def _get_swarm_stat(self):
leader = get_etcd_value("/lain/swarm/docker/swarm/leader")
cmd = [
"timeout", "3", "docker", "-H",
"swarm.lain:%d" % self._swarm_manager_port, "version"
]
is_healthy = 0
is_alive = 0
has_leader = 0
if leader != "":
has_leader = 1
with open("/dev/null", "w") as f:
if subprocess.call(cmd, stdout=f) == 0:
is_alive = 1
cmd[5] = "info"
if subprocess.call(cmd, stdout=f) == 0:
is_healthy = 1
self._result.append(
GraphiteData("lain.cluster.swarm.alive", self._endpoint, is_alive,
self._step, "status"))
self._result.append(
GraphiteData("lain.cluster.swarm.health", self._endpoint,
is_healthy, self._step, "status"))
self._result.append(
GraphiteData("lain.cluster.swarm.leaderkey", self._endpoint,
has_leader, self._step, "status"))
def _get_swarm_stat(self):
leader = get_etcd_value("/lain/swarm/docker/swarm/leader")
cmd = ["timeout", "3", "docker", "-H",
"swarm.lain:%d" % self._swarm_manager_port, "version"]
is_healthy = 0
is_alive = 0
has_leader = 0
if leader != "":
has_leader = 1
with open("/dev/null", "w") as f:
if subprocess.call(cmd, stdout=f) == 0:
is_alive = 1
cmd[5] = "info"
if subprocess.call(cmd, stdout=f) == 0:
is_healthy = 1
self._result.append(
GraphiteData("lain.cluster.swarm.alive", self._endpoint,
is_alive, self._step, "status"))
self._result.append(
GraphiteData("lain.cluster.swarm.health", self._endpoint,
is_healthy, self._step, "status"))
self._result.append(
GraphiteData("lain.cluster.swarm.leaderkey", self._endpoint,
has_leader, self._step, "status"))
def _collect_deployd_debug_info(self):
is_alive = 0
try:
domain = get_etcd_value("/lain/config/domain")
resp = requests.get(
"http://deployd.lain:%d/debug/vars" % self._deployd_port, timeout=1)
if resp.status_code == 200:
is_alive = 1
except Exception:
pass
self._result.append(
GraphiteData("lain.cluster.deployd.alive",
self._endpoint, is_alive, self._step, "status"))
def _collect_deployd_debug_info(self):
is_alive = 0
try:
domain = get_etcd_value("/lain/config/domain")
resp = requests.get("http://deployd.lain:%d/debug/vars" %
self._deployd_port,
timeout=1)
if resp.status_code == 200:
is_alive = 1
except Exception:
pass
self._result.append(
GraphiteData("lain.cluster.deployd.alive", self._endpoint,
is_alive, self._step, "status"))