def test_container_release_cores(test_db):
a = App.get_or_create('app', 'http://git.hunantv.com/group/app.git')
v = a.add_version(random_sha1())
p = Pod.create('pod', 'pod', 10, -1)
host = Host.create(p, random_ipv4(), random_string(), random_uuid(), 200, 0)
for core in host.cores:
assert core.host_id == host.id
assert core.remain == 10
containers = []
cores = sorted(host.cores, key=operator.attrgetter('label'))
for fcores, pcores in zip(chunked(cores[:100], 10), chunked(cores[100:], 10)):
used_cores = {'full': fcores, 'part': pcores}
host.occupy_cores(used_cores, 5)
c = Container.create(random_sha1(), host, v, random_string(), 'entrypoint', used_cores, 'env', nshare=5)
containers.append(c)
cores = sorted(host.cores, key=operator.attrgetter('label'))
for fcores, pcores in zip(chunked(cores[:100], 10), chunked(cores[100:], 10)):
for core in fcores:
assert core.remain == 0
for core in pcores:
assert core.remain == 5
for c in containers:
c.delete()
cores = sorted(host.cores, key=operator.attrgetter('label'))
for fcores, pcores in zip(chunked(cores[:100], 10), chunked(cores[100:], 10)):
for core in fcores:
assert core.remain == 10
for core in pcores:
assert core.remain == 10
def test_container_transform(test_db):
a = App.get_or_create('app', 'http://git.hunantv.com/group/app.git', '')
assert a is not None
v = a.add_version(random_sha1())
v2 = a.add_version(random_sha1())
assert v is not None
assert v.app.id == a.id
assert v.name == a.name
assert len(v.containers.all()) == 0
assert len(v.tasks.all()) == 0
g = Group.create('group', 'group')
p = Pod.create('pod', 'pod')
assert p.assigned_to_group(g)
hosts = [Host.create(p, random_ipv4(), random_string(prefix='host'),
random_uuid(), 4, 4096) for i in range(6)]
for host in hosts[:3]:
host.assigned_to_group(g)
assert g.get_max_containers(p, 3, 0) == 3
host_cores = g.get_free_cores(p, 3, 3, 0)
assert len(host_cores) == 3
containers = []
for (host, count), cores in host_cores.iteritems():
cores_per_container = len(cores) / count
for i in range(count):
cid = random_sha1()
used_cores = {'full': cores['full'][i*cores_per_container:(i+1)*cores_per_container]}
c = Container.create(cid, host, v, random_string(), 'entrypoint', used_cores, 'env')
assert c is not None
containers.append(c)
host.occupy_cores(cores, 0)
for host in g.private_hosts.all():
assert len(host.get_free_cores()[0]) == 1
assert len(host.containers.all()) == 1
assert host.count == 1
assert len(containers) == 3
assert len(v.containers.all()) == 3
cids = [c.container_id for c in containers]
for c in containers:
host = c.host
cid = c.container_id
c.transform(v2, random_sha1(), random_string())
assert c.container_id != cid
new_cids = [c.container_id for c in containers]
assert new_cids != cids
def test_container_release_cores(test_db):
a = App.get_or_create('app', 'http://git.hunantv.com/group/app.git')
v = a.add_version(random_sha1())
p = Pod.create('pod', 'pod', 10, -1)
host = Host.create(p, random_ipv4(), random_string(), random_uuid(), 200,
0)
for core in host.cores:
assert core.host_id == host.id
assert core.remain == 10
containers = []
cores = sorted(host.cores, key=operator.attrgetter('label'))
for fcores, pcores in zip(chunked(cores[:100], 10),
chunked(cores[100:], 10)):
used_cores = {'full': fcores, 'part': pcores}
host.occupy_cores(used_cores, 5)
c = Container.create(random_sha1(),
host,
v,
random_string(),
'entrypoint',
used_cores,
'env',
nshare=5)
containers.append(c)
cores = sorted(host.cores, key=operator.attrgetter('label'))
for fcores, pcores in zip(chunked(cores[:100], 10),
chunked(cores[100:], 10)):
for core in fcores:
assert core.remain == 0
for core in pcores:
assert core.remain == 5
for c in containers:
c.delete()
cores = sorted(host.cores, key=operator.attrgetter('label'))
for fcores, pcores in zip(chunked(cores[:100], 10),
chunked(cores[100:], 10)):
for core in fcores:
assert core.remain == 10
for core in pcores:
assert core.remain == 10
def create_test_suite():
appyaml = {
'appname': 'app',
'entrypoints': {
'web': {
'cmd': 'python app.py',
'ports': ['5000/tcp'],
},
'daemon': {
'cmd': 'python daemon.py',
},
'service': {
'cmd': 'python service.py'
},
},
'build': 'pip install -r ./requirements.txt',
