def verify(_):
def check_status():
check_d = treq.head(
append_segments(server_endpoint, 'servers', server_id),
headers=headers(auth_token))
check_d.addCallback(check_success, [404])
return check_d
start_time = clock.seconds()
# this is treating all errors as transient, so the only error that can
# occur is a CancelledError from timing out
verify_d = retry_and_timeout(check_status, timeout,
next_interval=repeating_interval(interval),
clock=clock)
def on_success(_):
time_delete = clock.seconds() - start_time
del_log.msg('Server deleted successfully: {time_delete} seconds.',
time_delete=time_delete)
verify_d.addCallback(on_success)
def on_timeout(_):
time_delete = clock.seconds() - start_time
del_log.err(None, timeout=timeout, time_delete=time_delete,
why=('Server {instance_id} failed to be deleted within '
'a {timeout} second timeout (it has been '
'{time_delete} seconds).'))
verify_d.addErrback(on_timeout)
def wait_for_stack_list(self, expected_states, timeout=180, period=10):
def check(content):
states = pbag([s['stack_status'] for s in content['stacks']])
if not (states == expected_states):
msg("Waiting for group {} to reach desired group state.\n"
"{} (actual) {} (expected)".format(self.group.group_id,
states,
expected_states))
raise TransientRetryError(
"Group states of {} did not match expected {})".format(
states, expected_states))
msg("Success: desired group state reached:\n{}".format(
expected_states))
return self.rcs
def poll():
return self.get_stack_list().addCallback(check)
expected_states = pbag(expected_states)
return retry_and_timeout(
poll,
timeout,
can_retry=terminal_errors_except(TransientRetryError),
next_interval=repeating_interval(period),
clock=reactor,
deferred_description=(
"Waiting for group {} to reach state {}".format(
self.group.group_id, str(expected_states))))
def wait_for_state(self, rcs, matcher, timeout=600, period=10, clock=None):
"""
Wait for the state on the scaling group to match the provided matchers,
specified by matcher.
:param rcs: a :class:`otter.integration.lib.resources.TestResources`
instance
:param matcher: A :mod:`testtool.matcher`, as specified in
module: testtools.matchers in
http://testtools.readthedocs.org/en/latest/api.html.
:param timeout: The amount of time to wait until this step is
considered failed.
:param period: How long to wait before polling again.
:param clock: a :class:`twisted.internet.interfaces.IReactorTime`
provider
:return: None, if the state is reached
:raises: :class:`TimedOutError` if the state is never reached within
the requisite amount of time.
Example usage:
```
matcher = MatchesAll(
IncludesServers(included_server_ids),
ExcludesServers(exclude_server_ids),
ContainsDict({
'pending': Equals(0),
'desired': Equals(5),
'status': Equals('ACTIVE')
})
)
..wait_for_state(rcs, matchers, timeout=60)
```
"""
def check(result):
response, group_state = result
mismatch = matcher.match(group_state['group'])
if mismatch:
msg("Waiting for group {} to reach desired group state.\n"
"Mismatch: {}"
.format(self.group_id, mismatch.describe()))
raise TransientRetryError(mismatch.describe())
msg("Success: desired group state reached:\n{}\nmatches:\n{}"
.format(group_state['group'], matcher))
return rcs
def poll():
return self.get_scaling_group_state(rcs, [200]).addCallback(check)
return retry_and_timeout(
poll, timeout,
can_retry=terminal_errors_except(TransientRetryError),
next_interval=repeating_interval(period),
clock=clock or reactor,
deferred_description=(
"Waiting for group {} to reach state {}"
.format(self.group_id, str(matcher)))
)
def wait_for_stack_list(self, expected_states, timeout=180, period=10):
def check(content):
states = pbag([s['stack_status'] for s in content['stacks']])
if not (states == expected_states):
msg("Waiting for group {} to reach desired group state.\n"
"{} (actual) {} (expected)"
.format(self.group.group_id, states, expected_states))
raise TransientRetryError(
"Group states of {} did not match expected {})"
.format(states, expected_states))
msg("Success: desired group state reached:\n{}"
.format(expected_states))
return self.rcs
def poll():
return self.get_stack_list().addCallback(check)
expected_states = pbag(expected_states)
return retry_and_timeout(
poll, timeout,
can_retry=terminal_errors_except(TransientRetryError),
next_interval=repeating_interval(period),
clock=reactor,
deferred_description=(
"Waiting for group {} to reach state {}".format(
self.group.group_id, str(expected_states))))
def verify(_):
def check_status():
check_d = treq.head(
append_segments(server_endpoint, 'servers', server_id),
headers=headers(auth_token))
check_d.addCallback(check_success, [404])
return check_d
start_time = clock.seconds()
timeout_description = (
"Waiting for Nova to actually delete server {0}".format(server_id))
verify_d = retry_and_timeout(check_status, timeout,
next_interval=repeating_interval(interval),
clock=clock,
deferred_description=timeout_description)
def on_success(_):
time_delete = clock.seconds() - start_time
del_log.msg('Server deleted successfully: {time_delete} seconds.',
time_delete=time_delete)
verify_d.addCallback(on_success)
verify_d.addErrback(del_log.err)
def wait_for_state(self, rcs, matcher, timeout=600, period=10, clock=None):
"""
Wait for the state on the scaling group to match the provided matchers,
specified by matcher.
:param rcs: a :class:`otter.integration.lib.resources.TestResources`
instance
:param matcher: A :mod:`testtool.matcher`, as specified in
module: testtools.matchers in
http://testtools.readthedocs.org/en/latest/api.html.
:param timeout: The amount of time to wait until this step is
considered failed.
