def download_charm(client, charm_id, charms_directory):
"""Retrieve a charm from the provider storage to the local machine.
"""
charm_state_manager = CharmStateManager(client)
charm_state = yield charm_state_manager.get_charm_state(charm_id)
# Calculate local charm path
checksum = yield charm_state.get_sha256()
charm_key = under.quote("%s:%s" % (charm_state.id, checksum))
local_charm_path = os.path.join(
charms_directory, charm_key)
# Retrieve charm from provider storage link
if charm_state.bundle_url.startswith("file://"):
file_path = charm_state.bundle_url[len("file://"):]
if not os.path.exists(file_path):
raise FileNotFound(charm_state.bundle_url)
shutil.copyfileobj(open(file_path), open(local_charm_path, "w"))
else:
try:
yield downloadPage(charm_state.bundle_url, local_charm_path)
except Error:
raise FileNotFound(charm_state.bundle_url)
returnValue(CharmBundle(local_charm_path))
def validate_hooks(client, unit_state, hook_names):
# Assemble a list of valid hooks for the charm.
valid_hooks = ["start", "stop", "install", "config-changed"]
service_manager = ServiceStateManager(client)
endpoints = yield service_manager.get_relation_endpoints(
unit_state.service_name)
endpoint_names = [ep.relation_name for ep in endpoints]
for endpoint_name in endpoint_names:
valid_hooks.extend([
endpoint_name + "-relation-joined",
endpoint_name + "-relation-changed",
endpoint_name + "-relation-departed",
endpoint_name + "-relation-broken",
])
# Verify the debug names.
for hook_name in hook_names:
if hook_name in valid_hooks:
continue
break
else:
returnValue(True)
# We dereference to the charm to give a fully qualified error
# message. I wish this was a little easier to dereference, the
# service_manager.get_relation_endpoints effectively does this
# already.
service_manager = ServiceStateManager(client)
service_state = yield service_manager.get_service_state(
unit_state.service_name)
charm_id = yield service_state.get_charm_id()
charm_manager = CharmStateManager(client)
charm = yield charm_manager.get_charm_state(charm_id)
raise InvalidCharmHook(charm.id, hook_name)
def download_charm(client, charm_id, charms_directory):
charm_state_manager = CharmStateManager(client)
charm_state = yield charm_state_manager.get_charm_state(charm_id)
# Calculate local charm path
checksum = yield charm_state.get_sha256()
charm_key = under.quote("%s:%s" % (charm_state.id, checksum))
local_charm_path = os.path.join(charms_directory, charm_key)
# Retrieve charm from provider storage link
if charm_state.bundle_url.startswith("file://"):
file_path = charm_state.bundle_url[len("file://"):]
if not os.path.exists(file_path):
raise FileNotFound(charm_state.bundle_url)
shutil.copyfileobj(open(file_path), open(local_charm_path, "w"))
else:
try:
yield downloadPage(charm_state.bundle_url, local_charm_path)
except Error:
raise FileNotFound(charm_state.bundle_url)
returnValue(CharmBundle(local_charm_path))
def test_deploy_upgrade(self):
"""A charm can be deployed and get the latest version"""
environment = self.config.get("firstenv")
yield deploy.deploy(
self.config, environment, self.unbundled_repo_path, "local:sample",
"myblog", logging.getLogger("deploy"), [])
yield deploy.deploy(
self.config, environment, self.unbundled_repo_path, "local:sample",
"myblog2", logging.getLogger("deploy"), [], upgrade=True)
services = ServiceStateManager(self.client)
service1 = yield services.get_service_state("myblog")
s1_charm_id = yield service1.get_charm_id()
service2 = yield services.get_service_state("myblog2")
s2_charm_id = yield service2.get_charm_id()
self.assertNotEqual(s1_charm_id, s2_charm_id)
charms = CharmStateManager(self.client)
charm1 = yield charms.get_charm_state(s1_charm_id)
charm2 = yield charms.get_charm_state(s2_charm_id)
self.assertEqual(charm1.revision + 1, charm2.revision)
class CharmStateManagerTest(StateTestBase):
@inlineCallbacks
def setUp(self):
yield super(CharmStateManagerTest, self).setUp()
self.charm_state_manager = CharmStateManager(self.client)
self.charm_id = local_charm_id(self.charm)
@inlineCallbacks
def test_add_charm(self):
"""
Adding a Charm into a CharmStateManager should register
the charm within the Zookeeper state, according to the
specification.
"""
charm_state = yield self.charm_state_manager.add_charm_state(
self.charm_id, self.charm, "http://example.com/abc")
self.assertEquals(charm_state.id, "local:series/dummy-1")
children = yield self.client.get_children("/charms")
self.assertEquals(children, ["local_3a_series_2f_dummy-1"])
content, stat = yield self.client.get(
"/charms/local_3a_series_2f_dummy-1")
charm_data = yaml.load(content)
self.assertEquals(
charm_data, {
"metadata": self.charm.metadata.get_serialization_data(),
"config": self.charm.config.get_serialization_data(),
"sha256": self.charm.get_sha256(),
"url": "http://example.com/abc"
})
@inlineCallbacks
def test_get_charm(self):
"""
A CharmState should be available if one get()s a charm
that was previously added into the manager.
"""
yield self.charm_state_manager.add_charm_state(self.charm_id,
self.charm, "")
charm_state = yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
self.assertEquals(charm_state.id, "local:series/dummy-1")
@inlineCallbacks
def test_charm_state_attributes(self):
"""
Verify that the basic (invariant) attributes of the
CharmState are correctly in place.
"""
yield self.charm_state_manager.add_charm_state(
self.charm_id, self.charm, "http://example.com/abc")
charm_state = yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
self.assertEquals(charm_state.name, "dummy")
self.assertEquals(charm_state.revision, 1)
self.assertEquals(charm_state.id, "local:series/dummy-1")
self.assertEquals(charm_state.bundle_url, "http://example.com/abc")
@inlineCallbacks
def test_charm_state_metadata(self):
"""
Check that the charm metadata was correctly saved and loaded.