}
app = App.get_or_create('app', 'http://git.hunantv.com/group/app.git')
version = app.add_version(random_sha1())
appconfig = version.appconfig
appconfig.update(**appyaml)
appconfig.save()
pod = Pod.create('pod', 'pod')
hosts = [
Host.create(pod, random_ipv4(), random_string(prefix='host'),
random_uuid(), 4, 4096) for i in range(4)
]
containers = []
for (host, count), cores in centralized_schedule(pod, 4, 4, 0).iteritems():
cores_per_container = len(cores) / count
for i in range(count):
cid = random_sha1()
used_cores = {
'full':
cores['full'][i * cores_per_container:(i + 1) *
cores_per_container]
}
c = Container.create(cid, host, version, random_string(), 'web',
used_cores, 'env')
containers.append(c)
host.occupy_cores(cores, 0)
return app, version, pod, hosts, containers
def create_test_suite():
appyaml = {
'appname': 'app',
'entrypoints': {
'web': {
'cmd': 'python app.py',
'ports': ['5000/tcp'],
},
'daemon': {
'cmd': 'python daemon.py',
},
'service': {
'cmd': 'python service.py'
},
},
'build': 'pip install -r ./requirements.txt',
}
app = App.get_or_create('app', 'http://git.hunantv.com/group/app.git', 'token')
version = app.add_version(random_sha1())
appconfig = version.appconfig
appconfig.update(**appyaml)
appconfig.save()
group = Group.create('group', 'group')
pod = Pod.create('pod', 'pod')
pod.assigned_to_group(group)
hosts = [Host.create(pod, random_ipv4(), random_string(prefix='host'),
random_uuid(), 4, 4096) for i in range(4)]
for host in hosts:
host.assigned_to_group(group)
containers = []
for (host, count), cores in group.get_free_cores(pod, 4, 4, 0).iteritems():
cores_per_container = len(cores) / count
for i in range(count):
cid = random_sha1()
used_cores = cores['full'][i*cores_per_container:(i+1)*cores_per_container]
c = Container.create(cid, host, version, random_string(), 'entrypoint', used_cores, 'env')
containers.append(c)
host.occupy_cores(cores, 0)
return app, version, group, pod, hosts, containers
def test_container(test_db):
a = App.get_or_create('app', 'http://git.hunantv.com/group/app.git', '')
assert a is not None
assert a.id == a.user_id
v = a.add_version(random_sha1())
assert v is not None
assert v.app.id == a.id
assert v.name == a.name
assert len(v.containers.all()) == 0
assert len(v.tasks.all()) == 0
g = Group.create('group', 'group')
p = Pod.create('pod', 'pod', 10, -1)
assert p.assigned_to_group(g)
hosts = [Host.create(p, random_ipv4(), random_string(prefix='host'),
random_uuid(), 4, 4096) for i in range(6)]
for host in hosts[:3]:
host.assigned_to_group(g)
host_ids1 = {h.id for h in hosts[:3]}
host_ids2 = {h.id for h in hosts[3:]}
host_cores = g.get_free_cores(p, 3, 3, 0)
#测试没有碎片核的情况
#获取核
containers = []
for (host, count), cores in host_cores.iteritems():
cores_per_container = len(cores['full']) / count
for i in range(count):
cid = random_sha1()
used_cores = {'full': cores['full'][i*cores_per_container:(i+1)*cores_per_container]}
# not using a port
c = Container.create(cid, host, v, random_string(), 'entrypoint', used_cores, 'env')
assert c is not None
containers.append(c)
host.occupy_cores(cores, 0)
for host in g.private_hosts.all():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 1
assert len(part_cores) == 0
assert len(host.containers.all()) == 1
assert host.count == 1
assert len(containers) == 3
assert len(v.containers.all()) == 3
for c in containers:
assert c.host_id in host_ids1
assert c.host_id not in host_ids2
assert c.app.id == a.id
assert c.version.id == v.id
assert c.is_alive
assert len(c.full_cores.all()) == 3
assert len(c.part_cores.all()) == 0
all_core_labels = sorted(['0', '1', '2', '3', ])
used_full_core_labels = [core.label for core in c.full_cores.all()]
used_part_core_labels = [core.label for core in c.part_cores.all()]
free_core_labels = [core.label for core in c.host.get_free_cores()[0]]
assert all_core_labels == sorted(used_full_core_labels + used_part_core_labels + free_core_labels)
#释放核
for c in containers:
c.delete()
assert len(v.containers.all()) == 0
assert g.get_max_containers(p, 3, 0) == 3
host_cores = g.get_free_cores(p, 3, 3, 0)
assert len(host_cores) == 3
for host in g.private_hosts.all():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 4
