def test_create_listening_endpoint(self): self.patch_cfg( 'pyon.container.procs.CFG', { 'container': { 'messaging': { 'endpoint': { 'proc_listening_type': 'pyon.container.test.test_procs.TestRPCServer' } } } }) fakecc = FakeContainer() fakecc.resource_registry = Mock() fakecc.resource_registry.create.return_value = ["ID", "rev"] with patch( 'pyon.container.procs.ProcManager._get_execution_engine_config', Mock(return_value={})): pm = ProcManager(fakecc) pm._get_execution_engine_config = Mock() ep = pm._create_listening_endpoint(node=sentinel.node, service=sentinel.service, process=sentinel.process) self.assertIsInstance(ep, TestRPCServer)
def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) self._is_started = False self._capabilities = [] # set container id and cc_agent name (as they are set in base class call) self.id = get_default_container_id() self.name = "cc_agent_%s" % self.id Container.instance = self from pyon.core import bootstrap bootstrap.container_instance = self log.debug("Container (sysname=%s) initializing ..." % bootstrap.get_sys_name()) # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = kwargs # Instantiate Directory and self-register # Has the additional side effect of either # bootstrapping the configuration into the # directory or read the configuration based # in the value of the auto_bootstrap setting self.directory = Directory() # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # File System - Interface to the OS File System, using correct path names and setups self.file_system = FileSystem(CFG) # Governance Controller - manages the governance related interceptors self.governance_controller = GovernanceController(self) # sFlow manager - controls sFlow stat emission self.sflow_manager = SFlowManager(self) # Coordinates the container start self._status = "INIT" # protection for when the container itself is used as a Process for clients self.container = self log.debug("Container initialized, OK.")
def setUp(self): self.container = Mock() self.pm = ProcManager(self.container) self.container.resource_registry.create.return_value = (sentinel.rid, sentinel.rev) self.container.resource_registry.find_resources.return_value = ([ sentinel.oid ], [sentinel.orev])
def test_procmanager_iso(self): fakecc = FakeContainer() fakecc.resource_registry = Mock() fakecc.resource_registry.create.return_value=["ID","rev"] pm = ProcManager(fakecc) self.assertTrue(hasattr(fakecc, "spawn_process")) pm.start() pm.stop()
def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) self._is_started = False # set id and name (as they are set in base class call) self.id = string.replace('%s_%d' % (os.uname()[1], os.getpid()), ".", "_") self.name = "cc_agent_%s" % self.id Container.instance = self # TODO: Bug: Replacing CFG instance not work because references are already public. Update directly dict_merge(CFG, kwargs, inplace=True) from pyon.core import bootstrap bootstrap.container_instance = self bootstrap.assert_configuration(CFG) log.debug("Container (sysname=%s) initializing ..." % bootstrap.get_sys_name()) # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = kwargs # Load object and service registry etc. bootstrap_pyon() # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() # File System - Interface to the OS File System, using correct path names and setups self.file_system = FileSystem(CFG) # Governance Controller - manages the governance related interceptors self.governance_controller = GovernanceController(self) # sFlow manager - controls sFlow stat emission self.sflow_manager = SFlowManager(self) # Coordinates the container start self._is_started = False self._capabilities = [] self._status = "INIT" # protection for when the container itself is used as a Process for clients self.container = self log.debug("Container initialized, OK.")
def test_create_listening_endpoint(self): self.patch_cfg('pyon.container.procs.CFG', {'container':{'messaging':{'endpoint':{'proc_listening_type':'pyon.container.test.test_procs.TestRPCServer'}}}}) fakecc = FakeContainer() fakecc.resource_registry = Mock() fakecc.resource_registry.create.return_value=["ID","rev"] pm = ProcManager(fakecc) ep = pm._create_listening_endpoint(node=sentinel.node, service=sentinel.service, process=sentinel.process) self.assertIsInstance(ep, TestRPCServer)
def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) # set id and name (as they are set in base class call) self.id = string.replace('%s_%d' % (os.uname()[1], os.getpid()), ".", "_") self.name = "cc_agent_%s" % self.id Container.instance = self # TODO: Bug: Replacing CFG instance not work because references are already public. Update directly dict_merge(CFG, kwargs) from pyon.core import bootstrap bootstrap.sys_name = CFG.system.name or bootstrap.sys_name log.debug("Container (sysname=%s) initializing ..." % bootstrap.sys_name) # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = DictModifier(CFG, kwargs) # Load object and service registry etc. bootstrap_pyon() # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() log.debug("Container initialized, OK.")
def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) self._is_started = False # set container id and cc_agent name (as they are set in base class call) self.id = get_default_container_id() self.name = "cc_agent_%s" % self.id self._capabilities = [] from pyon.core import bootstrap bootstrap.container_instance = self Container.instance = self log.debug("Container (sysname=%s) initializing ..." % bootstrap.get_sys_name()) # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = kwargs # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() # TODO: Do not start a capability here. Symmetric start/stop self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") # Instantiate Directory self.directory = Directory() # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # File System - Interface to the OS File System, using correct path names and setups self.file_system = FileSystem(CFG) # Governance Controller - manages the governance related interceptors self.governance_controller = GovernanceController(self) # sFlow manager - controls sFlow stat emission self.sflow_manager = SFlowManager(self) # Coordinates the container start self._status = "INIT" # protection for when the container itself is used as a Process for clients self.container = self log.debug("Container initialized, OK.")
def test_procmanager_iso(self): fakecc = FakeContainer() fakecc.resource_registry = Mock() fakecc.resource_registry.create.return_value = ["ID", "rev"] pm = ProcManager(fakecc) self.assertTrue(hasattr(fakecc, "spawn_process")) pm.start() pm.stop()
def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) self._is_started = False # set container id and cc_agent name (as they are set in base class call) self.id = get_default_container_id() self.name = "cc_agent_%s" % self.id self._capabilities = [] bootstrap.container_instance = self Container.instance = self log.debug("Container (sysname=%s) initializing ..." % bootstrap.get_sys_name()) # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = kwargs # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() # TODO: Do not start a capability here. Symmetric start/stop self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") # Instantiate Directory self.directory = Directory() # internal router self.local_router = None # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # File System - Interface to the OS File System, using correct path names and setups self.file_system = FileSystem(CFG) # Governance Controller - manages the governance related interceptors self.governance_controller = GovernanceController(self) # sFlow manager - controls sFlow stat emission self.sflow_manager = SFlowManager(self) # Coordinates the container start self._status = "INIT" # protection for when the container itself is used as a Process for clients self.container = self # publisher, initialized in start() self.event_pub = None # context-local storage self.context = LocalContextMixin() log.debug("Container initialized, OK.")
