class ProcessDispatcherEEAgentIntTest(ProcessDispatcherServiceIntTest):
"""Run the basic int tests again, with a different environment
"""
def setUp(self):
self.dashi = None
self._start_container()
from pyon.public import CFG
self.container_client = ContainerAgentClient(node=self.container.node,
name=self.container.name)
self.container = self.container_client._get_container_instance()
app = dict(name="process_dispatcher", processapp=("process_dispatcher",
"ion.services.cei.process_dispatcher_service",
"ProcessDispatcherService"))
self.container.start_app(app, config=pd_config)
self.rr_cli = self.container.resource_registry
self.pd_cli = ProcessDispatcherServiceClient(node=self.container.node)
self.process_definition = ProcessDefinition(name='test_process')
self.process_definition.executable = {'module': 'ion.services.cei.test.test_process_dispatcher',
'class': 'TestProcess'}
self.process_definition_id = self.pd_cli.create_process_definition(self.process_definition)
self._eea_pids = []
self._eea_pid_to_resource_id = {}
self._eea_pid_to_persistence_dir = {}
self._tmpdirs = []
self.dashi = get_dashi(uuid.uuid4().hex,
pd_config['processdispatcher']['dashi_uri'],
pd_config['processdispatcher']['dashi_exchange'],
sysname=CFG.get_safe("dashi.sysname")
)
#send a fake node_state message to PD's dashi binding.
self.node1_id = uuid.uuid4().hex
self._send_node_state("engine1", self.node1_id)
self._initial_eea_pid = self._start_eeagent(self.node1_id)
self.waiter = ProcessStateWaiter()
def _send_node_state(self, engine_id, node_id=None):
node_id = node_id or uuid.uuid4().hex
node_state = dict(node_id=node_id, state=InstanceState.RUNNING,
domain_id=domain_id_from_engine(engine_id))
self.dashi.fire(get_pd_dashi_name(), "node_state", args=node_state)
def _start_eeagent(self, node_id, resource_id=None, persistence_dir=None):
if not persistence_dir:
persistence_dir = tempfile.mkdtemp()
self._tmpdirs.append(persistence_dir)
resource_id = resource_id or uuid.uuid4().hex
agent_config = _get_eeagent_config(node_id, persistence_dir,
resource_id=resource_id)
pid = self.container_client.spawn_process(name="eeagent",
module="ion.agents.cei.execution_engine_agent",
cls="ExecutionEngineAgent", config=agent_config)
log.info('Agent pid=%s.', str(pid))
self._eea_pids.append(pid)
self._eea_pid_to_resource_id[pid] = resource_id
self._eea_pid_to_persistence_dir[pid] = persistence_dir
return pid
def _kill_eeagent(self, pid):
self.assertTrue(pid in self._eea_pids)
self.container.terminate_process(pid)
self._eea_pids.remove(pid)
del self._eea_pid_to_resource_id[pid]
del self._eea_pid_to_persistence_dir[pid]
def tearDown(self):
for pid in list(self._eea_pids):
self._kill_eeagent(pid)
for d in self._tmpdirs:
shutil.rmtree(d)
self.waiter.stop()
if self.dashi:
self.dashi.cancel()
def test_requested_ee(self):
# request non-default engine
process_target = ProcessTarget(execution_engine_id="engine2")
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.ALWAYS
process_schedule.target = process_target
pid = self.pd_cli.create_process(self.process_definition_id)
self.waiter.start()
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, process_id=pid)
self.waiter.await_state_event(pid, ProcessStateEnum.WAITING)
# request unknown engine, with NEVER queuing mode. The request
# should be rejected.
# verifies L4-CI-CEI-RQ52
process_target = ProcessTarget(execution_engine_id="not-a-real-ee")
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.NEVER
process_schedule.target = process_target
rejected_pid = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, process_id=rejected_pid)
self.waiter.await_state_event(rejected_pid, ProcessStateEnum.REJECTED)
# now add a node and eeagent for engine2. original process should leave
# queue and start running
node2_id = uuid.uuid4().hex
self._send_node_state("engine2", node2_id)
self._start_eeagent(node2_id)
self.waiter.await_state_event(pid, ProcessStateEnum.RUNNING)