:param period: How long to wait before polling again.
:param clock: a :class:`twisted.internet.interfaces.IReactorTime`
provider
:return: None, if the state is reached
:raises: :class:`TimedOutError` if the state is never reached within
the requisite amount of time.
Example usage:
```
matcher = MatchesAll(
IncludesServers(included_server_ids),
ExcludesServers(exclude_server_ids),
ContainsDict({
'pending': Equals(0),
'desired': Equals(5),
'status': Equals('ACTIVE')
})
)
..wait_for_state(rcs, matchers, timeout=60)
```
"""
def check(result):
response, group_state = result
mismatch = matcher.match(group_state['group'])
if mismatch:
msg("Waiting for group {} to reach desired group state.\n"
"Mismatch: {}".format(self.group_id, mismatch.describe()))
raise TransientRetryError(mismatch.describe())
msg("Success: desired group state reached:\n{}\nmatches:\n{}".
format(group_state['group'], matcher))
return rcs
def poll():
return self.get_scaling_group_state(rcs, [200]).addCallback(check)
return retry_and_timeout(
poll,
timeout,
can_retry=terminal_errors_except(TransientRetryError),
next_interval=repeating_interval(period),
clock=clock or reactor,
deferred_description=(
"Waiting for group {} to reach state {}".format(
self.group_id, str(matcher))))
def retrier(*args, **kwargs):
return retry_and_timeout(
partial(f, *args, **kwargs),
timeout,
can_retry=terminal_errors_except(TransientRetryError),
next_interval=repeating_interval(period),
clock=clock,
deferred_description=reason)
def test_repeating_interval_always_returns_interval(self):
"""
``repeating_interval`` returns the same interval no matter what the
failure
"""
next_interval = repeating_interval(3)
for exception in (DummyException(), NotImplementedError()):
self.assertEqual(next_interval(Failure(exception)), 3)
def retrier(*args, **kwargs):
return retry_and_timeout(
partial(f, *args, **kwargs), timeout,
can_retry=terminal_errors_except(TransientRetryError),
next_interval=repeating_interval(period),
clock=clock,
deferred_description=reason
)
def test_repeating_interval_always_returns_interval(self):
"""
``repeating_interval`` returns the same interval no matter what the
failure
"""
next_interval = repeating_interval(3)
for exception in (DummyException(), NotImplementedError()):
self.assertEqual(next_interval(Failure(exception)), 3)
def authenticate_tenant(self, tenant_id, log=None):
"""
see :meth:`IAuthenticator.authenticate_tenant`
"""
return retry(
partial(self._authenticator.authenticate_tenant, tenant_id, log=log),
can_retry=retry_times(self._max_retries),
next_interval=repeating_interval(self._retry_interval),
clock=self._reactor)
def authenticate_tenant(self, tenant_id, log=None):
"""
see :meth:`IAuthenticator.authenticate_tenant`
"""
return retry(partial(self._authenticator.authenticate_tenant,
tenant_id,
log=log),
can_retry=retry_times(self._max_retries),
next_interval=repeating_interval(self._retry_interval),
clock=self._reactor)
def verified_delete(log,
server_endpoint,
auth_token,
server_id,
interval=10,
timeout=3660,
clock=None):
"""
Attempt to delete a server from the server endpoint, and ensure that it is
deleted by trying again until deleting/getting the server results in a 404
or until ``OS-EXT-STS:task_state`` in server details is 'deleting',
indicating that Nova has acknowledged that the server is to be deleted
as soon as possible.
Time out attempting to verify deletes after a period of time and log an
error.
:param log: A bound logger.
:param str server_endpoint: Server endpoint URI.
:param str auth_token: Keystone Auth token.
:param str server_id: Opaque nova server id.
:param int interval: Deletion interval in seconds - how long until
verifying a delete is retried. Default: 5.
:param int timeout: Seconds after which the deletion will be logged as a
failure, if Nova fails to return a 404. Default is 3660, because if
the server is building, the delete will not happen until immediately
after it has finished building.
:return: Deferred that fires when the expected status has been seen.
"""
serv_log = log.bind(server_id=server_id)
serv_log.msg('Deleting server')
if clock is None: # pragma: no cover
from twisted.internet import reactor
clock = reactor
timeout_description = (
"Waiting for Nova to actually delete server {0} (or acknowledge delete)"
.format(server_id))
d = retry_and_timeout(
partial(delete_and_verify, serv_log, server_endpoint, auth_token, server_id),
timeout,
next_interval=repeating_interval(interval),
clock=clock,
deferred_description=timeout_description)
d.addCallback(log_with_time, clock, serv_log, clock.seconds(),
('Server deleted successfully (or acknowledged by Nova as '
'to-be-deleted) : {time_delete} seconds.'), 'time_delete')
d.addErrback(serv_log.err)
return d
def add_to_load_balancer(log, endpoint, auth_token, lb_config, ip_address, undo, clock=None):
"""
Add an IP addressed to a load balancer based on the lb_config.
TODO: Handle load balancer node metadata.
:param log: A bound logger
:param str endpoint: Load balancer endpoint URI.
:param str auth_token: Keystone Auth Token.
:param str lb_config: An lb_config dictionary.
:param str ip_address: The IP Address of the node to add to the load
balancer.
:param IUndoStack undo: An IUndoStack to push any reversable operations onto.
:return: Deferred that fires with the Add Node to load balancer response
as a dict.
