def test_multi_node(self):
     base_port = Configuration.get('ovs.ports.arakoon')[0]
     cluster = 'one'
     nodes = sorted(TestArakoonInstaller.nodes.keys())
     nodes = dict((node, SSHClient(node)) for node in nodes)
     first_node = nodes.keys()[0]
     ArakoonInstaller.create_cluster(cluster, first_node, [])
     for node in nodes[1:]:
         ArakoonInstaller.extend_cluster(first_node, node, cluster, [])
     expected = TestArakoonInstaller.expected_global.format(cluster, ','.join(TestArakoonInstaller.nodes[node] for node in nodes))
     for node in nodes:
         expected += TestArakoonInstaller.expected_base.format(TestArakoonInstaller.nodes[node], node, base_port, base_port + 1)
     expected = expected.strip()
     for node, client in nodes.iteritems():
         contents = client.file_read(self._get_config_path(cluster))
         self.assertEqual(contents.strip(), expected.strip())
     ArakoonInstaller.shrink_cluster(nodes[1], first_node, cluster)
     expected = TestArakoonInstaller.expected_global.format(cluster, ','.join(TestArakoonInstaller.nodes[node] for node in nodes[1:]))
     for node in nodes.keys()[1:]:
         expected += TestArakoonInstaller.expected_base.format(TestArakoonInstaller.nodes[node], node, base_port, base_port + 1)
     expected = expected.strip()
     for node, client in nodes.iteritems():
         if node == first_node:
             continue
         contents = client.file_read(self._get_config_path(cluster))
         self.assertEqual(contents.strip(), expected.strip())
 def on_demote(cluster_ip, master_ip, offline_node_ips=None):
     """
     Handles the demote for the StorageDrivers
     :param cluster_ip: IP of the node to demote
     :param master_ip: IP of the master node
     :param offline_node_ips: IPs of nodes which are offline
     """
     if offline_node_ips is None:
         offline_node_ips = []
     client = SSHClient(cluster_ip, username='******') if cluster_ip not in offline_node_ips else None
     servicetype = ServiceTypeList.get_by_name('Arakoon')
     current_service = None
     remaining_ips = []
     for service in servicetype.services:
         if service.name == 'arakoon-voldrv':
             if service.storagerouter.ip == cluster_ip:
                 current_service = service
             elif service.storagerouter.ip not in offline_node_ips:
                 remaining_ips.append(service.storagerouter.ip)
     if current_service is not None:
         print '* Shrink StorageDriver cluster'
         ArakoonInstaller.shrink_cluster(master_ip, cluster_ip, 'voldrv', offline_node_ips)
         if client is not None and ServiceManager.has_service(current_service.name, client=client) is True:
             ServiceManager.stop_service(current_service.name, client=client)
             ServiceManager.remove_service(current_service.name, client=client)
         ArakoonInstaller.restart_cluster_remove('voldrv', remaining_ips)
         current_service.delete()
         StorageDriverController._configure_arakoon_to_volumedriver(offline_node_ips)
Ejemplo n.º 3
0
 def on_demote(cluster_ip, master_ip, offline_node_ips=None):
     """
     Handles the demote for the StorageDrivers
     :param cluster_ip: IP of the node to demote
     :type cluster_ip: str
     :param master_ip: IP of the master node
     :type master_ip: str
     :param offline_node_ips: IPs of nodes which are offline
     :type offline_node_ips: list
     :return: None
     """
     _ = master_ip
     if offline_node_ips is None:
         offline_node_ips = []
     servicetype = ServiceTypeList.get_by_name(ServiceType.SERVICE_TYPES.ARAKOON)
     current_service = None
     remaining_ips = []
     for service in servicetype.services:
         if service.name == 'arakoon-voldrv' and service.is_internal is True:  # Externally managed arakoon cluster service does not have storage router
             if service.storagerouter.ip == cluster_ip:
                 current_service = service
             elif service.storagerouter.ip not in offline_node_ips:
                 remaining_ips.append(service.storagerouter.ip)
     if current_service is not None:
         if len(remaining_ips) == 0:
             raise RuntimeError('Could not find any remaining arakoon nodes for the voldrv cluster')
         StorageDriverController._logger.debug('* Shrink StorageDriver cluster')
         cluster_name = str(Configuration.get('/ovs/framework/arakoon_clusters|voldrv'))
         ArakoonInstaller.shrink_cluster(deleted_node_ip=cluster_ip,
                                         remaining_node_ips=remaining_ips,
                                         cluster_name=cluster_name,
                                         offline_nodes=offline_node_ips)
         current_service.delete()
         StorageDriverController._configure_arakoon_to_volumedriver(cluster_name=cluster_name)
Ejemplo n.º 4
0
 def on_demote(cluster_ip, master_ip):
     """
     Handles the demote for the StorageDrivers
     :param cluster_ip: IP of the node to demote
     :param master_ip: IP of the master node
     """
     client = SSHClient(cluster_ip, username='******')
     servicetype = ServiceTypeList.get_by_name('Arakoon')
     current_service = None
     remaining_ips = []
     for service in servicetype.services:
         if service.name == 'arakoon-voldrv':
             if service.storagerouter.ip == cluster_ip:
                 current_service = service
             else:
                 remaining_ips.append(service.storagerouter.ip)
     if current_service is not None:
         print '* Shrink StorageDriver cluster'
         ArakoonInstaller.shrink_cluster(master_ip, cluster_ip, 'voldrv')
         if ServiceManager.has_service(current_service.name, client=client) is True:
             ServiceManager.stop_service(current_service.name, client=client)
             ServiceManager.remove_service(current_service.name, client=client)
         ArakoonInstaller.restart_cluster_remove('voldrv', remaining_ips)
         current_service.delete()
         for storagerouter in StorageRouterList.get_storagerouters():
             ArakoonInstaller.deploy_to_slave(master_ip, storagerouter.ip, 'voldrv')
         StorageDriverController._configure_arakoon_to_volumedriver()
Ejemplo n.º 5
0
 def on_demote(cluster_ip, master_ip, offline_node_ips=None):
     """
     Handles the demote for the StorageDrivers
     :param cluster_ip: IP of the node to demote
     :param master_ip: IP of the master node
     :param offline_node_ips: IPs of nodes which are offline
     """
     _ = master_ip
     if offline_node_ips is None:
         offline_node_ips = []
     client = SSHClient(
         cluster_ip,
         username='******') if cluster_ip not in offline_node_ips else None
     servicetype = ServiceTypeList.get_by_name('Arakoon')
     current_service = None
     remaining_ips = []
     for service in servicetype.services:
         if service.name == 'arakoon-voldrv':
             if service.storagerouter.ip == cluster_ip:
                 current_service = service
             elif service.storagerouter.ip not in offline_node_ips:
                 remaining_ips.append(service.storagerouter.ip)
     if current_service is not None:
         print '* Shrink StorageDriver cluster'
         ArakoonInstaller.shrink_cluster(cluster_ip, 'voldrv',
                                         offline_node_ips)
         if client is not None and ServiceManager.has_service(
                 current_service.name, client=client) is True:
             ServiceManager.stop_service(current_service.name,
                                         client=client)
             ServiceManager.remove_service(current_service.name,
                                           client=client)
         ArakoonInstaller.restart_cluster_remove('voldrv', remaining_ips)
         current_service.delete()
         StorageDriverController._configure_arakoon_to_volumedriver()
Ejemplo n.º 6
0
    def on_demote(cluster_ip, master_ip, offline_node_ips=None):
        """
        Handles the demote for the StorageDrivers
        :param cluster_ip: IP of the node to demote
        :type cluster_ip: str

