fix: Fix up merging request changes and make sure that there is no racing condition errors, and UI always gets correct information.
ref: N25B-256
This commit is contained in:
Björn Otgaar
@@ -56,28 +56,29 @@ class RICommunicationAgent(BaseAgent):
|
||||
"we probably dont have any receivers... but let's check!"
|
||||
)
|
||||
|
||||
# Wait up to {seconds_to_wait_total/2} seconds for a reply:)
|
||||
# Wait up to {seconds_to_wait_total/2} seconds for a reply
|
||||
try:
|
||||
message = await asyncio.wait_for(
|
||||
self.agent._req_socket.recv_json(), timeout=seconds_to_wait_total / 2
|
||||
)
|
||||
|
||||
# We didnt get a reply :(
|
||||
# We didnt get a reply
|
||||
except TimeoutError:
|
||||
self.agent.logger.info(
|
||||
f"No ping retrieved in {seconds_to_wait_total} seconds, "
|
||||
"sending UI disconnection event and soft killing myself."
|
||||
"sending UI disconnection event and attempting to restart."
|
||||
)
|
||||
|
||||
# Make sure we dont retry receiving messages untill we're setup.
|
||||
self.agent.connected = False
|
||||
self.agent.remove_behaviour(self)
|
||||
|
||||
# Tell UI we're disconnected.
|
||||
topic = b"ping"
|
||||
data = json.dumps(False).encode()
|
||||
if self.agent.pub_socket is None:
|
||||
self.agent.logger.error(
|
||||
"communication agent pub socket not correctly initialized."
|
||||
"Communication agent pub socket not correctly initialized."
|
||||
)
|
||||
else:
|
||||
try:
|
||||
@@ -85,17 +86,20 @@ class RICommunicationAgent(BaseAgent):
|
||||
self.agent.pub_socket.send_multipart([topic, data]), 5
|
||||
)
|
||||
except TimeoutError:
|
||||
self.agent.logger.error(
|
||||
self.agent.logger.warning(
|
||||
"Initial connection ping for router timed"
|
||||
" out in ri_communication_agent."
|
||||
)
|
||||
|
||||
# Try to reboot.
|
||||
self.agent.logger.debug("Restarting communication agent.")
|
||||
await self.agent.setup()
|
||||
|
||||
self.agent.logger.debug(f'Received message "{message}" from RI.')
|
||||
if "endpoint" not in message:
|
||||
self.agent.logger.error("No received endpoint in message, excepted ping endpoint.")
|
||||
self.agent.logger.warning(
|
||||
"No received endpoint in message, expected ping endpoint."
|
||||
)
|
||||
return
|
||||
|
||||
# See what endpoint we received
|
||||
@@ -107,7 +111,7 @@ class RICommunicationAgent(BaseAgent):
|
||||
await self.agent.pub_socket.send_multipart([topic, data])
|
||||
await asyncio.sleep(1)
|
||||
case _:
|
||||
self.agent.logger.info(
|
||||
self.agent.logger.debug(
|
||||
"Received message with topic different than ping, while ping expected."
|
||||
)
|
||||
|
||||
@@ -143,16 +147,20 @@ class RICommunicationAgent(BaseAgent):
|
||||
if self._req_socket is None:
|
||||
continue
|
||||
|
||||
# Send our message and receive one back:)
|
||||
# Send our message and receive one back
|
||||
message = {"endpoint": "negotiate/ports", "data": {}}
|
||||
await self._req_socket.send_json(message)
|
||||
|
||||
retry_frequency = 1.0
|
||||
try:
|
||||
received_message = await asyncio.wait_for(self._req_socket.recv_json(), timeout=1.0)
|
||||
received_message = await asyncio.wait_for(
|
||||
self._req_socket.recv_json(), timeout=retry_frequency
|
||||
)
|
||||
|
||||
except TimeoutError:
|
||||
self.logger.warning(
|
||||
"No connection established in 20 seconds (attempt %d/%d)",
|
||||
"No connection established in %d seconds (attempt %d/%d)",
|
||||
retries * retry_frequency,
|
||||
retries + 1,
|
||||
max_retries,
|
||||
)
|
||||
@@ -160,21 +168,21 @@ class RICommunicationAgent(BaseAgent):
|
||||
continue
|
||||
|
||||
except Exception as e:
|
||||
self.logger.error("Unexpected error during negotiation: %s", e)
|
||||
self.logger.warning("Unexpected error during negotiation: %s", e)
|
||||
retries += 1
|
||||
continue
|
||||
|
||||
# Validate endpoint
|
||||
endpoint = received_message.get("endpoint")
|
||||
if endpoint != "negotiate/ports":
|
||||
# TODO: Should this send a message back?
|
||||
self.logger.error(
|
||||
self.logger.warning(
|
||||
"Invalid endpoint '%s' received (attempt %d/%d)",
|
||||
endpoint,
|
||||
retries + 1,
|
||||
max_retries,
|
||||
)
|
||||
retries += 1
|
||||
await asyncio.sleep(1)
|
||||
continue
|
||||
|
||||
# At this point, we have a valid response
|
||||
@@ -194,7 +202,7 @@ class RICommunicationAgent(BaseAgent):
|
||||
if addr != self._address:
|
||||
if not bind:
|
||||
self._req_socket.connect(addr)
|
||||
else: # TODO: Should this ever be the case?
|
||||
else:
|
||||
self._req_socket.bind(addr)
|
||||
case "actuation":
|
||||
ri_commands_agent = RICommandAgent(
|
||||
@@ -210,31 +218,32 @@ class RICommunicationAgent(BaseAgent):
|
||||
self.logger.warning("Unhandled negotiation id: %s", id)
|
||||
|
||||
except Exception as e:
|
||||
self.logger.error("Error unpacking negotiation data: %s", e)
|
||||
self.logger.warning("Error unpacking negotiation data: %s", e)
|
||||
retries += 1
|
||||
await asyncio.sleep(1)
|
||||
continue
|
||||
|
||||
# setup succeeded
|
||||
break
|
||||
|
||||
else:
|
||||
self.logger.error("Failed to set up RICommunicationAgent after %d retries", max_retries)
|
||||
self.logger.error("Failed to set up %s after %d retries", self.name, max_retries)
|
||||
return
|
||||
|
||||
# Set up ping behaviour
|
||||
listen_behaviour = self.ListenBehaviour()
|
||||
self.add_behaviour(listen_behaviour)
|
||||
|
||||
# Let UI know that we're connected >:)
|
||||
# Let UI know that we're connected
|
||||
topic = b"ping"
|
||||
data = json.dumps(True).encode()
|
||||
if self.pub_socket is None:
|
||||
self.logger.error("communication agent pub socket not correctly initialized.")
|
||||
self.logger.error("Communication agent pub socket not correctly initialized.")
|
||||
else:
|
||||
try:
|
||||
await asyncio.wait_for(self.pub_socket.send_multipart([topic, data]), 5)
|
||||
except TimeoutError:
|
||||
self.logger.error(
|
||||
self.logger.warning(
|
||||
"Initial connection ping for router timed out in ri_communication_agent."
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user