def update_asd(asd, update_data):
"""
Updates an ASD with the 'update_data' provided
:param asd: ASD to update
:type asd: source.dal.objects.asd.ASD
:param update_data: Data to update
:type update_data: dict
:raises ValueError: - When ASD configuration key is not present
- When an unsupported key is passed in via 'update_data'
:return: None
:rtype: NoneType
"""
key_map = {'ips': 'hosts'}
if not Configuration.exists(asd.config_key):
raise ValueError(
'Failed to the configuration at location {0}'.format(
asd.config_key))
config = Configuration.get(asd.config_key)
for key, value in update_data.iteritems():
if key not in key_map: # Only updating IPs is supported for now
raise ValueError(
'Unsupported property provided: {0}. Only IPs can be updated for now'
.format(key))
setattr(asd, key_map[key], value)
config[key] = value
asd.save()
Configuration.set(key=asd.config_key, value=config)
def get_logging_info(cls):
"""
Retrieve logging information from the Configuration management
:return: Dictionary containing logging information
:rtype: dict
"""
try:
return Configuration.get('/ovs/alba/logging')
except (IOError, NotFoundException):
return {}
def export(self):
"""
Exports the ASD information to a dict structure
:return: Representation of the ASD as dict
:rtype: dict
"""
if not self.has_config:
raise RuntimeError('No configuration found for ASD {0}'.format(
self.asd_id))
data = Configuration.get(self.config_key)
for prop in self._properties:
if prop.name == 'hosts':
data['ips'] = getattr(self, prop.name)
else:
data[prop.name] = getattr(self, prop.name)
if self.disk.state == 'MISSING':
data.update({'state': 'error', 'state_detail': 'missing'})
else:
output, error = ASD._local_client.run(
['ls', '{0}/{1}/'.format(self.disk.mountpoint, self.folder)],
allow_nonzero=True,
return_stderr=True)
output += error
if 'Input/output error' in output:
data.update({'state': 'error', 'state_detail': 'io_error'})
elif ASD._service_manager.has_service(self.service_name,
ASD._local_client):
service_state = ASD._service_manager.get_service_status(
self.service_name, ASD._local_client)
if service_state == 'activating':
data.update({
'state': 'warning',
'state_detail': 'service_activating'
})
elif service_state == 'active':
data.update({'state': 'ok', 'state_detail': None})
else:
data.update({
'state': 'error',
'state_detail': 'service_failure'
})
else:
data.update({
'state': 'error',
'state_detail': 'service_failure'
})
return data
def authorized(cls):
"""
Indicates whether a call is authenticated
"""
# For backwards compatibility we first try to retrieve the node ID by using the bootstrap file
try:
with open(BOOTSTRAP_FILE) as bstr_file:
node_id = json.load(bstr_file)['node_id']
except:
node_id = SettingList.get_setting_by_code(code='node_id').value
node_config = Configuration.get(
ASD_NODE_CONFIG_MAIN_LOCATION.format(node_id))
username = node_config['username']
password = node_config['password']
auth = request.authorization
return auth and auth.username == username and auth.password == password
def sync_disks():
# type: () -> None
"""
Syncs the disks
Changes made to this code should be reflected in the framework DiskController.sync_with_reality call.
:return: None
:rtype: NoneType
"""
node_id = SettingList.get_setting_by_code(code='node_id').value
s3 = Configuration.get(
ASD_NODE_CONFIG_MAIN_LOCATION_S3.format(node_id), default=False)
disks, name_alias_mapping = DiskTools.model_devices(s3=s3)
disks_by_name = dict((disk.name, disk) for disk in disks)
alias_name_mapping = name_alias_mapping.reverse_mapping()
# Specific for the asd-manager: handle unique constraint exception
DiskController._prepare_for_name_switch(disks)
# Sync the model
for disk in DiskList.get_disks():
generic_disk_model = None # type: GenericDisk
for alias in disk.aliases:
# IBS wont have alias
if alias in alias_name_mapping:
name = alias_name_mapping[alias].replace('/dev/', '')
if name in disks_by_name:
generic_disk_model = disks_by_name.pop(name)
break
# Partitioned loop, nvme devices no longer show up in alias_name_mapping
if generic_disk_model is None and disk.name in disks_by_name and (
disk.name.startswith(tuple(['fio', 'loop', 'nvme']))):
generic_disk_model = disks_by_name.pop(disk.name)
if not generic_disk_model:
# Remove disk / partitions if not reported by 'lsblk'
DiskController._remove_disk_model(disk)
else:
# Update existing disks and their partitions
DiskController._sync_disk_with_model(disk, generic_disk_model)
# Create all disks and their partitions not yet modeled
for disk_name, generic_disk_model in disks_by_name.iteritems():
DiskController._model_disk(generic_disk_model)
def migrate(cls):
# type: () -> None
"""
Execute the migration logic.