        :param master_ip: IP of the master node
        :type master_ip: str

        :param offline_node_ips: IPs of nodes which are offline
        :type offline_node_ips: list

        :return: None
        """
        _ = master_ip
        if offline_node_ips is None:
            offline_node_ips = []
        client = SSHClient(
            cluster_ip,
            username='******') if cluster_ip not in offline_node_ips else None
        servicetype = ServiceTypeList.get_by_name(
            ServiceType.SERVICE_TYPES.ARAKOON)
        current_service = None
        remaining_ips = []
        for service in servicetype.services:
            if service.name == 'arakoon-voldrv' and service.is_internal is True:  # Externally managed arakoon cluster service does not have storage router
                if service.storagerouter.ip == cluster_ip:
                    current_service = service
                elif service.storagerouter.ip not in offline_node_ips:
                    remaining_ips.append(service.storagerouter.ip)
        if current_service is not None:
            StorageDriverController._logger.debug(
                '* Shrink StorageDriver cluster')
            cluster_name = str(
                EtcdConfiguration.get(
                    '/ovs/framework/arakoon_clusters|voldrv'))
            ArakoonInstaller.shrink_cluster(deleted_node_ip=cluster_ip,
                                            cluster_name=cluster_name,
                                            offline_nodes=offline_node_ips)
            if client is not None and ServiceManager.has_service(
                    current_service.name, client=client) is True:
                ServiceManager.stop_service(current_service.name,
                                            client=client)
                ServiceManager.remove_service(current_service.name,
                                              client=client)
            ArakoonInstaller.restart_cluster_remove(cluster_name,
                                                    remaining_ips)
            current_service.delete()
            StorageDriverController._configure_arakoon_to_volumedriver(
                cluster_name=cluster_name)
 def test_multi_node(self):
     nodes = sorted(TestArakoonInstaller.nodes.keys())
     nodes = dict((node, SSHClient(node)) for node in nodes)
     first_node = nodes.keys()[0]
     result = ArakoonInstaller.create_cluster(TestArakoonInstaller.cluster_name, first_node, "/tmp/db")
     for node in nodes.keys()[1:]:
         ArakoonInstaller.extend_cluster(first_node, node, TestArakoonInstaller.cluster_name, "/tmp/db")
     expected = TestArakoonInstaller.expected_global.format(
         ",".join(TestArakoonInstaller.nodes[node] for node in nodes), TestArakoonInstaller.cluster_name
     )
     for node in nodes:
         expected += TestArakoonInstaller.expected_base.format(
             TestArakoonInstaller.nodes[node],
             result["client_port"],
             result["messaging_port"],
             TestArakoonInstaller.cluster_name,
             node,
         )
     expected = expected.strip()
     for node, client in nodes.iteritems():
         contents = client.file_read(TestArakoonInstaller.cluster_config_file)
         self.assertEqual(contents.strip(), expected.strip())
     ArakoonInstaller.shrink_cluster(first_node, TestArakoonInstaller.cluster_name)
     expected = TestArakoonInstaller.expected_global.format(
         ",".join(TestArakoonInstaller.nodes[node] for node in nodes[1:]), TestArakoonInstaller.cluster_name
     )
     for node in nodes.keys()[1:]:
         expected += TestArakoonInstaller.expected_base.format(
             TestArakoonInstaller.nodes[node],
             result["client_port"],
             result["messaging_port"],
             TestArakoonInstaller.cluster_name,
             node,
         )
     expected = expected.strip()
     for node, client in nodes.iteritems():
         if node == first_node:
             continue
         contents = client.file_read(TestArakoonInstaller.cluster_config_file)
         self.assertEqual(contents.strip(), expected.strip())
Ejemplo n.º 8
0
    def on_demote(cluster_ip, master_ip, offline_node_ips=None):
        """
        Handles the demote for the StorageDrivers
        :param cluster_ip: IP of the node to demote
        :type cluster_ip: str