# spawn another process. it should start immediately.
process_target = ProcessTarget(execution_engine_id="engine2")
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.NEVER
process_schedule.target = process_target
pid2 = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, process_id=pid2)
self.waiter.await_state_event(pid2, ProcessStateEnum.RUNNING)
# one more with node exclusive
process_target = ProcessTarget(execution_engine_id="engine2",
node_exclusive="hats")
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.NEVER
process_schedule.target = process_target
pid3 = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, process_id=pid3)
self.waiter.await_state_event(pid3, ProcessStateEnum.RUNNING)
# kill the processes for good
self.pd_cli.cancel_process(pid)
self.waiter.await_state_event(pid, ProcessStateEnum.TERMINATED)
self.pd_cli.cancel_process(pid2)
self.waiter.await_state_event(pid2, ProcessStateEnum.TERMINATED)
self.pd_cli.cancel_process(pid3)
self.waiter.await_state_event(pid3, ProcessStateEnum.TERMINATED)
def test_node_exclusive(self):
# the node_exclusive constraint is used to ensure multiple processes
# of the same "kind" each get a VM exclusive of each other. Other
# processes may run on these VMs, just not processes with the same
# node_exclusive tag. Since we cannot directly query the contents
# of each node in this test, we prove the capability by scheduling
# processes one by one and checking their state.
# verifies L4-CI-CEI-RQ121
# verifies L4-CI-CEI-RQ57
# first off, setUp() created a single node and eeagent.
# We schedule two processes with the same "abc" node_exclusive
# tag. Since there is only one node, the first process should run
# and the second should be queued.
process_target = ProcessTarget(execution_engine_id="engine1")
process_target.node_exclusive = "abc"
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.ALWAYS
process_schedule.target = process_target
pid1 = self.pd_cli.create_process(self.process_definition_id)
self.waiter.start()
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, process_id=pid1)
self.waiter.await_state_event(pid1, ProcessStateEnum.RUNNING)
pid2 = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, process_id=pid2)
self.waiter.await_state_event(pid2, ProcessStateEnum.WAITING)
# now demonstrate that the node itself is not full by launching
# a third process without a node_exclusive tag -- it should start
# immediately
process_target.node_exclusive = None
pid3 = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, process_id=pid3)
self.waiter.await_state_event(pid3, ProcessStateEnum.RUNNING)
# finally, add a second node to the engine. pid2 should be started
# since there is an exclusive "abc" node free.
node2_id = uuid.uuid4().hex
self._send_node_state("engine1", node2_id)
self._start_eeagent(node2_id)
self.waiter.await_state_event(pid2, ProcessStateEnum.RUNNING)
# kill the processes for good
self.pd_cli.cancel_process(pid1)
self.waiter.await_state_event(pid1, ProcessStateEnum.TERMINATED)
self.pd_cli.cancel_process(pid2)
self.waiter.await_state_event(pid2, ProcessStateEnum.TERMINATED)
self.pd_cli.cancel_process(pid3)
self.waiter.await_state_event(pid3, ProcessStateEnum.TERMINATED)
def test_code_download(self):
# create a process definition that has no URL; only module and class.
process_definition_no_url = ProcessDefinition(name='test_process_nodownload')
process_definition_no_url.executable = {'module': 'ion.my.test.process',
'class': 'TestProcess'}
process_definition_id_no_url = self.pd_cli.create_process_definition(process_definition_no_url)
# create another that has a URL of the python file (this very file)
# verifies L4-CI-CEI-RQ114
url = "file://%s" % os.path.join(os.path.dirname(__file__), 'test_process_dispatcher.py')
process_definition = ProcessDefinition(name='test_process_download')
process_definition.executable = {'module': 'ion.my.test.process',
'class': 'TestProcess', 'url': url}
process_definition_id = self.pd_cli.create_process_definition(process_definition)
process_target = ProcessTarget()
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.ALWAYS
process_schedule.target = process_target
self.waiter.start()
# Test a module with no download fails
pid_no_url = self.pd_cli.create_process(process_definition_id_no_url)
self.pd_cli.schedule_process(process_definition_id_no_url,
process_schedule, process_id=pid_no_url)
self.waiter.await_state_event(pid_no_url, ProcessStateEnum.FAILED)
# Test a module with a URL runs
pid = self.pd_cli.create_process(process_definition_id)
self.pd_cli.schedule_process(process_definition_id,
process_schedule, process_id=pid)
self.waiter.await_state_event(pid, ProcessStateEnum.RUNNING)
def _add_test_process(self, restart_mode=None):
process_schedule = ProcessSchedule()
if restart_mode is not None:
process_schedule.restart_mode = restart_mode
pid = self.pd_cli.create_process(self.process_definition_id)
pid_listen_name = "PDtestproc_%s" % uuid.uuid4().hex
config = {'process': {'listen_name': pid_listen_name}}