:return: None
:rtype: NoneType
"""
with file_mutex('package_update_pu'):
local_client = SSHClient(endpoint='127.0.0.1', username='******')
# Override the created openvstorage_sdm_id during package install, with currently available SDM ID
if local_client.file_exists(BOOTSTRAP_FILE):
with open(BOOTSTRAP_FILE) as bstr_file:
node_id = json.load(bstr_file)['node_id']
local_client.file_write(filename='/etc/openvstorage_sdm_id',
contents=node_id + '\n')
else:
with open('/etc/openvstorage_sdm_id', 'r') as id_file:
node_id = id_file.read().strip()
key = '{0}/versions'.format(
ASD_NODE_CONFIG_LOCATION.format(node_id))
version = Configuration.get(key) if Configuration.exists(
key) else 0
asd_manager_service_name = 'asd-manager'
if cls.service_manager.has_service(
asd_manager_service_name,
local_client) and cls.service_manager.get_service_status(
asd_manager_service_name, local_client) == 'active':
cls.logger.info('Stopping asd-manager service')
cls.service_manager.stop_service(asd_manager_service_name,
local_client)
# @TODO: Move these migrations to alba_node.client.update_execute_migration_code()
if version < cls.CURRENT_VERSION:
try:
# DB migrations
from source.controllers.asd import ASDController
from source.controllers.disk import DiskController
from source.dal.asdbase import ASDBase
from source.dal.lists.asdlist import ASDList
from source.dal.lists.disklist import DiskList
from source.dal.objects.asd import ASD
if not local_client.file_exists('{0}/main.db'.format(
ASDBase.DATABASE_FOLDER)):
local_client.dir_create([ASDBase.DATABASE_FOLDER])
asd_map = dict(
(asd.asd_id, asd) for asd in ASDList.get_asds())
DiskController.sync_disks()
for disk in DiskList.get_usable_disks():
if disk.state == 'MISSING' or disk.mountpoint is None:
continue
for asd_id in local_client.dir_list(disk.mountpoint):
if asd_id in asd_map:
asd = asd_map[asd_id]
else:
asd = ASD()
asd.disk = disk
asd.asd_id = asd_id
asd.folder = asd_id
if asd.has_config:
if asd.port is None or asd.hosts is None:
config = Configuration.get(
key=asd.config_key)
asd.port = config['port']
asd.hosts = config.get('ips', [])
asd.save()
# Adjustment of open file descriptors for ASD/maintenance services to 8192
asd_service_names = list(ASDController.list_asd_services())
maintenance_service_names = list(
MaintenanceController.get_services())
for service_name in asd_service_names + maintenance_service_names:
if cls.service_manager.has_service(
name=service_name, client=local_client):
if cls.service_manager.__class__ == Systemd:
path = '/lib/systemd/system/{0}.service'.format(
service_name)
check = 'LimitNOFILE=8192'
else:
path = '/etc/init/{0}.conf'.format(
service_name)
check = 'limit nofile 8192 8192'
restart_required = False
if os.path.exists(path):
with open(path, 'r') as system_file:
if check not in system_file.read():
restart_required = True
if restart_required is False:
continue
configuration_key = ServiceFactory.SERVICE_CONFIG_KEY.format(
node_id, service_name)
if Configuration.exists(configuration_key):
# Rewrite the service file
cls.service_manager.add_service(
name=ASDController.ASD_PREFIX
if service_name in asd_service_names else
MaintenanceController.MAINTENANCE_PREFIX,
client=local_client,
params=Configuration.get(
configuration_key),
target_name=service_name)
# Let the update know that the ASD / maintenance services need to be restarted
# Inside `if Configuration.exists`, because useless to rapport restart if we haven't rewritten service file
ExtensionsToolbox.edit_version_file(
client=local_client,
package_name='alba',
old_run_file='{0}/{1}.version'.format(
ServiceFactory.RUN_FILE_DIR,
service_name))
if cls.service_manager.__class__ == Systemd:
local_client.run(['systemctl', 'daemon-reload'])
# Version 3: Addition of 'ExecReload' for ASD/maintenance SystemD services
if cls.service_manager.__class__ == Systemd: # Upstart does not have functionality to reload a process' configuration
reload_daemon = False
asd_service_names = list(
ASDController.list_asd_services())
maintenance_service_names = list(
MaintenanceController.get_services())
for service_name in asd_service_names + maintenance_service_names:
if not cls.service_manager.has_service(
name=service_name, client=local_client):
continue
path = '/lib/systemd/system/{0}.service'.format(
service_name)
if os.path.exists(path):
with open(path, 'r') as system_file:
if 'ExecReload' not in system_file.read():
reload_daemon = True
configuration_key = ServiceFactory.SERVICE_CONFIG_KEY.format(
node_id, service_name)
if Configuration.exists(
configuration_key):
# No need to edit the service version file, since this change only requires a daemon-reload
cls.service_manager.add_service(
name=ASDController.ASD_PREFIX
if service_name
in asd_service_names else
MaintenanceController.