        :param master_ip: IP of the master node
        :type master_ip: str

        :param offline_node_ips: IPs of nodes which are offline
        :type offline_node_ips: list

        :return: None
        """
        _ = master_ip
        if offline_node_ips is None:
            offline_node_ips = []
        client = SSHClient(cluster_ip, username='******') if cluster_ip not in offline_node_ips else None
        servicetype = ServiceTypeList.get_by_name(ServiceType.SERVICE_TYPES.ARAKOON)
        current_service = None
        remaining_ips = []
        for service in servicetype.services:
            if service.name == 'arakoon-voldrv' and service.is_internal is True:  # Externally managed arakoon cluster service does not have storage router
                if service.storagerouter.ip == cluster_ip:
                    current_service = service
                elif service.storagerouter.ip not in offline_node_ips:
                    remaining_ips.append(service.storagerouter.ip)
        if current_service is not None:
            StorageDriverController._logger.debug('* Shrink StorageDriver cluster')
            cluster_name = str(EtcdConfiguration.get('/ovs/framework/arakoon_clusters|voldrv'))
            ArakoonInstaller.shrink_cluster(deleted_node_ip=cluster_ip,
                                            cluster_name=cluster_name,
                                            offline_nodes=offline_node_ips)
            if client is not None and ServiceManager.has_service(current_service.name, client=client) is True:
                ServiceManager.stop_service(current_service.name, client=client)
                ServiceManager.remove_service(current_service.name, client=client)
            ArakoonInstaller.restart_cluster_remove(cluster_name, remaining_ips)
            current_service.delete()
            StorageDriverController._configure_arakoon_to_volumedriver(cluster_name=cluster_name)
 def test_multi_node(self):
     nodes = sorted(TestArakoonInstaller.nodes.keys())
     nodes = dict((node, SSHClient(node)) for node in nodes)
     first_node = nodes.keys()[0]
     result = ArakoonInstaller.create_cluster(TestArakoonInstaller.cluster_name, first_node, '/tmp/db')
     for node in nodes.keys()[1:]:
         ArakoonInstaller.extend_cluster(first_node, node, TestArakoonInstaller.cluster_name, '/tmp/db')
     expected = TestArakoonInstaller.expected_global.format(','.join(TestArakoonInstaller.nodes[node] for node in nodes), TestArakoonInstaller.cluster_name)
     for node in nodes:
         expected += TestArakoonInstaller.expected_base.format(TestArakoonInstaller.nodes[node], result['client_port'], result['messaging_port'], TestArakoonInstaller.cluster_name, node)
     expected = expected.strip()
     for node, client in nodes.iteritems():
         contents = client.file_read(TestArakoonInstaller.cluster_config_file)
         self.assertEqual(contents.strip(), expected.strip())
     ArakoonInstaller.shrink_cluster(first_node, TestArakoonInstaller.cluster_name)
     expected = TestArakoonInstaller.expected_global.format(','.join(TestArakoonInstaller.nodes[node] for node in nodes[1:]), TestArakoonInstaller.cluster_name)
     for node in nodes.keys()[1:]:
         expected += TestArakoonInstaller.expected_base.format(TestArakoonInstaller.nodes[node], result['client_port'], result['messaging_port'], TestArakoonInstaller.cluster_name, node)
     expected = expected.strip()
     for node, client in nodes.iteritems():
         if node == first_node:
             continue
         contents = client.file_read(TestArakoonInstaller.cluster_config_file)
         self.assertEqual(contents.strip(), expected.strip())
Ejemplo n.º 10
0
    def test_cluster_maintenance(self):
        """
        Validates whether a cluster can be correctly created
        """
        Configuration.set('/ovs/framework/hosts/1/ports', {'arakoon': [10000, 10100]})
        Configuration.set('/ovs/framework/hosts/2/ports', {'arakoon': [20000, 20100]})

        structure = Helper.build_service_structure(
            {'storagerouters': [1, 2]}
        )
        storagerouters = structure['storagerouters']
        System._machine_id = {storagerouters[1].ip: '1',
                              storagerouters[2].ip: '2'}