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, process_id=pid, configuration=config)
client = TestClient(to_name=pid_listen_name)
return pid, client
def test_restart(self):
self.waiter.start()
restartable_pids = []
nonrestartable_pids = []
clients = {}
# start 10 processes with RestartMode.ALWAYS
for _ in range(10):
pid, client = self._add_test_process(ProcessRestartMode.ALWAYS)
restartable_pids.append(pid)
clients[pid] = client
# and 10 processes with RestartMode.ABNORMAL
for _ in range(10):
pid, client = self._add_test_process(ProcessRestartMode.ABNORMAL)
restartable_pids.append(pid)
clients[pid] = client
# and 10 with RestartMode.NEVER
for _ in range(10):
pid, client = self._add_test_process(ProcessRestartMode.NEVER)
nonrestartable_pids.append(pid)
clients[pid] = client
all_pids = restartable_pids + nonrestartable_pids
self.waiter.await_many_state_events(all_pids, ProcessStateEnum.RUNNING)
for pid in all_pids:
client = clients[pid]
self.assertFalse(client.is_restart())
self.assertEqual(client.count(), 1)
# now kill the whole eeagent and restart it. processes should
# show up as FAILED in the next heartbeat.
resource_id = self._eea_pid_to_resource_id[self._initial_eea_pid]
persistence_dir = self._eea_pid_to_persistence_dir[self._initial_eea_pid]
log.debug("Restarting eeagent %s", self._initial_eea_pid)
self._kill_eeagent(self._initial_eea_pid)
# manually kill the processes to simulate a real container failure
for pid in all_pids:
self.container.terminate_process(pid)
self._start_eeagent(self.node1_id, resource_id=resource_id,
persistence_dir=persistence_dir)
# wait for restartables to restart
self.waiter.await_many_state_events(restartable_pids, ProcessStateEnum.RUNNING)
# query the processes again. it should have restart mode config
for pid in restartable_pids:
client = clients[pid]
self.assertTrue(client.is_restart())
self.assertEqual(client.count(), 1)
# meanwhile some procs should not have restarted
for pid in nonrestartable_pids:
proc = self.pd_cli.read_process(pid)
self.assertEqual(proc.process_state, ProcessStateEnum.FAILED)
# guard against extraneous events we were receiving as part of a bug:
# processes restarting again after they were already restarted
self.waiter.await_nothing(timeout=5)
def test_idempotency(self):
# ensure every operation can be safely retried
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.ALWAYS
proc_name = 'myreallygoodname'
pid = self.pd_cli.create_process(self.process_definition_id)
self.waiter.start(pid)
# note: if we import UNSCHEDULED state into ProcessStateEnum,
# this assertion will need to change.
proc = self.pd_cli.read_process(pid)
self.assertEqual(proc.process_id, pid)
self.assertEqual(proc.process_state, ProcessStateEnum.REQUESTED)
pid2 = self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, configuration={}, process_id=pid, name=proc_name)
self.assertEqual(pid, pid2)
self.waiter.await_state_event(pid, ProcessStateEnum.RUNNING)
# repeating schedule is harmless
pid2 = self.pd_cli.schedule_process(self.process_definition_id,
process_schedule, configuration={}, process_id=pid, name=proc_name)
self.assertEqual(pid, pid2)
proc = self.pd_cli.read_process(pid)
self.assertEqual(proc.process_id, pid)
self.assertEqual(proc.process_configuration, {})
self.assertEqual(proc.process_state, ProcessStateEnum.RUNNING)
self.pd_cli.cancel_process(pid)
self.waiter.await_state_event(pid, ProcessStateEnum.TERMINATED)
# repeating cancel is harmless
self.pd_cli.cancel_process(pid)
proc = self.pd_cli.read_process(pid)
self.assertEqual(proc.process_id, pid)
self.assertEqual(proc.process_configuration, {})
self.assertEqual(proc.process_state, ProcessStateEnum.TERMINATED)
class ProcessDispatcherEEAgentIntTest(ProcessDispatcherServiceIntTest):
"""Run the basic int tests again, with a different environment
"""
def setUp(self):
self.dashi = None
self._start_container()
from pyon.public import CFG
self.container_client = ContainerAgentClient(node=self.container.node,
name=self.container.name)
self.container = self.container_client._get_container_instance()
app = dict(name="process_dispatcher",
processapp=("process_dispatcher",
"ion.services.cei.process_dispatcher_service",
"ProcessDispatcherService"))
self.container.start_app(app, config=pd_config)
self.rr_cli = self.container.resource_registry
self.pd_cli = ProcessDispatcherServiceClient(node=self.container.node)
self.process_definition = ProcessDefinition(name='test_process')
self.process_definition.executable = {
'module': 'ion.services.cei.test.test_process_dispatcher',
'class': 'TestProcess'
}
self.process_definition_id = self.pd_cli.create_process_definition(
self.process_definition)
self._eea_pids = []
self._eea_pid_to_resource_id = {}
self._eea_pid_to_persistence_dir = {}
self._tmpdirs = []
self.dashi = get_dashi(
uuid.uuid4().hex,
pd_config['processdispatcher']['dashi_uri'],
pd_config['processdispatcher']['dashi_exchange'],
sysname=CFG.get_safe("dashi.sysname"))
#send a fake node_state message to PD's dashi binding.
