added gossip style failure detection

test1
This commit is contained in:
Aryan Singh committed 2026-09-26 00:42:10 -05:00
1 parent 13a2033b5f
commit a63bff32f9
1 file changed
+84 -29
+84 -29
View File
@@ -12,9 +12,11 @@ INTRODUCER_IP = "172.22.156.3"
IP = socket.gethostbyname(socket.gethostname())
SERVER_PORT = 8989
INTRODUCER_FLAG = (INTRODUCER_IP == IP)
HEART_BEAT_INTERVAL = 5
GOSSIP_INTERVAL = 10.0
FAILURE_TIMEOUT = 3.0
HEART_BEAT_INTERVAL = 0.5
GOSSIP_INTERVAL = 0.5
FAILURE_TIMEOUT = 2.5
FAILURE_T_CLEANUP = 10.0
FANOUT = 1
members_list = {}
members_list_lock = threading.Lock()
@@ -22,7 +24,10 @@ members_list_lock = threading.Lock()
SELF_ID = (IP, SERVER_PORT, utc_time())
members_list[SELF_ID] = {
"heart_beat": 1,
"last_seen": utc_time()
"local_last_seen": utc_time(),
"state": "alive",
"local_failed_since" : None,
"local_suspected_since" : None
}
print(f"Starting peer: IP={IP}, port={SERVER_PORT}, introducer={INTRODUCER_FLAG}, self_id={SELF_ID}")
@@ -38,7 +43,6 @@ def worker_1():
"type": "JOIN",
"sender_id": SELF_ID
}
time.sleep(5) #just so the introducer has time to start the socket
server_socket.sendto(json.dumps(join_payload).encode('utf-8'), (INTRODUCER_IP, SERVER_PORT))
while not stop_event.is_set():
@@ -58,7 +62,10 @@ def worker_1():
if join_id not in members_list:
members_list[join_id] = {
"heart_beat": 1,
"last_seen": utc_time()
"local_last_seen": utc_time(),
"state": "alive",
"local_failed_since" : None,
"local_suspected_since" : None
}
print(f"Added member: {join_id}")
@@ -68,19 +75,20 @@ def worker_1():
"ip": m_id[0],
"port": m_id[1],
"timestamp": m_id[2],
"heart_beat": m_data["heart_beat"]
"heart_beat": m_data["heart_beat"],
"state": m_data["state"]
})
print(f"New member joined: {join_id}")
# Send the initial state directly to the joining node - decreases the overhead of having that init state propogate through the cluster
# Send the initial state directly to the joining node
reply_payload = {
"type": "GOSSIP",
"sender_id": SELF_ID,
"members": members_payload
}
server_socket.sendto(json.dumps(reply_payload).encode('utf-8'), (join_ip, join_port))
print(f"Sent initial membership state to {(join_ip, join_port)}: {reply_payload}")
print(f"Sent initial membership state to {(join_ip, join_port)}")
elif msg_type == "GOSSIP":
members_json = msg.get('members', [])
@@ -88,21 +96,39 @@ def worker_1():
with members_list_lock:
for member in members_json:
member_id = (member['ip'], member['port'], member['timestamp'])
member_hb = member['heart_beat']
if member_id == SELF_ID:
continue
incoming_state = member['state']
incoming_hb = member['heart_beat']
if member_id not in members_list:
members_list[member_id] = {
"heart_beat": member_hb,
"last_seen": utc_time()
"heart_beat": incoming_hb,
"local_last_seen": utc_time(),
"state": incoming_state,
"local_failed_since" : utc_time() if incoming_state == "failed" else None,
"local_suspected_since" : None
}
print(f"Added gossip member: {member_id}, heartbeat={member_hb}")
elif member_hb > members_list[member_id]["heart_beat"]:
members_list[member_id]["heart_beat"] = member_hb
members_list[member_id]["last_seen"] = utc_time()
print(f"Updated member: {member_id}, heartbeat={member_hb}")
print(f"Added gossip member: {member_id}, state={incoming_state}")
continue
local_member = members_list[member_id]
# Prevent stale ALIVE messages from resurrecting a node we already marked FAILED
if local_member["state"] == "failed":
continue
# Merge FAILED state independently of heartbeat comparison
if incoming_state == "failed":
local_member["state"] = "failed"
local_member["local_failed_since"] = utc_time()
print(f"Member {member_id} marked as failed due to gossip.")
elif incoming_hb > local_member["heart_beat"]:
local_member["heart_beat"] = incoming_hb
local_member["local_last_seen"] = utc_time()
print(f"Updated member: {member_id}, heartbeat={incoming_hb}")
except socket.timeout:
continue
@@ -117,7 +143,7 @@ def worker_2():
def worker_3():
while not stop_event.is_set():
with members_list_lock:
peers = list(members_list.keys())
peers = [member_id for member_id in members_list.keys() if member_id != SELF_ID and members_list[member_id]["state"] == "alive"]
members_payload = []
for m_id, m_data in members_list.items():
@@ -125,15 +151,15 @@ def worker_3():
"ip": m_id[0],
"port": m_id[1],
"timestamp": m_id[2],
"heart_beat": m_data["heart_beat"]
"heart_beat": m_data["heart_beat"],
"state": m_data["state"]
})
if SELF_ID in peers:
peers.remove(SELF_ID)
if peers:
random_peer = random.choice(peers)
peer_ip, peer_port, _ = random_peer
selected_peers = random.sample(peers, min(FANOUT, len(peers)))
for peer in selected_peers:
peer_ip, peer_port, _ = peer
payload = {
"type": "GOSSIP",
@@ -142,12 +168,40 @@ def worker_3():
}
json_string = json.dumps(payload)
print(f"Sending GOSSIP to {(peer_ip, peer_port)}: {payload}")
print(f"Sending GOSSIP to {(peer_ip, peer_port)}")
server_socket.sendto(json_string.encode('utf-8'), (peer_ip, peer_port))
time.sleep(GOSSIP_INTERVAL)
#only for debuging
def worker_4():
while not stop_event.is_set():
current_time = utc_time()
to_remove = []
with members_list_lock:
for member_id, member_data in members_list.items():
if member_id == SELF_ID:
continue
if member_data["state"] == "alive":
last_seen = member_data["local_last_seen"]
time_since_last_seen = current_time - last_seen
if time_since_last_seen > FAILURE_TIMEOUT:
member_data["state"] = "failed"
member_data["local_failed_since"] = current_time
last_failed_since = member_data.get("local_failed_since")
if last_failed_since is not None and (current_time - last_failed_since) > FAILURE_T_CLEANUP:
to_remove.append(member_id)
for member_id in to_remove:
del members_list[member_id]
print(f"Purged member {member_id} from list.")
time.sleep(0.5)
# Only for debugging
def worker_monitoring():
while not stop_event.is_set():
with members_list_lock:
@@ -161,17 +215,17 @@ def worker_monitoring():
f.write(f"{member_id}: {member_data}\n")
time.sleep(0.5)
t1 = threading.Thread(target=worker_1)
t2 = threading.Thread(target=worker_2)
t3 = threading.Thread(target=worker_3)
#only for debugging
t4 = threading.Thread(target=worker_monitoring)
t4 = threading.Thread(target=worker_4)
t5 = threading.Thread(target=worker_monitoring)
t1.start()
t2.start()
t3.start()
t4.start()
t5.start()
try:
while not stop_event.is_set():
@@ -184,4 +238,5 @@ t1.join()
t2.join()
t3.join()
t4.join()
t5.join()
server_socket.close()