class TestProcManager(PyonTestCase): def setUp(self): self.container = Mock() self.pm = ProcManager(self.container) self.container.resource_registry.create.return_value = (sentinel.rid, sentinel.rev) self.container.resource_registry.find_resources.return_value = ([ sentinel.oid ], [sentinel.orev]) def test_start(self): self.pm.start() self.assertEquals(self.pm.cc_id, sentinel.rid) def test_start_with_org(self): self.patch_cfg('pyon.container.procs.CFG', {'container': { 'org_name': 'NOT_DEFAULT' }}) self.pm.start() self.container.resource_registry.create_association.assert_called_once_with( sentinel.oid, PRED.hasResource, sentinel.rid) @patch('pyon.datastore.couchdb.couchdb_datastore.CouchDB_DataStore._stats') def test_stop(self, statsmock): self.pm.start() self.pm.stop() self.assertEquals(statsmock.get_stats.call_count, 2) def test__cleanup_method(self): ep = Mock() ep._chan._queue_auto_delete = False self.pm._cleanup_method(sentinel.queue, ep) ch = self.container.node.channel.return_value ch._destroy_queue.assert_called_once_with() self.assertIsInstance(ch._recv_name, NameTrio) self.assertIn(str(sentinel.queue), str(ch._recv_name)) @patch('pyon.container.procs.log') def test__cleanup_method_raises_error(self, mocklog): ep = Mock() ep._chan._queue_auto_delete = False ch = self.container.node.channel.return_value ch._destroy_queue.side_effect = TransportError self.pm._cleanup_method(sentinel.queue, ep) self.assertEquals(mocklog.warn.call_count, 1) @patch('pyon.datastore.couchdb.couchdb_datastore.CouchDB_DataStore._stats', Mock()) @patch('pyon.container.procs.log') def test_stop_with_error(self, mocklog): self.pm.start() self.pm.terminate_process = Mock(side_effect=BadRequest) procmock = Mock() procmock._proc_start_time = 0 procmock.id = sentinel.pid self.pm.procs[sentinel.pid] = procmock self.pm.procs_by_name['dummy'] = procmock self.pm.stop() self.pm.terminate_process.assert_called_once_with(sentinel.pid) mocklog.warn.assert_has_calls([ call("Failed to terminate process (%s): %s", sentinel.pid, ANY), call("ProcManager procs not empty: %s", self.pm.procs), call("ProcManager procs_by_name not empty: %s", self.pm.procs_by_name) ]) def test_list_local_processes(self): pmock = Mock() pmock.process_type = sentinel.ptype pmock2 = Mock() pmock2.process_type = sentinel.ptype2 self.pm.procs = {'one': pmock, 'two': pmock2, 'three': pmock} self.assertEquals(self.pm.list_local_processes(), [pmock, pmock2, pmock]) def test_list_local_processes_proc_type_filter(self): pmock = Mock() pmock.process_type = sentinel.ptype pmock2 = Mock() pmock2.process_type = sentinel.ptype2 self.pm.procs = {'one': pmock, 'two': pmock2, 'three': pmock} self.assertEquals(self.pm.list_local_processes(sentinel.ptype2), [pmock2]) def test_get_a_local_process(self): pmock = Mock() pmock.name = sentinel.name pmock2 = Mock() pmock2.name = sentinel.name2 self.pm.procs = {'one': pmock, 'two': pmock2} self.assertEquals(self.pm.get_a_local_process(sentinel.name2), pmock2) def test_get_a_local_process_for_agent_res_id(self): pmock = Mock() pmock.process_type = 'agent' pmock.resource_type = sentinel.rtype pmock2 = Mock() pmock2.process_type = 'agent' pmock2.resource_type = sentinel.rtype2 self.pm.procs = {'one': pmock, 'two': pmock2} self.assertEquals(self.pm.get_a_local_process(sentinel.rtype2), pmock2) def test_get_a_local_process_no_match(self): self.assertIsNone(self.pm.get_a_local_process()) def test_is_local_service_process(self): pmock = Mock() pmock.name = sentinel.name pmock.process_type = 'simple' pmock2 = Mock() pmock2.name = sentinel.name2 pmock2.process_type = 'service' pmock3 = Mock() pmock3.name = sentinel.name3 pmock3.process_type = 'service' self.pm.procs = {'one': pmock, 'two': pmock2, 'three': pmock3} self.assertTrue(self.pm.is_local_service_process(sentinel.name3)) def test_is_local_service_process_name_matches_but_type_doesnt(self): pmock = Mock() pmock.name = sentinel.name pmock.process_type = 'simple' pmock2 = Mock() pmock2.name = sentinel.name2 pmock2.process_type = 'notservice' pmock3 = Mock() pmock3.name = sentinel.name3 pmock3.process_type = 'notservice' self.pm.procs = {'one': pmock, 'two': pmock2, 'three': pmock3} self.assertFalse(self.pm.is_local_service_process(sentinel.name3)) def test_is_local_agent_process(self): # agent is similar to above, but checks resource_type instead pmock = Mock() pmock.name = sentinel.name pmock.process_type = 'simple' pmock2 = Mock() pmock2.resource_type = sentinel.name2 pmock2.process_type = 'agent' pmock3 = Mock() pmock3.name = sentinel.name3 pmock3.process_type = 'notservice' self.pm.procs = {'one': pmock, 'two': pmock2, 'three': pmock3} self.assertTrue(self.pm.is_local_agent_process(sentinel.name2)) def test_is_local_agent_process_not_found(self): self.assertFalse(self.pm.is_local_agent_process(sentinel.one)) def test__unregister_process_errors(self): pmock = Mock() pmock._proc_name = '1' pmock._proc_type = 'service' pmock._proc_res_id = sentinel.presid pmock._proc_svc_id = sentinel.psvcid self.container.resource_registry.delete.side_effect = NotFound self.container.resource_registry.find_objects.side_effect = ResourceNotFound self.pm.procs[sentinel.pid] = pmock self.pm.procs_by_name['1'] = pmock self.pm._unregister_process(sentinel.pid, pmock) # show we tried to interact with the RR self.container.resource_registry.delete.assert_call( sentinel.presid, del_associations=True) self.container.resource_registry.find_objects.assert_called_once_with( sentinel.psvcid, "hasProcess", "Process", id_only=True) self.assertEquals(self.pm.procs, {}) self.assertEquals(self.pm.procs_by_name, {}) # NEXT: find_objects works fine and gives us an error deletion self.container.resource_registry.delete.reset_mock() self.container.resource_registry.find_objects.reset_mock() self.container.resource_registry.find_objects.side_effect = None self.container.resource_registry.find_objects.return_value = ([ sentinel.svcid ], [None]) self.pm.procs[sentinel.pid] = pmock self.pm.procs_by_name['1'] = pmock self.pm._unregister_process(sentinel.pid, pmock) self.container.resource_registry.delete.assert_calls([ call(sentinel.presid, del_associations=True), call(sentinel.psvcid, del_associations=True) ]) # NEXT: agent pmock = Mock() pmock.id = sentinel.pid pmock._proc_name = '1' pmock._proc_type = 'agent' self.pm.procs[sentinel.pid] = pmock self.pm.procs_by_name['1'] = pmock self.pm._unregister_process(sentinel.pid, pmock) self.container.directory.unregister_safe.assert_called_once_with( "/Agents", sentinel.pid) def test__create_listening_endpoint_with_cfg(self): self.patch_cfg('pyon.container.procs.CFG', container=dict(messaging=dict(endpoint=dict( proc_listening_type= 'pyon.container.test.test_procs.TestRPCServer')))) ep = self.pm._create_listening_endpoint(process=sentinel.process) self.assertIsInstance(ep, TestRPCServer) def test__create_listening_endpoint_without_cfg_and_no_conv(self): self.patch_cfg( 'pyon.container.procs.CFG', container=dict(messaging=dict(endpoint=dict( proc_listening_type=None, rpc_conversation_enabled=False)))) ep = self.pm._create_listening_endpoint(process=sentinel.process) self.assertIsInstance(ep, ProcessRPCServer) def test__create_listening_endpoint_without_cfg_and_conv(self): self.patch_cfg( 'pyon.container.procs.CFG', container=dict(messaging=dict(endpoint=dict( proc_listening_type=None, rpc_conversation_enabled=True)))) ep = self.pm._create_listening_endpoint(process=sentinel.process) self.assertIsInstance(ep, ConversationRPCServer)