assert len(part_cores) == 0
assert len(host.containers.all()) == 0
assert host.count == 0
#测试有碎片的情况
#获取核
host_cores = g.get_free_cores(p, 3, 3, 4)
containers = []
for (host, count), cores in host_cores.iteritems():
cores_per_container = len(cores['full']) / count
for i in range(count):
cid = random_sha1()
used_cores = {'full': cores['full'][i*cores_per_container:(i+1)*cores_per_container],
'part': cores['part']}
# not using a port
c = Container.create(cid, host, v, random_string(), 'entrypoint', used_cores, 'env')
assert c is not None
containers.append(c)
host.occupy_cores(cores, 4)
for host in g.private_hosts.all():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 0
assert len(part_cores) == 1
assert part_cores[0].used == 4
assert len(host.containers.all()) == 1
assert host.count == 1
assert len(containers) == 3
assert len(v.containers.all()) == 3
for c in containers:
assert c.host_id in host_ids1
assert c.host_id not in host_ids2
assert c.app.id == a.id
assert c.version.id == v.id
assert c.is_alive
assert len(c.full_cores.all()) == 3
assert len(c.part_cores.all()) == 1
all_core_labels = sorted(['0', '1', '2', '3', ])
used_full_core_labels = [core.label for core in c.full_cores.all()]
used_part_core_labels = [core.label for core in c.part_cores.all()]
free_core_labels = [core.label for core in c.host.get_free_cores()[0]]
assert all_core_labels == sorted(used_full_core_labels + used_part_core_labels + free_core_labels)
#释放核
for c in containers:
c.delete(4)
assert len(v.containers.all()) == 0
assert g.get_max_containers(p, 3, 0) == 3
host_cores = g.get_free_cores(p, 3, 3, 0)
assert len(host_cores) == 3
for host in g.private_hosts.all():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 4
assert len(host.containers.all()) == 0
assert host.count == 0
#获取
host_cores = g.get_free_cores(p, 6, 1, 5)
containers = []
for (host, count), cores in host_cores.iteritems():
cores_per_container = len(cores['full']) / count
for i in range(count):
cid = random_sha1()
used_cores = {'full': cores['full'][i*cores_per_container:(i+1)*cores_per_container],
'part': cores['part'][i:i+1]}
# not using a port
c = Container.create(cid, host, v, random_string(), 'entrypoint', used_cores, 'env')
assert c is not None
containers.append(c)
host.occupy_cores(used_cores, 5)
for host in g.private_hosts.all():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 1
assert len(part_cores) == 0
assert len(host.containers.all()) == 2
assert host.count == 2
assert len(containers) == 6
assert len(v.containers.all()) == 6
for c in containers:
assert c.host_id in host_ids1
assert c.host_id not in host_ids2
assert c.app.id == a.id
assert c.version.id == v.id
assert c.is_alive
assert len(c.full_cores.all()) == 1
assert len(c.part_cores.all()) == 1
##释放核
for c in containers:
c.delete(5)
assert len(v.containers.all()) == 0
assert g.get_max_containers(p, 3, 0) == 3
host_cores = g.get_free_cores(p, 3, 3, 0)
assert len(host_cores) == 3
for host in g.private_hosts.all():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 4
assert len(part_cores) == 0
assert len(host.containers.all()) == 0
assert host.count == 0
def create_containers_with_macvlan(task_id, ncontainer, core_ids, network_ids):
"""
执行task_id的任务. 部署ncontainer个容器, 占用core_ids这些核, 绑定到networks这些子网
"""
task = Task.get(task_id)
if not task:
return
networks = Network.get_multi(network_ids)
notifier = TaskNotifier(task)
host = task.host
version = task.version
entrypoint = task.props['entrypoint']
env = task.props['env']
used_cores = Core.get_multi(core_ids)
pub_agent_vlan_key = 'eru:agent:%s:vlan' % host.name
feedback_key = 'eru:agent:%s:feedback' % task_id
cids = []
for cores in more_itertools.chunked(used_cores, len(core_ids)/ncontainer):
try:
cid, cname = dockerjob.create_one_container(host, version,
entrypoint, env, cores)
except:
host.release_cores(cores)
continue
ips = [n.acquire_ip() for n in networks]
ip_dict = {ip.vlan_address: ip for ip in ips}
if ips:
ident_id = cname.split('_')[-1]
values = [str(task_id), cid, ident_id] + ['{0}:{1}'.format(ip.vlan_seq_id, ip.vlan_address) for ip in ips]
rds.publish(pub_agent_vlan_key, '|'.join(values))