"""
lb_id = lb_config['loadBalancerId']
port = lb_config['port']
path = append_segments(endpoint, 'loadbalancers', str(lb_id), 'nodes')
lb_log = log.bind(loadbalancer_id=lb_id)
def add():
d = treq.post(path, headers=headers(auth_token),
data=json.dumps({"nodes": [{"address": ip_address,
"port": port,
"condition": "ENABLED",
"type": "PRIMARY"}]}),
log=lb_log)
d.addCallback(check_success, [200, 202])
d.addErrback(log_lb_unexpected_errors, path, lb_log, 'add_node')
return d
d = retry(
add,
can_retry=retry_times(config_value('worker.lb_max_retries') or LB_MAX_RETRIES),
next_interval=repeating_interval(
config_value('worker.lb_retry_interval') or LB_RETRY_INTERVAL),
clock=clock)
def when_done(result):
lb_log.msg('Added to load balancer')
undo.push(remove_from_load_balancer,
lb_log,
endpoint,
auth_token,
lb_id,
result['nodes'][0]['id'])
return result
return d.addCallback(treq.json_content).addCallback(when_done)
def wait_for_servers(rcs,
pool,
matcher,
group=None,
timeout=600,
period=10,
clock=None,
_treq=treq):
"""
Wait until Nova reaches a particular state (as described by the given
matcher) - if a group is provided, then match only the servers for the
given group.
:param rcs: an instance of
:class:`otter.integration.lib.resources.TestResources`
:param pool: a :class:`twisted.web.client.HTTPConnectionPool`
:param matcher: a :mod:`testtools.matcher` matcher that describes the
desired state of the servers belonging to the autoscaling group.
:param group: a :class:`otter.integration.lib.autoscale.ScalingGroup` that
specifies which autoscaling group's servers we are looking at. This
group should already exist, and have a `group_id` attribute. If not
provided, the matcher will apply to all servers.
"""
message = "Waiting for {0} Nova servers".format(
"all" if group is None else "group {0} 's".format(group.group_id))
@inlineCallbacks
def do_work():
servers = yield list_servers(rcs, pool, _treq=_treq)
servers = servers['servers']
if group is not None:
servers = [
server for server in servers
if (group.group_id == server['metadata'].get(
"rax:autoscale:group:id", None))
]
mismatch = matcher.match(servers)
if mismatch:
msg("{0}.\nMismatch: {1}".format(message, mismatch.describe()))
raise TransientRetryError(mismatch.describe())
returnValue(servers)
return retry_and_timeout(
do_work,
timeout,
can_retry=terminal_errors_except(TransientRetryError),
next_interval=repeating_interval(period),
clock=clock or reactor,
deferred_description=("{0} to reach state {1}".format(
message, str(matcher))))
def test_retry_sequence_fails_if_mismatch_sequence(self):
"""
Fail if the wrong number of performers are given.
"""
r = Retry(
effect=Effect(1),
should_retry=ShouldDelayAndRetry(
can_retry=retry_times(5),
next_interval=repeating_interval(10)))
seq = [
retry_sequence(r, [lambda _: raise_(Exception()),
lambda _: raise_(Exception())])
]
self.assertRaises(AssertionError,
perform_sequence, seq, Effect(r))
def test_retry_sequence_retries_without_delays(self):
"""
Perform the wrapped effect with the performers given,
without any delay even if the original intent had a delay.
"""
r = Retry(
effect=Effect(1),
should_retry=ShouldDelayAndRetry(
can_retry=retry_times(5),
next_interval=repeating_interval(10)))
seq = [
retry_sequence(r, [lambda _: raise_(Exception()),
lambda _: raise_(Exception()),
lambda _: "yay done"])
]
self.assertEqual(perform_sequence(seq, Effect(r)), "yay done")
def test_do_not_have_to_expect_an_exact_can_retry(self):
"""
The expected retry intent does not actually have to specify the
exact ``can_retry`` function, since it might just be a lambda,
which is hard to compare or hash.
"""
expected = Retry(effect=Effect(1), should_retry=ANY)
actual = Retry(effect=Effect(1), should_retry=ShouldDelayAndRetry(
can_retry=lambda _: False,
next_interval=repeating_interval(10)))
seq = [
retry_sequence(expected, [lambda _: raise_(Exception())])
]
self.assertRaises(Exception,
perform_sequence, seq, Effect(actual))
def wait_for_servers(rcs, pool, matcher, group=None, timeout=600, period=10,
clock=None, _treq=treq):
"""
Wait until Nova reaches a particular state (as described by the given
matcher) - if a group is provided, then match only the servers for the
given group.
:param rcs: an instance of
:class:`otter.integration.lib.resources.TestResources`
:param pool: a :class:`twisted.web.client.HTTPConnectionPool`
:param matcher: a :mod:`testtools.matcher` matcher that describes the
desired state of the servers belonging to the autoscaling group.
:param group: a :class:`otter.integration.lib.autoscale.ScalingGroup` that
specifies which autoscaling group's servers we are looking at. This
group should already exist, and have a `group_id` attribute. If not
provided, the matcher will apply to all servers.
"""
message = "Waiting for {0} Nova servers".format(
"all" if group is None else "group {0} 's".format(group.group_id))
@inlineCallbacks
def do_work():
servers = yield list_servers(rcs, pool, _treq=_treq)
servers = servers['servers']
if group is not None:
servers = [
server for server in servers
if (group.group_id ==
server['metadata'].get("rax:autoscale:group:id", None))
]
mismatch = matcher.match(servers)
if mismatch:
msg("{0}.\nMismatch: {1}".format(message, mismatch.describe()))
raise TransientRetryError(mismatch.describe())
returnValue(servers)
return retry_and_timeout(
do_work, timeout,
can_retry=terminal_errors_except(TransientRetryError),
next_interval=repeating_interval(period),
clock=clock or reactor,
deferred_description=(
"{0} to reach state {1}".format(message, str(matcher)))
)
def test_fallback(self):
"""
Accept a ``fallback`` dispatcher that will be used if a performer
returns an effect for an intent that is not covered by the base
dispatcher.