"""
yield self.charm_state_manager.add_charm_state(self.charm_id,
self.charm, "")
charm_state = yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
metadata = yield charm_state.get_metadata()
self.assertEquals(metadata.name, "dummy")
@inlineCallbacks
def test_charm_state_config_options(self):
"""Verify ConfigOptions present and correct."""
from juju.charm.tests.test_config import sample_yaml_data
yield self.charm_state_manager.add_charm_state(self.charm_id,
self.charm, "")
charm_state = yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
config = yield charm_state.get_config()
self.assertEquals(config.get_serialization_data(), sample_yaml_data)
@inlineCallbacks
def test_get_non_existing_charm_prior_to_initialization(self):
"""
Getting a charm before the charms node was even
initialized should raise an error about the charm not
being present.
"""
try:
yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
except CharmStateNotFound, e:
self.assertEquals(e.charm_id, "local:series/dummy-1")
else:
def collect(scope, machine_provider, client, log):
"""Extract status information into nested dicts for rendering.
`scope`: an optional list of name specifiers. Globbing based
wildcards supported. Defaults to all units, services and
relations.
`machine_provider`: machine provider for the environment
`client`: ZK client connection
`log`: a Python stdlib logger.
"""
service_manager = ServiceStateManager(client)
relation_manager = RelationStateManager(client)
machine_manager = MachineStateManager(client)
charm_manager = CharmStateManager(client)
service_data = {}
machine_data = {}
state = dict(services=service_data, machines=machine_data)
seen_machines = set()
filter_services, filter_units = digest_scope(scope)
services = yield service_manager.get_all_service_states()
for service in services:
if len(filter_services):
found = False
for filter_service in filter_services:
if fnmatch(service.service_name, filter_service):
found = True
break
if not found:
continue
unit_data = {}
relation_data = {}
charm_id = yield service.get_charm_id()
charm = yield charm_manager.get_charm_state(charm_id)
service_data[service.service_name] = dict(units=unit_data,
charm=charm.id,
relations=relation_data)
exposed = yield service.get_exposed_flag()
if exposed:
service_data[service.service_name].update(exposed=exposed)
units = yield service.get_all_unit_states()
unit_matched = False
relations = yield relation_manager.get_relations_for_service(service)
for unit in units:
if len(filter_units):
found = False
for filter_unit in filter_units:
if fnmatch(unit.unit_name, filter_unit):
found = True
break
if not found:
continue
u = unit_data[unit.unit_name] = dict()
machine_id = yield unit.get_assigned_machine_id()
u["machine"] = machine_id
unit_workflow_client = WorkflowStateClient(client, unit)
unit_state = yield unit_workflow_client.get_state()
if not unit_state:
u["state"] = "pending"
else:
unit_connected = yield unit.has_agent()
u["state"] = unit_state if unit_connected else "down"
if exposed:
open_ports = yield unit.get_open_ports()
u["open-ports"] = ["{port}/{proto}".format(**port_info)
for port_info in open_ports]
u["public-address"] = yield unit.get_public_address()
# indicate we should include information about this
# machine later
seen_machines.add(machine_id)
unit_matched = True
# collect info on each relation for the service unit
relation_status = {}
for relation in relations:
try:
relation_unit = yield relation.get_unit_state(unit)
except UnitRelationStateNotFound:
# This exception will occur when relations are
# established between services without service
# units, and therefore never have any
# corresponding service relation units. This
# scenario does not occur in actual deployments,
# but can happen in test circumstances. In
# particular, it will happen with a misconfigured
# provider, which exercises this codepath.
continue # should not occur, but status should not fail
relation_workflow_client = WorkflowStateClient(
client, relation_unit)
relation_workflow_state = \
yield relation_workflow_client.get_state()
relation_status[relation.relation_name] = dict(
state=relation_workflow_state)
u["relations"] = relation_status
# after filtering units check if any matched or remove the
# service from the output
if filter_units and not unit_matched:
del service_data[service.service_name]
continue
for relation in relations:
rel_services = yield relation.get_service_states()
# A single related service implies a peer relation. More
# imply a bi-directional provides/requires relationship.
# In the later case we omit the local side of the relation
# when reporting.
if len(rel_services) > 1:
# Filter out self from multi-service relations.
rel_services = [
rsn for rsn in rel_services if rsn.service_name !=
service.service_name]
if len(rel_services) > 1:
raise ValueError("Unexpected relationship with more "
"than 2 endpoints")
rel_service = rel_services[0]
relation_data[relation.relation_name] = rel_service.service_name
machines = yield machine_manager.get_all_machine_states()
for machine_state in machines:
if (filter_services or filter_units) and \
machine_state.id not in seen_machines:
continue
instance_id = yield machine_state.get_instance_id()
m = {"instance-id": instance_id \
if instance_id is not None else "pending"}
if instance_id is not None:
try:
pm = yield machine_provider.get_machine(instance_id)
m["dns-name"] = pm.dns_name
m["instance-state"] = pm.state
if (yield machine_state.has_agent()):
# if the agent's connected, we're fine
m["state"] = "running"
else:
units = (yield machine_state.get_all_service_unit_states())
for unit in units:
unit_workflow_client = WorkflowStateClient(client, unit)
if (yield unit_workflow_client.get_state()):
# for unit to have a state, its agent must have
# run, which implies the machine agent must have
# been running correctly at some point in the past
m["state"] = "down"
break
else:
# otherwise we're probably just still waiting
m["state"] = "not-started"
except ProviderError:
# The provider doesn't have machine information
log.error(
"Machine provider information missing: machine %s" % (
machine_state.id))
machine_data[machine_state.id] = m
returnValue(state)
class MachineAgent(BaseAgent):
"""An juju Machine Agent.
The machine agent is responsible for monitoring service units
assigned to a machine. If a new unit is assigned to machine, the
machine agent will download the charm, create a working
space for the service unit agent, and then launch it.
Additionally the machine agent will monitor the running service
unit agents on the machine, via their ephemeral nodes, and
restart them if they die.
"""
name = "juju-machine-agent"
unit_agent_module = "juju.agents.unit"
@property
def charms_directory(self):
return os.path.join(self.config["juju_directory"], "charms")
@property
def units_directory(self):
return os.path.join(self.config["juju_directory"], "units")
@property
def unit_state_directory(self):
return os.path.join(self.config["juju_directory"], "state")
@inlineCallbacks
def start(self):
"""Start the machine agent.