for _ in ips:
# timeout 15s
rv = rds.blpop(feedback_key, 15)
if rv is None:
break
# rv is like (feedback_key, 'succ|container_id|vethname|ip')
succ, _, vethname, vlan_address = rv[1].split('|')
if succ == '0':
break
ip = ip_dict.get(vlan_address, None)
if ip:
ip.set_vethname(vethname)
else:
logger.info('Creating container with cid %s and ips %s' % (cid, ips))
c = Container.create(cid, host, version, cname, entrypoint, cores, env)
for ip in ips:
ip.assigned_to_container(c)
notifier.notify_agent(cid)
add_container_for_agent(c)
add_container_backends(c)
cids.append(cid)
# 略过清理工作
continue
# 清理掉失败的容器, 释放核, 释放ip
logger.info('Cleaning failed container with cid %s' % cid)
dockerjob.remove_container_by_cid([cid], host)
host.release_cores(cores)
[ip.release() for ip in ips]
publish_to_service_discovery(version.name)
task.finish_with_result(code.TASK_SUCCESS, container_ids=cids)
notifier.pub_success()
def create_containers_with_macvlan_public(task_id, ncontainer, nshare, network_ids, spec_ips=None):
"""
执行task_id的任务. 部署ncontainer个容器, 绑定到networks这些子网
"""
current_flask.logger.info('Task<id=%s>: Started', task_id)
task = Task.get(task_id)
if not task:
current_flask.logger.error('Task (id=%s) not found, quit', task_id)
return
if spec_ips is None:
spec_ips = []
networks = Network.get_multi(network_ids)
notifier = TaskNotifier(task)
host = task.host
version = task.version
entrypoint = task.props['entrypoint']
env = task.props['env']
# use raw
image = task.props['image']
cpu_shares = 1024
pub_agent_vlan_key = 'eru:agent:%s:vlan' % host.name
feedback_key = 'eru:agent:%s:feedback' % task_id
cids = []
for _ in range(ncontainer):
try:
cid, cname = dockerjob.create_one_container(host, version,
entrypoint, env, cores=None, cpu_shares=cpu_shares, image=image)
except Exception as e:
print e # 同上
continue
if spec_ips:
ips = [n.acquire_specific_ip(ip) for n, ip in zip(networks, spec_ips)]
else:
ips = [n.acquire_ip() for n in networks]
ips = [i for i in ips if i]
ip_dict = {ip.vlan_address: ip for ip in ips}
if ips:
ident_id = cname.split('_')[-1]
values = [str(task_id), cid, ident_id] + ['{0}:{1}'.format(ip.vlan_seq_id, ip.vlan_address) for ip in ips]
rds.publish(pub_agent_vlan_key, '|'.join(values))
for _ in ips:
# timeout 15s
rv = rds.blpop(feedback_key, 15)
if rv is None:
break
# rv is like (feedback_key, 'succ|container_id|vethname|ip')
succ, _, vethname, vlan_address = rv[1].split('|')
if succ == '0':
break
ip = ip_dict.get(vlan_address, None)
if ip:
ip.set_vethname(vethname)
else:
current_flask.logger.info('Creating container (cid=%s, ips=%s)', cid, ips)
c = Container.create(cid, host, version, cname, entrypoint, {}, env, nshare)
for ip in ips:
ip.assigned_to_container(c)
notifier.notify_agent(cid)
add_container_for_agent(c)
add_container_backends(c)
cids.append(cid)
# 略过清理工作
continue
# 清理掉失败的容器, 释放核, 释放ip
current_flask.logger.info('Cleaning failed container (cid=%s)', cid)
dockerjob.remove_container_by_cid([cid], host)
[ip.release() for ip in ips]
# 失败了就得清理掉这个key
rds.delete(feedback_key)
publish_to_service_discovery(version.name)
task.finish_with_result(consts.TASK_SUCCESS, container_ids=cids)
notifier.pub_success()
current_flask.logger.info('Task<id=%s>: Done', task_id)
def create_containers_with_macvlan(task_id, ncontainer, nshare, cores, network_ids, spec_ips=None):
"""
执行task_id的任务. 部署ncontainer个容器, 占用*_core_ids这些核, 绑定到networks这些子网
"""
current_flask.logger.info('Task<id=%s>: Started', task_id)
task = Task.get(task_id)
if not task:
current_flask.logger.error('Task (id=%s) not found, quit', task_id)
return
if spec_ips is None:
spec_ips = []
need_network = bool(network_ids)
networks = Network.get_multi(network_ids)
notifier = TaskNotifier(task)
host = task.host
version = task.version
entrypoint = task.props['entrypoint']
env = task.props['env']
ports = task.props['ports']
args = task.props['args']
# use raw
route = task.props['route']
image = task.props['image']
callback_url = task.props['callback_url']
cpu_shares = int(float(nshare) / host.pod.core_share * 1024) if nshare else 1024
pub_agent_vlan_key = 'eru:agent:%s:vlan' % host.name
pub_agent_route_key = 'eru:agent:%s:route' % host.name
feedback_key = 'eru:agent:%s:feedback' % task_id
