def test_average_schedule(test_db):
# 4个16核, 不限制共享数
group, pod = _create_data(10, -1, 4)
assert len(average_schedule(group, pod, ncontainer=100, ncore=2)) == 0
assert len(average_schedule(group, pod, ncontainer=130, ncore=0, nshare=5)) == 0
r = average_schedule(group, pod, ncontainer=10, ncore=1)
assert len(r) == 4
assert sum(i[1] for i in r.keys()) == 10
for (host, count), cores in r.iteritems():
assert count in (2, 3)
if count == 2:
assert len(cores['full']) == 2
assert len(cores['part']) == 0
if count == 3:
assert len(cores['full']) == 3
assert len(cores['part']) == 0
r = average_schedule(group, pod, ncontainer=9, ncore=2)
assert len(r) == 4
assert sum(i[1] for i in r.keys()) == 9
for (host, count), cores in r.iteritems():
assert count in (2, 3)
if count == 2:
assert len(cores['full']) == 4
assert len(cores['part']) == 0
if count == 3:
assert len(cores['full']) == 6
assert len(cores['part']) == 0
r = average_schedule(group, pod, ncontainer=40, ncore=1, nshare=5)
assert len(r) == 4
assert sum(i[1] for i in r.keys()) == 40
for (host, count), cores in r.iteritems():
assert count == 10
assert len(cores['full']) == 10
assert len(cores['part']) == 10
assert len(set(cores['part'])) == 5
r = average_schedule(group, pod, ncontainer=22, ncore=2, nshare=3)
assert len(r) == 4
assert sum(i[1] for i in r.keys()) == 22
for (host, count), cores in r.iteritems():
assert count in (5, 6)
if count == 5:
assert len(cores['full']) == 10
assert len(cores['part']) == 5
assert len(set(cores['part'])) == 2
if count == 6:
assert len(cores['full']) == 12
assert len(cores['part']) == 6
assert len(set(cores['part'])) == 2
def test_average_schedule(test_db):
# 4个16核, 不限制共享数
pod = _create_data(10, -1, 4)
assert len(average_schedule(pod, ncontainer=100, ncore=2)) == 0
assert len(average_schedule(pod, ncontainer=130, ncore=0, nshare=5)) == 0
r = average_schedule(pod, ncontainer=10, ncore=1)
assert len(r) == 4
assert sum(i[1] for i in r.keys()) == 10
for (host, count), cores in r.iteritems():
assert count in (2, 3)
if count == 2:
assert len(cores['full']) == 2
assert len(cores['part']) == 0
if count == 3:
assert len(cores['full']) == 3
assert len(cores['part']) == 0
r = average_schedule(pod, ncontainer=9, ncore=2)
assert len(r) == 4
assert sum(i[1] for i in r.keys()) == 9
for (host, count), cores in r.iteritems():
assert count in (2, 3)
if count == 2:
assert len(cores['full']) == 4
assert len(cores['part']) == 0
if count == 3:
assert len(cores['full']) == 6
assert len(cores['part']) == 0
r = average_schedule(pod, ncontainer=40, ncore=1, nshare=5)
assert len(r) == 4
assert sum(i[1] for i in r.keys()) == 40
for (host, count), cores in r.iteritems():
assert count == 10
assert len(cores['full']) == 10
assert len(cores['part']) == 10
assert len(set(cores['part'])) == 5
r = average_schedule(pod, ncontainer=22, ncore=2, nshare=3)
assert len(r) == 4
assert sum(i[1] for i in r.keys()) == 22
for (host, count), cores in r.iteritems():
assert count in (5, 6)
if count == 5:
assert len(cores['full']) == 10
assert len(cores['part']) == 5
assert len(set(cores['part'])) == 2
if count == 6:
assert len(cores['full']) == 12
assert len(cores['part']) == 6
assert len(set(cores['part'])) == 2
def migrate_container(container_id, need_to_remove=True):
container = Container.get_by_container_id(container_id)
if not container:
_log.error('container %s is not found, ignore migration', container_id)
return
ncore, nshare = container.host.pod.get_core_allocation(container.ncore)
host_cores = average_schedule(container.host.pod, 1, ncore, nshare, None)
if not host_cores:
_log.error('not enough cores to migrate')
return
cids = [container.id]
spec_ips = cidrs = container.get_ips()
(host, container_count), cores = next(host_cores.iteritems())
props = {
'ncontainer': 1,
'entrypoint': container.entrypoint,
'env': container.env,
'full_cores': [c.label for c in cores.get('full', [])],
'part_cores': [c.label for c in cores.get('part', [])],
'ports': None,
'args': None,
'nshare': nshare,
'networks': cidrs,
'image': None,
'route': '',
'callback_url': container.callback_url,
'container_ids': cids,
}
task = Task.create(consts.TASK_MIGRATE, container.version, host, props)
if not task:
_log.error('create migrate task error')
return
_log.info('start migration...')