self.node1_id = uuid.uuid4().hex
self._send_node_state("engine1", self.node1_id)
self._initial_eea_pid = self._start_eeagent(self.node1_id)
self.waiter = ProcessStateWaiter()
def _send_node_state(self, engine_id, node_id=None):
node_id = node_id or uuid.uuid4().hex
node_state = dict(node_id=node_id,
state=InstanceState.RUNNING,
domain_id=domain_id_from_engine(engine_id))
self.dashi.fire(get_pd_dashi_name(), "node_state", args=node_state)
def _start_eeagent(self, node_id, resource_id=None, persistence_dir=None):
if not persistence_dir:
persistence_dir = tempfile.mkdtemp()
self._tmpdirs.append(persistence_dir)
resource_id = resource_id or uuid.uuid4().hex
agent_config = _get_eeagent_config(node_id,
persistence_dir,
resource_id=resource_id)
pid = self.container_client.spawn_process(
name="eeagent",
module="ion.agents.cei.execution_engine_agent",
cls="ExecutionEngineAgent",
config=agent_config)
log.info('Agent pid=%s.', str(pid))
self._eea_pids.append(pid)
self._eea_pid_to_resource_id[pid] = resource_id
self._eea_pid_to_persistence_dir[pid] = persistence_dir
return pid
def _kill_eeagent(self, pid):
self.assertTrue(pid in self._eea_pids)
self.container.terminate_process(pid)
self._eea_pids.remove(pid)
del self._eea_pid_to_resource_id[pid]
del self._eea_pid_to_persistence_dir[pid]
def tearDown(self):
for pid in list(self._eea_pids):
self._kill_eeagent(pid)
for d in self._tmpdirs:
shutil.rmtree(d)
self.waiter.stop()
if self.dashi:
self.dashi.cancel()
def test_requested_ee(self):
# request non-default engine
process_target = ProcessTarget(execution_engine_id="engine2")
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.ALWAYS
process_schedule.target = process_target
pid = self.pd_cli.create_process(self.process_definition_id)
self.waiter.start()
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
process_id=pid)
self.waiter.await_state_event(pid, ProcessStateEnum.WAITING)
# request unknown engine, with NEVER queuing mode. The request
# should be rejected.
# verifies L4-CI-CEI-RQ52
process_target = ProcessTarget(execution_engine_id="not-a-real-ee")
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.NEVER
process_schedule.target = process_target
rejected_pid = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
process_id=rejected_pid)
self.waiter.await_state_event(rejected_pid, ProcessStateEnum.REJECTED)
# now add a node and eeagent for engine2. original process should leave
# queue and start running
node2_id = uuid.uuid4().hex
self._send_node_state("engine2", node2_id)
self._start_eeagent(node2_id)
self.waiter.await_state_event(pid, ProcessStateEnum.RUNNING)