Creates state directories on the machine, retrieves the machine state,
and enables watch on assigned units.
"""
# Initialize directory paths.
if not os.path.exists(self.charms_directory):
os.makedirs(self.charms_directory)
if not os.path.exists(self.units_directory):
os.makedirs(self.units_directory)
if not os.path.exists(self.unit_state_directory):
os.makedirs(self.unit_state_directory)
# Get state managers we'll be utilizing.
self.service_state_manager = ServiceStateManager(self.client)
self.charm_state_manager = CharmStateManager(self.client)
# Retrieve the machine state for the machine we represent.
machine_manager = MachineStateManager(self.client)
self.machine_state = yield machine_manager.get_machine_state(
self.get_machine_id())
# Watch assigned units for the machine.
if self.get_watch_enabled():
self.machine_state.watch_assigned_units(
self.watch_service_units)
# Find out what provided the machine, and how to deploy units.
settings = GlobalSettingsStateManager(self.client)
self.provider_type = yield settings.get_provider_type()
self.deploy_factory = get_deploy_factory(self.provider_type)
# Connect the machine agent, broadcasting presence to the world.
yield self.machine_state.connect_agent()
log.info("Machine agent started id:%s deploy:%r provider:%r" % (
self.get_machine_id(), self.deploy_factory, self.provider_type))
def download_charm(self, charm_state):
"""Retrieve a charm from the provider storage to the local machine.
Utilizes a local charm cache to avoid repeated downloading of the
same charm.
"""
log.debug("Downloading charm %s to %s",
charm_state.id, self.charms_directory)
return download_charm(
self.client, charm_state.id, self.charms_directory)
@inlineCallbacks
def watch_service_units(self, old_units, new_units):
"""Callback invoked when the assigned service units change.
"""
if old_units is None:
old_units = set()
log.debug(
"Units changed old:%s new:%s", old_units, new_units)
stopped = old_units - new_units
started = new_units - old_units
for unit_name in stopped:
log.debug("Stopping service unit: %s ...", unit_name)
try:
yield self.kill_service_unit(unit_name)
except Exception:
log.exception("Error stopping unit: %s", unit_name)
for unit_name in started:
log.debug("Starting service unit: %s ...", unit_name)
try:
yield self.start_service_unit(unit_name)
except Exception:
log.exception("Error starting unit: %s", unit_name)
@inlineCallbacks
def start_service_unit(self, service_unit_name):
"""Start a service unit on the machine.
Downloads the charm, and extract it to the service unit directory,
and launch the service unit agent within the unit directory.
"""
# Retrieve the charm state to get at the charm.
unit_state = yield self.service_state_manager.get_unit_state(
service_unit_name)
charm_id = yield unit_state.get_charm_id()
charm_state = yield self.charm_state_manager.get_charm_state(
charm_id)
# Download the charm.
bundle = yield self.download_charm(charm_state)
# Use deployment to setup the workspace and start the unit agent.
deployment = self.deploy_factory(
service_unit_name, self.config["juju_directory"])
running = yield deployment.is_running()
if not running:
log.debug("Starting service unit %s", service_unit_name)
yield deployment.start(
self.get_machine_id(), self.client.servers, bundle)
log.info("Started service unit %s", service_unit_name)
def kill_service_unit(self, service_unit_name):
"""Stop service unit and destroy disk state, ala SIGKILL or lxc-destroy
"""
deployment = self.deploy_factory(
service_unit_name, self.config["juju_directory"])
log.info("Stopping service unit %s...", service_unit_name)
return deployment.destroy()
def get_machine_id(self):
"""Get the id of the machine as known within the zk state."""
return self.config["machine_id"]
def get_agent_name(self):
return "Machine:%s" % (self.get_machine_id())
def configure(self, options):
super(MachineAgent, self).configure(options)
if not options.get("machine_id"):
msg = ("--machine-id must be provided in the command line, "
"or $JUJU_MACHINE_ID in the environment")
raise JujuError(msg)
@classmethod
def setup_options(cls, parser):
super(MachineAgent, cls).setup_options(parser)
machine_id = os.environ.get("JUJU_MACHINE_ID", "")
parser.add_argument(
"--machine-id", default=machine_id)
return parser
class StatusCommand(object):
def __init__(self, client, provider, log):
"""
Callable status command object.
`client`: ZK client connection
`provider`: machine provider for the environment
`log`: a Python stdlib logger.
"""
self.client = client
self.provider = provider
self.log = log
self.service_manager = ServiceStateManager(client)
self.relation_manager = RelationStateManager(client)
self.machine_manager = MachineStateManager(client)
self.charm_manager = CharmStateManager(client)
self._reset()
def _reset(self, scope=None):
# init per-run state
# self.state is assembled by the various process methods
# intermediate access to state is made more convenient
# using these references to its internals.
self.service_data = {} # service name: service info
self.machine_data = {} # machine id: machine state
self.unit_data = {} # unit_name :unit_info
# used in collecting subordinate (which are added to state in a two
# phase pass)
self.subordinates = {} # service : set(principal service names)
self.state = dict(services=self.service_data,
machines=self.machine_data)
# Filtering info
self.seen_machines = set()
self.filter_services, self.filter_units = digest_scope(scope)
@inlineCallbacks
def __call__(self, scope=None):
"""Extract status information into nested dicts for rendering.
`scope`: an optional list of name specifiers. Globbing based wildcards
supported. Defaults to all units, services and relations.
"""
self._reset(scope)
# Pass 1 Gather Data (including principals and subordinates)
# this builds unit info and container relationships
# which is assembled in pass 2 below
yield self._process_services()
# Pass 2: Nest information according to principal/subordinates
# rules
self._process_subordinates()
yield self._process_machines()
returnValue(self.state)
@inlineCallbacks
def _process_services(self):
"""
For each service gather the following information::
<service name>:
charm: <charm name>
exposed: <expose boolean>
relations:
<relation info -- see _process_relations>
units:
<unit info -- see _process_units>
"""
services = yield self.service_manager.get_all_service_states()
for service in services:
if len(self.filter_services):
found = False
for filter_service in self.filter_services:
if fnmatch(service.service_name, filter_service):
found = True
break
if not found:
continue
yield self._process_service(service)
@inlineCallbacks
def _process_service(self, service):
"""
Gather the service info (described in _process_services).