class TestProcManager(PyonTestCase): def setUp(self): self.container = Mock() self.pm = ProcManager(self.container) self.container.resource_registry.create.return_value = (sentinel.rid, sentinel.rev) self.container.resource_registry.find_resources.return_value = ([sentinel.oid], [sentinel.orev]) def test_start(self): self.pm.start() self.assertEquals(self.pm.cc_id, sentinel.rid) def test_start_with_org(self): self.patch_cfg('pyon.container.procs.CFG', {'container':{'org_name':'NOT_DEFAULT'}}) self.pm.start() self.container.resource_registry.create_association.assert_called_once_with(sentinel.oid, PRED.hasResource, sentinel.rid) @patch('pyon.container.procs.log') def test_stop_with_error(self, mocklog): self.pm.start() self.pm.terminate_process = Mock(side_effect=BadRequest) procmock = Mock() procmock._proc_start_time = 0 procmock.id = sentinel.pid self.pm.procs[sentinel.pid] = procmock self.pm.procs_by_name['dummy'] = procmock self.pm.stop() self.pm.terminate_process.assert_called_once_with(sentinel.pid) mocklog.warn.assert_has_calls([call("Failed to terminate process (%s): %s", sentinel.pid, ANY), call("ProcManager procs not empty: %s", self.pm.procs), call("ProcManager procs_by_name not empty: %s", self.pm.procs_by_name)]) def test_list_local_processes(self): pmock = Mock() pmock.process_type = sentinel.ptype pmock2 = Mock() pmock2.process_type = sentinel.ptype2 self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock} self.assertEquals(self.pm.list_local_processes(), [pmock, pmock2, pmock]) def test_list_local_processes_proc_type_filter(self): pmock = Mock() pmock.process_type = sentinel.ptype pmock2 = Mock() pmock2.process_type = sentinel.ptype2 self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock} self.assertEquals(self.pm.list_local_processes(sentinel.ptype2), [pmock2]) def test_get_a_local_process(self): pmock = Mock() pmock.name = sentinel.name pmock2 = Mock() pmock2.name = sentinel.name2 self.pm.procs = {'one':pmock, 'two':pmock2} self.assertEquals(self.pm.get_a_local_process(sentinel.name2), pmock2) def test_get_a_local_process_for_agent_res_id(self): pmock = Mock() pmock.process_type = 'agent' pmock.resource_type = sentinel.rtype pmock2 = Mock() pmock2.process_type = 'agent' pmock2.resource_type = sentinel.rtype2 self.pm.procs = {'one':pmock, 'two':pmock2} self.assertEquals(self.pm.get_a_local_process(sentinel.rtype2), pmock2) def test_get_a_local_process_no_match(self): self.assertIsNone(self.pm.get_a_local_process()) def test_is_local_service_process(self): pmock = Mock() pmock.name = sentinel.name pmock.process_type = 'simple' pmock2 = Mock() pmock2.name = sentinel.name2 pmock2.process_type = 'service' pmock3 = Mock() pmock3.name = sentinel.name3 pmock3.process_type = 'service' self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock3} self.assertTrue(self.pm.is_local_service_process(sentinel.name3)) def test_is_local_service_process_name_matches_but_type_doesnt(self): pmock = Mock() pmock.name = sentinel.name pmock.process_type = 'simple' pmock2 = Mock() pmock2.name = sentinel.name2 pmock2.process_type = 'notservice' pmock3 = Mock() pmock3.name = sentinel.name3 pmock3.process_type = 'notservice' self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock3} self.assertFalse(self.pm.is_local_service_process(sentinel.name3)) def test_is_local_agent_process(self): # agent is similar to above, but checks resource_type instead pmock = Mock() pmock.name = sentinel.name pmock.process_type = 'simple' pmock2 = Mock() pmock2.resource_type = sentinel.name2 pmock2.process_type = 'agent' pmock3 = Mock() pmock3.name = sentinel.name3 pmock3.process_type = 'notservice' self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock3} self.assertTrue(self.pm.is_local_agent_process(sentinel.name2)) def test_is_local_agent_process_not_found(self): self.assertFalse(self.pm.is_local_agent_process(sentinel.one)) def test__unregister_process_errors(self): pmock = Mock() pmock._proc_name = '1' pmock._proc_type = 'service' pmock._proc_res_id = sentinel.presid pmock._proc_svc_id = sentinel.psvcid self.container.resource_registry.delete.side_effect = NotFound self.pm.procs[sentinel.pid] = pmock self.pm.procs_by_name['1'] = pmock self.container.resource_registry.find_objects.reset_mock() self.container.resource_registry.find_objects.side_effect = None self.container.resource_registry.find_objects.return_value = ([],[]) self.pm._unregister_process(sentinel.pid, pmock) # show we tried to interact with the RR self.container.resource_registry.delete.assert_call(sentinel.presid, del_associations=True) self.container.resource_registry.find_objects.assert_called_once_with(sentinel.psvcid, "hasProcess", "Process", id_only=True) self.assertEquals(self.pm.procs, {}) self.assertEquals(self.pm.procs_by_name, {}) # NEXT: find_objects works fine and gives us an error deletion self.container.resource_registry.delete.reset_mock() self.container.resource_registry.find_objects.reset_mock() self.container.resource_registry.find_objects.side_effect = None self.container.resource_registry.find_objects.return_value = ([sentinel.svcid],[None]) self.pm.procs[sentinel.pid] = pmock self.pm.procs_by_name['1'] = pmock self.pm._unregister_process(sentinel.pid, pmock) self.container.resource_registry.delete.assert_calls([call(sentinel.presid, del_associations=True), call(sentinel.psvcid, del_associations=True)]) # NEXT: agent pmock = Mock() pmock.id = sentinel.pid pmock._proc_name = '1' pmock._proc_type = 'agent' self.pm.procs[sentinel.pid] = pmock self.pm.procs_by_name['1'] = pmock self.pm._unregister_process(sentinel.pid, pmock) self.container.directory.unregister_safe.assert_called_once_with("/Agents", sentinel.pid) def test__create_listening_endpoint_with_cfg(self): self.patch_cfg('pyon.container.procs.CFG', container=dict(messaging=dict(endpoint=dict(proc_listening_type='pyon.container.test.test_procs.TestRPCServer')))) ep = self.pm._create_listening_endpoint(process=sentinel.process) self.assertIsInstance(ep, TestRPCServer) def test__create_listening_endpoint_without_cfg_and_no_conv(self): self.patch_cfg('pyon.container.procs.CFG', container=dict(messaging=dict(endpoint=dict(proc_listening_type=None)))) ep = self.pm._create_listening_endpoint(process=sentinel.process) self.assertIsInstance(ep, ProcessRPCServer) def test_failed_process(self): self.pm.start() self.container.fail_fast = Mock() self.pm.procs['pid1'] = Mock() proc2 = BadProcess() self.pm.proc_sup.spawn(name="bad", service=proc2, target=proc2.fail_target) gevent.sleep(0) # Allow the new thread to fail and trigger the chain self.assertFalse(self.container.fail_fast.called) del self.pm.procs['pid1'] proc3 = BadProcess() self.pm.proc_sup.spawn(name="bad", service=proc3, target=proc3.fail_target) gevent.sleep(0) # Allow the new thread to fail and trigger the chain self.assertTrue(self.container.fail_fast.called)
def test_procmanager_iso(self): fakecc = FakeContainer() pm = ProcManager(fakecc) self.assertTrue(hasattr(fakecc, "spawn_process")) pm.start() pm.stop()
class Container(BaseContainerAgent): """ The Capability Container. Its purpose is to spawn/monitor processes and services that do the bulk of the work in the ION system. It also manages connections to the Exchange and the various forms of datastores in the systems. """ # Singleton static variables #node = None id = None name = None pidfile = None instance = None def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) self._is_started = False self._capabilities = [] # set container id and cc_agent name (as they are set in base class call) self.id = get_default_container_id() self.name = "cc_agent_%s" % self.id Container.instance = self from pyon.core import bootstrap bootstrap.container_instance = self log.debug("Container (sysname=%s) initializing ..." % bootstrap.get_sys_name()) # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = kwargs # Instantiate Directory and self-register # Has the additional side effect of either # bootstrapping the configuration into the # directory or read the configuration based # in the value of the auto_bootstrap setting self.directory = Directory() # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # File System - Interface to the OS File System, using correct path names and setups self.file_system = FileSystem(CFG) # Governance Controller - manages the governance related interceptors self.governance_controller = GovernanceController(self) # sFlow manager - controls sFlow stat emission self.sflow_manager = SFlowManager(self) # Coordinates the container start self._status = "INIT" # protection for when the container itself is used as a Process for clients self.container = self log.debug("Container initialized, OK.") def start(self): log.debug("Container starting...") if self._is_started: raise ContainerError("Container already started") # Check if this UNIX process already runs a Container. self.pidfile = "cc-pid-%d" % os.getpid() if os.path.exists(self.pidfile): raise ContainerError("Container.on_start(): Container is a singleton per UNIX process. Existing pid file found: %s" % self.pidfile) # write out a PID file containing our agent messaging name with open(self.pidfile, 'w') as f: pid_contents = {'messaging': dict(CFG.server.amqp), 'container-agent': self.name, 'container-xp': bootstrap.get_sys_name() } f.write(msgpack.dumps(pid_contents)) atexit.register(self._cleanup_pid) self._capabilities.append("PID_FILE") # set up abnormal termination handler for this container def handl(signum, frame): try: self._cleanup_pid() # cleanup the pidfile first self.quit() # now try to quit - will not error on second cleanup pidfile call finally: signal.signal(signal.SIGTERM, self._normal_signal) os.kill(os.getpid(), signal.SIGTERM) self._normal_signal = signal.signal(signal.SIGTERM, handl) self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") # Self-register with Directory self.directory.register("/Containers", self.id, cc_agent=self.name) self.directory.register("/Containers/%s" % self.id, "Processes") self._capabilities.append("DIRECTORY") # Event repository self.event_repository = EventRepository() self.event_pub = EventPublisher() self._capabilities.append("EVENT_REPOSITORY") # Local resource registry self.resource_registry = ResourceRegistry() self._capabilities.append("RESOURCE_REGISTRY") # Persistent objects self.datastore_manager.get_datastore("objects", DataStore.DS_PROFILE.OBJECTS) # State repository self.state_repository = StateRepository() self._capabilities.append("STATE_REPOSITORY") # Start ExchangeManager, which starts the node (broker connection) self.ex_manager.start() self._capabilities.append("EXCHANGE_MANAGER") self.proc_manager.start() self._capabilities.append("PROC_MANAGER") self.app_manager.start() self._capabilities.append("APP_MANAGER") self.governance_controller.start() self._capabilities.append("GOVERNANCE_CONTROLLER") if CFG.container.get('sflow', {}).get('enabled', False): self.sflow_manager.start() self._capabilities.append("SFLOW_MANAGER") # Start the CC-Agent API rsvc = ProcessRPCServer(node=self.node, from_name=self.name, service=self, process=self) # Start an ION process with the right kind of endpoint factory proc = self.proc_manager.proc_sup.spawn(name=self.name, listeners=[rsvc], service=self) self.proc_manager.proc_sup.ensure_ready(proc) self._capabilities.append("CONTAINER_AGENT") self.event_pub.publish_event(event_type="ContainerLifecycleEvent", origin=self.id, origin_type="CapabilityContainer", sub_type="START", state=ContainerStateEnum.START) self._is_started = True self._status = "RUNNING" log.info("Container started, OK.") @property def node(self): """ Returns the active/default Node that should be used for most communication in the system. Defers to exchange manager, but only if it has been started, otherwise returns None. """ if "EXCHANGE_MANAGER" in self._capabilities: return self.ex_manager.default_node return None @contextmanager def _push_status(self, new_status): """ Temporarily sets the internal status flag. Use this as a decorator or in a with-statement before calling a temporary status changing method, like start_rel_from_url. """ curstatus = self._status self._status = new_status try: yield finally: self._status = curstatus def serve_forever(self): """ Run the container until killed. """ log.debug("In Container.serve_forever") if not self.proc_manager.proc_sup.running: self.start() # serve forever short-circuits if immediate is on and children len is ok num_procs = len(self.proc_manager.proc_sup.children) immediate = CFG.system.get('immediate', False) if not (immediate and num_procs == 1): # only spawned greenlet is the CC-Agent # print a warning just in case if immediate and num_procs != 1: log.warn("CFG.system.immediate=True but number of spawned processes is not 1 (%d)", num_procs) try: # This just waits in this Greenlet for all child processes to complete, # which is triggered somewhere else. self.proc_manager.proc_sup.join_children() except (KeyboardInterrupt, SystemExit) as ex: log.info('Received a kill signal, shutting down the container.') watch_parent = CFG.system.get('watch_parent', None) if watch_parent: watch_parent.kill() except: log.exception('Unhandled error! Forcing container shutdown') else: log.debug("Container.serve_forever short-circuiting due to CFG.system.immediate") self.proc_manager.proc_sup.shutdown(CFG.cc.timeout.shutdown) def status(self): """ Returns the internal status. """ return self._status def _cleanup_pid(self): if self.pidfile: log.debug("Cleanup pidfile: %s", self.pidfile) try: os.remove(self.pidfile) except Exception, e: log.warn("Pidfile could not be deleted: %s" % str(e)) self.pidfile = None
class Container(BaseContainerAgent): """ The Capability Container. Its purpose is to spawn/monitor processes and services that do the bulk of the work in the ION system. It also manages connections to the Exchange and the various forms of datastores in the systems. """ # Singleton static variables #node = None id = None name = None pidfile = None instance = None def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) self._is_started = False # set container id and cc_agent name (as they are set in base class call) self.id = get_default_container_id() self.name = "cc_agent_%s" % self.id self._capabilities = [] bootstrap.container_instance = self Container.instance = self log.debug("Container (sysname=%s) initializing ..." % bootstrap.get_sys_name()) # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = kwargs # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() # TODO: Do not start a capability here. Symmetric start/stop self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") # Instantiate Directory self.directory = Directory() # internal router self.local_router = None # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # File System - Interface to the OS File System, using correct path names and setups self.file_system = FileSystem(CFG) # Governance Controller - manages the governance related interceptors self.governance_controller = GovernanceController(self) # sFlow manager - controls sFlow stat emission self.sflow_manager = SFlowManager(self) # Coordinates the container start self._status = "INIT" # protection for when the container itself is used as a Process for clients self.container = self # publisher, initialized in start() self.event_pub = None # context-local storage self.context = LocalContextMixin() log.debug("Container initialized, OK.") def start(self): log.debug("Container starting...") if self._is_started: raise ContainerError("Container already started") # Check if this UNIX process already runs a Container. self.pidfile = "cc-pid-%d" % os.getpid() if os.path.exists(self.pidfile): raise ContainerError("Container.on_start(): Container is a singleton per UNIX process. Existing pid file found: %s" % self.pidfile) # write out a PID file containing our agent messaging name with open(self.pidfile, 'w') as f: pid_contents = {'messaging': dict(CFG.server.amqp), 'container-agent': self.name, 'container-xp': bootstrap.get_sys_name()} f.write(msgpack.dumps(pid_contents)) atexit.register(self._cleanup_pid) self._capabilities.append("PID_FILE") # set up abnormal termination handler for this container def handl(signum, frame): try: self._cleanup_pid() # cleanup the pidfile first self.quit() # now try to quit - will not error on second cleanup pidfile call finally: signal.signal(signal.SIGTERM, self._normal_signal) os.kill(os.getpid(), signal.SIGTERM) self._normal_signal = signal.signal(signal.SIGTERM, handl) # set up greenlet debugging signal handler gevent.signal(signal.SIGUSR2, self._handle_sigusr2) self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") self._capabilities.append("DIRECTORY") # Event repository self.event_repository = EventRepository() self.event_pub = EventPublisher() self._capabilities.append("EVENT_REPOSITORY") # Local resource registry self.resource_registry = ResourceRegistry() self._capabilities.append("RESOURCE_REGISTRY") # Persistent objects self.datastore_manager.get_datastore("objects", DataStore.DS_PROFILE.OBJECTS) # State repository self.state_repository = StateRepository() self._capabilities.append("STATE_REPOSITORY") # internal router for local transports self.local_router = LocalRouter(bootstrap.get_sys_name()) self.local_router.start() self.local_router.ready.wait(timeout=2) self._capabilities.append("LOCAL_ROUTER") # Start ExchangeManager, which starts the node (broker connection) self.ex_manager.start() self._capabilities.append("EXCHANGE_MANAGER") self.proc_manager.start() self._capabilities.append("PROC_MANAGER") self.app_manager.start() self._capabilities.append("APP_MANAGER") self.governance_controller.start() self._capabilities.append("GOVERNANCE_CONTROLLER") if CFG.get_safe('container.sflow.enabled', False): self.sflow_manager.start() self._capabilities.append("SFLOW_MANAGER") # Start the CC-Agent API rsvc = ProcessRPCServer(node=self.node, from_name=self.name, service=self, process=self) cleanup = lambda _: self.proc_manager._cleanup_method(self.name, rsvc) # Start an ION process with the right kind of endpoint factory proc = self.proc_manager.proc_sup.spawn(name=self.name, listeners=[rsvc], service=self, cleanup_method=cleanup) self.proc_manager.proc_sup.ensure_ready(proc) proc.start_listeners() self._capabilities.append("CONTAINER_AGENT") self.event_pub.publish_event(event_type="ContainerLifecycleEvent", origin=self.id, origin_type="CapabilityContainer", sub_type="START", state=ContainerStateEnum.START) self._is_started = True self._status = "RUNNING" log.info("Container (%s) started, OK." , self.id) def _handle_sigusr2(self):#, signum, frame): """ Handles SIGUSR2, prints debugging greenlet information. """ gls = GreenletLeak.get_greenlets() allgls = [] for gl in gls: status = GreenletLeak.format_greenlet(gl) # build formatted output: # Greenlet at 0xdeadbeef # self: <EndpointUnit at 0x1ffcceef> # func: bound, EndpointUnit.some_func status[0].insert(0, "%s at %s:" % (gl.