        # Create new cluster
        mountpoint = storagerouters[1].disks[0].partitions[0].mountpoint
        if os.path.exists(mountpoint) and mountpoint != '/':
            shutil.rmtree(mountpoint)
        base_dir = mountpoint + '/test_create_cluster'
        info = ArakoonInstaller.create_cluster('test', ServiceType.ARAKOON_CLUSTER_TYPES.FWK, storagerouters[1].ip, base_dir)

        reality = Helper.extract_dir_structure(base_dir)
        expected = {'dirs': {'arakoon': {'dirs': {'test': {'dirs': {'tlogs': {'dirs': {},
                                                                              'files': []},
                                                                    'db': {'dirs': {},
                                                                           'files': []}},
                                                           'files': []}},
                                         'files': []}},
                    'files': []}
        self.assertDictEqual(reality, expected)
        expected = '{0}\n\n{1}\n\n'.format(ArakoonInstallerTester.EXPECTED_CLUSTER_CONFIG.format('1', 'test', ''),
                                           ArakoonInstallerTester.EXPECTED_NODE_CONFIG.format(
                                               '1', storagerouters[1].ip, 10000, base_dir, '1', 10001
                                           ))
        self.assertEqual(Configuration.get(ArakoonInstaller.CONFIG_KEY.format('test'), raw=True), expected)
        # @TODO: assert service availability here. It should be stopped

        ArakoonInstaller.start_cluster('test', storagerouters[1].ip, filesystem=False)
        # @TODO: assert the service is running

        config = ArakoonClusterConfig('test', filesystem=False)
        config.load_config(storagerouters[1].ip)
        client = ArakoonInstaller.build_client(config)
        reality = client.get(ArakoonInstaller.INTERNAL_CONFIG_KEY)
        self.assertEqual(reality, expected)
        self.assertFalse(client.exists(ArakoonInstaller.METADATA_KEY))

        ArakoonInstaller.claim_cluster('test', storagerouters[1].ip, filesystem=False, metadata=info['metadata'])

        reality = json.loads(client.get(ArakoonInstaller.METADATA_KEY))
        expected = {'cluster_name': 'test',
                    'cluster_type': 'FWK',
                    'in_use': True,
                    'internal': True}
        self.assertDictEqual(reality, expected)

        # Extending cluster
        mountpoint = storagerouters[2].disks[0].partitions[0].mountpoint
        if os.path.exists(mountpoint) and mountpoint != '/':
            shutil.rmtree(mountpoint)
        base_dir2 = mountpoint + '/test_extend_cluster'
        ArakoonInstaller.extend_cluster(storagerouters[1].ip, storagerouters[2].ip, 'test', base_dir2)
        reality = Helper.extract_dir_structure(base_dir)
        expected = {'dirs': {'arakoon': {'dirs': {'test': {'dirs': {'tlogs': {'dirs': {},
                                                                              'files': []},
                                                                    'db': {'dirs': {},
                                                                           'files': []}},
                                                           'files': []}},
                                         'files': []}},
                    'files': []}
        self.assertDictEqual(reality, expected)
        expected = '{0}\n\n{1}\n\n{2}\n\n'.format(ArakoonInstallerTester.EXPECTED_CLUSTER_CONFIG.format('1,2', 'test', ''),
                                                  ArakoonInstallerTester.EXPECTED_NODE_CONFIG.format(
                                                      '1', storagerouters[1].ip, 10000, base_dir, '1', 10001
                                                  ),
                                                  ArakoonInstallerTester.EXPECTED_NODE_CONFIG.format(
                                                      '2', storagerouters[2].ip, 20000, base_dir2, '2', 20001
                                                  ))
        self.assertEqual(Configuration.get(ArakoonInstaller.CONFIG_KEY.format('test'), raw=True), expected)
        # @TODO: assert service availability here. It should be stopped

        catchup_command = 'arakoon --node 2 -config file://opt/OpenvStorage/config/framework.json?key=/ovs/arakoon/test/config -catchup-only'
        SSHClient._run_returns[catchup_command] = None
        SSHClient._run_recordings = []
        ArakoonInstaller.restart_cluster_add('test', [storagerouters[1].ip], storagerouters[2].ip, filesystem=False)
        self.assertIn(catchup_command, SSHClient._run_recordings)
        # @TODO: assert the service is running

        config = ArakoonClusterConfig('test', filesystem=False)
        config.load_config(storagerouters[2].ip)
        client = ArakoonInstaller.build_client(config)
        reality = client.get(ArakoonInstaller.INTERNAL_CONFIG_KEY)
        self.assertEqual(reality, expected)

        reality = json.loads(client.get(ArakoonInstaller.METADATA_KEY))
        expected = {'cluster_name': 'test',
                    'cluster_type': 'FWK',
                    'in_use': True,
                    'internal': True}
        self.assertDictEqual(reality, expected)

        # Shrinking cluster
        ArakoonInstaller.shrink_cluster(storagerouters[1].ip, storagerouters[2].ip, 'test')
        reality = Helper.extract_dir_structure(base_dir)
        expected = {'dirs': {'arakoon': {'dirs': {'test': {'dirs': {}, 'files': []}},
                                         'files': []}},
                    'files': []}
        self.assertDictEqual(reality, expected)
        expected = '{0}\n\n{1}\n\n'.format(ArakoonInstallerTester.EXPECTED_CLUSTER_CONFIG.format('2', 'test', ''),
                                           ArakoonInstallerTester.EXPECTED_NODE_CONFIG.format(
                                               '2', storagerouters[2].ip, 20000, base_dir2, '2', 20001
                                           ))
        self.assertEqual(Configuration.get(ArakoonInstaller.CONFIG_KEY.format('test'), raw=True), expected)
        # @TODO: assert service availability here. It should be stopped