`service`: ServiceState instance
"""
relation_data = {}
service_data = self.service_data
charm_id = yield service.get_charm_id()
charm = yield self.charm_manager.get_charm_state(charm_id)
service_data[service.service_name] = (
dict(units={},
charm=charm.id,
relations=relation_data))
if (yield service.is_subordinate()):
service_data[service.service_name]["subordinate"] = True
yield self._process_expose(service)
relations, rel_svc_map = yield self._process_relation_map(
service)
unit_matched = yield self._process_units(service,
relations,
rel_svc_map)
# after filtering units check if any matched or remove the
# service from the output
if self.filter_units and not unit_matched:
del service_data[service.service_name]
return
yield self._process_relations(service, relations, rel_svc_map)
@inlineCallbacks
def _process_units(self, service, relations, rel_svc_map):
"""
Gather unit information for a service::
<unit name>:
agent-state: <started|pendding|etc>
machine: <machine id>
open-ports: ["port/protocol", ...]
public-address: <public dns name or ip>
subordinates:
<optional nested units of subordinate services>
`service`: ServiceState intance
`relations`: list of ServiceRelationState instance for this service
`rel_svc_map`: maps relation internal ids to the remote endpoint
service name. This references the name of the remote
endpoint and so is generated per service.
"""
units = yield service.get_all_unit_states()
unit_matched = False
for unit in units:
if len(self.filter_units):
found = False
for filter_unit in self.filter_units:
if fnmatch(unit.unit_name, filter_unit):
found = True
break
if not found:
continue
yield self._process_unit(service, unit, relations, rel_svc_map)
unit_matched = True
returnValue(unit_matched)
@inlineCallbacks
def _process_unit(self, service, unit, relations, rel_svc_map):
""" Generate unit info for a single unit of a single service.
`unit`: ServiceUnitState
see `_process_units` for an explanation of other arguments.
"""
u = self.unit_data[unit.unit_name] = dict()
container = yield unit.get_container()
if container:
u["container"] = container.unit_name
self.subordinates.setdefault(unit.service_name,
set()).add(container.service_name)
machine_id = yield unit.get_assigned_machine_id()
u["machine"] = machine_id
unit_workflow_client = WorkflowStateClient(self.client, unit)
unit_state = yield unit_workflow_client.get_state()
if not unit_state:
u["agent-state"] = "pending"
else:
unit_connected = yield unit.has_agent()
u["agent-state"] = unit_state.replace("_", "-") \
if unit_connected else "down"
exposed = self.service_data[service.service_name].get("exposed")
if exposed:
open_ports = yield unit.get_open_ports()
u["open-ports"] = ["{port}/{proto}".format(**port_info)
for port_info in open_ports]
u["public-address"] = yield unit.get_public_address()
# indicate we should include information about this
# machine later
self.seen_machines.add(machine_id)
# collect info on each relation for the service unit
yield self._process_unit_relations(service, unit,
relations, rel_svc_map)
@inlineCallbacks
def _process_relation_map(self, service):
"""Generate a mapping from a services relations to the service name of
the remote endpoints.
returns: ([ServiceRelationState, ...], mapping)
"""
relation_data = self.service_data[service.service_name]["relations"]
relation_mgr = self.relation_manager
relations = yield relation_mgr.get_relations_for_service(service)
rel_svc_map = {}
for relation in relations:
rel_services = yield relation.get_service_states()
# A single related service implies a peer relation. More
# imply a bi-directional provides/requires relationship.
# In the later case we omit the local side of the relation
# when reporting.
if len(rel_services) > 1:
# Filter out self from multi-service relations.
rel_services = [
rsn for rsn in rel_services if rsn.service_name !=
service.service_name]
if len(rel_services) > 1:
raise ValueError("Unexpected relationship with more "
"than 2 endpoints")
rel_service = rel_services[0]
relation_data.setdefault(relation.relation_name, set()).add(
rel_service.service_name)
rel_svc_map[relation.internal_relation_id] = (
rel_service.service_name)
returnValue((relations, rel_svc_map))
@inlineCallbacks
def _process_relations(self, service, relations, rel_svc_map):
"""Generate relation information for a given service
Each service with relations will have a relations dict
nested under it with one or more relations described::
relations:
<relation name>:
- <remote service name>
"""
relation_data = self.service_data[service.service_name]["relations"]
for relation in relations:
rel_services = yield relation.get_service_states()
# A single related service implies a peer relation. More
# imply a bi-directional provides/requires relationship.
# In the later case we omit the local side of the relation
# when reporting.
if len(rel_services) > 1:
# Filter out self from multi-service relations.
rel_services = [
rsn for rsn in rel_services if rsn.service_name !=
service.service_name]
if len(rel_services) > 1:
raise ValueError("Unexpected relationship with more "
"than 2 endpoints")
rel_service = rel_services[0]
relation_data.setdefault(
relation.relation_name, set()).add(
rel_service.service_name)
rel_svc_map[relation.internal_relation_id] = (
rel_service.service_name)
# Normalize the sets back to lists
for r in relation_data:
relation_data[r] = sorted(relation_data[r])
@inlineCallbacks
def _process_unit_relations(self, service, unit, relations, rel_svc_map):
"""Collect UnitRelationState information per relation and per unit.
Includes information under each unit for its relations including
its relation state and information about any possible errors.
see `_process_relations` for argument information
"""
u = self.unit_data[unit.unit_name]
relation_errors = {}
for relation in relations:
try:
relation_unit = yield relation.get_unit_state(unit)
except UnitRelationStateNotFound:
# This exception will occur when relations are
# established between services without service
# units, and therefore never have any
# corresponding service relation units.
# UPDATE: common with subordinate services, and
# some testing scenarios.
continue
relation_workflow_client = WorkflowStateClient(
self.client, relation_unit)
workflow_state = yield relation_workflow_client.get_state()
rel_svc_name = rel_svc_map.get(relation.internal_relation_id)
if rel_svc_name and workflow_state not in ("up", None):
relation_errors.setdefault(
relation.relation_name, set()).add(rel_svc_name)
if relation_errors:
# Normalize sets and store.
u["relation-errors"] = dict(
[(r, sorted(relation_errors[r])) for r in relation_errors])
def _process_subordinates(self):
"""Properly nest subordinate units under their principal service's
unit nodes. Services and units are generated in one pass, then
iterated by this method to structure the output data to reflect
actual unit containment.
