def domain_n_preserve_remove_node_test(self):
site = "site1"
fake_site, lc = self.make_fake_libcloud_site(site)
dt_name = self._load_dtrs(site, fake_site)
n = 3
dt = _make_domain_def(n, dt_name, site)
def_id = str(uuid.uuid4())
self.epum_client.add_domain_definition(def_id, example_definition)
domain_id = str(uuid.uuid4())
log.debug("Launching domain %s", domain_id)
self.epum_client.add_domain(domain_id, def_id, dt, caller=self.user)
self._wait_states(n, lc, states=[NodeState.RUNNING])
# wait a little while to make sure query thread detects all VMs
time.sleep(10)
nodes = get_valid_nodes(lc)
lc.destroy_node(random.choice(nodes))
wait(lambda: len(get_valid_nodes(lc)) == n, timeout=60)
self._wait_remove_domain(domain_id)
self._wait_for_all_terminated(lc)
def test_initialized_system_boot_with_procs(self):
# tests the case where just the doctor dies in the middle of system boot
# but after a doctor has already declared the system initialized. In this
# case we have processes in the UNSCHEDULED_PENDING state that should be
# rescheduled once system boot ends.
self.store.set_system_boot(True)
self.store.set_initialized()
self.store.set_pd_state(ProcessDispatcherState.SYSTEM_BOOTING)
p0 = ProcessRecord.new(None, "proc0", {}, ProcessState.UNSCHEDULED_PENDING)
self.store.add_process(p0)
p1 = ProcessRecord.new(None, "proc1", {}, ProcessState.UNSCHEDULED_PENDING)
self.store.add_process(p1)
p2 = ProcessRecord.new(None, "proc2", {}, ProcessState.UNSCHEDULED_PENDING)
self.store.add_process(p2)
restartable_procs = ["proc0", "proc1", "proc2"]
self._run_in_thread()
# now end system boot
self.store.set_system_boot(False)
wait(lambda: self.store.get_pd_state() == ProcessDispatcherState.OK)
# check that pending processes were correctly rescheduled
self.assertEqual(len(self.store.get_queued_processes()), len(restartable_procs))
for proc in restartable_procs:
self.assertEqual(self.store.get_process(None, proc).state,
ProcessState.REQUESTED)
def domain_n_preserve_alter_state_test(self):
site = uuid.uuid4().hex
fake_site, lc = self.make_fake_libcloud_site(site)
dt_name = self._load_dtrs(site, fake_site)
n = 3
dt = _make_domain_def(n, dt_name, site)
def_id = str(uuid.uuid4())
self.epum_client.add_domain_definition(def_id, example_definition)
domain_id = str(uuid.uuid4())
log.debug("Launching domain %s", domain_id)
self.epum_client.add_domain(domain_id, def_id, dt, caller=self.user)
wait(lambda: len(get_valid_nodes(lc)) == n, timeout=60)
nodes = get_valid_nodes(lc)
lc.set_node_state(nodes[0], NodeState.TERMINATED)
self._wait_states(n, lc, states=[NodeState.RUNNING, NodeState.PENDING])
print "terminating"
self._wait_remove_domain(domain_id)
# check the node list
nodes = lc.list_nodes(immediate=True)
for nd in nodes:
# verify that any node that is still around is terminated
self.assertEqual(nd.state, NodeState.TERMINATED)
def test_uninitialized_system_boot_without_state(self):
self.store.set_system_boot(True)
self._run_in_thread()
assert self.store.wait_initialized(timeout=10)
self.assertEqual(self.store.get_pd_state(),
ProcessDispatcherState.SYSTEM_BOOTING)
self.store.set_system_boot(False)
wait(lambda: self.store.get_pd_state() == ProcessDispatcherState.OK)
def _wait_for_domain(self, domain_id):
def waiter():
try:
domain = self.epum_client.describe_domain(domain_id)
return domain is not None
except NotFoundError:
return False
wait(waiter, timeout=30)
def _wait_for_all_terminated(self, lc):