# spawn another process. it should start immediately.
process_target = ProcessTarget(execution_engine_id="engine2")
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.NEVER
process_schedule.target = process_target
pid2 = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
process_id=pid2)
self.waiter.await_state_event(pid2, ProcessStateEnum.RUNNING)
# one more with node exclusive
process_target = ProcessTarget(execution_engine_id="engine2",
node_exclusive="hats")
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.NEVER
process_schedule.target = process_target
pid3 = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
process_id=pid3)
self.waiter.await_state_event(pid3, ProcessStateEnum.RUNNING)
# kill the processes for good
self.pd_cli.cancel_process(pid)
self.waiter.await_state_event(pid, ProcessStateEnum.TERMINATED)
self.pd_cli.cancel_process(pid2)
self.waiter.await_state_event(pid2, ProcessStateEnum.TERMINATED)
self.pd_cli.cancel_process(pid3)
self.waiter.await_state_event(pid3, ProcessStateEnum.TERMINATED)
def test_node_exclusive(self):
# the node_exclusive constraint is used to ensure multiple processes
# of the same "kind" each get a VM exclusive of each other. Other
# processes may run on these VMs, just not processes with the same
# node_exclusive tag. Since we cannot directly query the contents
# of each node in this test, we prove the capability by scheduling
# processes one by one and checking their state.
# verifies L4-CI-CEI-RQ121
# verifies L4-CI-CEI-RQ57
# first off, setUp() created a single node and eeagent.
# We schedule two processes with the same "abc" node_exclusive
# tag. Since there is only one node, the first process should run
# and the second should be queued.
process_target = ProcessTarget(execution_engine_id="engine1")
process_target.node_exclusive = "abc"
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.ALWAYS
process_schedule.target = process_target
pid1 = self.pd_cli.create_process(self.process_definition_id)
self.waiter.start()
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
process_id=pid1)
self.waiter.await_state_event(pid1, ProcessStateEnum.RUNNING)
pid2 = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
process_id=pid2)
self.waiter.await_state_event(pid2, ProcessStateEnum.WAITING)
# now demonstrate that the node itself is not full by launching
# a third process without a node_exclusive tag -- it should start
# immediately
process_target.node_exclusive = None
pid3 = self.pd_cli.create_process(self.process_definition_id)
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
process_id=pid3)
self.waiter.await_state_event(pid3, ProcessStateEnum.RUNNING)
# finally, add a second node to the engine. pid2 should be started
# since there is an exclusive "abc" node free.
node2_id = uuid.uuid4().hex
self._send_node_state("engine1", node2_id)
self._start_eeagent(node2_id)
self.waiter.await_state_event(pid2, ProcessStateEnum.RUNNING)
# kill the processes for good
self.pd_cli.cancel_process(pid1)
self.waiter.await_state_event(pid1, ProcessStateEnum.TERMINATED)
self.pd_cli.cancel_process(pid2)
self.waiter.await_state_event(pid2, ProcessStateEnum.TERMINATED)
self.pd_cli.cancel_process(pid3)
self.waiter.await_state_event(pid3, ProcessStateEnum.TERMINATED)
def test_code_download(self):
# create a process definition that has no URL; only module and class.
process_definition_no_url = ProcessDefinition(
name='test_process_nodownload')
process_definition_no_url.executable = {
'module': 'ion.my.test.process',
'class': 'TestProcess'
}
process_definition_id_no_url = self.pd_cli.create_process_definition(
process_definition_no_url)
# create another that has a URL of the python file (this very file)
# verifies L4-CI-CEI-RQ114
url = "file://%s" % os.path.join(os.path.dirname(__file__),
'test_process_dispatcher.py')
process_definition = ProcessDefinition(name='test_process_download')
process_definition.executable = {
'module': 'ion.my.test.process',
'class': 'TestProcess',
'url': url
}
process_definition_id = self.pd_cli.create_process_definition(
process_definition)
process_target = ProcessTarget()
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.ALWAYS
process_schedule.target = process_target
self.waiter.start()
# Test a module with no download fails
pid_no_url = self.pd_cli.create_process(process_definition_id_no_url)
self.pd_cli.schedule_process(process_definition_id_no_url,