Subordinate units will include the follow::
subordinate: true
subordinate-to:
- <principal service names>
Principal services that have subordinates will include::
subordinates:
<subordinate unit name>:
agent-state: <agent state>
"""
service_data = self.service_data
for unit_name, u in self.unit_data.iteritems():
container = u.get("container")
if container:
d = self.unit_data[container].setdefault("subordinates", {})
d[unit_name] = u
# remove keys that don't appear in output or come from container
for key in ("container", "machine", "public-address"):
u.pop(key, None)
else:
service_name = parse_service_name(unit_name)
service_data[service_name]["units"][unit_name] = u
for sub_service, principal_services in self.subordinates.iteritems():
service_data[sub_service]["subordinate-to"] = sorted(principal_services)
service_data[sub_service].pop("units", None)
@inlineCallbacks
def _process_expose(self, service):
"""Indicate if a service is exposed or not."""
exposed = yield service.get_exposed_flag()
if exposed:
self.service_data[service.service_name].update(exposed=exposed)
returnValue(exposed)
@inlineCallbacks
def _process_machines(self):
"""Gather machine information.
machines:
<machine id>:
agent-state: <agent state>
dns-name: <dns name>
instance-id: <provider specific instance id>
instance-state: <instance state>
"""
machines = yield self.machine_manager.get_all_machine_states()
for machine_state in machines:
if (self.filter_services or self.filter_units) and \
machine_state.id not in self.seen_machines:
continue
yield self._process_machine(machine_state)
@inlineCallbacks
def _process_machine(self, machine_state):
"""
`machine_state`: MachineState instance
"""
instance_id = yield machine_state.get_instance_id()
m = {"instance-id": instance_id \
if instance_id is not None else "pending"}
if instance_id is not None:
try:
pm = yield self.provider.get_machine(instance_id)
m["dns-name"] = pm.dns_name
m["instance-state"] = pm.state
if (yield machine_state.has_agent()):
# if the agent's connected, we're fine
m["agent-state"] = "running"
else:
units = (
yield machine_state.get_all_service_unit_states())
for unit in units:
unit_workflow_client = WorkflowStateClient(
self.client, unit)
if (yield unit_workflow_client.get_state()):
# for unit to have a state, its agent must
# have run, which implies the machine agent
# must have been running correctly at some
# point in the past
m["agent-state"] = "down"
break
else:
# otherwise we're probably just still waiting
m["agent-state"] = "not-started"
except ProviderError:
# The provider doesn't have machine information
self.log.error(
"Machine provider information missing: machine %s" % (
machine_state.id))
self.machine_data[machine_state.id] = m
class CharmPublisher(object):
"""
Publishes a charm to an environment.
"""
def __init__(self, client, storage):
self._client = client
self._storage = storage
self._charm_state_manager = CharmStateManager(self._client)
self._charm_add_queue = []
self._charm_state_cache = {}
@classmethod
@inlineCallbacks
def for_environment(cls, environment):
provider = environment.get_machine_provider()
storage = provider.get_file_storage()
client = yield provider.connect()
returnValue(cls(client, storage))
@inlineCallbacks
def add_charm(self, charm_id, charm):
"""Schedule a charm for addition to an juju environment.
Returns true if the charm is scheduled for upload, false if
the charm is already present in juju.
"""
self._charm_add_queue.append((charm_id, charm))
if charm_id in self._charm_state_cache:
returnValue(False)
try:
state = yield self._charm_state_manager.get_charm_state(
charm_id)
except CharmStateNotFound:
pass
else:
log.info("Using cached charm version of %s" % charm.metadata.name)
self._charm_state_cache[charm_id] = state
returnValue(False)
returnValue(True)
def _publish_one(self, charm_id, charm):
if charm_id in self._charm_state_cache:
return succeed(self._charm_state_cache[charm_id])
bundle = charm.as_bundle()
charm_file = open(bundle.path, "rb")
charm_store_path = under.quote(
"%s:%s" % (charm_id, bundle.get_sha256()))
def close_charm_file(passthrough):
charm_file.close()
return passthrough
def get_charm_url(result):
return self._storage.get_url(charm_store_path)
d = self._storage.put(charm_store_path, charm_file)
d.addBoth(close_charm_file)
d.addCallback(get_charm_url)
d.addCallback(self._cb_store_charm_state, charm_id, bundle)
d.addErrback(self._eb_verify_duplicate, charm_id, bundle)
return d
def publish(self):
"""Publish all added charms to provider storage and zookeeper.
Returns the charm_state of all scheduled charms.
"""
publish_deferreds = []
for charm_id, charm in self._charm_add_queue:
publish_deferreds.append(self._publish_one(charm_id, charm))
publish_deferred = DeferredList(publish_deferreds,
fireOnOneErrback=1,
consumeErrors=1)
# callbacks and deferreds to unwind the dlist
publish_deferred.addCallback(self._cb_extract_charm_state)
publish_deferred.addErrback(self._eb_extract_error)
return publish_deferred
def _cb_extract_charm_state(self, result):
return [r[1] for r in result]
def _eb_extract_error(self, failure):
failure.trap(FirstError)
return failure.value.subFailure
def _cb_store_charm_state(self, charm_url, charm_id, charm):
return self._charm_state_manager.add_charm_state(
charm_id, charm, charm_url)
@inlineCallbacks
def _eb_verify_duplicate(self, failure, charm_id, charm):
"""Detects duplicates vs. conflicts, raises stateerror on conflict."""
failure.trap(NodeExistsException)
try:
charm_state = \
yield self._charm_state_manager.get_charm_state(charm_id)
except NoNodeException:
# Check if the state goes away due to concurrent removal
msg = "Charm removed concurrently during publish, please retry."
raise StateChanged(msg)
if charm_state.get_sha256() != charm.get_sha256():
msg = "Concurrent upload of charm has different checksum %s" % (
charm_id)
raise StateChanged(msg)
class CharmStateManagerTest(StateTestBase):
@inlineCallbacks
def setUp(self):
yield super(CharmStateManagerTest, self).setUp()
self.charm_state_manager = CharmStateManager(self.client)
self.charm_id = local_charm_id(self.charm)
@inlineCallbacks
def test_add_charm(self):
"""
Adding a Charm into a CharmStateManager should register
the charm within the Zookeeper state, according to the
specification.