if need_to_remove:
remove_containers.apply(args=(task.id, cids, False),
task_id='task:%s' % task.id)
create_containers.apply(args=(task.id, 1, nshare, cores, cidrs, spec_ips),
task_id='task:%s' % task.id)
_log.info('migration done')
def migrate_container(container_id, need_to_remove=True):
container = Container.get_by_container_id(container_id)
if not container:
_log.error('container %s is not found, ignore migration', container_id)
return
ncore, nshare= container.host.pod.get_core_allocation(container.ncore)
host_cores = average_schedule(container.host.pod, 1, ncore, nshare, None)
if not host_cores:
_log.error('not enough cores to migrate')
return
cids = [container.id]
spec_ips = cidrs = container.get_ips()
(host, container_count), cores = next(host_cores.iteritems())
props = {
'ncontainer': 1,
'entrypoint': container.entrypoint,
'env': container.env,
'full_cores': [c.label for c in cores.get('full', [])],
'part_cores': [c.label for c in cores.get('part', [])],
'ports': None,
'args': None,
'nshare': nshare,
'networks': cidrs,
'image': None,
'route': '',
'callback_url': container.callback_url,
'container_ids': cids,
}
task = Task.create(consts.TASK_MIGRATE, container.version, host, props)
if not task:
_log.error('create migrate task error')
return
_log.info('start migration...')
if need_to_remove:
remove_containers.apply(args=(task.id, cids, False), task_id='task:%s' % task.id)
create_containers.apply(args=(task.id, 1, nshare, cores, cidrs, spec_ips), task_id='task:%s' % task.id)
_log.info('migration done')
def create_private(group_name, pod_name, appname):
"""ncore: 需要的核心数, 可以是小数, 例如1.5个"""
data = request.get_json()
vstr = data['version']
group, pod, application, version = validate_instance(group_name,
pod_name, appname, vstr)
# TODO check if group has this pod
core_require = int(float(data['ncore']) * pod.core_share) # 是说一个容器要几个核...
ncore = core_require / pod.core_share
nshare = core_require % pod.core_share
ncontainer = int(data['ncontainer'])
networks = Network.get_multi(data.get('networks', []))
spec_ips = data.get('spec_ips', [])
appconfig = version.appconfig
strategy = data.get('strategy', 'average')
# 指定的host, 如果没有则按照编排分配host
hostname = data.get('hostname', '')
host = hostname and Host.get_by_name(hostname) or None
if host and not (host.group_id == group.id and host.pod_id == pod.id):
current_app.logger.error('Host must belong to pod/group (hostname=%s, pod=%s, group=%s)',
host, pod_name, group_name)
raise EruAbortException(consts.HTTP_BAD_REQUEST, 'Host must belong to this pod and group')
if not data['entrypoint'] in appconfig.entrypoints:
current_app.logger.error('Entrypoint not in app.yaml (entry=%s, name=%s, version=%s)',
data['entrypoint'], appname, version.short_sha)
raise EruAbortException(consts.HTTP_BAD_REQUEST, 'Entrypoint %s not in app.yaml' % data['entrypoint'])
ts, keys = [], []
with rds.lock('%s:%s' % (group_name, pod_name)):
if strategy == 'average':
host_cores = average_schedule(group, pod, ncontainer, ncore, nshare, spec_host=host)
elif strategy == 'centralized':
host_cores = centralized_schedule(group, pod, ncontainer, ncore, nshare, spec_host=host)
else:
raise EruAbortException(consts.HTTP_BAD_REQUEST, 'strategy %s not supported' % strategy)
if not host_cores:
current_app.logger.error('Not enough cores (name=%s, version=%s, ncore=%s)',
appname, version.short_sha, data['ncore'])
raise EruAbortException(consts.HTTP_BAD_REQUEST, 'Not enough core resources')
for (host, container_count), cores in host_cores.iteritems():
t = _create_task(
version,
host,
container_count,
cores,
nshare,
networks,
spec_ips,
data['entrypoint'],
data['env'],
image=data.get('image', ''),
)
if not t:
continue
host.occupy_cores(cores, nshare)
ts.append(t.id)
keys.append(t.result_key)