def wait_terminated():
nodes = lc.list_nodes(immediate=True)
return all(node.state == NodeState.TERMINATED for node in nodes)
try:
wait(wait_terminated, timeout=60)
except wait.TimeOutWaitingFor:
nodes = [n.id for n in lc.list_nodes(immediate=True) if n.state != NodeState.TERMINATED]
self.fail("Timed out waiting for all nodes to be terminated. Remaining: %s" % nodes)
def _wait_remove_many_domains(self, domains, delay=0.1):
domains = set(domains)
for domain_id in domains:
self.epum_client.remove_domain(domain_id)
# wait for intersection between sets to be empty (no domains left)
try:
wait(lambda: len(domains & set(self.epum_client.list_domains())) == 0, timeout=60, wait=delay)
except wait.TimeOutWaitingFor:
remaining = domains & set(self.epum_client.list_domains())
self.fail("Timed out waiting for domains to exit. domains: %s" % list(remaining))
def _get_contender(self, path, ndx=0):
"""returns name, hostname, pid tuple"""
assert ndx < self.epum_replica_count
contenders = []
election = self.kazoo.Election(path)
def getem():
contenders[:] = election.contenders()
return len(contenders) == self.epum_replica_count
# retry getting contenders. may take them a while to emerge
wait(getem, timeout=20)
name, hostname, pid = contenders[ndx].split(':')
return name, hostname, int(pid)
def test_monitor_thread(self):
self._run_in_thread()
assert self.store.wait_initialized(timeout=10)
self.assertEqual(self.store.get_pd_state(),
ProcessDispatcherState.OK)
self.assertIsNotNone(self.doctor.monitor)
monitor_thread = self.doctor.monitor_thread
self.assertIsNotNone(monitor_thread)
self.assertTrue(monitor_thread.is_alive())
# now cancel doctor. monitor should stop too
self.doctor.cancel()
wait(lambda: not monitor_thread.is_alive())
def test_initialized_system_boot_without_procs(self):
# tests the case where just the doctor dies in the middle of system boot
# but after a doctor has already declared the system initialized. In this
# case we have no processes to schedule on system boot completion.
self.store.set_system_boot(True)
self.store.set_initialized()
self.store.set_pd_state(ProcessDispatcherState.SYSTEM_BOOTING)
self._run_in_thread()
# now end system boot
self.store.set_system_boot(False)
wait(lambda: self.store.get_pd_state() == ProcessDispatcherState.OK)
def domain_add_check_n_remove_test(self):
site = uuid.uuid4().hex
fake_site, lc = self.make_fake_libcloud_site(site)
dt_name = self._load_dtrs(site, fake_site)
n = 3
dt = _make_domain_def(n, dt_name, site)
def_id = str(uuid.uuid4())
self.epum_client.add_domain_definition(def_id, example_definition)
domain_id = str(uuid.uuid4())
log.debug("Launching domain %s", domain_id)
self.epum_client.add_domain(domain_id, def_id, dt, caller=self.user)
wait(lambda: len(get_valid_nodes(lc)) == n, timeout=60)
self._wait_remove_domain(domain_id)
self._wait_for_all_terminated(lc)
def _wait_states(self, n, lc, states=None, timeout=20, delay=0.1):
if states is None:
states = [NodeState.RUNNING, NodeState.PENDING]
print "Waiting for %d nodes in states: %s" % (n, states)
def wait_running_count():
nodes = lc.list_nodes(immediate=True)
running_count = 0
found_states = defaultdict(int)
for nd in nodes:
found_states[nd.state] += 1
if nd.state in states:
running_count += 1
print "Found %d nodes in states: %s" % (len(nodes),
" ".join("%s:%s" % pair for pair in found_states.iteritems()))
return running_count == n
