def rebuild_container_pool():
hosts = Host.query.all()
for host in hosts:
key = 'eru:agent:%s:containers:meta' % host.name
rds.delete(key)
for c in host.containers.all():
add_container_for_agent(c)
def fix_core(host):
containers = Container.get_multi_by_host(host)
# 没有的话, 直接销毁重建
if not containers:
rds.delete(host._cores_key)
_create_cores_on_host(host, host.ncore)
return
data = {str(i): host.core_share for i in xrange(host.ncore)}
for c in containers:
cores = c.cores
nshare = int(cores.get("nshare", "0"))
for e in cores.get("full", []):
e.remain = host.core_share
data.pop(e.label)
for s in cores.get("part", []):
s.remain = host.core_share - nshare
data[s.label] -= nshare
c.cores = cores
rds.delete(host._cores_key)
if data:
rds.zadd(host._cores_key, **data)
print "done", host
def kill_container(cid):
c = Container.get_by_container_id(cid)
if c:
c.kill()
key = consts.ERU_AGENT_DIE_REASON % c.container_id
r = rds.get(key)
rds.delete(key)
if r is not None:
c.set_props({'oom': 1})
current_app.logger.info('Kill container (container_id=%s)', cid[:7])
return {'r': 0, 'msg': consts.OK}
def _bind_container_ip():
rds.publish(pub_agent_vlan_key, '|'.join(values))
for _ in ips:
rv = rds.blpop(feedback_key, 15)
if rv is None:
break
succ = rv[1].split('|')[0]
if succ == '0':
break
else:
return True
rds.delete(feedback_key)
return False
def remove_containers(task_id, cids, rmi=False):
task = Task.get(task_id)
if not task:
current_flask.logger.error('Task (id=%s) not found, quit', task_id)
return
current_flask.logger.info('Task<id=%s>: Start on host %s' % (task_id, task.host.ip))
notifier = TaskNotifier(task)
containers = Container.get_multi(cids)
container_ids = [c.container_id for c in containers if c]
host = task.host
version = task.version
try:
# flag, don't report these
flags = {'eru:agent:%s:container:flag' % cid: 1 for cid in container_ids}
rds.mset(**flags)
for c in containers:
remove_container_backends(c)
current_flask.logger.info('Task<id=%s>: Container (cid=%s) backends removed',
task_id, c.container_id[:7])
appnames = {c.appname for c in containers}
publish_to_service_discovery(*appnames)
time.sleep(3)
dockerjob.remove_host_containers(containers, host)
current_flask.logger.info('Task<id=%s>: Containers (cids=%s) removed', task_id, cids)
if rmi:
try:
dockerjob.remove_image(task.version, host)
except Exception as e:
current_flask.logger.error('Task<id=%s>: Exception (e=%s), fail to remove image', task_id, e)
except Exception as e:
task.finish_with_result(consts.TASK_FAILED)
notifier.pub_fail()
current_flask.logger.error('Task<id=%s>: Exception (e=%s)', task_id, e)
else:
for c in containers:
c.delete()
task.finish_with_result(consts.TASK_SUCCESS)
notifier.pub_success()
if container_ids:
rds.hdel('eru:agent:%s:containers:meta' % host.name, *container_ids)
rds.delete(*flags.keys())
current_flask.logger.info('Task<id=%s>: Done', task_id)
if not version.containers.count():
falcon_remove_alarms(version)
def _bind_container_ip_http(task_id, container, ips, nid=None):
agent = get_agent(container.host)
feedback_key = 'eru:agent:%s:feedback' % task_id
ip_list = [(nid or ip.vlan_seq_id, ip.vlan_address) for ip in ips]
agent.add_container_vlan(container.container_id, str(task_id), ip_list)
for _ in ips:
rv = rds.blpop(feedback_key, 15)
if rv is None:
break
succ = rv[1].split('|')[0]
if succ == '0':
break
else:
return True
rds.delete(feedback_key)
return False
def _bind_container_ip_pubsub(task_id, container, ips, nid=None):
pub_agent_vlan_key = 'eru:agent:%s:vlan' % container.host.name
feedback_key = 'eru:agent:%s:feedback' % task_id
values = [task_id, container.container_id]
values += ['{0}:{1}'.format(nid or ip.vlan_seq_id, ip.vlan_address) for ip in ips]
rds.publish(pub_agent_vlan_key, '|'.join(values))
for _ in ips:
rv = rds.blpop(feedback_key, 15)
if rv is None:
break
succ = rv[1].split('|')[0]
if succ == '0':
break
else:
return True
rds.delete(feedback_key)
return False
def _del_falcon(self):
rds.delete("eru:falcon:version:%s:expression" % self.id)
def _set_falcon(self, expr_ids):
rds.delete("eru:falcon:version:%s:expression" % self.id)
for i in expr_ids:
rds.sadd("eru:falcon:version:%s:expression" % self.id, i)
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)
dockerjob.remove_host_containers(containers, host)
current_flask.logger.info('Task<id=%s>: Containers (cids=%s) removed', task_id, cids)
if rmi:
dockerjob.remove_image(task.version, host)
except Exception, e:
task.finish_with_result(consts.TASK_FAILED)
notifier.pub_fail()
current_flask.logger.error('Task<id=%s>: Exception (e=%s)', task_id, e)
else:
for c in containers:
c.delete()
task.finish_with_result(consts.TASK_SUCCESS)
notifier.pub_success()
if container_ids:
rds.srem('eru:agent:%s:containers' % host.name, *container_ids)
rds.delete(*flags.keys())
current_flask.logger.info('Task<id=%s>: Done', task_id)
@current_app.task()
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:
def cores(self):
rds.delete(self._cores_key)
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 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)