MAINTENANCE_PREFIX,
client=local_client,
params=Configuration.get(
configuration_key),
target_name=service_name)
if reload_daemon is True:
local_client.run(['systemctl', 'daemon-reload'])
# Version 6: Introduction of Active Drive
all_local_ips = OSFactory.get_manager().get_ip_addresses(
client=local_client)
for asd in ASDList.get_asds():
if asd.has_config:
asd_config = Configuration.get(asd.config_key)
if 'multicast' not in asd_config:
asd_config['multicast'] = None
if 'ips' in asd_config:
asd_ips = asd_config['ips'] or all_local_ips
else:
asd_ips = all_local_ips
asd.hosts = asd_ips
asd_config['ips'] = asd_ips
Configuration.set(asd.config_key, asd_config)
asd.save()
# Version 7: Moving flask certificate files to config dir
for file_name in [
'passphrase', 'server.crt', 'server.csr',
'server.key'
]:
if local_client.file_exists(
'/opt/asd-manager/source/{0}'.format(
file_name)):
local_client.file_move(
source_file_name='/opt/asd-manager/source/{0}'.
format(file_name),
destination_file_name=
'/opt/asd-manager/config/{0}'.format(
file_name))
except:
cls.logger.exception(
'Error while executing post-update code on node {0}'.
format(node_id))
Configuration.set(key, cls.CURRENT_VERSION)
if cls.service_manager.has_service(
asd_manager_service_name,
local_client) and cls.service_manager.get_service_status(
asd_manager_service_name, local_client) != 'active':
cls.logger.info('Starting asd-manager service')
cls.service_manager.start_service(asd_manager_service_name,
local_client)
cls.logger.info('Post-update logic executed')
def execute_migration_code(cls):
# type: () -> None
"""
Run some migration code after an update has been done
:return: None
:rtype: NoneType
"""
cls._logger.info('Starting out of band migrations for SDM nodes')
###########################
# Start crucial migration #
###########################
# Removal of bootstrap file and store API IP, API port and node ID in SQLite DB
try:
if cls._local_client.file_exists(BOOTSTRAP_FILE):
cls._logger.info('Bootstrap file still exists. Retrieving node ID')
with open(BOOTSTRAP_FILE) as bstr_file:
node_id = json.load(bstr_file)['node_id']
else:
node_id = SettingList.get_setting_by_code(code='node_id').value
except Exception:
cls._logger.exception('Unable to determine the node ID, cannot migrate')
raise
try:
api_settings_map = {'api_ip': 'ip', 'api_port': 'port'} # Map settings code to keys in the Config management
required_settings = ['node_id', 'migration_version'] + api_settings_map.keys()
for settings_code in required_settings:
try:
_ = SettingList.get_setting_by_code(settings_code)
except ObjectNotFoundException:
cls._logger.info('Missing required settings: {0}'.format(settings_code))
if settings_code == 'node_id':
value = node_id
elif settings_code in api_settings_map.keys():
# Information must be extracted from Configuration
main_config = Configuration.get(ASD_NODE_CONFIG_MAIN_LOCATION.format(node_id))
value = main_config[api_settings_map[settings_code]]
elif settings_code == 'migration_version':
# Introduce version for ASD Manager migration code
value = 0
else:
raise NotImplementedError('No action implemented for setting {0}'.format(settings_code))
cls._logger.info('Modeling Setting with code {0} and value {1}'.format(settings_code, value))
setting = Setting()
setting.code = settings_code
setting.value = value
setting.save()
if cls._local_client.file_exists(BOOTSTRAP_FILE):
cls._logger.info('Removing the bootstrap file')
cls._local_client.file_delete(BOOTSTRAP_FILE)
except Exception:
cls._logger.exception('Error during migration of code settings. Unable to proceed')
raise