"""
def dispatch_2(intent):
if intent == 2:
return sync_performer(lambda d, i: "yay done")
r = Retry(
effect=Effect(1),
should_retry=ShouldDelayAndRetry(
can_retry=retry_times(5),
next_interval=repeating_interval(10)))
seq = [
retry_sequence(r, [lambda _: Effect(2)],
fallback_dispatcher=ComposedDispatcher(
[dispatch_2, base_dispatcher]))
]
self.assertEqual(perform_sequence(seq, Effect(r)), "yay done")
def delete(self, rcs):
"""
Delete the server.
:param rcs: an instance of
:class:`otter.integration.lib.resources.TestResources`
"""
def try_delete():
d = self.treq.delete(
"{}/servers/{}".format(rcs.endpoints["nova"], self.id),
headers=headers(str(rcs.token)),
pool=self.pool)
d.addCallback(check_success, [404], _treq=self.treq)
d.addCallback(self.treq.content)
return d
return retry_and_timeout(
try_delete, 120,
can_retry=terminal_errors_except(APIError),
next_interval=repeating_interval(5),
clock=self.clock,
deferred_description=(
"Waiting for server {} to get deleted".format(self.id)))
def verify(_):
def check_status():
check_d = treq.head(append_segments(server_endpoint, 'servers',
server_id),
headers=headers(auth_token))
check_d.addCallback(check_success, [404])
return check_d
start_time = clock.seconds()
# this is treating all errors as transient, so the only error that can
# occur is a CancelledError from timing out
verify_d = retry_and_timeout(
check_status,
timeout,
next_interval=repeating_interval(interval),
clock=clock)
def on_success(_):
time_delete = clock.seconds() - start_time
del_log.msg('Server deleted successfully: {time_delete} seconds.',
time_delete=time_delete)
verify_d.addCallback(on_success)
def on_timeout(_):
time_delete = clock.seconds() - start_time
del_log.err(
None,
timeout=timeout,
time_delete=time_delete,
why=('Server {instance_id} failed to be deleted within '
'a {timeout} second timeout (it has been '
'{time_delete} seconds).'))
verify_d.addErrback(on_timeout)
def delete(self, rcs):
"""
Delete the server.
:param rcs: an instance of
:class:`otter.integration.lib.resources.TestResources`
"""
def try_delete():
d = self.treq.delete("{}/servers/{}".format(
rcs.endpoints["nova"], self.id),
headers=headers(str(rcs.token)),
pool=self.pool)
d.addCallback(check_success, [404], _treq=self.treq)
d.addCallback(self.treq.content)
return d
return retry_and_timeout(
try_delete,
120,
can_retry=terminal_errors_except(APIError),
next_interval=repeating_interval(5),
clock=self.clock,
deferred_description=(
"Waiting for server {} to get deleted".format(self.id)))
def create_server(server_endpoint,
auth_token,
server_config,
log=None,
clock=None,
retries=3,
create_failure_delay=5,
_treq=None):
"""
Create a new server. If there is an error from Nova from this call,
checks to see if the server was created anyway. If not, will retry the
create ``retries`` times (checking each time if a server).
If the error from Nova is a 400, does not retry, because that implies that
retrying will just result in another 400 (bad args).
If checking to see if the server is created also results in a failure,
does not retry because there might just be something wrong with Nova.
:param str server_endpoint: Server endpoint URI.
:param str auth_token: Keystone Auth Token.
:param dict server_config: Nova server config.
:param: int retries: Number of tries to retry the create.
:param: int create_failure_delay: how much time in seconds to wait after
a create server failure before checking Nova to see if a server
was created
:param log: logger
:type log: :class:`otter.log.bound.BoundLog`
:param _treq: To be used for testing - what treq object to use
:type treq: something with the same api as :obj:`treq`
:return: Deferred that fires with the CreateServer response as a dict.
"""
path = append_segments(server_endpoint, 'servers')
if _treq is None: # pragma: no cover
_treq = treq
if clock is None: # pragma: no cover
from twisted.internet import reactor
clock = reactor
def _check_results(result, propagated_f):
"""
Return the original failure, if checking a server resulted in a
failure too. Returns a wrapped propagated failure, if there were no
servers created, so that the retry utility knows that server creation
can be retried.
"""
if isinstance(result, Failure):
log.msg(
"Attempt to find a created server in nova resulted in "
"{failure}. Propagating the original create error instead.",
failure=result)
return propagated_f
if result is None:
raise _NoCreatedServerFound(propagated_f)
return result
def _check_server_created(f):
"""
If creating a server failed with anything other than a 400, see if
Nova created a server anyway (a 400 means that the server creation args
were bad, and there is no point in retrying).
If Nova created a server, just return it and pretend that the error
never happened. If it didn't, or if checking resulted in another
failure response, return a failure of some type.
"""
f.trap(APIError)
if f.value.code == 400:
return f
d = deferLater(clock,
create_failure_delay,
find_server,
server_endpoint,
auth_token,
server_config,
log=log)
d.addBoth(_check_results, f)
return d
def _create_with_delay(to_delay):
d = _treq.post(path,
headers=headers(auth_token),
data=json.dumps({'server': server_config}),
log=log)
if to_delay:
# Add 1 second delay to space 1 second between server creations
d.addCallback(delay, clock, 1)
return d
def _create_server():
"""
Attempt to create a server, handling spurious non-400 errors from Nova
by seeing if Nova created a server anyway in spite of the error. If so
then create server succeeded.