cids = []
for fcores, pcores in _iter_cores(cores, ncontainer):
cores_for_one_container = {'full': fcores, 'part': pcores}
try:
cid, cname = dockerjob.create_one_container(host, version,
entrypoint, env, fcores+pcores, ports=ports, args=args,
cpu_shares=cpu_shares, image=image, need_network=need_network)
except Exception as e:
# 写给celery日志看
print e
host.release_cores(cores_for_one_container, nshare)
continue
if spec_ips:
ips = [n.acquire_specific_ip(ip) for n, ip in zip(networks, spec_ips)]
else:
ips = [n.acquire_ip() for n in networks]
ips = [i for i in ips if i]
ip_dict = {ip.vlan_address: ip for ip in ips}
if ips:
if ERU_AGENT_API == 'pubsub':
values = [str(task_id), cid] + ['{0}:{1}'.format(ip.vlan_seq_id, ip.vlan_address) for ip in ips]
rds.publish(pub_agent_vlan_key, '|'.join(values))
elif ERU_AGENT_API == 'http':
agent = get_agent(host)
ip_list = [(ip.vlan_seq_id, ip.vlan_address) for ip in ips]
agent.add_container_vlan(cid, str(task_id), ip_list)
for _ in ips:
# timeout 15s
rv = rds.blpop(feedback_key, 15)
if rv is None:
break
# rv is like (feedback_key, 'succ|container_id|vethname|ip')
succ, _, vethname, vlan_address = rv[1].split('|')
if succ == '0':
break
ip = ip_dict.get(vlan_address, None)
if ip:
ip.set_vethname(vethname)
if route:
rds.publish(pub_agent_route_key, '%s|%s' % (cid, route))
else:
current_flask.logger.info('Creating container (cid=%s, ips=%s)', cid, ips)
c = Container.create(cid, host, version, cname, entrypoint,
cores_for_one_container, env, nshare, callback_url)
for ip in ips:
ip.assigned_to_container(c)
notifier.notify_agent(c)
add_container_for_agent(c)
add_container_backends(c)
cids.append(cid)
# 略过清理工作
continue
# 清理掉失败的容器, 释放核, 释放ip
current_flask.logger.info('Cleaning failed container (cid=%s)', cid)
dockerjob.remove_container_by_cid([cid], host)
host.release_cores(cores_for_one_container, nshare)
[ip.release() for ip in ips]
# 失败了就得清理掉这个key
rds.delete(feedback_key)
publish_to_service_discovery(version.name)
task.finish_with_result(consts.TASK_SUCCESS, container_ids=cids)
notifier.pub_success()
# 有IO, 丢最后面算了
falcon_all_graphs(version)
falcon_all_alarms(version)
current_flask.logger.info('Task<id=%s>: Done', task_id)
def test_container(test_db):
a = App.get_or_create('app', 'http://git.hunantv.com/group/app.git')
assert a is not None
assert a.id == a.user_id
v = a.add_version(random_sha1())
assert v is not None
assert v.app.id == a.id
assert v.name == a.name
assert len(v.containers.all()) == 0
assert len(v.tasks.all()) == 0
p = Pod.create('pod', 'pod', 10, -1)
hosts = [
Host.create(p, random_ipv4(), random_string(prefix='host'),
random_uuid(), 4, 4096) for i in range(6)
]
for host in hosts[3:]:
host.set_public()
host_ids1 = {h.id for h in hosts[:3]}
host_ids2 = {h.id for h in hosts[3:]}
host_cores = centralized_schedule(p, 3, 3, 0)
#测试没有碎片核的情况
#获取核
containers = []
for (host, count), cores in host_cores.iteritems():
host.occupy_cores(cores, 0)
cores_per_container = len(cores['full']) / count
for i in range(count):
cid = random_sha1()
used_cores = {
'full':
cores['full'][i * cores_per_container:(i + 1) *
cores_per_container]
}
c = Container.create(cid,
host,
v,
random_string(),
'entrypoint',
used_cores,
'env',
nshare=0)
assert c is not None
containers.append(c)
for host in p.get_private_hosts():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 1
assert len(part_cores) == 0
assert len(host.containers.all()) == 1
assert host.count == 1
assert len(containers) == 3
assert len(v.containers.all()) == 3
for c in containers:
assert c.host_id in host_ids1
assert c.host_id not in host_ids2
assert c.app.id == a.id
assert c.version.id == v.id
assert c.is_alive
assert len(c.full_cores) == 3
assert len(c.part_cores) == 0
all_core_labels = sorted([
'0',
'1',
'2',
'3',
])
used_full_core_labels = [core.label for core in c.full_cores]
used_part_core_labels = [core.label for core in c.part_cores]
free_core_labels = [core.label for core in c.host.get_free_cores()[0]]
assert all_core_labels == sorted(used_full_core_labels +
used_part_core_labels +
free_core_labels)
#释放核
for c in containers:
c.delete()
assert len(v.containers.all()) == 0
assert get_max_container_count(p, 3, 0) == 3
host_cores = centralized_schedule(p, 3, 3, 0)
assert len(host_cores) == 3
for host in p.get_private_hosts():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 4
assert len(part_cores) == 0
assert len(host.containers.all()) == 0
assert host.count == 4
#测试有碎片的情况
#获取核
host_cores = centralized_schedule(p, 3, 3, 4)
containers = []
for (host, count), cores in host_cores.iteritems():
cores_per_container = len(cores['full']) / count
host.occupy_cores(cores, 4)
for i in range(count):
cid = random_sha1()
used_cores = {
'full':
cores['full'][i * cores_per_container:(i + 1) *
cores_per_container],
'part':
cores['part']
}
# not using a port
c = Container.create(cid,
host,
v,
random_string(),
'entrypoint',
used_cores,
'env',
nshare=4)
assert c is not None
containers.append(c)
for host in p.get_private_hosts():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 0
assert len(part_cores) == 1
assert part_cores[0].remain == 6
assert len(host.containers.all()) == 1
assert host.count == D('0.6')
assert len(containers) == 3
assert len(v.containers.all()) == 3
for c in containers:
assert c.host_id in host_ids1
assert c.host_id not in host_ids2
assert c.app.id == a.id
assert c.version.id == v.id
assert c.is_alive
assert len(c.full_cores) == 3
assert len(c.part_cores) == 1
all_core_labels = sorted([
'0',
'1',
'2',
'3',
])
used_full_core_labels = [core.label for core in c.full_cores]
used_part_core_labels = [core.label for core in c.part_cores]
free_core_labels = [core.label for core in c.host.get_free_cores()[0]]
assert all_core_labels == sorted(used_full_core_labels +
used_part_core_labels +
free_core_labels)
#释放核
for c in containers:
c.delete()
assert len(v.containers.all()) == 0
assert get_max_container_count(p, 3, 0) == 3
host_cores = centralized_schedule(p, 3, 3, 0)
assert len(host_cores) == 3
for host in p.get_private_hosts():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 4
assert len(host.containers.all()) == 0
assert host.count == 4
#获取
host_cores = centralized_schedule(p, 6, 1, 5)
containers = []
for (host, count), cores in host_cores.iteritems():
cores_per_container = len(cores['full']) / count
for i in range(count):
cid = random_sha1()
used_cores = {
'full':
cores['full'][i * cores_per_container:(i + 1) *
cores_per_container],
'part':
cores['part'][i:i + 1],
}
host.occupy_cores(used_cores, 5)
# not using a port
c = Container.create(cid,
host,
v,
random_string(),
'entrypoint',
used_cores,
'env',
nshare=5)
assert c is not None
containers.append(c)
for host in p.get_private_hosts():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 1
assert len(part_cores) == 0
assert len(host.containers.all()) == 2
assert host.count == D('1')
assert len(containers) == 6
assert len(v.containers.all()) == 6
for c in containers:
assert c.host_id in host_ids1
assert c.host_id not in host_ids2
assert c.app.id == a.id
assert c.version.id == v.id
assert c.is_alive
assert len(c.full_cores) == 1
assert len(c.part_cores) == 1
##释放核
for c in containers:
c.delete()
assert len(v.containers.all()) == 0
assert get_max_container_count(p, 3, 0) == 3
host_cores = centralized_schedule(p, 3, 3, 0)
assert len(host_cores) == 3
for host in p.get_private_hosts():
full_cores, part_cores = host.get_free_cores()
assert len(full_cores) == 4
assert len(part_cores) == 0
assert len(host.containers.all()) == 0
assert host.count == 4
def create_containers(task_id, ncontainer, nshare, cores, network_ids, spec_ips=None):
"""
执行task_id的任务. 部署ncontainer个容器, 占用*_core_ids这些核, 绑定到networks这些子网
"""
_log.info('Task<id=%s>: Started', task_id)
task = Task.get(task_id)
if not task:
_log.error('Task (id=%s) not found, quit', task_id)
return
if spec_ips is None:
spec_ips = []
need_network = bool(network_ids)
networks = [ipam.get_pool(n) for n in network_ids]
notifier = TaskNotifier(task)
host = task.host
version = task.version
entrypoint = task.props['entrypoint']
env = task.props['env']
ports = task.props['ports']
args = task.props['args']
# use raw
image = task.props['image']
callback_url = task.props['callback_url']
cpu_shares = int(float(nshare) / host.pod.core_share * 1024) if nshare else 1024
cids = []
backends = []
