create_container(wallet, rule=placement_rule, basic_acl=PUBLIC_ACL)
+    oid = put_object(wallet, generate_file, cid)
+
+    with allure.step(f'Return storage nodes'):
+        nodes = wait_object_replication(wallet, cid, oid, 2)
+        for node in nodes:
+            panic_reboot_host(ip=node.split(':')[-2])
+            if sequence:
+                wait_object_replication(wallet, cid, oid, 2)
+
+        if not sequence:
+            wait_object_replication(wallet, cid, oid, 2)
+
+        got_file_path = get_object(wallet, cid, oid)
+        assert get_file_hash(generate_file) == get_file_hash(got_file_path)

robot/resources/lib/python_keywords/aws_cli_client.py

@@ -89,7 +89,7 @@ class AwsCliClient:
         version = f' --version-id {VersionId}' if VersionId else ''
         cmd = f' aws --no-verify-ssl s3api get-object --bucket {Bucket} ' \
               f'--key {Key} {version} {file_path} --endpoint {S3_GATE}'
-        output = _cmd_run(cmd)
+        output = _cmd_run(cmd, timeout=90)
         return self._to_json(output)
 
     def delete_objects(self, Bucket: str, Delete: dict) -> dict: