def _stop_old_agent(new_agent, old_agent, timeout):
ctx.logger.info('Stopping old Cloudify agent...')
_, script_url = stop_agent_script_download_link(
new_agent, old_agent_name=old_agent['name'])
_run_script(new_agent, script_url, timeout)
ctx.logger.info('Old Cloudify agent stopped')
update_agent_record(old_agent, AgentState.STOPPED)
def stop(cloudify_agent, installer, **_):
"""
Only called in "remote" mode - other modes stop via AMQP
"""
ctx.logger.info('Stopping Agent {0}'.format(cloudify_agent['name']))
update_agent_record(cloudify_agent, AgentState.STOPPING)
installer.stop_agent()
update_agent_record(cloudify_agent, AgentState.STOPPED)
script.cleanup_scripts()
def stop(cloudify_agent, installer, **_):
"""
Only called in "remote" mode - other modes stop via AMQP
"""
ctx.logger.info('Stopping Agent {0}'.format(cloudify_agent['name']))
update_agent_record(cloudify_agent, AgentState.STOPPING)
installer.stop_agent()
update_agent_record(cloudify_agent, AgentState.STOPPED)
script.cleanup_scripts()
def _stop_old_agent(new_agent, old_agent, timeout):
ctx.logger.info('Stopping old Cloudify agent...')
_, script_url = stop_agent_script_download_link(
new_agent,
old_agent_name=old_agent['name']
)
_run_script(new_agent, script_url, timeout)
ctx.logger.info('Old Cloudify agent stopped')
update_agent_record(old_agent, AgentState.STOPPED)
def delete(cloudify_agent, installer, **_):
update_agent_record(cloudify_agent, AgentState.DELETING)
# delete the runtime properties set on create
if cloudify_agent.has_installer:
ctx.logger.info('Deleting Agent {0}'.format(cloudify_agent['name']))
installer.delete_agent()
ctx.instance.runtime_properties.pop('cloudify_agent', None)
ctx.instance.update()
update_agent_record(cloudify_agent, AgentState.DELETED)
# TODO: Delete the RabbitMQ queue after deleting the agent
delete_agent_rabbitmq_user(cloudify_agent)
def delete(cloudify_agent, installer, **_):
update_agent_record(cloudify_agent, AgentState.DELETING)
# delete the runtime properties set on create
if cloudify_agent.has_installer:
ctx.logger.info('Deleting Agent {0}'.format(cloudify_agent['name']))
installer.delete_agent()
ctx.instance.runtime_properties.pop('cloudify_agent', None)
ctx.instance.update()
update_agent_record(cloudify_agent, AgentState.DELETED)
# TODO: Delete the RabbitMQ queue after deleting the agent
delete_agent_rabbitmq_user(cloudify_agent)
def create_agent_amqp(install_agent_timeout=300, manager_ip=None,
manager_certificate=None, stop_old_agent=False, **_):
"""
Installs a new agent on a host machine.
:param install_agent_timeout: operation's timeout.
:param manager_ip: the private IP of the current leader (master) Manager.
This IP is used to connect to the Manager's RabbitMQ.
(relevant only in HA cluster)
:param manager_certificate: the SSL certificate of the current leader
(master) Manager. (relevant only in HA cluster)
:param stop_old_agent: if set, stop the old agent after successfully
installing the new one
"""
old_agent = _validate_agent()
update_agent_record(old_agent, AgentState.UPGRADING)
_update_broker_config(old_agent, manager_ip, manager_certificate)
agents = _run_install_script(old_agent, install_agent_timeout)
new_agent = agents['new']
ctx.logger.info('Installed agent {0}'.format(new_agent['name']))
create_agent_record(new_agent, AgentState.STARTING)
result = _validate_current_amqp(new_agent)
if not result['agent_alive']:
update_agent_record(new_agent, AgentState.FAILED)
raise RecoverableError('New agent did not start and connect')
update_agent_record(new_agent, AgentState.STARTED)
update_agent_record(old_agent, AgentState.UPGRADED)
if stop_old_agent:
_stop_old_diamond(old_agent, install_agent_timeout)
_stop_old_agent(new_agent, old_agent, install_agent_timeout)
# Setting old_cloudify_agent in order to uninstall it later.
ctx.instance.runtime_properties['old_cloudify_agent'] = agents['old']
update_agent_runtime_properties(new_agent)
def delete(cloudify_agent, installer, **_):
update_agent_record(cloudify_agent, AgentState.DELETING)
# delete the runtime properties set on create
if cloudify_agent.has_installer:
ctx.logger.info('Deleting Agent {0}'.format(cloudify_agent['name']))
installer.delete_agent()
ctx.instance.runtime_properties.pop('cloudify_agent', None)
ctx.instance.update()
update_agent_record(cloudify_agent, AgentState.DELETED)
delete_agent_rabbitmq_user(cloudify_agent)
try:
delete_agent_exchange(cloudify_agent)
delete_agent_queues(cloudify_agent)
except KeyError as e:
# this would happen for malformed cloudify_agent which is missing
# eg. the tenant data. We'd be possibly leaving queues around,
# but without that info, we can't really do anything else
ctx.logger.error('Could not delete agent queues: %s', e)
def create(cloudify_agent, installer, **_):
# When not in "remote" mode, this operation is called only to set the
# agent_config dict in the runtime properties
create_agent_record(cloudify_agent)
if cloudify_agent.has_installer:
with script.install_script_path(cloudify_agent) as script_path:
ctx.logger.info('Creating Agent {0}'.format(
cloudify_agent['name']))
try:
installer.runner.run_script(script_path)
except (CommandExecutionError, CommandExecutionException):
ctx.logger.error("Failed creating agent; marking agent as "
"failed")
update_agent_record(cloudify_agent, AgentState.FAILED)
raise
ctx.logger.info(
'Agent created, configured and started successfully'
)
update_agent_record(cloudify_agent, AgentState.STARTED)
elif cloudify_agent.is_proxied:
ctx.logger.info('Working in "proxied" mode')
elif cloudify_agent.is_provided:
ctx.logger.info('Working in "provided" mode')
_, install_script_download_link = script.install_script_download_link(
cloudify_agent
)
ctx.logger.info(
'Agent config created. To configure/start the agent, download the '
'following script: {0}'.format(install_script_download_link)
)
cloudify_agent['install_script_download_link'] = \
install_script_download_link
update_agent_runtime_properties(cloudify_agent)
update_agent_record(cloudify_agent, AgentState.CREATED)
def create(cloudify_agent, installer, **_):
# When not in "remote" mode, this operation is called only to set the
# agent_config dict in the runtime properties
create_agent_record(cloudify_agent)
if cloudify_agent.has_installer:
with script.install_script_path(cloudify_agent) as script_path:
ctx.logger.info('Creating Agent {0}'.format(
cloudify_agent['name']))
try:
installer.runner.run_script(script_path)
except (CommandExecutionError, CommandExecutionException):
ctx.logger.error("Failed creating agent; marking agent as "
"failed")
update_agent_record(cloudify_agent, AgentState.FAILED)
raise
ctx.logger.info(
'Agent created, configured and started successfully')
update_agent_record(cloudify_agent, AgentState.STARTED)
elif cloudify_agent.is_proxied:
ctx.logger.info('Working in "proxied" mode')
elif cloudify_agent.is_provided:
ctx.logger.info('Working in "provided" mode')
_, install_script_download_link = script.install_script_download_link(
cloudify_agent)
ctx.logger.info(
'Agent config created. To configure/start the agent, download the '
'following script: {0}'.format(install_script_download_link))
cloudify_agent['install_script_download_link'] = \
install_script_download_link
update_agent_runtime_properties(cloudify_agent)
update_agent_record(cloudify_agent, AgentState.CREATED)
def start(cloudify_agent, **_):
"""
Only called in "init_script"/"plugin" mode, where the agent is started
externally (e.g. userdata script), and all we have to do is wait for it
"""
update_agent_record(cloudify_agent, AgentState.STARTING)
tenant = cloudify_utils.get_tenant()
client = get_client(
amqp_user=tenant['rabbitmq_username'],
amqp_pass=tenant['rabbitmq_password'],
amqp_vhost=tenant['rabbitmq_vhost']
)
agent_alive = utils.is_agent_alive(cloudify_agent['queue'], client)
if not agent_alive:
if ctx.operation.retry_number > 3:
ctx.logger.warning('Waiting too long for Agent to start')
update_agent_record(cloudify_agent, AgentState.NONRESPONSIVE)
return ctx.operation.retry(
message='Waiting for Agent to start...')
ctx.logger.info('Agent has started')
update_agent_record(cloudify_agent, AgentState.STARTED)
if not cloudify_agent.is_provided:
script.cleanup_scripts()
def create_agent_amqp(install_agent_timeout=300,
manager_ip=None,
manager_certificate=None,
stop_old_agent=False,
**_):
"""
Installs a new agent on a host machine.