__class__.__name__, hex(id(gl)))) # indent anything in status a second time prefmt = [s.replace("\t", "\t\t") for s in status[0]] prefmt.append("traceback:") for line in status[1]: for subline in line.split("\n")[0:2]: prefmt.append(subline) glstr = "\n\t".join(prefmt) allgls.append(glstr) # print it out! print >>sys.stderr, "\n\n".join(allgls) with open("gls-%s" % os.getpid(), "w") as f: f.write("\n\n".join(allgls)) @property def node(self): """ Returns the active/default Node that should be used for most communication in the system. Defers to exchange manager, but only if it has been started, otherwise returns None. """ if "EXCHANGE_MANAGER" in self._capabilities: return self.ex_manager.default_node return None @contextmanager def _push_status(self, new_status): """ Temporarily sets the internal status flag. Use this as a decorator or in a with-statement before calling a temporary status changing method, like start_rel_from_url. """ curstatus = self._status self._status = new_status try: yield finally: self._status = curstatus def serve_forever(self): """ Run the container until killed. """ log.debug("In Container.serve_forever") if not self.proc_manager.proc_sup.running: self.start() # serve forever short-circuits if immediate is on and children len is ok num_procs = len(self.proc_manager.proc_sup.children) immediate = CFG.system.get('immediate', False) if not (immediate and num_procs == 1): # only spawned greenlet is the CC-Agent # print a warning just in case if immediate and num_procs != 1: log.warn("CFG.system.immediate=True but number of spawned processes is not 1 (%d)", num_procs) try: # This just waits in this Greenlet for all child processes to complete, # which is triggered somewhere else. self.proc_manager.proc_sup.join_children() except (KeyboardInterrupt, SystemExit) as ex: log.info('Received a kill signal, shutting down the container.') if hasattr(self, 'gl_parent_watch') and self.gl_parent_watch is not None: self.gl_parent_watch.kill() except: log.exception('Unhandled error! Forcing container shutdown') else: log.debug("Container.serve_forever short-circuiting due to CFG.system.immediate") self.proc_manager.proc_sup.shutdown(CFG.cc.timeout.shutdown) def status(self): """ Returns the internal status. """ return self._status def _cleanup_pid(self): if self.pidfile: log.debug("Cleanup pidfile: %s", self.pidfile) try: os.remove(self.pidfile) except Exception, e: log.warn("Pidfile could not be deleted: %s" % str(e)) self.pidfile = None
class Container(BaseContainerAgent): """ The Capability Container. Its purpose is to spawn/monitor processes and services that do the bulk of the work in the ION system. """ # Singleton static variables node = None id = None name = None pidfile = None instance = None def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) self._is_started = False # set id and name (as they are set in base class call) self.id = string.replace('%s_%d' % (os.uname()[1], os.getpid()), ".", "_") self.name = "cc_agent_%s" % self.id Container.instance = self # TODO: Bug: Replacing CFG instance not work because references are already public. Update directly dict_merge(CFG, kwargs, inplace=True) from pyon.core import bootstrap bootstrap.container_instance = self bootstrap.assert_configuration(CFG) bootstrap.sys_name = CFG.system.name or bootstrap.sys_name log.debug("Container (sysname=%s) initializing ..." % bootstrap.sys_name) # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = kwargs # Load object and service registry etc. bootstrap_pyon() # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() # File System - Interface to the OS File System, using correct path names and setups self.file_system = FileSystem(CFG) # Coordinates the container start self._is_started = False self._capabilities = [] log.debug("Container initialized, OK.") def start(self): log.debug("Container starting...") if self._is_started: raise ContainerError("Container already started") # Check if this UNIX process already runs a Container. self.pidfile = "cc-pid-%d" % os.getpid() if os.path.exists(self.pidfile): raise ContainerError("Container.on_start(): Container is a singleton per UNIX process. Existing pid file found: %s" % self.pidfile) # write out a PID file containing our agent messaging name with open(self.pidfile, 'w') as f: from pyon.core.bootstrap import get_sys_name pid_contents = {'messaging': dict(CFG.server.amqp), 'container-agent': self.name, 'container-xp': get_sys_name() } f.write(msgpack.dumps(pid_contents)) atexit.register(self._cleanup_pid) self._capabilities.append("PID_FILE") # set up abnormal termination handler for this container def handl(signum, frame): try: self._cleanup_pid() # cleanup the pidfile first self.quit() # now try to quit - will not error on second cleanup pidfile call finally: signal.signal(signal.SIGTERM, self._normal_signal) os.kill(os.getpid(), signal.SIGTERM) self._normal_signal = signal.signal(signal.SIGTERM, handl) self._capabilities.append("EXCHANGE_CONNECTION") self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") # Instantiate Directory and self-register self.directory = Directory() self.directory.register("/Containers", self.id, cc_agent=self.name) self._capabilities.append("DIRECTORY") # Create other repositories to make sure they are there and clean if needed self.datastore_manager.get_datastore("resources", DataStore.DS_PROFILE.RESOURCES) self.datastore_manager.get_datastore("objects", DataStore.DS_PROFILE.OBJECTS) self.state_repository = StateRepository() self._capabilities.append("STATE_REPOSITORY") self.event_repository = EventRepository() self._capabilities.append("EVENT_REPOSITORY") # Start ExchangeManager. In particular establish broker connection self.ex_manager.start() # TODO: Move this in ExchangeManager - but there is an error self.node, self.ioloop = messaging.make_node() # TODO: shortcut hack self._capabilities.append("EXCHANGE_MANAGER") self.proc_manager.start() self._capabilities.append("PROC_MANAGER") self.app_manager.start() self._capabilities.append("APP_MANAGER") # Start the CC-Agent API rsvc = ProcessRPCServer(node=self.node, name=self.name, service=self, process=self) # Start an ION process with the right kind of endpoint factory proc = self.proc_manager.proc_sup.spawn((CFG.cc.proctype or 'green', None), listener=rsvc) self.proc_manager.proc_sup.ensure_ready(proc) self._capabilities.append("CONTAINER_AGENT") self._is_started = True log.info("Container started, OK.") def serve_forever(self): """ Run the container until killed. """ log.debug("In Container.serve_forever") if not self.proc_manager.proc_sup.running: self.start() try: # This just waits in this Greenlet for all child processes to complete, # which is triggered somewhere else. self.proc_manager.proc_sup.join_children() except (KeyboardInterrupt, SystemExit) as ex: log.info('Received a kill signal, shutting down the container.') except: log.exception('Unhandled error! Forcing container shutdown') self.proc_manager.proc_sup.shutdown(CFG.cc.timeout.shutdown) def _cleanup_pid(self): if self.pidfile: log.debug("Cleanup pidfile: %s", self.pidfile) try: os.remove(self.pidfile) except Exception, e: log.warn("Pidfile could not be deleted: %s" % str(e)) self.pidfile = None