        ArakoonInstaller.restart_cluster_remove('test', [storagerouters[2].ip], filesystem=False)
        # @TODO: assert the service is running

        config = ArakoonClusterConfig('test', filesystem=False)
        config.load_config(storagerouters[2].ip)
        client = ArakoonInstaller.build_client(config)
        reality = client.get(ArakoonInstaller.INTERNAL_CONFIG_KEY)
        self.assertEqual(reality, expected)

        reality = json.loads(client.get(ArakoonInstaller.METADATA_KEY))
        expected = {'cluster_name': 'test',
                    'cluster_type': 'FWK',
                    'in_use': True,
                    'internal': True}
        self.assertDictEqual(reality, expected)
    def ar_0001_validate_create_extend_shrink_delete_cluster_test():
        """
        Validate extending and shrinking of arakoon clusters
        """
        storagerouters = GeneralStorageRouter.get_storage_routers()
        if not len(storagerouters) >= 3:
            TestArakoon.logger.info('Environment has only {0} node(s)'.format(len(storagerouters)))
            return

        cluster_name = 'ar_0001'
        cluster_basedir = '/var/tmp/'
        first_sr = storagerouters[0]
        second_sr = storagerouters[1]
        third_sr = storagerouters[2]
        first_root_client = SSHClient(first_sr, username='******')
        second_root_client = SSHClient(second_sr, username='******')
        third_root_client = SSHClient(third_sr, username='******')

        TestArakoon.logger.info('===================================================')
        TestArakoon.logger.info('setup and validate single node cluster')
        create_info = ArakoonInstaller.create_cluster(cluster_name, ServiceType.ARAKOON_CLUSTER_TYPES.FWK, first_sr.ip,
                                                      cluster_basedir, filesystem=False)
        TestArakoon.logger.info('create_info: \n{0}'.format(create_info))
        ArakoonInstaller.start_cluster(cluster_name, first_sr.ip, False)
        ArakoonInstaller.claim_cluster(cluster_name, first_sr, False, metadata=create_info['metadata'])
        TestArakoon.validate_arakoon_config_files([first_sr], cluster_name)
        TestArakoon.verify_arakoon_structure(first_root_client, cluster_name, True, True)

        TestArakoon.logger.info('===================================================')
        TestArakoon.logger.info('setup and validate two node cluster')
        ArakoonInstaller.extend_cluster(first_sr.ip, second_sr.ip, cluster_name, cluster_basedir)
        TestArakoon.validate_arakoon_config_files([first_sr, second_sr], cluster_name)
        TestArakoon.verify_arakoon_structure(first_root_client, cluster_name, True, True)
        TestArakoon.verify_arakoon_structure(second_root_client, cluster_name, True, True)

        TestArakoon.logger.info('===================================================')
        TestArakoon.logger.info('setup and validate three node cluster')
        ArakoonInstaller.extend_cluster(first_sr.ip, third_sr.ip, cluster_name, cluster_basedir)
        TestArakoon.validate_arakoon_config_files([first_sr, second_sr, third_sr], cluster_name)

        for client in [first_root_client, second_root_client, third_root_client]:
            TestArakoon.verify_arakoon_structure(client, cluster_name, True, True)

        TestArakoon.logger.info('===================================================')
        TestArakoon.logger.info('reduce and validate three node to two node cluster')
        ArakoonInstaller.shrink_cluster(second_sr.ip, first_sr.ip, cluster_name)
        TestArakoon.validate_arakoon_config_files([first_sr, third_sr], cluster_name)
        TestArakoon.verify_arakoon_structure(first_root_client, cluster_name, True, True)
        TestArakoon.verify_arakoon_structure(second_root_client, cluster_name, True, False)
        TestArakoon.verify_arakoon_structure(third_root_client, cluster_name, True, True)

        TestArakoon.logger.info('===================================================')
        TestArakoon.logger.info('reduce and validate two node to one node cluster')
        ArakoonInstaller.shrink_cluster(first_sr.ip, third_sr.ip, cluster_name)
        TestArakoon.validate_arakoon_config_files([third_sr], cluster_name)

        TestArakoon.verify_arakoon_structure(first_root_client, cluster_name, True, False)
        TestArakoon.verify_arakoon_structure(second_root_client, cluster_name, True, False)
        TestArakoon.verify_arakoon_structure(third_root_client, cluster_name, True, True)

        TestArakoon.logger.info('===================================================')
        TestArakoon.logger.info('remove cluster')
        ArakoonInstaller.delete_cluster(cluster_name, third_sr.ip, False)

        for client in [first_root_client, second_root_client, third_root_client]:
            TestArakoon.verify_arakoon_structure(client, cluster_name, False, False)

        GeneralArakoon.delete_config(cluster_name)
Ejemplo n.º 12
0
    def demote_node(cluster_ip, master_ip, ip_client_map, unique_id, unconfigure_memcached, unconfigure_rabbitmq, offline_nodes=None):
        """
        Demotes a given node
        """
        from ovs.dal.lists.storagerouterlist import StorageRouterList

        Toolbox.log(logger=NodeTypeController._logger, messages='Demoting node', title=True)
        if offline_nodes is None:
            offline_nodes = []

        if unconfigure_memcached is True and len(offline_nodes) == 0:
            if NodeTypeController._validate_local_memcache_servers(ip_client_map) is False:
                raise RuntimeError('Not all memcache nodes can be reached which is required for demoting a node.')