If not, and if no further errors occur, server creation can be retried.
"""
sem = get_sempahore("create_server", "worker.create_server_limit")
if sem is not None:
d = sem.run(_create_with_delay, True)
else:
d = _create_with_delay(False)
d.addCallback(check_success, [202], _treq=_treq)
d.addCallback(_treq.json_content)
d.addErrback(_check_server_created)
return d
def _unwrap_NoCreatedServerFound(f):
"""
The original failure was wrapped in a :class:`_NoCreatedServerFound`
for ease of retry, but that should not be the final error propagated up
by :func:`create_server`.
This errback unwraps the :class:`_NoCreatedServerFound` error and
returns the original failure.
"""
f.trap(_NoCreatedServerFound)
return f.value.original
d = retry(_create_server,
can_retry=compose_retries(
retry_times(retries),
terminal_errors_except(_NoCreatedServerFound)),
next_interval=repeating_interval(15),
clock=clock)
d.addErrback(_unwrap_NoCreatedServerFound)
d.addErrback(wrap_request_error, path, 'server_create')
return d
def wait_for_active(log,
server_endpoint,
auth_token,
server_id,
interval=20,
timeout=7200,
clock=None):
"""
Wait until the server specified by server_id's status is 'ACTIVE'
:param log: A bound logger.
:param str server_endpoint: Server endpoint URI.
:param str auth_token: Keystone Auth token.
:param str server_id: Opaque nova server id.
:param int interval: Polling interval in seconds. Default: 20.
:param int timeout: timeout to poll for the server status in seconds.
Default 7200 (2 hours).
:return: Deferred that fires when the expected status has been seen.
"""
log.msg("Checking instance status every {interval} seconds",
interval=interval)
if clock is None: # pragma: no cover
from twisted.internet import reactor
clock = reactor
start_time = clock.seconds()
def poll():
def check_status(server):
status = server['server']['status']
time_building = clock.seconds() - start_time
if status == 'ACTIVE':
log.msg(("Server changed from 'BUILD' to 'ACTIVE' within "
"{time_building} seconds"),
time_building=time_building)
return server
elif status != 'BUILD':
log.msg("Server changed to '{status}' in {time_building} seconds",
time_building=time_building, status=status)
raise UnexpectedServerStatus(
server_id,
status,
'ACTIVE')
else:
raise TransientRetryError() # just poll again
sd = server_details(server_endpoint, auth_token, server_id, log=log)
sd.addCallback(check_status)
return sd
timeout_description = ("Waiting for server <{0}> to change from BUILD "
"state to ACTIVE state").format(server_id)
return retry_and_timeout(
poll, timeout,
can_retry=transient_errors_except(UnexpectedServerStatus, ServerDeleted),
next_interval=repeating_interval(interval),
clock=clock,
deferred_description=timeout_description)
def launch_server(log, request_bag, scaling_group, launch_config, undo,
clock=None):
"""
Launch a new server given the launch config auth tokens and service
catalog. Possibly adding the newly launched server to a load balancer.
:param BoundLog log: A bound logger.
:param request_bag: An object with a bunch of useful data on it, including
a callable to re-auth and get a new token.
:param IScalingGroup scaling_group: The scaling group to add the launched
server to.
:param dict launch_config: A launch_config args structure as defined for
the launch_server_v1 type.
:param IUndoStack undo: The stack that will be rewound if undo fails.
:return: Deferred that fires with a 2-tuple of server details and the
list of load balancer responses from add_to_load_balancers.
"""
launch_config = prepare_launch_config(scaling_group.uuid, launch_config)
cloudServersOpenStack = config_value('cloudServersOpenStack')
server_endpoint = public_endpoint_url(request_bag.service_catalog,
cloudServersOpenStack,
request_bag.region)
lb_config = launch_config.get('loadBalancers', [])
server_config = launch_config['server']
log = log.bind(server_name=server_config['name'])
ilog = [None]
def check_metadata(server):
# sanity check to make sure the metadata didn't change - can probably
# be removed after a while if we do not see any log messages from this
# function
expected = launch_config['server']['metadata']
result = server['server'].get('metadata')
if result != expected:
ilog[0].msg('Server metadata has changed.',
sanity_check=True,
expected_metadata=expected,
nova_metadata=result)
return server
def wait_for_server(server, new_request_bag):
server_id = server['server']['id']
# NOTE: If server create is retried, each server delete will be pushed
# to undo stack even after it will be deleted in check_error which is
# fine since verified_delete succeeds on deleted server
undo.push(
verified_delete, log, server_endpoint, new_request_bag, server_id)
ilog[0] = log.bind(server_id=server_id)
return wait_for_active(
ilog[0],
server_endpoint,
new_request_bag.auth_token,
server_id).addCallback(check_metadata)
def add_lb(server, new_request_bag):
if lb_config:
lbd = add_to_load_balancers(
ilog[0], new_request_bag, lb_config, server, undo)
lbd.addCallback(lambda lb_response: (server, lb_response))
return lbd
return (server, [])
def _real_create_server(new_request_bag):
auth_token = new_request_bag.auth_token
d = create_server(server_endpoint, auth_token, server_config, log=log)
d.addCallback(wait_for_server, new_request_bag)
d.addCallback(add_lb, new_request_bag)
return d
def _create_server():
return request_bag.re_auth().addCallback(_real_create_server)
def check_error(f):
f.trap(UnexpectedServerStatus)
if f.value.status == 'ERROR':
log.msg('{server_id} errored, deleting and creating new '
'server instead', server_id=f.value.server_id)
# trigger server delete and return True to allow retry
verified_delete(log, server_endpoint, request_bag,
f.value.server_id)
return True
else:
return False
d = retry(_create_server,
can_retry=compose_retries(retry_times(3), check_error),
next_interval=repeating_interval(15), clock=clock)
return d
def create_server(server_endpoint, auth_token, server_config, log=None,
clock=None, retries=3, create_failure_delay=5, _treq=None):
"""
Create a new server. If there is an error from Nova from this call,
checks to see if the server was created anyway. If not, will retry the
create ``retries`` times (checking each time if a server).