:param install_agent_timeout: operation's timeout.
:param manager_ip: the private IP of the current leader (master) Manager.
This IP is used to connect to the Manager's RabbitMQ.
(relevant only in HA cluster)
:param manager_certificate: the SSL certificate of the current leader
(master) Manager. (relevant only in HA cluster)
:param stop_old_agent: if set, stop the old agent after successfully
installing the new one
"""
old_agent = _validate_agent()
original_agent = copy.deepcopy(old_agent)
update_agent_record(old_agent, AgentState.UPGRADING)
_update_broker_config(old_agent, manager_ip, manager_certificate)
try:
agents = _run_install_script(old_agent, install_agent_timeout)
except Exception:
update_agent_runtime_properties(original_agent)
raise
new_agent = agents['new']
ctx.logger.info('Installed agent {0}'.format(new_agent['name']))
create_agent_record(new_agent, AgentState.STARTING)
result = _validate_current_amqp(new_agent)
if not result['agent_alive']:
update_agent_record(new_agent, AgentState.FAILED)
update_agent_runtime_properties(original_agent)
raise RecoverableError('New agent did not start and connect')
update_agent_record(new_agent, AgentState.STARTED)
update_agent_record(old_agent, AgentState.UPGRADED)
if stop_old_agent:
_stop_old_diamond(old_agent, install_agent_timeout)
_stop_old_agent(new_agent, old_agent, install_agent_timeout)
# Setting old_cloudify_agent in order to uninstall it later.
ctx.instance.runtime_properties['old_cloudify_agent'] = agents['old']
update_agent_runtime_properties(new_agent)
def configure(cloudify_agent, installer, **_):
ctx.logger.info('Configuring Agent {0}'.format(cloudify_agent['name']))
update_agent_record(cloudify_agent, AgentState.CONFIGURING)
try:
installer.configure_agent()
except CommandExecutionError as e:
ctx.logger.error(str(e))
update_agent_record(cloudify_agent, AgentState.FAILED)
raise
update_agent_record(cloudify_agent, AgentState.CONFIGURED)
def configure(cloudify_agent, installer, **_):
ctx.logger.info('Configuring Agent {0}'.format(cloudify_agent['name']))
update_agent_record(cloudify_agent, AgentState.CONFIGURING)
try:
installer.configure_agent()
except CommandExecutionError as e:
ctx.logger.error(str(e))
update_agent_record(cloudify_agent, AgentState.FAILED)
raise
update_agent_record(cloudify_agent, AgentState.CONFIGURED)
def restart(cloudify_agent, installer, **_):
# no need to handling remote_execution False because this operation is
# not invoked in that case
ctx.logger.info('Restarting Agent {0}'.format(cloudify_agent['name']))
update_agent_record(cloudify_agent, AgentState.RESTARTING)
try:
installer.restart_agent()
except CommandExecutionError as e:
ctx.logger.error(str(e))
update_agent_record(cloudify_agent, AgentState.FAILED)
raise
update_agent_record(cloudify_agent, AgentState.RESTARTED)
def restart(cloudify_agent, installer, **_):
# no need to handling remote_execution False because this operation is
# not invoked in that case
ctx.logger.info('Restarting Agent {0}'.format(cloudify_agent['name']))
update_agent_record(cloudify_agent, AgentState.RESTARTING)
try:
installer.restart_agent()
except CommandExecutionError as e:
ctx.logger.error(str(e))
update_agent_record(cloudify_agent, AgentState.FAILED)
raise
update_agent_record(cloudify_agent, AgentState.RESTARTED)
def start(cloudify_agent, **_):
"""
Only called in "init_script"/"plugin" mode, where the agent is started
externally (e.g. userdata script), and all we have to do is wait for it
"""
update_agent_record(cloudify_agent, AgentState.STARTING)
tenant = cloudify_utils.get_tenant()
client = get_client(amqp_user=tenant['rabbitmq_username'],
amqp_pass=tenant['rabbitmq_password'],
amqp_vhost=tenant['rabbitmq_vhost'])
agent_alive = utils.is_agent_alive(cloudify_agent['queue'], client)
if not agent_alive:
if ctx.operation.retry_number > 3:
ctx.logger.warning('Waiting too long for Agent to start')
update_agent_record(cloudify_agent, AgentState.NONRESPONSIVE)
return ctx.operation.retry(message='Waiting for Agent to start...')
ctx.logger.info('Agent has started')
update_agent_record(cloudify_agent, AgentState.STARTED)
if not cloudify_agent.is_provided:
script.cleanup_scripts()