return {'r': 0, 'msg': 'ok', 'tasks': ts, 'watch_keys': keys}
def test_scheduler(test_db):
# 10000个16核, 不限制共享数
pod = _create_data(10, -1, 10000)
def test_max_container_count(ncore, nshare, expected):
start = time.time()
assert get_max_container_count(pod, ncore, nshare) == expected
delta = time.time() - start
_log.debug('test_max_container_count with ncore={}, nshare={}, expected={} takes {}'.format(ncore, nshare, expected, delta))
test_max_container_count_cases = (
(1, 0, 160000),
(2, 0, 80000),
(3, 0, 50000),
(4, 0, 40000),
(5, 0, 30000),
(1, 5, 100000),
(2, 5, 60000),
(3, 5, 40000),
(1, 1, 140000),
(2, 1, 70000),
)
for case in test_max_container_count_cases:
test_max_container_count(*case)
start = time.time()
assert len(average_schedule(pod, ncontainer=100, ncore=2)) == 100
print time.time() - start
start = time.time()
assert len(average_schedule(pod, ncontainer=130, ncore=0, nshare=5)) == 130
print time.time() - start
start = time.time()
r = average_schedule(pod, ncontainer=10, ncore=1)
print time.time() - start
assert len(r) == 10
assert sum(i[1] for i in r.keys()) == 10
for (host, count), cores in r.iteritems():
assert len(cores['full']) == 1
assert len(cores['part']) == 0
start = time.time()
r = average_schedule(pod, ncontainer=10000, ncore=2)
print time.time() - start
assert len(r) == 10000
assert sum(i[1] for i in r.keys()) == 10000
for (host, count), cores in r.iteritems():
assert count == 1
assert len(cores['full']) == 2
assert len(cores['part']) == 0
start = time.time()
r = average_schedule(pod, ncontainer=10000, ncore=1, nshare=5)
print time.time() - start
assert len(r) == 10000
assert sum(i[1] for i in r.keys()) == 10000
for (host, count), cores in r.iteritems():
assert count == 1
assert len(cores['full']) == 1
assert len(cores['part']) == 1
start = time.time()
r = average_schedule(pod, ncontainer=10000, ncore=2, nshare=3)
print time.time() - start
assert len(r) == 10000
assert sum(i[1] for i in r.keys()) == 10000
for (host, count), cores in r.iteritems():
assert count == 1
assert len(cores['full']) == 2
assert len(cores['part']) == 1
start = time.time()
assert len(centralized_schedule(pod, ncontainer=100, ncore=2)) == 13
print time.time() - start
start = time.time()
assert len(centralized_schedule(pod, ncontainer=130, ncore=0, nshare=5)) == 5
print time.time() - start
start = time.time()
r = centralized_schedule(pod, ncontainer=10000, ncore=1)
print time.time() - start
assert len(r) == 625
assert sum(i[1] for i in r.keys()) == 10000
for (host, count), cores in r.iteritems():
assert count == 16
assert len(cores['full']) == 16
assert len(cores['part']) == 0
start = time.time()
r = centralized_schedule(pod, ncontainer=100, ncore=2)
print time.time() - start
assert len(r) == 13
assert sum(i[1] for i in r.keys()) == 100
for (host, count), cores in r.iteritems():
assert count in (4, 8)
if count == 4:
assert len(cores['full']) == 8
assert len(cores['part']) == 0
if count == 8:
assert len(cores['full']) == 16
assert len(cores['part']) == 0
start = time.time()
r = centralized_schedule(pod, ncontainer=30, ncore=1, nshare=5)
print time.time() - start
assert len(r) == 3
assert sum(i[1] for i in r.keys()) == 30
for (host, count), cores in r.iteritems():
assert count == 10
assert len(cores['full']) == 10
assert len(cores['part']) == 10
assert len(set(cores['part'])) == 5
start = time.time()
r = centralized_schedule(pod, ncontainer=20, ncore=2, nshare=3)
print time.time() - start
assert len(r) == 4
assert sum(i[1] for i in r.keys()) == 20
for (host, count), cores in r.iteritems():
assert count in (2, 6)
if count == 2:
assert len(cores['full']) == 4
assert len(cores['part']) == 2
assert len(set(cores['part'])) == 1
if count == 6:
assert len(cores['full']) == 12
assert len(cores['part']) == 6
assert len(set(cores['part'])) == 2