If the error from Nova is a 400, does not retry, because that implies that
retrying will just result in another 400 (bad args).
If checking to see if the server is created also results in a failure,
does not retry because there might just be something wrong with Nova.
:param str server_endpoint: Server endpoint URI.
:param str auth_token: Keystone Auth Token.
:param dict server_config: Nova server config.
:param: int retries: Number of tries to retry the create.
:param: int create_failure_delay: how much time in seconds to wait after
a create server failure before checking Nova to see if a server
was created
:param log: logger
:type log: :class:`otter.log.bound.BoundLog`
:param _treq: To be used for testing - what treq object to use
:type treq: something with the same api as :obj:`treq`
:return: Deferred that fires with the CreateServer response as a dict.
"""
path = append_segments(server_endpoint, 'servers')
if _treq is None: # pragma: no cover
_treq = treq
if clock is None: # pragma: no cover
from twisted.internet import reactor
clock = reactor
def _check_results(result, propagated_f):
"""
Return the original failure, if checking a server resulted in a
failure too. Returns a wrapped propagated failure, if there were no
servers created, so that the retry utility knows that server creation
can be retried.
"""
if isinstance(result, Failure):
log.msg("Attempt to find a created server in nova resulted in "
"{failure}. Propagating the original create error instead.",
failure=result)
return propagated_f
if result is None:
raise _NoCreatedServerFound(propagated_f)
return result
def _check_server_created(f):
"""
If creating a server failed with anything other than a 400, see if
Nova created a server anyway (a 400 means that the server creation args
were bad, and there is no point in retrying).
If Nova created a server, just return it and pretend that the error
never happened. If it didn't, or if checking resulted in another
failure response, return a failure of some type.
"""
f.trap(APIError)
if f.value.code == 400:
return f
d = deferLater(clock, create_failure_delay, find_server,
server_endpoint, auth_token, server_config, log=log)
d.addBoth(_check_results, f)
return d
def _create_with_delay(to_delay):
d = _treq.post(path, headers=headers(auth_token),
data=json.dumps({'server': server_config}), log=log)
if to_delay:
# Add 1 second delay to space 1 second between server creations
d.addCallback(delay, clock, 1)
return d
def _create_server():
"""
Attempt to create a server, handling spurious non-400 errors from Nova
by seeing if Nova created a server anyway in spite of the error. If so
then create server succeeded.
If not, and if no further errors occur, server creation can be retried.
"""
sem = get_sempahore("create_server", "worker.create_server_limit")
if sem is not None:
d = sem.run(_create_with_delay, True)
else:
d = _create_with_delay(False)
d.addCallback(check_success, [202], _treq=_treq)
d.addCallback(_treq.json_content)
d.addErrback(_check_server_created)
return d
def _unwrap_NoCreatedServerFound(f):
"""
The original failure was wrapped in a :class:`_NoCreatedServerFound`
for ease of retry, but that should not be the final error propagated up
by :func:`create_server`.
This errback unwraps the :class:`_NoCreatedServerFound` error and
returns the original failure.
"""
f.trap(_NoCreatedServerFound)
return f.value.original
d = retry(
_create_server,
can_retry=compose_retries(
retry_times(retries),
terminal_errors_except(_NoCreatedServerFound)),
next_interval=repeating_interval(15), clock=clock)
d.addErrback(_unwrap_NoCreatedServerFound)
d.addErrback(wrap_request_error, path, 'server_create')
return d
def launch_server(log, region, scaling_group, service_catalog, auth_token,
launch_config, undo, clock=None):
"""
Launch a new server given the launch config auth tokens and service catalog.
Possibly adding the newly launched server to a load balancer.
:param BoundLog log: A bound logger.
:param str region: A rackspace region as found in the service catalog.
:param IScalingGroup scaling_group: The scaling group to add the launched
server to.
:param list service_catalog: A list of services as returned by the auth apis.
:param str auth_token: The user's auth token.
:param dict launch_config: A launch_config args structure as defined for
the launch_server_v1 type.
:param IUndoStack undo: The stack that will be rewound if undo fails.
:return: Deferred that fires with a 2-tuple of server details and the
list of load balancer responses from add_to_load_balancers.