wait(wait_running_count, timeout=timeout, wait=delay)
def _add_many_domains_terminate_all(self, kill_func=None, places_to_kill=None, n=1, nodes_per_domain=3):
test_pc = 1
def_name = str(uuid.uuid4())
self.epum_client.add_domain_definition(def_name, example_definition)
domain = _example_domain(nodes_per_domain)
domains_started = []
for i in range(n):
name = "dom%d" % (i)
self.epum_client.add_domain(name, def_name, domain)
domains_started.append(name)
test_pc = self._kill_cb(test_pc, places_to_kill, kill_func)
domains = self.epum_client.list_domains()
domains_started.sort()
domains.sort()
self.assertEqual(domains, domains_started)
self.wait_for_libcloud_nodes(n * nodes_per_domain)
test_pc = self._kill_cb(test_pc, places_to_kill, kill_func)
self.wait_for_all_domains()
state = self.provisioner_client.terminate_all()
self.assertFalse(state) # cannot all be terminated this quickly
test_pc = self._kill_cb(test_pc, places_to_kill, kill_func)
# wait a little while until hopefully termination is underway
time.sleep(2)
test_pc = self._kill_cb(test_pc, places_to_kill, kill_func)
# this will return true when everything is terminated
wait(self.provisioner_client.terminate_all, timeout=20)
self.assertFalse(self.get_valid_libcloud_nodes())
for name in domains_started:
self.epum_client.remove_domain(name)
self.wait_for_domain_set([])
def wait_for_domain_set(self, expected, timeout=30):
expected = set(expected)
wait(lambda: set(self.epum_client.list_domains()) == expected,
timeout=timeout)
def test_uninitialized_system_boot_with_state(self):
self.store.set_system_boot(True)
self.core.node_state("node1", domain_id_from_engine("engine1"),
InstanceState.RUNNING)
resource_id = "eeagent_1"
self.core.ee_heartbeat(resource_id, make_beat("node1"))
p0 = ProcessRecord.new(None, "proc0", {}, ProcessState.RUNNING,
configuration=nosystemrestart_process_config(),
assigned=resource_id,
restart_mode=RestartMode.ALWAYS)
self.store.add_process(p0)
p1 = ProcessRecord.new(None, "proc1", {}, ProcessState.RUNNING,
assigned=resource_id)
self.store.add_process(p1)
p2 = ProcessRecord.new(None, "proc2", {}, ProcessState.PENDING,
assigned=resource_id)
self.store.add_process(p2)
p3 = ProcessRecord.new(None, "proc3", {}, ProcessState.TERMINATING,
assigned=resource_id)
self.store.add_process(p3)
# this one shouldn't restart
p4 = ProcessRecord.new(None, "proc4", {}, ProcessState.RUNNING,
configuration=nosystemrestart_process_config(),
assigned=resource_id,
restart_mode=RestartMode.ABNORMAL)
self.store.add_process(p4)
# non-running proceses should also potentially be restarted on boot
p5 = ProcessRecord.new(None, "proc5", {}, ProcessState.WAITING)
self.store.add_process(p5)
self.store.enqueue_process(*p5.key)
p6 = ProcessRecord.new(None, "proc6", {}, ProcessState.REQUESTED)
self.store.add_process(p6)
# not this one, due to RestartMode
p7 = ProcessRecord.new(None, "proc7", {}, ProcessState.REQUESTED,
configuration=nosystemrestart_process_config(),
restart_mode=RestartMode.ALWAYS)
self.store.add_process(p7)
self.store.enqueue_process(*p7.key)
resource = self.store.get_resource(resource_id)
resource.assigned = [p0.key, p1.key, p2.key, p3.key, p4.key]
self.store.update_resource(resource)
restartable_procs = ["proc1", "proc2", "proc5", "proc6"]
dead_procs = ["proc0", "proc4", "proc7"]
self._run_in_thread()
assert self.store.wait_initialized(timeout=10)
self.assertEqual(len(self.store.get_queued_processes()), 0)