entry = version.appconfig.entrypoints[entrypoint]
for fcores, pcores in _iter_cores(cores, ncontainer):
cores_for_one_container = {'full': fcores, 'part': pcores}
# 在宿主机上创建容器
try:
cid, cname = dockerjob.create_one_container(host,
version,
entrypoint,
env,
fcores + pcores,
ports=ports, args=args,
cpu_shares=cpu_shares,
image=image,
need_network=need_network)
except Exception as e:
# 写给celery日志看
_log.exception(e)
host.release_cores(cores_for_one_container, nshare)
continue
# 容器记录下来
c = Container.create(cid, host, version, cname, entrypoint, cores_for_one_container, env, nshare, callback_url)
# 为容器创建网络栈
# 同时把各种信息都记录下来
# 如果失败, 清除掉所有记录和宿主机上的容器
# 循环下一次尝试
cidrs = [n.netspace for n in networks]
if not ipam.allocate_ips(cidrs, cid, spec_ips):
_clean_failed_containers(cid)
continue
notifier.notify_agent(c)
add_container_for_agent(host, c)
add_container_backends(c)
cids.append(cid)
backends.extend(c.get_backends())
c.callback_report(status='start')
health_check = entry.get('health_check', '')
if health_check and backends:
urls = [b + health_check for b in backends]
if not wait_health_check(urls):
# TODO 这里要么回滚要么报警
_log.info('Task<id=%s>: Done, but something went error', task_id)
return
publish_to_service_discovery(version.name)
task.finish(consts.TASK_SUCCESS)
task.reason = 'ok'
task.container_ids = cids
notifier.pub_success()
_log.info('Task<id=%s>: Done', task_id)
def create_containers(task_id,
ncontainer,
nshare,
cores,
network_ids,
spec_ips=None):
"""
执行task_id的任务. 部署ncontainer个容器, 占用*_core_ids这些核, 绑定到networks这些子网
"""
_log.info('Task<id=%s>: Started', task_id)
task = Task.get(task_id)
if not task:
_log.error('Task (id=%s) not found, quit', task_id)
return
if spec_ips is None:
spec_ips = []
need_network = bool(network_ids)
networks = [ipam.get_pool(n) for n in network_ids]
notifier = TaskNotifier(task)
host = task.host
version = task.version
entrypoint = task.props['entrypoint']
env = task.props['env']
ports = task.props['ports']
args = task.props['args']
# use raw
image = task.props['image']
callback_url = task.props['callback_url']
cpu_shares = int(float(nshare) / host.pod.core_share *
1024) if nshare else 1024
cids = []
backends = []
entry = version.appconfig.entrypoints[entrypoint]
for fcores, pcores in _iter_cores(cores, ncontainer):
cores_for_one_container = {'full': fcores, 'part': pcores}
# 在宿主机上创建容器
try:
cid, cname = dockerjob.create_one_container(
host,
version,
entrypoint,
env,
fcores + pcores,
ports=ports,
args=args,
cpu_shares=cpu_shares,
image=image,
need_network=need_network)
except Exception as e:
# 写给celery日志看
_log.exception(e)
host.release_cores(cores_for_one_container, nshare)
continue
# 容器记录下来
c = Container.create(cid, host, version, cname, entrypoint,
cores_for_one_container, env, nshare,
callback_url)
# 为容器创建网络栈
# 同时把各种信息都记录下来
# 如果失败, 清除掉所有记录和宿主机上的容器
# 循环下一次尝试
cidrs = [n.netspace for n in networks]
if not ipam.allocate_ips(cidrs, cid, spec_ips):
_clean_failed_containers(cid)
continue
notifier.notify_agent(c)
add_container_for_agent(host, c)
add_container_backends(c)
cids.append(cid)
backends.extend(c.get_backends())
c.callback_report(status='start')
health_check = entry.get('health_check', '')
if health_check and backends:
urls = [b + health_check for b in backends]
if not wait_health_check(urls):
# TODO 这里要么回滚要么报警
_log.info('Task<id=%s>: Done, but something went error', task_id)
return
publish_to_service_discovery(version.name)
task.finish(consts.TASK_SUCCESS)
task.reason = 'ok'
task.container_ids = cids
notifier.pub_success()
_log.info('Task<id=%s>: Done', task_id)
def create_containers_with_macvlan(task_id, ncontainer, nshare, cores, network_ids, spec_ips=None):
"""
执行task_id的任务. 部署ncontainer个容器, 占用*_core_ids这些核, 绑定到networks这些子网
"""
# TODO support part core
current_flask.logger.info("Task<id=%s>: Started", task_id)
task = Task.get(task_id)
if not task:
current_flask.logger.error("Task (id=%s) not found, quit", task_id)
return
if spec_ips is None:
spec_ips = []
networks = Network.get_multi(network_ids)
notifier = TaskNotifier(task)
host = task.host
version = task.version
entrypoint = task.props["entrypoint"]
env = task.props["env"]
# use raw
image = task.props["image"]