"""
launch_config = prepare_launch_config(scaling_group.uuid, launch_config)
lb_region = config_value('regionOverrides.cloudLoadBalancers') or region
cloudLoadBalancers = config_value('cloudLoadBalancers')
cloudServersOpenStack = config_value('cloudServersOpenStack')
lb_endpoint = public_endpoint_url(service_catalog,
cloudLoadBalancers,
lb_region)
server_endpoint = public_endpoint_url(service_catalog,
cloudServersOpenStack,
region)
lb_config = launch_config.get('loadBalancers', [])
server_config = launch_config['server']
log = log.bind(server_name=server_config['name'])
ilog = [None]
def wait_for_server(server):
server_id = server['server']['id']
# NOTE: If server create is retried, each server delete will be pushed
# to undo stack even after it will be deleted in check_error which is fine
# since verified_delete succeeds on deleted server
undo.push(
verified_delete, log, server_endpoint, auth_token, server_id)
ilog[0] = log.bind(server_id=server_id)
return wait_for_active(
ilog[0],
server_endpoint,
auth_token,
server_id)
def add_lb(server):
ip_address = private_ip_addresses(server)[0]
lbd = add_to_load_balancers(
ilog[0], lb_endpoint, auth_token, lb_config, ip_address, undo)
lbd.addCallback(lambda lb_response: (server, lb_response))
return lbd
def _create_server():
d = create_server(server_endpoint, auth_token, server_config, log=log)
d.addCallback(wait_for_server)
d.addCallback(add_lb)
return d
def check_error(f):
f.trap(UnexpectedServerStatus)
if f.value.status == 'ERROR':
log.msg('{server_id} errored, deleting and creating new server instead',
server_id=f.value.server_id)
# trigger server delete and return True to allow retry
verified_delete(log, server_endpoint, auth_token, f.value.server_id)
return True
else:
return False
d = retry(_create_server, can_retry=compose_retries(retry_times(3), check_error),
next_interval=repeating_interval(15), clock=clock)
return d
def verified_delete(log,
server_endpoint,
auth_token,
server_id,
interval=10,
timeout=3660,
clock=None):
"""
Attempt to delete a server from the server endpoint, and ensure that it is
deleted by trying again until deleting the server results in a 404.
Time out attempting to verify deletes after a period of time and log an
error.
:param log: A bound logger.
:param str server_endpoint: Server endpoint URI.
:param str auth_token: Keystone Auth token.
:param str server_id: Opaque nova server id.
:param int interval: Deletion interval in seconds - how long until
verifying a delete is retried. Default: 5.
:param int timeout: Seconds after which the deletion will be logged as a
failure, if Nova fails to return a 404. Default is 3660, because if
the server is building, the delete will not happen until immediately
after it has finished building.
:return: Deferred that fires when the expected status has been seen.
"""
serv_log = log.bind(server_id=server_id)
serv_log.msg('Deleting server')
path = append_segments(server_endpoint, 'servers', server_id)
if clock is None: # pragma: no cover
from twisted.internet import reactor
clock = reactor
# just delete over and over until a 404 is received
def delete():
del_d = treq.delete(path, headers=headers(auth_token), log=serv_log)
del_d.addCallback(check_success, [404])
del_d.addCallback(treq.content)
return del_d
start_time = clock.seconds()
timeout_description = (
"Waiting for Nova to actually delete server {0}".format(server_id))
d = retry_and_timeout(delete, timeout,
next_interval=repeating_interval(interval),
clock=clock,
deferred_description=timeout_description)
def on_success(_):
time_delete = clock.seconds() - start_time
serv_log.msg('Server deleted successfully: {time_delete} seconds.',
time_delete=time_delete)
d.addCallback(on_success)
d.addErrback(serv_log.err)
return d
def launch_server(log,
request_bag,
scaling_group,
launch_config,
undo,
clock=None):
"""
Launch a new server given the launch config auth tokens and service
catalog. Possibly adding the newly launched server to a load balancer.
:param BoundLog log: A bound logger.
:param request_bag: An object with a bunch of useful data on it, including
a callable to re-auth and get a new token.
:param IScalingGroup scaling_group: The scaling group to add the launched
server to.
:param dict launch_config: A launch_config args structure as defined for
the launch_server_v1 type.
:param IUndoStack undo: The stack that will be rewound if undo fails.
:return: Deferred that fires with a 2-tuple of server details and the
list of load balancer responses from add_to_load_balancers.
"""
launch_config = prepare_launch_config(scaling_group.uuid, launch_config)
cloudServersOpenStack = config_value('cloudServersOpenStack')
server_endpoint = public_endpoint_url(request_bag.service_catalog,
cloudServersOpenStack,
request_bag.region)
lb_config = launch_config.get('loadBalancers', [])
server_config = launch_config['server']
log = log.bind(server_name=server_config['name'])
ilog = [None]
def check_metadata(server):
# sanity check to make sure the metadata didn't change - can probably
# be removed after a while if we do not see any log messages from this
# function
expected = launch_config['server']['metadata']
result = server['server'].get('metadata')
if result != expected:
ilog[0].msg('Server metadata has changed.',
sanity_check=True,
expected_metadata=expected,
nova_metadata=result)
return server
def wait_for_server(server, new_request_bag):
server_id = server['server']['id']
# NOTE: If server create is retried, each server delete will be pushed
# to undo stack even after it will be deleted in check_error which is
# fine since verified_delete succeeds on deleted server
undo.push(verified_delete, log, server_endpoint, new_request_bag,
server_id)
ilog[0] = log.bind(server_id=server_id)
return wait_for_active(ilog[0], server_endpoint,
new_request_bag.auth_token,
server_id).addCallback(check_metadata)
def add_lb(server, new_request_bag):
if lb_config:
lbd = add_to_load_balancers(ilog[0], new_request_bag, lb_config,
server, undo)
lbd.addCallback(lambda lb_response: (server, lb_response))
return lbd
return (server, [])
def _real_create_server(new_request_bag):
auth_token = new_request_bag.auth_token
d = create_server(server_endpoint, auth_token, server_config, log=log)
d.addCallback(wait_for_server, new_request_bag)
d.addCallback(add_lb, new_request_bag)
return d
def _create_server():
return request_bag.re_auth().addCallback(_real_create_server)
def check_error(f):
f.trap(UnexpectedServerStatus)
if f.value.status == 'ERROR':
log.msg(
'{server_id} errored, deleting and creating new '
'server instead',
server_id=f.value.server_id)
# trigger server delete and return True to allow retry
verified_delete(log, server_endpoint, request_bag,
f.value.server_id)
return True
else:
return False
d = retry(_create_server,
can_retry=compose_retries(retry_times(3), check_error),
next_interval=repeating_interval(15),
clock=clock)
return d
def wait_for_active(log,
server_endpoint,
auth_token,
server_id,
interval=5,
timeout=3600,
clock=None):
"""
Wait until the server specified by server_id's status is 'ACTIVE'
:param log: A bound logger.
