def _check_machine(self, machine: Machine):
"""Method checks if the agent process inside of the container is running."""
hostname = machine.get_hostname()
logging.debug(f"Checking connection with machine {hostname}")
connection = socket.socket()
connection.settimeout(self._start_timeout)
retry_counter = 5
for i in range(retry_counter):
logging.debug(
f"Connecting to {machine.get_hostname()}, retry counter: {i}")
try:
connection.connect((hostname, machine._port))
except (ConnectionRefusedError, ConnectionAbortedError):
sleep(1)
continue
logging.debug(f"Connected to agent process at machine {hostname}")
break # successfully connected
else:
raise PoolManagerError(f"Could not connect to machine {hostname}")
err = connection.getsockopt(socket.SOL_SOCKET, socket.SO_ERROR)
if err:
connection.close()
raise PoolManagerError(f"Could not connect to machine {hostname}")
connection.shutdown(socket.SHUT_RDWR)
connection.close()
logging.info(f"Agent process is running at {hostname}")
def _create_container(self, name: str, reqs: dict):
logging.info("Creating container " + name)
if "rpc_port" in reqs:
rpc_port = reqs["rpc_port"]
else:
rpc_port = None
self._pool[name] = {
"interfaces": {},
"params": {
"hostname": "",
"rpc_port": rpc_port
},
"security": {
"auth_type": "none"
},
"available": True,
}
try:
container = self._podman_client.containers.create(self.image,
hostname=name,
privileged=True)
except APIError as e:
raise PoolManagerError(f"Could not create container {name}: {e}")
machine = Machine(
name,
"",
self._msg_dispatcher,
self._ctl_config,
None,
rpc_port,
self._pool[name]["security"],
reqs,
) # to get hostname the container needs to run
self._start_container(container, machine)
if self._pool_check:
self._check_machine(
machine) # checks if the agent process is already running
self._pool[name]["params"]["hostname"] = machine.get_hostname()
return container, machine