process_schedule,
process_id=pid_no_url)
self.waiter.await_state_event(pid_no_url, ProcessStateEnum.FAILED)
# Test a module with a URL runs
pid = self.pd_cli.create_process(process_definition_id)
self.pd_cli.schedule_process(process_definition_id,
process_schedule,
process_id=pid)
self.waiter.await_state_event(pid, ProcessStateEnum.RUNNING)
def _add_test_process(self, restart_mode=None):
process_schedule = ProcessSchedule()
if restart_mode is not None:
process_schedule.restart_mode = restart_mode
pid = self.pd_cli.create_process(self.process_definition_id)
pid_listen_name = "PDtestproc_%s" % uuid.uuid4().hex
config = {'process': {'listen_name': pid_listen_name}}
self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
process_id=pid,
configuration=config)
client = TestClient(to_name=pid_listen_name)
return pid, client
def test_restart(self):
self.waiter.start()
restartable_pids = []
nonrestartable_pids = []
clients = {}
# start 10 processes with RestartMode.ALWAYS
for _ in range(10):
pid, client = self._add_test_process(ProcessRestartMode.ALWAYS)
restartable_pids.append(pid)
clients[pid] = client
# and 10 processes with RestartMode.ABNORMAL
for _ in range(10):
pid, client = self._add_test_process(ProcessRestartMode.ABNORMAL)
restartable_pids.append(pid)
clients[pid] = client
# and 10 with RestartMode.NEVER
for _ in range(10):
pid, client = self._add_test_process(ProcessRestartMode.NEVER)
nonrestartable_pids.append(pid)
clients[pid] = client
all_pids = restartable_pids + nonrestartable_pids
self.waiter.await_many_state_events(all_pids, ProcessStateEnum.RUNNING)
for pid in all_pids:
client = clients[pid]
self.assertFalse(client.is_restart())
self.assertEqual(client.count(), 1)
# now kill the whole eeagent and restart it. processes should
# show up as FAILED in the next heartbeat.
resource_id = self._eea_pid_to_resource_id[self._initial_eea_pid]
persistence_dir = self._eea_pid_to_persistence_dir[
self._initial_eea_pid]
log.debug("Restarting eeagent %s", self._initial_eea_pid)
self._kill_eeagent(self._initial_eea_pid)
# manually kill the processes to simulate a real container failure
for pid in all_pids:
self.container.terminate_process(pid)
self._start_eeagent(self.node1_id,
resource_id=resource_id,
persistence_dir=persistence_dir)
# wait for restartables to restart
self.waiter.await_many_state_events(restartable_pids,
ProcessStateEnum.RUNNING)
# query the processes again. it should have restart mode config
for pid in restartable_pids:
client = clients[pid]
self.assertTrue(client.is_restart())
self.assertEqual(client.count(), 1)
# meanwhile some procs should not have restarted
for pid in nonrestartable_pids:
proc = self.pd_cli.read_process(pid)
self.assertEqual(proc.process_state, ProcessStateEnum.FAILED)
# guard against extraneous events we were receiving as part of a bug:
# processes restarting again after they were already restarted
self.waiter.await_nothing(timeout=5)
def test_idempotency(self):
# ensure every operation can be safely retried
process_schedule = ProcessSchedule()
process_schedule.queueing_mode = ProcessQueueingMode.ALWAYS
proc_name = 'myreallygoodname'
pid = self.pd_cli.create_process(self.process_definition_id)
self.waiter.start(pid)
# note: if we import UNSCHEDULED state into ProcessStateEnum,
# this assertion will need to change.
proc = self.pd_cli.read_process(pid)
self.assertEqual(proc.process_id, pid)
self.assertEqual(proc.process_state, ProcessStateEnum.REQUESTED)
pid2 = self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
configuration={},
process_id=pid,
name=proc_name)
self.assertEqual(pid, pid2)
self.waiter.await_state_event(pid, ProcessStateEnum.RUNNING)
# repeating schedule is harmless
pid2 = self.pd_cli.schedule_process(self.process_definition_id,
process_schedule,
configuration={},
process_id=pid,
name=proc_name)
self.assertEqual(pid, pid2)
proc = self.pd_cli.read_process(pid)
self.assertEqual(proc.process_id, pid)
self.assertEqual(proc.process_configuration, {})
self.assertEqual(proc.process_state, ProcessStateEnum.RUNNING)
self.pd_cli.cancel_process(pid)
self.waiter.await_state_event(pid, ProcessStateEnum.TERMINATED)
# repeating cancel is harmless
self.pd_cli.cancel_process(pid)
proc = self.pd_cli.read_process(pid)
self.assertEqual(proc.process_id, pid)
self.assertEqual(proc.process_configuration, {})
self.assertEqual(proc.process_state, ProcessStateEnum.TERMINATED)