###############################
# Start non-crucial migration #
###############################
errors = []
migration_setting = SettingList.get_setting_by_code(code='migration_version')
# Add installed package_name in version files and additional string replacements in service files
try:
if migration_setting.value < 1:
cls._logger.info('Adding additional information to service files')
edition = Configuration.get_edition()
if edition == PackageFactory.EDITION_ENTERPRISE:
for version_file_name in cls._local_client.file_list(directory=ServiceFactory.RUN_FILE_DIR):
version_file_path = '{0}/{1}'.format(ServiceFactory.RUN_FILE_DIR, version_file_name)
contents = cls._local_client.file_read(filename=version_file_path)
if '{0}='.format(PackageFactory.PKG_ALBA) in contents:
contents = contents.replace(PackageFactory.PKG_ALBA, PackageFactory.PKG_ALBA_EE)
cls._local_client.file_write(filename=version_file_path, contents=contents)
node_id = SettingList.get_setting_by_code(code='node_id').value
asd_services = list(ASDController.list_asd_services())
maint_services = list(MaintenanceController.get_services())
for service_name in asd_services + maint_services:
config_key = ServiceFactory.SERVICE_CONFIG_KEY.format(node_id, service_name)
if Configuration.exists(key=config_key):
config = Configuration.get(key=config_key)
if 'RUN_FILE_DIR' in config:
continue
config['RUN_FILE_DIR'] = ServiceFactory.RUN_FILE_DIR
config['ALBA_PKG_NAME'] = PackageFactory.PKG_ALBA_EE
config['ALBA_VERSION_CMD'] = PackageFactory.VERSION_CMD_ALBA
Configuration.set(key=config_key, value=config)
cls._service_manager.regenerate_service(name=ASDController.ASD_PREFIX if service_name in asd_services else MaintenanceController.MAINTENANCE_PREFIX,
client=cls._local_client,
target_name=service_name)
except Exception as ex:
cls._logger.exception('Failed to regenerate the ASD and Maintenance services')
errors.append(ex)
try:
if migration_setting.value < 2:
if System.get_component_identifier() not in Configuration.get(Configuration.get_registration_key(), default=[]):
Configuration.register_usage(System.get_component_identifier())
except Exception as ex:
cls._logger.exception('Failed to register the asd-manager')
errors.append(ex)
if len(errors) == 0:
cls._logger.info('No errors during non-crucial migration. Saving the migration setting')
# Save migration settings when no errors occurred
migration_setting = SettingList.get_setting_by_code(code='migration_version')
migration_setting.value = 2
migration_setting.save()
cls._logger.info('Finished out of band migrations for SDM nodes')
def create_asd(disk):
"""
Creates and starts an ASD on a given disk
:param disk: Disk on which to create an ASD
:type disk: source.dal.objects.disk.Disk
:return: None
:rtype: NoneType
"""
# Validations
if disk.state == 'MISSING':
raise RuntimeError(
'Cannot create an ASD on missing disk {0}'.format(disk.name))
_node_id = SettingList.get_setting_by_code(code='node_id').value
ipaddresses = Configuration.get('{0}|ips'.format(
ASD_NODE_CONFIG_NETWORK_LOCATION.format(_node_id)))
if len(ipaddresses) == 0:
ipaddresses = OSFactory.get_manager().get_ip_addresses(
client=ASDController._local_client)
if len(ipaddresses) == 0:
raise RuntimeError('Could not find any IP on the local node')
alba_pkg_name, alba_version_cmd = PackageFactory.get_package_and_version_cmd_for(
component='alba'
) # Call here, because this potentially raises error, which should happen before actually making changes
# Fetch disk information
disk_size = int(
ASDController._local_client.run(
['df', '-B', '1', '--output=size', disk.mountpoint],
timeout=5).splitlines()[1])