class Container(BaseContainerAgent): """ The Capability Container. Its purpose is to spawn/monitor processes and services that do the bulk of the work in the ION system. """ # Singleton static variables node = None id = None name = None pidfile = None instance = None def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) self._is_started = False # set id and name (as they are set in base class call) self.id = string.replace('%s_%d' % (os.uname()[1], os.getpid()), ".", "_") self.name = "cc_agent_%s" % self.id Container.instance = self # TODO: Bug: Replacing CFG instance not work because references are already public. Update directly dict_merge(CFG, kwargs, inplace=True) from pyon.core import bootstrap bootstrap.container_instance = self bootstrap.assert_configuration(CFG) log.debug("Container (sysname=%s) initializing ..." % bootstrap.get_sys_name()) # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = kwargs # Load object and service registry etc. bootstrap_pyon() # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() # File System - Interface to the OS File System, using correct path names and setups self.file_system = FileSystem(CFG) # Governance Controller - manages the governance related interceptors self.governance_controller = GovernanceController(self) # sFlow manager - controls sFlow stat emission self.sflow_manager = SFlowManager(self) # Coordinates the container start self._is_started = False self._capabilities = [] self._status = "INIT" # protection for when the container itself is used as a Process for clients self.container = self log.debug("Container initialized, OK.") def start(self): log.debug("Container starting...") if self._is_started: raise ContainerError("Container already started") # Check if this UNIX process already runs a Container. self.pidfile = "cc-pid-%d" % os.getpid() if os.path.exists(self.pidfile): raise ContainerError("Container.on_start(): Container is a singleton per UNIX process. Existing pid file found: %s" % self.pidfile) # write out a PID file containing our agent messaging name with open(self.pidfile, 'w') as f: pid_contents = {'messaging': dict(CFG.server.amqp), 'container-agent': self.name, 'container-xp': bootstrap.get_sys_name() } f.write(msgpack.dumps(pid_contents)) atexit.register(self._cleanup_pid) self._capabilities.append("PID_FILE") # set up abnormal termination handler for this container def handl(signum, frame): try: self._cleanup_pid() # cleanup the pidfile first self.quit() # now try to quit - will not error on second cleanup pidfile call finally: signal.signal(signal.SIGTERM, self._normal_signal) os.kill(os.getpid(), signal.SIGTERM) self._normal_signal = signal.signal(signal.SIGTERM, handl) self._capabilities.append("EXCHANGE_CONNECTION") self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") # Instantiate Directory and self-register self.directory = Directory() self.directory.register("/Containers", self.id, cc_agent=self.name) self.directory.register("/Containers/%s" % self.id, "Processes") self._capabilities.append("DIRECTORY") # Event repository self.event_repository = EventRepository() self.event_pub = EventPublisher() self._capabilities.append("EVENT_REPOSITORY") # Local resource registry self.resource_registry = ResourceRegistry() self._capabilities.append("RESOURCE_REGISTRY") # Persistent objects self.datastore_manager.get_datastore("objects", DataStore.DS_PROFILE.OBJECTS) # State repository self.state_repository = StateRepository() self._capabilities.append("STATE_REPOSITORY") # Start ExchangeManager, which starts the node (broker connection) self.node, self.ioloop = self.ex_manager.start() self._capabilities.append("EXCHANGE_MANAGER") self.proc_manager.start() self._capabilities.append("PROC_MANAGER") self.app_manager.start() self._capabilities.append("APP_MANAGER") self.governance_controller.start() self._capabilities.append("GOVERNANCE_CONTROLLER") if CFG.container.get('sflow', {}).get('enabled', False): self.sflow_manager.start() self._capabilities.append("SFLOW_MANAGER") # Start the CC-Agent API rsvc = ProcessRPCServer(node=self.node, from_name=self.name, service=self, process=self) # Start an ION process with the right kind of endpoint factory proc = self.proc_manager.proc_sup.spawn((CFG.cc.proctype or 'green', None), listener=rsvc) self.proc_manager.proc_sup.ensure_ready(proc) self._capabilities.append("CONTAINER_AGENT") self.event_pub.publish_event(event_type="ContainerLifecycleEvent", origin=self.id, origin_type="CapabilityContainer", sub_type="START", state=ContainerStateEnum.START) self._is_started = True self._status = "RUNNING" log.info("Container started, OK.") @contextmanager def _push_status(self, new_status): """ Temporarily sets the internal status flag. Use this as a decorator or in a with-statement before calling a temporary status changing method, like start_rel_from_url. """ curstatus = self._status self._status = new_status try: yield finally: self._status = curstatus def serve_forever(self): """ Run the container until killed. """ log.debug("In Container.serve_forever") if not self.proc_manager.proc_sup.running: self.start() # serve forever short-circuits if immediate is on and children len is ok num_procs = len(self.proc_manager.proc_sup.children) immediate = CFG.system.get('immediate', False) if not (immediate and num_procs == 1): # only spawned greenlet is the CC-Agent # print a warning just in case if immediate and num_procs != 1: log.warn("CFG.system.immediate=True but number of spawned processes is not 1 (%d)", num_procs) try: # This just waits in this Greenlet for all child processes to complete, # which is triggered somewhere else. self.proc_manager.proc_sup.join_children() except (KeyboardInterrupt, SystemExit) as ex: log.info('Received a kill signal, shutting down the container.') watch_parent = CFG.system.get('watch_parent', None) if watch_parent: watch_parent.kill() except: log.exception('Unhandled error! Forcing container shutdown') else: log.debug("Container.serve_forever short-circuiting due to CFG.system.immediate") self.proc_manager.proc_sup.shutdown(CFG.cc.timeout.shutdown) def status(self): """ Returns the internal status. """ return self._status def _cleanup_pid(self): if self.pidfile: log.debug("Cleanup pidfile: %s", self.pidfile) try: os.remove(self.pidfile) except Exception, e: log.warn("Pidfile could not be deleted: %s" % str(e)) self.pidfile = None
class Container(BaseContainerAgent): """ The Capability Container. Its purpose is to spawn/monitor processes and services that do the bulk of the work in the ION system. It also manages connections to the Exchange and the various forms of datastores in the systems. """ # Singleton static variables #node = None id = None name = None pidfile = None instance = None def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) self._is_started = False # set container id and cc_agent name (as they are set in base class call) self.id = get_default_container_id() self.name = "cc_agent_%s" % self.id self._capabilities = [] bootstrap.container_instance = self Container.instance = self log.debug("Container (sysname=%s) initializing ..." % bootstrap.get_sys_name()) # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = kwargs # DatastoreManager - controls access to Datastores (both mock and couch backed) self.datastore_manager = DatastoreManager() # TODO: Do not start a capability here. Symmetric start/stop self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") # Instantiate Directory self.directory = Directory() # internal router self.local_router = None # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) # File System - Interface to the OS File System, using correct path names and setups self.file_system = FileSystem(CFG) # Governance Controller - manages the governance related interceptors self.governance_controller = GovernanceController(self) # sFlow manager - controls sFlow stat emission self.sflow_manager = SFlowManager(self) # Coordinates the container start self._status = "INIT" # protection for when the container itself is used as a Process for clients self.container = self # publisher, initialized in start() self.event_pub = None # context-local storage self.context = LocalContextMixin() log.debug("Container initialized, OK.") def start(self): log.debug("Container starting...") if self._is_started: raise ContainerError("Container already started") # Check if this UNIX process already runs a Container. self.pidfile = "cc-pid-%d" % os.getpid() if os.path.exists(self.pidfile): raise ContainerError( "Container.on_start(): Container is a singleton per UNIX process. Existing pid file found: %s" % self.pidfile) # write out a PID file containing our agent messaging name with open(self.pidfile, 'w') as f: pid_contents = { 'messaging': dict(CFG.server.amqp), 'container-agent': self.name, 'container-xp': bootstrap.get_sys_name() } f.write(msgpack.dumps(pid_contents)) atexit.register(self._cleanup_pid) self._capabilities.append("PID_FILE") # set up abnormal termination handler for this container def handl(signum, frame): try: self._cleanup_pid() # cleanup the pidfile first self.quit( ) # now try to quit - will not error on second cleanup pidfile call finally: signal.signal(signal.SIGTERM, self._normal_signal) os.kill(os.getpid(), signal.SIGTERM) self._normal_signal = signal.signal(signal.SIGTERM, handl) # set up greenlet debugging signal handler gevent.signal(signal.SIGUSR2, self._handle_sigusr2) self.datastore_manager.start() self._capabilities.append("DATASTORE_MANAGER") self._capabilities.append("DIRECTORY") # Event repository self.event_repository = EventRepository() self.event_pub = EventPublisher() self._capabilities.append("EVENT_REPOSITORY") # Local resource registry self.resource_registry = ResourceRegistry() self._capabilities.append("RESOURCE_REGISTRY") # Persistent objects self.datastore_manager.get_datastore("objects", DataStore.DS_PROFILE.OBJECTS) # State repository self.state_repository = StateRepository() self._capabilities.append("STATE_REPOSITORY") # internal router for local transports self.local_router = LocalRouter(bootstrap.get_sys_name()) self.local_router.start() self.local_router.ready.wait(timeout=2) self._capabilities.append("LOCAL_ROUTER") # Start ExchangeManager, which starts the node (broker connection) self.ex_manager.start() self._capabilities.append("EXCHANGE_MANAGER") self.proc_manager.start() self._capabilities.append("PROC_MANAGER") self.app_manager.start() self._capabilities.append("APP_MANAGER") self.governance_controller.start() self._capabilities.append("GOVERNANCE_CONTROLLER") if CFG.get_safe('container.sflow.enabled', False): self.sflow_manager.start() self._capabilities.append("SFLOW_MANAGER") # Start the CC-Agent API rsvc = ProcessRPCServer(node=self.node, from_name=self.name, service=self, process=self) cleanup = lambda _: self.proc_manager._cleanup_method(self.name, rsvc) # Start an ION process with the right kind of endpoint factory proc = self.proc_manager.proc_sup.spawn(name=self.name, listeners=[rsvc], service=self, cleanup_method=cleanup) self.proc_manager.proc_sup.ensure_ready(proc) proc.start_listeners() self._capabilities.append("CONTAINER_AGENT") self.event_pub.publish_event(event_type="ContainerLifecycleEvent", origin=self.id, origin_type="CapabilityContainer", sub_type="START", state=ContainerStateEnum.START) self._is_started = True self._status = "RUNNING" log.info("Container (%s) started, OK.", self.id) def _handle_sigusr2(self): #, signum, frame): """ Handles SIGUSR2, prints debugging greenlet information. """ gls = GreenletLeak.get_greenlets() allgls = [] for gl in gls: status = GreenletLeak.format_greenlet(gl) # build formatted output: # Greenlet at 0xdeadbeef # self: <EndpointUnit at 0x1ffcceef> # func: bound, EndpointUnit.some_func status[0].insert( 0, "%s at %s:" % (gl.__class__.__name__, hex(id(gl)))) # indent anything in status a second time prefmt = [s.replace("\t", "\t\t") for s in status[0]] prefmt.append("traceback:") for line in status[1]: for subline in line.split("\n")[0:2]: prefmt.append(subline) glstr = "\n\t".join(prefmt) allgls.append(glstr) # print it out! print >> sys.stderr, "\n\n".join(allgls) with open("gls-%s" % os.getpid(), "w") as f: f.write("\n\n".join(allgls)) @property def node(self): """ Returns the active/default Node that should be used for most communication in the system. Defers to exchange manager, but only if it has been started, otherwise returns None. """ if "EXCHANGE_MANAGER" in self._capabilities: return self.ex_manager.default_node return None @contextmanager def _push_status(self, new_status): """ Temporarily sets the internal status flag. Use this as a decorator or in a with-statement before calling a temporary status changing method, like start_rel_from_url. """ curstatus = self._status self._status = new_status try: yield finally: self._status = curstatus def serve_forever(self): """ Run the container until killed. """ log.debug("In Container.serve_forever") if not self.proc_manager.proc_sup.running: self.start() # serve forever short-circuits if immediate is on and children len is ok num_procs = len(self.proc_manager.proc_sup.children) immediate = CFG.system.get('immediate', False) if not (immediate and num_procs == 1): # only spawned greenlet is the CC-Agent # print a warning just in case if immediate and num_procs != 1: log.warn( "CFG.system.immediate=True but number of spawned processes is not 1 (%d)", num_procs) try: # This just waits in this Greenlet for all child processes to complete, # which is triggered somewhere else. self.proc_manager.proc_sup.join_children() except (KeyboardInterrupt, SystemExit) as ex: log.info( 'Received a kill signal, shutting down the container.') if hasattr(self, 'gl_parent_watch' ) and self.gl_parent_watch is not None: self.gl_parent_watch.kill() except: log.exception('Unhandled error! Forcing container shutdown') else: log.debug( "Container.serve_forever short-circuiting due to CFG.system.immediate" ) self.proc_manager.proc_sup.shutdown(CFG.cc.timeout.shutdown) def status(self): """ Returns the internal status. """ return self._status def _cleanup_pid(self): if self.pidfile: log.debug("Cleanup pidfile: %s", self.pidfile) try: os.remove(self.pidfile) except Exception, e: log.warn("Pidfile could not be deleted: %s" % str(e)) self.pidfile = None
class TestProcManager(PyonTestCase): def setUp(self): self.container = Mock() self.pm = ProcManager(self.container) self.container.resource_registry.create.return_value = (sentinel.rid, sentinel.rev) self.container.resource_registry.find_resources.return_value = ([sentinel.oid], [sentinel.orev]) def test_start(self): self.pm.start() self.assertEquals(self.pm.cc_id, sentinel.rid) def test_start_with_org(self): self.patch_cfg('pyon.container.procs.CFG', {'container':{'org_name':'NOT_DEFAULT'}}) self.pm.start() self.container.resource_registry.create_association.assert_called_once_with(sentinel.oid, PRED.hasResource, sentinel.rid) # @patch('pyon.datastore.couchdb.couchdb_datastore.CouchDB_DataStore._stats') # def test_stop(self, statsmock): # self.pm.start() # # self.pm.stop() # # self.assertEquals(statsmock.get_stats.call_count, 2) def test__cleanup_method(self): ep = Mock() ep._chan._queue_auto_delete = False self.pm._cleanup_method(sentinel.queue, ep) ch = self.container.node.channel.return_value ch._destroy_queue.assert_called_once_with() self.assertIsInstance(ch._recv_name, NameTrio) self.assertIn(str(sentinel.queue), str(ch._recv_name)) @patch('pyon.container.procs.log') def test__cleanup_method_raises_error(self, mocklog): ep = Mock() ep._chan._queue_auto_delete = False ch = self.container.node.channel.return_value ch._destroy_queue.side_effect = TransportError self.pm._cleanup_method(sentinel.queue, ep) self.assertEquals(mocklog.warn.call_count, 1) # @patch('pyon.datastore.couchdb.couchdb_datastore.CouchDB_DataStore._stats', Mock()) @patch('pyon.container.procs.log') def test_stop_with_error(self, mocklog): self.pm.start() self.pm.terminate_process = Mock(side_effect=BadRequest) procmock = Mock() procmock._proc_start_time = 0 procmock.id = sentinel.pid self.pm.procs[sentinel.pid] = procmock self.pm.procs_by_name['dummy'] = procmock self.pm.stop() self.pm.terminate_process.assert_called_once_with(sentinel.pid) mocklog.warn.assert_has_calls([call("Failed to terminate process (%s): %s", sentinel.pid, ANY), call("ProcManager procs not empty: %s", self.pm.procs), call("ProcManager procs_by_name not empty: %s", self.pm.procs_by_name)]) def test_list_local_processes(self): pmock = Mock() pmock.process_type = sentinel.ptype pmock2 = Mock() pmock2.process_type = sentinel.ptype2 self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock} self.assertEquals(self.pm.list_local_processes(), [pmock, pmock2, pmock]) def test_list_local_processes_proc_type_filter(self): pmock = Mock() pmock.process_type = sentinel.ptype pmock2 = Mock() pmock2.process_type = sentinel.ptype2 self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock} self.assertEquals(self.pm.list_local_processes(sentinel.ptype2), [pmock2]) def test_get_a_local_process(self): pmock = Mock() pmock.name = sentinel.name pmock2 = Mock() pmock2.name = sentinel.name2 self.pm.procs = {'one':pmock, 'two':pmock2} self.assertEquals(self.pm.get_a_local_process(sentinel.name2), pmock2) def test_get_a_local_process_for_agent_res_id(self): pmock = Mock() pmock.process_type = 'agent' pmock.resource_type = sentinel.rtype pmock2 = Mock() pmock2.process_type = 'agent' pmock2.resource_type = sentinel.rtype2 self.pm.procs = {'one':pmock, 'two':pmock2} self.assertEquals(self.pm.get_a_local_process(sentinel.rtype2), pmock2) def test_get_a_local_process_no_match(self): self.assertIsNone(self.pm.get_a_local_process()) def test_is_local_service_process(self): pmock = Mock() pmock.name = sentinel.name pmock.process_type = 'simple' pmock2 = Mock() pmock2.name = sentinel.name2 pmock2.process_type = 'service' pmock3 = Mock() pmock3.name = sentinel.name3 pmock3.process_type = 'service' self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock3} self.assertTrue(self.pm.is_local_service_process(sentinel.name3)) def test_is_local_service_process_name_matches_but_type_doesnt(self): pmock = Mock() pmock.name = sentinel.name pmock.process_type = 'simple' pmock2 = Mock() pmock2.name = sentinel.name2 pmock2.process_type = 'notservice' pmock3 = Mock() pmock3.name = sentinel.name3 pmock3.process_type = 'notservice' self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock3} self.assertFalse(self.pm.is_local_service_process(sentinel.name3)) def test_is_local_agent_process(self): # agent is similar to above, but checks resource_type instead pmock = Mock() pmock.name = sentinel.name pmock.process_type = 'simple' pmock2 = Mock() pmock2.resource_type = sentinel.name2 pmock2.process_type = 'agent' pmock3 = Mock() pmock3.name = sentinel.name3 pmock3.process_type = 'notservice' self.pm.procs = {'one':pmock, 'two':pmock2, 'three':pmock3} self.assertTrue(self.pm.is_local_agent_process(sentinel.name2)) def test_is_local_agent_process_not_found(self): self.assertFalse(self.pm.is_local_agent_process(sentinel.one)) def test__unregister_process_errors(self): pmock = Mock() pmock._proc_name = '1' pmock._proc_type = 'service' pmock._proc_res_id = sentinel.presid pmock._proc_svc_id = sentinel.psvcid self.container.resource_registry.delete.side_effect = NotFound self.container.resource_registry.find_objects.side_effect = ResourceNotFound self.pm.procs[sentinel.pid] = pmock self.pm.procs_by_name['1'] = pmock self.pm._unregister_process(sentinel.pid, pmock) # show we tried to interact with the RR self.container.resource_registry.delete.assert_call(sentinel.presid, del_associations=True) self.container.resource_registry.find_objects.assert_called_once_with(sentinel.psvcid, "hasProcess", "Process", id_only=True) self.assertEquals(self.pm.procs, {}) self.assertEquals(self.pm.procs_by_name, {}) # NEXT: find_objects works fine and gives us an error deletion self.container.resource_registry.delete.reset_mock() self.container.resource_registry.find_objects.reset_mock() self.container.resource_registry.find_objects.side_effect = None self.container.resource_registry.find_objects.return_value = ([sentinel.svcid],[None]) self.pm.procs[sentinel.pid] = pmock self.pm.procs_by_name['1'] = pmock self.pm._unregister_process(sentinel.pid, pmock) self.container.resource_registry.delete.assert_calls([call(sentinel.presid, del_associations=True), call(sentinel.psvcid, del_associations=True)]) # NEXT: agent pmock = Mock() pmock.id = sentinel.pid pmock._proc_name = '1' pmock._proc_type = 'agent' self.pm.procs[sentinel.pid] = pmock self.pm.procs_by_name['1'] = pmock self.pm._unregister_process(sentinel.pid, pmock) self.container.directory.unregister_safe.assert_called_once_with("/Agents", sentinel.pid) def test__create_listening_endpoint_with_cfg(self): self.patch_cfg('pyon.container.procs.CFG', container=dict(messaging=dict(endpoint=dict(proc_listening_type='pyon.container.test.test_procs.TestRPCServer')))) ep = self.pm._create_listening_endpoint(process=sentinel.process) self.assertIsInstance(ep, TestRPCServer) def test__create_listening_endpoint_without_cfg_and_no_conv(self): self.patch_cfg('pyon.container.procs.CFG', container=dict(messaging=dict(endpoint=dict(proc_listening_type=None, rpc_conversation_enabled=False)))) ep = self.pm._create_listening_endpoint(process=sentinel.process) self.assertIsInstance(ep, ProcessRPCServer) def test__create_listening_endpoint_without_cfg_and_conv(self): self.patch_cfg('pyon.container.procs.CFG', container=dict(messaging=dict(endpoint=dict(proc_listening_type=None, rpc_conversation_enabled=True)))) ep = self.pm._create_listening_endpoint(process=sentinel.process) self.assertIsInstance(ep, ConversationRPCServer) def test_failed_process(self): self.pm.start() self.container.fail_fast = Mock() self.pm.procs['pid1'] = Mock() proc2 = BadProcess() self.pm.proc_sup.spawn(name="bad", service=proc2, target=proc2.fail_target) gevent.sleep(0) # Allow the new thread to fail and trigger the chain self.assertFalse(self.container.fail_fast.called) del self.pm.procs['pid1'] proc3 = BadProcess() self.pm.proc_sup.spawn(name="bad", service=proc3, target=proc3.fail_target) gevent.sleep(0) # Allow the new thread to fail and trigger the chain self.assertTrue(self.container.fail_fast.called)
def setUp(self): self.container = Mock() self.pm = ProcManager(self.container) self.container.resource_registry.create.return_value = (sentinel.rid, sentinel.rev) self.container.resource_registry.find_resources.return_value = ([sentinel.oid], [sentinel.orev])
class Container(BaseContainerAgent): """ The Capability Container. Its purpose is to spawn/monitor processes and services that do the bulk of the work in the ION system. """ node = None id = None name = None pidfile = None instance = None def __init__(self, *args, **kwargs): BaseContainerAgent.__init__(self, *args, **kwargs) # set id and name (as they are set in base class call) self.id = string.replace('%s_%d' % (os.uname()[1], os.getpid()), ".", "_") self.name = "cc_agent_%s" % self.id Container.instance = self # TODO: Bug: Replacing CFG instance not work because references are already public. Update directly dict_merge(CFG, kwargs) from pyon.core import bootstrap bootstrap.sys_name = CFG.system.name or bootstrap.sys_name log.debug("Container (sysname=%s) initializing ..." % bootstrap.sys_name) # Keep track of the overrides from the command-line, so they can trump app/rel file data self.spawn_args = DictModifier(CFG, kwargs) # Load object and service registry bootstrap_pyon() # Create this Container's specific ExchangeManager instance self.ex_manager = ExchangeManager(self) # Create this Container's specific ProcManager instance self.proc_manager = ProcManager(self) # Create this Container's specific AppManager instance self.app_manager = AppManager(self) log.debug("Container initialized, OK.") def start(self): log.debug("Container starting...") # Check if this UNIX process already runs a Container. self.pidfile = "cc-pid-%d" % os.getpid() if os.path.exists(self.pidfile): raise Exception("Container.on_start(): Container is a singleton per UNIX process. Existing pid file found: %s" % self.pidfile) # write out a PID file containing our agent messaging name with open(self.pidfile, 'w') as f: from pyon.core.bootstrap import sys_name pid_contents = {'messaging': dict(CFG.server.amqp), 'container-agent': self.name, 'container-xp': sys_name } f.write(msgpack.dumps(pid_contents)) atexit.register(self._cleanup_pid) # set up abnormal termination handler for this container def handl(signum, frame): try: self._cleanup_pid() # cleanup the pidfile first self.quit() # now try to quit - will not error on second cleanup pidfile call finally: signal.signal(signal.SIGTERM, self._normal_signal) os.kill(os.getpid(), signal.SIGTERM) self._normal_signal = signal.signal(signal.SIGTERM, handl) # Start ExchangeManager. In particular establish broker connection self.ex_manager.start() # TODO: Move this in ExchangeManager - but there is an error self.node, self.ioloop = messaging.make_node() # TODO: shortcut hack # Instantiate Directory singleton and self-register # TODO: At this point, there is no special config override self.directory = Directory() self.directory.register("/Containers", self.id, cc_agent=self.name) self.proc_manager.start() self.app_manager.start() # Start the CC-Agent API rsvc = ProcessRPCServer(node=self.node, name=self.name, service=self, process=self) # Start an ION process with the right kind of endpoint factory self.proc_manager.proc_sup.spawn((CFG.cc.proctype or 'green', None), listener=rsvc) rsvc.get_ready_event().wait(timeout=10) # @TODO: no hardcode log.info("Container started, OK.") def serve_forever(self): """ Run the container until killed. """ log.debug("In Container.serve_forever") if not self.proc_manager.proc_sup.running: self.start() try: # This just waits in this Greenlet for all child processes to complete, # which is triggered somewhere else. self.proc_manager.proc_sup.join_children() except (KeyboardInterrupt, SystemExit) as ex: log.info('Received a kill signal, shutting down the container.') except: log.exception('Unhandled error! Forcing container shutdown') self.proc_manager.proc_sup.shutdown(CFG.cc.timeout.shutdown) def _cleanup_pid(self): if self.pidfile: log.debug("Cleanup pidfile: %s", self.pidfile) try: os.remove(self.pidfile) except Exception, e: log.warn("Pidfile could not be deleted: %s" % str(e)) self.pidfile = None