        # Find other (arakoon) master nodes
        arakoon_cluster_name = str(Configuration.get('/ovs/framework/arakoon_clusters|ovsdb'))
        arakoon_metadata = ArakoonInstaller.get_arakoon_metadata_by_cluster_name(cluster_name=arakoon_cluster_name)
        config = ArakoonClusterConfig(cluster_id=arakoon_cluster_name, filesystem=False)
        config.load_config()
        master_node_ips = [node.ip for node in config.nodes]
        if cluster_ip in master_node_ips:
            master_node_ips.remove(cluster_ip)
        if len(master_node_ips) == 0:
            raise RuntimeError('There should be at least one other master node')

        storagerouter = StorageRouterList.get_by_machine_id(unique_id)
        storagerouter.node_type = 'EXTRA'
        storagerouter.save()

        offline_node_ips = [node.ip for node in offline_nodes]
        if arakoon_metadata['internal'] is True:
            Toolbox.log(logger=NodeTypeController._logger, messages='Leaving Arakoon {0} cluster'.format(arakoon_cluster_name))
            ArakoonInstaller.shrink_cluster(deleted_node_ip=cluster_ip,
                                            remaining_node_ips=master_node_ips,
                                            cluster_name=arakoon_cluster_name,
                                            offline_nodes=offline_node_ips)

        try:
            external_config = Configuration.get('/ovs/framework/external_config')
            if external_config is None:
                config_store = Configuration.get_store()
                if config_store == 'arakoon':
                    Toolbox.log(logger=NodeTypeController._logger, messages='Leaving Arakoon config cluster')
                    ArakoonInstaller.shrink_cluster(deleted_node_ip=cluster_ip,
                                                    remaining_node_ips=master_node_ips,
                                                    cluster_name='config',
                                                    offline_nodes=offline_node_ips,
                                                    filesystem=True)
                else:
                    from ovs.extensions.db.etcd.installer import EtcdInstaller
                    Toolbox.log(logger=NodeTypeController._logger, messages='Leaving Etcd cluster')
                    EtcdInstaller.shrink_cluster(master_ip, cluster_ip, 'config', offline_node_ips)
        except Exception as ex:
            Toolbox.log(logger=NodeTypeController._logger, messages=['\nFailed to leave configuration cluster', ex], loglevel='exception')

        Toolbox.log(logger=NodeTypeController._logger, messages='Update configurations')
        try:
            if unconfigure_memcached is True:
                endpoints = Configuration.get('/ovs/framework/memcache|endpoints')
                endpoint = '{0}:{1}'.format(cluster_ip, 11211)
                if endpoint in endpoints:
                    endpoints.remove(endpoint)
                Configuration.set('/ovs/framework/memcache|endpoints', endpoints)
            if unconfigure_rabbitmq is True:
                endpoints = Configuration.get('/ovs/framework/messagequeue|endpoints')
                endpoint = '{0}:{1}'.format(cluster_ip, 5672)
                if endpoint in endpoints:
                    endpoints.remove(endpoint)
                Configuration.set('/ovs/framework/messagequeue|endpoints', endpoints)
        except Exception as ex:
            Toolbox.log(logger=NodeTypeController._logger, messages=['\nFailed to update configurations', ex], loglevel='exception')

        if arakoon_metadata['internal'] is True:
            Toolbox.log(logger=NodeTypeController._logger, messages='Restarting master node services')
            remaining_nodes = ip_client_map.keys()[:]
            if cluster_ip in remaining_nodes:
                remaining_nodes.remove(cluster_ip)

            PersistentFactory.store = None
            VolatileFactory.store = None

            for service in storagerouter.services:
                if service.name == 'arakoon-ovsdb':
                    service.delete()

        target_client = None
        if storagerouter in offline_nodes:
            if unconfigure_rabbitmq is True:
                Toolbox.log(logger=NodeTypeController._logger, messages='Removing/unconfiguring offline RabbitMQ node')
                client = ip_client_map[master_ip]
                try:
                    client.run(['rabbitmqctl', 'forget_cluster_node', 'rabbit@{0}'.format(storagerouter.name)])
                except Exception as ex:
                    Toolbox.log(logger=NodeTypeController._logger, messages=['\nFailed to forget RabbitMQ cluster node', ex], loglevel='exception')
        else:
            target_client = ip_client_map[cluster_ip]
            if unconfigure_rabbitmq is True:
                Toolbox.log(logger=NodeTypeController._logger, messages='Removing/unconfiguring RabbitMQ')
                try:
                    if ServiceManager.has_service('rabbitmq-server', client=target_client):
                        Toolbox.change_service_state(target_client, 'rabbitmq-server', 'stop', NodeTypeController._logger)
                        target_client.run(['rabbitmq-server', '-detached'])
                        time.sleep(5)
                        target_client.run(['rabbitmqctl', 'stop_app'])
                        time.sleep(5)
                        target_client.run(['rabbitmqctl', 'reset'])
                        time.sleep(5)
                        target_client.run(['rabbitmqctl', 'stop'])
                        time.sleep(5)
                        target_client.file_unlink("/var/lib/rabbitmq/.erlang.cookie")
                        Toolbox.change_service_state(target_client, 'rabbitmq-server', 'stop', NodeTypeController._logger)  # To be sure
                except Exception as ex:
                    Toolbox.log(logger=NodeTypeController._logger, messages=['\nFailed to remove/unconfigure RabbitMQ', ex], loglevel='exception')