"""
charm_state = yield self.charm_state_manager.add_charm_state(
self.charm_id, self.charm, "http://example.com/abc")
self.assertEquals(charm_state.id, "local:series/dummy-1")
children = yield self.client.get_children("/charms")
self.assertEquals(children, ["local_3a_series_2f_dummy-1"])
content, stat = yield self.client.get(
"/charms/local_3a_series_2f_dummy-1")
charm_data = yaml.load(content)
self.assertEquals(charm_data, {
"metadata": self.charm.metadata.get_serialization_data(),
"config": self.charm.config.get_serialization_data(),
"sha256": self.charm.get_sha256(),
"url": "http://example.com/abc"
})
@inlineCallbacks
def test_get_charm(self):
"""
A CharmState should be available if one get()s a charm
that was previously added into the manager.
"""
yield self.charm_state_manager.add_charm_state(
self.charm_id, self.charm, "")
charm_state = yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
self.assertEquals(charm_state.id, "local:series/dummy-1")
@inlineCallbacks
def test_charm_state_attributes(self):
"""
Verify that the basic (invariant) attributes of the
CharmState are correctly in place.
"""
yield self.charm_state_manager.add_charm_state(
self.charm_id, self.charm, "http://example.com/abc")
charm_state = yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
self.assertEquals(charm_state.name, "dummy")
self.assertEquals(charm_state.revision, 1)
self.assertEquals(charm_state.id, "local:series/dummy-1")
self.assertEquals(charm_state.bundle_url, "http://example.com/abc")
@inlineCallbacks
def test_charm_state_metadata(self):
"""
Check that the charm metadata was correctly saved and loaded.
"""
yield self.charm_state_manager.add_charm_state(
self.charm_id, self.charm, "")
charm_state = yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
metadata = yield charm_state.get_metadata()
self.assertEquals(metadata.name, "dummy")
@inlineCallbacks
def test_charm_state_config_options(self):
"""Verify ConfigOptions present and correct."""
from juju.charm.tests.test_config import sample_yaml_data
yield self.charm_state_manager.add_charm_state(
self.charm_id, self.charm, "")
charm_state = yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
config = yield charm_state.get_config()
self.assertEquals(config.get_serialization_data(),
sample_yaml_data)
@inlineCallbacks
def test_get_non_existing_charm_prior_to_initialization(self):
"""
Getting a charm before the charms node was even
initialized should raise an error about the charm not
being present.
"""
try:
yield self.charm_state_manager.get_charm_state(
"local:series/dummy-1")
except CharmStateNotFound, e:
self.assertEquals(e.charm_id, "local:series/dummy-1")
else:
class CharmPublisher(object):
"""
Publishes a charm to an environment.
"""
def __init__(self, client, storage):
self._client = client
self._storage = storage
self._charm_state_manager = CharmStateManager(self._client)
self._charm_add_queue = []
self._charm_state_cache = {}
@classmethod
@inlineCallbacks
def for_environment(cls, environment):
provider = environment.get_machine_provider()
storage = provider.get_file_storage()
client = yield provider.connect()
returnValue(cls(client, storage))
@inlineCallbacks
def add_charm(self, charm_id, charm):
"""Schedule a charm for addition to an juju environment.
Returns true if the charm is scheduled for upload, false if
the charm is already present in juju.
"""
self._charm_add_queue.append((charm_id, charm))
if charm_id in self._charm_state_cache:
returnValue(False)
try:
state = yield self._charm_state_manager.get_charm_state(charm_id)
except CharmStateNotFound:
pass
else:
self._charm_state_cache[charm_id] = state
returnValue(False)
returnValue(True)
def _publish_one(self, charm_id, charm):
if charm_id in self._charm_state_cache:
return succeed(self._charm_state_cache[charm_id])
bundle = charm.as_bundle()
charm_file = open(bundle.path, "rb")
charm_store_path = under.quote("%s:%s" %
(charm_id, bundle.get_sha256()))
def close_charm_file(passthrough):
charm_file.close()
return passthrough
def get_charm_url(result):
return self._storage.get_url(charm_store_path)
d = self._storage.put(charm_store_path, charm_file)
d.addBoth(close_charm_file)
d.addCallback(get_charm_url)
d.addCallback(self._cb_store_charm_state, charm_id, bundle)
d.addErrback(self._eb_verify_duplicate, charm_id, bundle)
return d
def publish(self):
"""Publish all added charms to provider storage and zookeeper.
Returns the charm_state of all scheduled charms.
"""
publish_deferreds = []
for charm_id, charm in self._charm_add_queue:
publish_deferreds.append(self._publish_one(charm_id, charm))
publish_deferred = DeferredList(publish_deferreds,
fireOnOneErrback=1,
consumeErrors=1)
# callbacks and deferreds to unwind the dlist
publish_deferred.addCallback(self._cb_extract_charm_state)
publish_deferred.addErrback(self._eb_extract_error)
return publish_deferred
def _cb_extract_charm_state(self, result):
return [r[1] for r in result]
def _eb_extract_error(self, failure):
failure.trap(FirstError)
return failure.value.subFailure
def _cb_store_charm_state(self, charm_url, charm_id, charm):
return self._charm_state_manager.add_charm_state(
charm_id, charm, charm_url)
@inlineCallbacks
def _eb_verify_duplicate(self, failure, charm_id, charm):
"""Detects duplicates vs. conflicts, raises stateerror on conflict."""
failure.trap(NodeExistsException)
try:
charm_state = \
yield self._charm_state_manager.get_charm_state(charm_id)
except NoNodeException:
# Check if the state goes away due to concurrent removal
msg = "Charm removed concurrently during publish, please retry."
raise StateChanged(msg)
if charm_state.get_sha256() != charm.get_sha256():
msg = "Concurrent upload of charm has different checksum %s" % (
charm_id)
raise StateChanged(msg)
def collect(scope, machine_provider, client, log):
"""Extract status information into nested dicts for rendering.