# Find out appropriate disk size
asd_size = int(math.floor(disk_size / (len(disk.asds) + 1)))
for asd in disk.asds:
if asd.has_config:
config = Configuration.get(asd.config_key)
config['capacity'] = asd_size
cache_size = ASDController.calculate_rocksdb_cache_size(
is_ssd=disk.is_ssd)
if cache_size:
config.update({'rocksdb_block_cache_size': cache_size})
Configuration.set(asd.config_key, config)
try:
ASDController._service_manager.send_signal(
asd.service_name, signal.SIGUSR1,
ASDController._local_client)
except Exception as ex:
ASDController._logger.info(
'Could not send signal to ASD for reloading the quota: {0}'
.format(ex))
used_ports = []
for asd in ASDList.get_asds():
if asd.has_config:
config = Configuration.get(asd.config_key)
used_ports.append(config['port'])
if 'rora_port' in config:
used_ports.append(config['rora_port'])
# Prepare & start service
ASDController._logger.info('Setting up service for disk {0}'.format(
disk.name))
asd_id = ''.join(
random.choice(string.ascii_letters + string.digits)
for _ in range(32))
homedir = '{0}/{1}'.format(disk.mountpoint, asd_id)
base_port = Configuration.get('{0}|port'.format(
ASD_NODE_CONFIG_NETWORK_LOCATION.format(_node_id)))
asd_port = base_port
rora_port = base_port + 1
while asd_port in used_ports:
asd_port += 1
used_ports.append(asd_port)
while rora_port in used_ports:
rora_port += 1
asd_config = {
'ips': ipaddresses,
'home': homedir,
'port': asd_port,
'asd_id': asd_id,
'node_id': _node_id,
'capacity': asd_size,
'multicast': None,
'transport': 'tcp',
'log_level': 'info'
}
cache_size = ASDController.calculate_rocksdb_cache_size(
is_ssd=disk.is_ssd)
if cache_size:
asd_config.update({'rocksdb_block_cache_size': cache_size})
if Configuration.get('/ovs/framework/rdma'):
asd_config['rora_port'] = rora_port
asd_config['rora_transport'] = 'rdma'
if Configuration.exists('{0}/extra'.format(
ASD_NODE_CONFIG_LOCATION.format(_node_id))):
data = Configuration.get('{0}/extra'.format(
ASD_NODE_CONFIG_LOCATION.format(_node_id)))
asd_config.update(data)
asd = ASD()
asd.disk = disk
asd.port = asd_port
asd.hosts = ipaddresses
asd.asd_id = asd_id
asd.folder = asd_id
asd.save()
Configuration.set(asd.config_key, asd_config)
params = {
'LOG_SINK': Logger.get_sink_path('alba-asd_{0}'.format(asd_id)),
'CONFIG_PATH':
Configuration.get_configuration_path(asd.config_key),
'SERVICE_NAME': asd.service_name,
'ALBA_PKG_NAME': alba_pkg_name,
'ALBA_VERSION_CMD': alba_version_cmd
}
os.mkdir(homedir)
ASDController._local_client.run(['chown', '-R', 'alba:alba', homedir])
ASDController._service_manager.add_service(
name=ASDController.ASD_PREFIX,
client=ASDController._local_client,
params=params,
target_name=asd.service_name)
ASDController.start_asd(asd)
def _sync_disks():
from source.controllers.disk import DiskController
while True:
DiskController.sync_disks()
time.sleep(60)
try:
node_id = SettingList.get_setting_by_code(code='node_id').value
except:
# For backwards compatibility
# After update SettingList has not been populated yet and post-update script of package will restart asd-manager
with open(BOOTSTRAP_FILE) as bstr_file:
node_id = json.load(bstr_file)['node_id']
try:
asd_manager_config = Configuration.get(
ASD_NODE_CONFIG_MAIN_LOCATION.format(node_id))
except:
raise RuntimeError('Configuration management unavailable')
if 'ip' not in asd_manager_config or 'port' not in asd_manager_config:
raise RuntimeError(
'IP and/or port not available in configuration for ALBA node {0}'.
format(node_id))
from source.app import app
@app.before_first_request
def setup_logging():
"""
Configure logging
:return: None