            Toolbox.log(logger=NodeTypeController._logger, messages='Stopping services')
            services = ['memcached', 'rabbitmq-server']
            if unconfigure_rabbitmq is False:
                services.remove('rabbitmq-server')
            if unconfigure_memcached is False:
                services.remove('memcached')
            for service in services:
                if ServiceManager.has_service(service, client=target_client):
                    Toolbox.log(logger=NodeTypeController._logger, messages='Stopping service {0}'.format(service))
                    try:
                        Toolbox.change_service_state(target_client, service, 'stop', NodeTypeController._logger)
                    except Exception as ex:
                        Toolbox.log(logger=NodeTypeController._logger, messages=['\nFailed to stop service'.format(service), ex], loglevel='exception')

            Toolbox.log(logger=NodeTypeController._logger, messages='Removing services')
            services = ['scheduled-tasks', 'webapp-api', 'volumerouter-consumer']
            for service in services:
                if ServiceManager.has_service(service, client=target_client):
                    Toolbox.log(logger=NodeTypeController._logger, messages='Removing service {0}'.format(service))
                    try:
                        Toolbox.change_service_state(target_client, service, 'stop', NodeTypeController._logger)
                        ServiceManager.remove_service(service, client=target_client)
                    except Exception as ex:
                        Toolbox.log(logger=NodeTypeController._logger, messages=['\nFailed to remove service'.format(service), ex], loglevel='exception')

            if ServiceManager.has_service('workers', client=target_client):
                ServiceManager.add_service(name='workers',
                                           client=target_client,
                                           params={'WORKER_QUEUE': '{0}'.format(unique_id)})
        try:
            NodeTypeController._configure_amqp_to_volumedriver()
        except Exception as ex:
            Toolbox.log(logger=NodeTypeController._logger, messages=['\nFailed to configure AMQP to Storage Driver', ex], loglevel='exception')

        Toolbox.log(logger=NodeTypeController._logger, messages='Restarting services')
        NodeTypeController.restart_framework_and_memcache_services(clients=ip_client_map, logger=NodeTypeController._logger, offline_node_ips=offline_node_ips)

        if Toolbox.run_hooks(component='nodetype',
                             sub_component='demote',
                             logger=NodeTypeController._logger,
                             cluster_ip=cluster_ip,
                             master_ip=master_ip,
                             offline_node_ips=offline_node_ips):
            Toolbox.log(logger=NodeTypeController._logger, messages='Restarting services')
            NodeTypeController.restart_framework_and_memcache_services(clients=ip_client_map, logger=NodeTypeController._logger, offline_node_ips=offline_node_ips)

        if storagerouter not in offline_nodes:
            target_client = ip_client_map[cluster_ip]
            node_name, _ = target_client.get_hostname()
            if NodeTypeController.avahi_installed(client=target_client, logger=NodeTypeController._logger) is True:
                NodeTypeController.configure_avahi(client=target_client, node_name=node_name, node_type='extra', logger=NodeTypeController._logger)
        Configuration.set('/ovs/framework/hosts/{0}/type'.format(storagerouter.machine_id), 'EXTRA')

        if target_client is not None and target_client.file_exists('/tmp/ovs_rollback'):
            target_client.file_write('/tmp/ovs_rollback', 'rollback')

        Toolbox.log(logger=NodeTypeController._logger, messages='Demote complete', title=True)
Ejemplo n.º 13
0
    def test_cluster_maintenance(self):
        """
        Validates whether a cluster can be correctly created
        """
        Configuration.set('/ovs/framework/hosts/1/ports',
                          {'arakoon': [10000, 10100]})
        Configuration.set('/ovs/framework/hosts/2/ports',
                          {'arakoon': [20000, 20100]})

        structure = Helper.build_service_structure({'storagerouters': [1, 2]})
        storagerouters = structure['storagerouters']
        System._machine_id = {
            storagerouters[1].ip: '1',
            storagerouters[2].ip: '2'
        }

        # Create new cluster
        mountpoint = storagerouters[1].disks[0].partitions[0].mountpoint
        if os.path.exists(mountpoint) and mountpoint != '/':
            shutil.rmtree(mountpoint)
        base_dir = mountpoint + '/test_create_cluster'
        info = ArakoonInstaller.create_cluster(
            'test', ServiceType.ARAKOON_CLUSTER_TYPES.FWK,
            storagerouters[1].ip, base_dir)

        reality = Helper.extract_dir_structure(base_dir)
        expected = {
            'dirs': {
                'arakoon': {
                    'dirs': {
                        'test': {
                            'dirs': {
                                'tlogs': {
                                    'dirs': {},
                                    'files': []
                                },
                                'db': {
                                    'dirs': {},
                                    'files': []
                                }
                            },
                            'files': []
                        }
                    },
                    'files': []
                }
            },
            'files': []
        }
        self.assertDictEqual(reality, expected)
        expected = '{0}\n\n{1}\n\n'.format(
            ArakoonInstallerTester.EXPECTED_CLUSTER_CONFIG.format(
                '1', 'test', ''),
            ArakoonInstallerTester.EXPECTED_NODE_CONFIG.format(
                '1', storagerouters[1].ip, 10000, base_dir, '1', 10001))
        self.assertEqual(
            Configuration.get(ArakoonInstaller.CONFIG_KEY.format('test'),
                              raw=True), expected)
        # @TODO: assert service availability here. It should be stopped