cpu_shares = int(float(nshare) / host.pod.core_share * 1024) if nshare else 1024
pub_agent_vlan_key = "eru:agent:%s:vlan" % host.name
feedback_key = "eru:agent:%s:feedback" % task_id
cids = []
full_cores, part_cores = cores.get("full", []), cores.get("part", [])
for fcores, pcores in izip_longest(
chunked(full_cores, len(full_cores) / ncontainer),
chunked(part_cores, len(part_cores) / ncontainer),
fillvalue=[],
):
cores_for_one_container = {"full": fcores, "part": pcores}
try:
cid, cname = dockerjob.create_one_container(
host, version, entrypoint, env, fcores + pcores, cpu_shares, image=image
)
except:
host.release_cores(cores_for_one_container, nshare)
continue
if spec_ips:
ips = [n.acquire_specific_ip(ip) for n, ip in zip(networks, spec_ips)]
else:
ips = [n.acquire_ip() for n in networks]
ips = [i for i in ips if i]
ip_dict = {ip.vlan_address: ip for ip in ips}
if ips:
ident_id = cname.split("_")[-1]
values = [str(task_id), cid, ident_id] + ["{0}:{1}".format(ip.vlan_seq_id, ip.vlan_address) for ip in ips]
rds.publish(pub_agent_vlan_key, "|".join(values))
for _ in ips:
# timeout 15s
rv = rds.blpop(feedback_key, 15)
if rv is None:
break
# rv is like (feedback_key, 'succ|container_id|vethname|ip')
succ, _, vethname, vlan_address = rv[1].split("|")
if succ == "0":
break
ip = ip_dict.get(vlan_address, None)
if ip:
ip.set_vethname(vethname)
else:
current_flask.logger.info("Creating container (cid=%s, ips=%s)", cid, ips)
c = Container.create(cid, host, version, cname, entrypoint, cores_for_one_container, env, nshare)
for ip in ips:
ip.assigned_to_container(c)
notifier.notify_agent(cid)
add_container_for_agent(c)
add_container_backends(c)
cids.append(cid)
# 略过清理工作
continue
# 清理掉失败的容器, 释放核, 释放ip
current_flask.logger.info("Cleaning failed container (cid=%s)", cid)
dockerjob.remove_container_by_cid([cid], host)
host.release_cores(cores_for_one_container, nshare)
[ip.release() for ip in ips]
# 失败了就得清理掉这个key
rds.delete(feedback_key)
publish_to_service_discovery(version.name)
task.finish_with_result(consts.TASK_SUCCESS, container_ids=cids)
notifier.pub_success()
current_flask.logger.info("Task<id=%s>: Done", task_id)
def test_container(test_db):
a = App.get_or_create('app', 'http://git.hunantv.com/group/app.git', '')
assert a is not None
v = a.add_version(random_sha1())
assert v is not None
assert v.app.id == a.id
assert v.name == a.name
assert len(v.containers.all()) == 0
assert len(v.tasks.all()) == 0
g = Group.create('group', 'group')
p = Pod.create('pod', 'pod')
assert p.assigned_to_group(g)
hosts = [Host.create(p, random_ipv4(), random_string(prefix='host'),
random_uuid(), 4, 4096) for i in range(6)]
for host in hosts[:3]:
host.assigned_to_group(g)
host_ids1 = {h.id for h in hosts[:3]}
host_ids2 = {h.id for h in hosts[3:]}
assert g.get_max_containers(p, 3) == 3
host_cores = g.get_free_cores(p, 3, 3)
assert len(host_cores) == 3
containers = []
for (host, count), cores in host_cores.iteritems():
cores_per_container = len(cores) / count
for i in range(count):
cid = random_sha1()
used_cores = cores[i*cores_per_container:(i+1)*cores_per_container]
# not using a port
c = Container.create(cid, host, v, random_string(), 'entrypoint', used_cores, 'env')
assert c is not None
containers.append(c)
host.occupy_cores(cores)
for host in g.private_hosts.all():
assert len(host.get_free_cores()) == 1
assert len(host.containers.all()) == 1
assert host.count == 1
assert len(containers) == 3
assert len(v.containers.all()) == 3
for c in containers:
assert c.host_id in host_ids1
assert c.host_id not in host_ids2
assert c.app.id == a.id
assert c.version.id == v.id
assert c.is_alive
assert len(c.cores.all()) == 3
all_core_labels = sorted(['0', '1', '2', '3', ])
used_core_labels = [core.label for core in c.cores.all()]
free_core_labels = [core.label for core in c.host.get_free_cores()]
assert all_core_labels == sorted(used_core_labels + free_core_labels)
for c in containers:
c.delete()
assert len(v.containers.all()) == 0
assert g.get_max_containers(p, 3) == 3
host_cores = g.get_free_cores(p, 3, 3)
assert len(host_cores) == 3
for host in g.private_hosts.all():
assert len(host.get_free_cores()) == 4
assert len(host.containers.all()) == 0
assert host.count == 0