:param str server_endpoint: Server endpoint URI.
:param str auth_token: Keystone Auth token.
:param str server_id: Opaque nova server id.
:param int interval: Polling interval in seconds. Default: 5.
:param int timeout: timeout to poll for the server status in seconds.
Default 3600 (1 hour)
:return: Deferred that fires when the expected status has been seen.
"""
log.msg("Checking instance status every {interval} seconds",
interval=interval)
if clock is None: # pragma: no cover
from twisted.internet import reactor
clock = reactor
start_time = clock.seconds()
def poll():
def check_status(server):
status = server['server']['status']
if status == 'ACTIVE':
time_building = clock.seconds() - start_time
log.msg(("Server changed from 'BUILD' to 'ACTIVE' within "
"{time_building} seconds"),
time_building=time_building)
return server
elif status != 'BUILD':
raise UnexpectedServerStatus(server_id, status, 'ACTIVE')
else:
raise TransientRetryError() # just poll again
sd = server_details(server_endpoint, auth_token, server_id)
sd.addCallback(check_status)
return sd
d = retry_and_timeout(
poll,
timeout,
can_retry=transient_errors_except(UnexpectedServerStatus),
next_interval=repeating_interval(interval),
clock=clock)
def on_error(f):
if f.check(CancelledError):
time_building = clock.seconds() - start_time
log.msg(
('Server {instance_id} failed to change from BUILD state '
'to ACTIVE within a {timeout} second timeout (it has been '
'{time_building} seconds).'),
timeout=timeout,
time_building=time_building)
return f
d.addErrback(on_error)
return d
def wait_for_active(log,
server_endpoint,
auth_token,
server_id,
interval=20,
timeout=7200,
clock=None):
"""
Wait until the server specified by server_id's status is 'ACTIVE'
:param log: A bound logger.
:param str server_endpoint: Server endpoint URI.
:param str auth_token: Keystone Auth token.
:param str server_id: Opaque nova server id.
:param int interval: Polling interval in seconds. Default: 20.
:param int timeout: timeout to poll for the server status in seconds.
Default 7200 (2 hours).
:return: Deferred that fires when the expected status has been seen.
"""
log.msg("Checking instance status every {interval} seconds",
interval=interval)
if clock is None: # pragma: no cover
from twisted.internet import reactor
clock = reactor
start_time = clock.seconds()
def poll():
def check_status(server):
status = server['server']['status']
time_building = clock.seconds() - start_time
if status == 'ACTIVE':
log.msg(("Server changed from 'BUILD' to 'ACTIVE' within "
"{time_building} seconds"),
time_building=time_building)
return server
elif status != 'BUILD':
log.msg(
"Server changed to '{status}' in {time_building} seconds",
time_building=time_building,
status=status)
raise UnexpectedServerStatus(server_id, status, 'ACTIVE')
else:
raise TransientRetryError() # just poll again
sd = server_details(server_endpoint, auth_token, server_id, log=log)
sd.addCallback(check_status)
return sd
timeout_description = ("Waiting for server <{0}> to change from BUILD "
"state to ACTIVE state").format(server_id)
return retry_and_timeout(poll,
timeout,
can_retry=transient_errors_except(
UnexpectedServerStatus, ServerDeleted),
next_interval=repeating_interval(interval),
clock=clock,
deferred_description=timeout_description)
def wait_for_active(log,
server_endpoint,
auth_token,
server_id,
interval=5,
timeout=3600,
clock=None):
"""
Wait until the server specified by server_id's status is 'ACTIVE'
:param log: A bound logger.
:param str server_endpoint: Server endpoint URI.
:param str auth_token: Keystone Auth token.
:param str server_id: Opaque nova server id.
:param int interval: Polling interval in seconds. Default: 5.
:param int timeout: timeout to poll for the server status in seconds.
Default 3600 (1 hour)
:return: Deferred that fires when the expected status has been seen.
"""
log.msg("Checking instance status every {interval} seconds",
interval=interval)
if clock is None: # pragma: no cover
from twisted.internet import reactor
clock = reactor
start_time = clock.seconds()
def poll():
def check_status(server):
status = server['server']['status']
if status == 'ACTIVE':
time_building = clock.seconds() - start_time
log.msg(("Server changed from 'BUILD' to 'ACTIVE' within "
"{time_building} seconds"),
time_building=time_building)
return server
elif status != 'BUILD':
raise UnexpectedServerStatus(
server_id,
status,
'ACTIVE')
else:
raise TransientRetryError() # just poll again
sd = server_details(server_endpoint, auth_token, server_id)
sd.addCallback(check_status)
return sd
d = retry_and_timeout(
poll, timeout,
can_retry=transient_errors_except(UnexpectedServerStatus),
next_interval=repeating_interval(interval),
clock=clock)
def on_error(f):
if f.check(CancelledError):
time_building = clock.seconds() - start_time
log.msg(('Server {instance_id} failed to change from BUILD state '
'to ACTIVE within a {timeout} second timeout (it has been '
'{time_building} seconds).'),
timeout=timeout, time_building=time_building)
return f
d.addErrback(on_error)
return d