        ArakoonInstaller.start_cluster('test',
                                       storagerouters[1].ip,
                                       filesystem=False)
        # @TODO: assert the service is running

        config = ArakoonClusterConfig('test', filesystem=False)
        config.load_config(storagerouters[1].ip)
        client = ArakoonInstaller.build_client(config)
        reality = client.get(ArakoonInstaller.INTERNAL_CONFIG_KEY)
        self.assertEqual(reality, expected)
        self.assertFalse(client.exists(ArakoonInstaller.METADATA_KEY))

        ArakoonInstaller.claim_cluster('test',
                                       storagerouters[1].ip,
                                       filesystem=False,
                                       metadata=info['metadata'])

        reality = json.loads(client.get(ArakoonInstaller.METADATA_KEY))
        expected = {
            'cluster_name': 'test',
            'cluster_type': 'FWK',
            'in_use': True,
            'internal': True
        }
        self.assertDictEqual(reality, expected)

        # Extending cluster
        mountpoint = storagerouters[2].disks[0].partitions[0].mountpoint
        if os.path.exists(mountpoint) and mountpoint != '/':
            shutil.rmtree(mountpoint)
        base_dir2 = mountpoint + '/test_extend_cluster'
        ArakoonInstaller.extend_cluster(storagerouters[1].ip,
                                        storagerouters[2].ip, 'test',
                                        base_dir2)
        reality = Helper.extract_dir_structure(base_dir)
        expected = {
            'dirs': {
                'arakoon': {
                    'dirs': {
                        'test': {
                            'dirs': {
                                'tlogs': {
                                    'dirs': {},
                                    'files': []
                                },
                                'db': {
                                    'dirs': {},
                                    'files': []
                                }
                            },
                            'files': []
                        }
                    },
                    'files': []
                }
            },
            'files': []
        }
        self.assertDictEqual(reality, expected)
        expected = '{0}\n\n{1}\n\n{2}\n\n'.format(
            ArakoonInstallerTester.EXPECTED_CLUSTER_CONFIG.format(
                '1,2', 'test', ''),
            ArakoonInstallerTester.EXPECTED_NODE_CONFIG.format(
                '1', storagerouters[1].ip, 10000, base_dir, '1', 10001),
            ArakoonInstallerTester.EXPECTED_NODE_CONFIG.format(
                '2', storagerouters[2].ip, 20000, base_dir2, '2', 20001))
        self.assertEqual(
            Configuration.get(ArakoonInstaller.CONFIG_KEY.format('test'),
                              raw=True), expected)
        # @TODO: assert service availability here. It should be stopped

        catchup_command = 'arakoon --node 2 -config file://opt/OpenvStorage/config/framework.json?key=/ovs/arakoon/test/config -catchup-only'
        SSHClient._run_returns[catchup_command] = None
        SSHClient._run_recordings = []
        ArakoonInstaller.restart_cluster_add('test', [storagerouters[1].ip],
                                             storagerouters[2].ip,
                                             filesystem=False)
        self.assertIn(catchup_command, SSHClient._run_recordings)
        # @TODO: assert the service is running

        config = ArakoonClusterConfig('test', filesystem=False)
        config.load_config(storagerouters[2].ip)
        client = ArakoonInstaller.build_client(config)
        reality = client.get(ArakoonInstaller.INTERNAL_CONFIG_KEY)
        self.assertEqual(reality, expected)

        reality = json.loads(client.get(ArakoonInstaller.METADATA_KEY))
        expected = {
            'cluster_name': 'test',
            'cluster_type': 'FWK',
            'in_use': True,
            'internal': True
        }
        self.assertDictEqual(reality, expected)

        # Shrinking cluster
        ArakoonInstaller.shrink_cluster(storagerouters[1].ip,
                                        storagerouters[2].ip, 'test')
        reality = Helper.extract_dir_structure(base_dir)
        expected = {
            'dirs': {
                'arakoon': {
                    'dirs': {
                        'test': {
                            'dirs': {},
                            'files': []
                        }
                    },
                    'files': []
                }
            },
            'files': []
        }
        self.assertDictEqual(reality, expected)
        expected = '{0}\n\n{1}\n\n'.format(
            ArakoonInstallerTester.EXPECTED_CLUSTER_CONFIG.format(
                '2', 'test', ''),
            ArakoonInstallerTester.EXPECTED_NODE_CONFIG.format(
                '2', storagerouters[2].ip, 20000, base_dir2, '2', 20001))
        self.assertEqual(
            Configuration.get(ArakoonInstaller.CONFIG_KEY.format('test'),
                              raw=True), expected)
        # @TODO: assert service availability here. It should be stopped

        ArakoonInstaller.restart_cluster_remove('test', [storagerouters[2].ip],
                                                filesystem=False)
        # @TODO: assert the service is running

        config = ArakoonClusterConfig('test', filesystem=False)
        config.load_config(storagerouters[2].ip)
        client = ArakoonInstaller.build_client(config)
        reality = client.get(ArakoonInstaller.INTERNAL_CONFIG_KEY)
        self.assertEqual(reality, expected)

        reality = json.loads(client.get(ArakoonInstaller.METADATA_KEY))
        expected = {
            'cluster_name': 'test',
            'cluster_type': 'FWK',
            'in_use': True,
            'internal': True
        }
        self.assertDictEqual(reality, expected)