self.assertEqual(len(self.store.get_node_ids()), 0)
self.assertEqual(len(self.store.get_resource_ids()), 0)
for proc in restartable_procs:
self.assertEqual(self.store.get_process(None, proc).state,
ProcessState.UNSCHEDULED_PENDING)
for proc in dead_procs:
self.assertEqual(self.store.get_process(None, proc).state,
ProcessState.TERMINATED)
self.assertEqual(self.store.get_process(None, "proc3").state,
ProcessState.TERMINATED)
self.assertEqual(self.store.get_pd_state(),
ProcessDispatcherState.SYSTEM_BOOTING)
# now end system boot
self.store.set_system_boot(False)
wait(lambda: self.store.get_pd_state() == ProcessDispatcherState.OK)
# check that pending processes were correctly rescheduled
self.assertEqual(len(self.store.get_queued_processes()), len(restartable_procs))
for proc in restartable_procs:
self.assertEqual(self.store.get_process(None, proc).state,
ProcessState.REQUESTED)
def domain_sensor_engine_test(self):
site = uuid.uuid4().hex
fake_site, lc = self.make_fake_libcloud_site(site)
dt_name = self._load_dtrs(site, fake_site)
minimum_n = 1
maximum_n = 3
scale_up_threshold = 2.0
scale_up_n_vms = 1
scale_down_threshold = 0.5
scale_down_n_vms = 1
scale_down_sensor_data = [0, 0, 0]
scale_up_sensor_data = [3, 3, 5]
metric = 'load'
sample_function = 'Average'
dt = _make_sensor_domain_def(metric, sample_function, minimum_n,
maximum_n, scale_up_threshold,
scale_up_n_vms, scale_down_threshold, scale_down_n_vms,
scale_down_sensor_data, dt_name, site)
def_id = str(uuid.uuid4())
self.epum_client.add_domain_definition(def_id, sensor_definition)
domain_id = str(uuid.uuid4())
log.debug("Launching domain %s", domain_id)
self.epum_client.add_domain(domain_id, def_id, dt, caller=self.user)
# make sure we hit the minimum number of nodes
wait(lambda: len(get_valid_nodes(lc)) >= minimum_n, timeout=60)
# Now get it to scale up
print "reconfiguring with sensor data: %s" % scale_up_sensor_data
new_config = {'engine_conf': {'sensor_data': scale_up_sensor_data}}
self.epum_client.reconfigure_domain(domain_id, new_config, caller=self.user)
# make sure we hit the maximum number of nodes
wait(lambda: len(get_valid_nodes(lc)) == maximum_n, timeout=60)
# Now get it to scale down
print "reconfiguring with sensor data: %s" % scale_down_sensor_data
new_config = {'engine_conf': {'sensor_data': scale_down_sensor_data}}
self.epum_client.reconfigure_domain(domain_id, new_config, caller=self.user)
wait(lambda: len(get_valid_nodes(lc)) == minimum_n, timeout=60)
# Now test the cooldown
new_config = {
'engine_conf': {
'sensor_data': scale_up_sensor_data,
'cooldown_period': 100,
}
}
self.epum_client.reconfigure_domain(domain_id, new_config, caller=self.user)
# Wait 10s for a few decides to happen:
time.sleep(10)
# And ensure we're still a minimum scaling
nodes = get_valid_nodes(lc)
self.assertEqual(len(nodes), minimum_n)
# Now set cooldown to 10s (which have already passed)
new_config = {'engine_conf': {'cooldown_period': 10}}
self.epum_client.reconfigure_domain(domain_id, new_config, caller=self.user)
# And watch it scale up
wait(lambda: len(get_valid_nodes(lc)) == maximum_n, timeout=60)
self._wait_remove_domain(domain_id)
self._wait_for_all_terminated(lc)
def wait_for_libcloud_nodes(self, count, timeout=60):
wait(lambda: len(self.get_valid_nodes()) == count,
timeout=timeout)
return self.get_valid_nodes()
def wait_for_all_domains(self, timeout=30):
wait(self.verify_all_domain_instances, timeout=timeout)