`scope`: an optional list of name specifiers. Globbing based
wildcards supported. Defaults to all units, services and
relations.
`machine_provider`: machine provider for the environment
`client`: ZK client connection
`log`: a Python stdlib logger.
"""
service_manager = ServiceStateManager(client)
relation_manager = RelationStateManager(client)
machine_manager = MachineStateManager(client)
charm_manager = CharmStateManager(client)
service_data = {}
machine_data = {}
state = dict(services=service_data, machines=machine_data)
seen_machines = set()
filter_services, filter_units = digest_scope(scope)
services = yield service_manager.get_all_service_states()
for service in services:
if len(filter_services):
found = False
for filter_service in filter_services:
if fnmatch(service.service_name, filter_service):
found = True
break
if not found:
continue
unit_data = {}
relation_data = {}
charm_id = yield service.get_charm_id()
charm = yield charm_manager.get_charm_state(charm_id)
service_data[service.service_name] = dict(units=unit_data,
charm=charm.id,
relations=relation_data)
exposed = yield service.get_exposed_flag()
if exposed:
service_data[service.service_name].update(exposed=exposed)
units = yield service.get_all_unit_states()
unit_matched = False
relations = yield relation_manager.get_relations_for_service(service)
for unit in units:
if len(filter_units):
found = False
for filter_unit in filter_units:
if fnmatch(unit.unit_name, filter_unit):
found = True
break
if not found:
continue
u = unit_data[unit.unit_name] = dict()
machine_id = yield unit.get_assigned_machine_id()
u["machine"] = machine_id
unit_workflow_client = WorkflowStateClient(client, unit)
unit_state = yield unit_workflow_client.get_state()
if not unit_state:
u["state"] = "pending"
else:
unit_connected = yield unit.has_agent()
u["state"] = unit_state if unit_connected else "down"
if exposed:
open_ports = yield unit.get_open_ports()
u["open-ports"] = [
"{port}/{proto}".format(**port_info)
for port_info in open_ports
]
u["public-address"] = yield unit.get_public_address()
# indicate we should include information about this
# machine later
seen_machines.add(machine_id)
unit_matched = True
# collect info on each relation for the service unit
relation_status = {}
for relation in relations:
try:
relation_unit = yield relation.get_unit_state(unit)
except UnitRelationStateNotFound:
# This exception will occur when relations are
# established between services without service
# units, and therefore never have any
# corresponding service relation units. This
# scenario does not occur in actual deployments,
# but can happen in test circumstances. In
# particular, it will happen with a misconfigured
# provider, which exercises this codepath.
continue # should not occur, but status should not fail
relation_workflow_client = WorkflowStateClient(
client, relation_unit)
relation_workflow_state = \
yield relation_workflow_client.get_state()
relation_status[relation.relation_name] = dict(
state=relation_workflow_state)
u["relations"] = relation_status
# after filtering units check if any matched or remove the
# service from the output
if filter_units and not unit_matched:
del service_data[service.service_name]
continue
for relation in relations:
rel_services = yield relation.get_service_states()
# A single related service implies a peer relation. More
# imply a bi-directional provides/requires relationship.
# In the later case we omit the local side of the relation
# when reporting.
if len(rel_services) > 1:
# Filter out self from multi-service relations.
rel_services = [
rsn for rsn in rel_services
if rsn.service_name != service.service_name
]
if len(rel_services) > 1:
raise ValueError("Unexpected relationship with more "
"than 2 endpoints")
rel_service = rel_services[0]
relation_data[relation.relation_name] = rel_service.service_name
machines = yield machine_manager.get_all_machine_states()
for machine_state in machines:
if (filter_services or filter_units) and \
machine_state.id not in seen_machines:
continue
instance_id = yield machine_state.get_instance_id()
m = {"instance-id": instance_id \
if instance_id is not None else "pending"}
if instance_id is not None:
try:
pm = yield machine_provider.get_machine(instance_id)
m["dns-name"] = pm.dns_name
m["instance-state"] = pm.state
if (yield machine_state.has_agent()):
# if the agent's connected, we're fine
m["state"] = "running"
else:
units = (yield machine_state.get_all_service_unit_states())
for unit in units:
unit_workflow_client = WorkflowStateClient(
client, unit)
if (yield unit_workflow_client.get_state()):
# for unit to have a state, its agent must have
# run, which implies the machine agent must have
# been running correctly at some point in the past
m["state"] = "down"
break
else:
# otherwise we're probably just still waiting
m["state"] = "not-started"
except ProviderError:
# The provider doesn't have machine information
log.error("Machine provider information missing: machine %s" %
(machine_state.id))
machine_data[machine_state.id] = m
returnValue(state)
class MachineAgent(BaseAgent):
"""An juju Machine Agent.
The machine agent is responsible for monitoring service units
assigned to a machine. If a new unit is assigned to machine, the
machine agent will download the charm, create a working
space for the service unit agent, and then launch it.
Additionally the machine agent will monitor the running service
unit agents on the machine, via their ephemeral nodes, and
restart them if they die.
"""
name = "juju-machine-agent"
unit_agent_module = "juju.agents.unit"
@property
def charms_directory(self):
return os.path.join(self.config["juju_directory"], "charms")
@property
def units_directory(self):
return os.path.join(self.config["juju_directory"], "units")
@property
def unit_state_directory(self):
return os.path.join(self.config["juju_directory"], "state")
@inlineCallbacks
def start(self):
"""Start the machine agent.
Creates state directories on the machine, retrieves the machine state,
and enables watch on assigned units.
"""
# Initialize directory paths.
if not os.path.exists(self.charms_directory):
os.makedirs(self.charms_directory)
if not os.path.exists(self.units_directory):
os.makedirs(self.units_directory)
if not os.path.exists(self.unit_state_directory):
os.makedirs(self.unit_state_directory)
# Get state managers we'll be utilizing.
self.service_state_manager = ServiceStateManager(self.client)
self.charm_state_manager = CharmStateManager(self.client)
# Retrieve the machine state for the machine we represent.
machine_manager = MachineStateManager(self.client)
self.machine_state = yield machine_manager.get_machine_state(
self.get_machine_id())
# Watch assigned units for the machine.
if self.get_watch_enabled():
self.machine_state.watch_assigned_units(self.watch_service_units)
# Find out what provided the machine, and how to deploy units.
settings = GlobalSettingsStateManager(self.client)
self.provider_type = yield settings.get_provider_type()
self.deploy_factory = get_deploy_factory(self.provider_type)
# Connect the machine agent, broadcasting presence to the world.
yield self.machine_state.connect_agent()
log.info(
"Machine agent started id:%s deploy:%r provider:%r" %
(self.get_machine_id(), self.deploy_factory, self.provider_type))
def download_charm(self, charm_state):
"""Retrieve a charm from the provider storage to the local machine.
Utilizes a local charm cache to avoid repeated downloading of the
same charm.
"""
log.debug("Downloading charm %s to %s", charm_state.id,
self.charms_directory)
return download_charm(self.client, charm_state.id,
self.charms_directory)
@inlineCallbacks
def watch_service_units(self, old_units, new_units):
"""Callback invoked when the assigned service units change.
"""
if old_units is None:
old_units = set()
log.debug("Units changed old:%s new:%s", old_units, new_units)
stopped = old_units - new_units
started = new_units - old_units
for unit_name in stopped:
log.debug("Stopping service unit: %s ...", unit_name)
try:
yield self.kill_service_unit(unit_name)
except Exception:
log.exception("Error stopping unit: %s", unit_name)
for unit_name in started:
log.debug("Starting service unit: %s ...", unit_name)
try:
yield self.start_service_unit(unit_name)
except Exception:
log.exception("Error starting unit: %s", unit_name)
@inlineCallbacks
def start_service_unit(self, service_unit_name):
"""Start a service unit on the machine.
Downloads the charm, and extract it to the service unit directory,
and launch the service unit agent within the unit directory.
"""
# Retrieve the charm state to get at the charm.
unit_state = yield self.service_state_manager.get_unit_state(
service_unit_name)
charm_id = yield unit_state.get_charm_id()
charm_state = yield self.charm_state_manager.get_charm_state(charm_id)
# Download the charm.
bundle = yield self.download_charm(charm_state)
# Use deployment to setup the workspace and start the unit agent.
deployment = self.deploy_factory(service_unit_name,
self.config["juju_directory"])
running = yield deployment.is_running()
if not running:
log.debug("Starting service unit %s", service_unit_name)
yield deployment.start(self.get_machine_id(), self.client.servers,
bundle)
log.info("Started service unit %s", service_unit_name)
def kill_service_unit(self, service_unit_name):
"""Stop service unit and destroy disk state, ala SIGKILL or lxc-destroy
"""
deployment = self.deploy_factory(service_unit_name,
self.config["juju_directory"])
log.info("Stopping service unit %s...", service_unit_name)
return deployment.destroy()
def get_machine_id(self):
"""Get the id of the machine as known within the zk state."""
return self.config["machine_id"]
def get_agent_name(self):
return "Machine:%s" % (self.get_machine_id())
def configure(self, options):
super(MachineAgent, self).configure(options)
if not options.get("machine_id"):
msg = ("--machine-id must be provided in the command line, "
"or $JUJU_MACHINE_ID in the environment")
raise JujuError(msg)
@classmethod
def setup_options(cls, parser):
super(MachineAgent, cls).setup_options(parser)
machine_id = os.environ.get("JUJU_MACHINE_ID", "")
parser.add_argument("--machine-id", default=machine_id)
return parser
class UnitDeployer(object):
"""Manages service unit deployment for an agent.
"""
def __init__(self, client, machine_id, juju_directory):
"""Initialize a Unit Deployer.
:param client: A connected zookeeper client.
:param str machine_id: the ID of the machine the agent is being run on.
:param str juju_directory: the directory the agent is running in.
"""
self.client = client
self.machine_id = machine_id
self.juju_directory = juju_directory
self.service_state_manager = ServiceStateManager(self.client)
self.charm_state_manager = CharmStateManager(self.client)
@property
def charms_directory(self):
return os.path.join(self.juju_directory, "charms")
@inlineCallbacks
def start(self, provider_type=None):
"""Starts the unit deployer."""
# Find out what provided the machine, and how to deploy units.
if provider_type is None:
settings = GlobalSettingsStateManager(self.client)
provider_type = yield settings.get_provider_type()
self.deploy_factory = get_deploy_factory(provider_type)
if not os.path.exists(self.charms_directory):
os.makedirs(self.charms_directory)
def download_charm(self, charm_state):
"""Retrieve a charm from the provider storage to the local machine.
:param charm_state: Charm to be downloaded
"""
log.debug("Downloading charm %s to %s",
charm_state.id, self.charms_directory)
return download_charm(
self.client, charm_state.id, self.charms_directory)
@inlineCallbacks
def start_service_unit(self, service_unit_name):
"""Start a service unit on the machine.
Downloads the charm, and extract it to the service unit directory,
and launch the service unit agent within the unit directory.
:param str service_unit_name: Service unit name to be started
"""
# Retrieve the charm state to get at the charm.
unit_state = yield self.service_state_manager.get_unit_state(
service_unit_name)
charm_id = yield unit_state.get_charm_id()
charm_state = yield self.charm_state_manager.get_charm_state(
charm_id)
# Download the charm.
bundle = yield self.download_charm(charm_state)
# Use deployment to setup the workspace and start the unit agent.
deployment = self.deploy_factory(
service_unit_name, self.juju_directory)
log.debug("Using %r for %s in %s",
deployment,
service_unit_name,
self.juju_directory)
running = yield deployment.is_running()
if not running:
log.debug("Starting service unit %s...", service_unit_name)
yield deployment.start(
self.machine_id, self.client.servers, bundle)
log.info("Started service unit %s", service_unit_name)
@inlineCallbacks
def kill_service_unit(self, service_unit_name):
"""Stop service unit and destroy disk state, ala SIGKILL or lxc-destroy
:param str service_unit_name: Service unit name to be killed
"""
deployment = self.deploy_factory(
service_unit_name, self.juju_directory)
log.info("Stopping service unit %s...", service_unit_name)
yield deployment.destroy()
log.info("Stopped service unit %s", service_unit_name)