负载均衡重构: 子网感知选择 + 节点健康监控 + 节点注册上报API
- 新增 subnet-aware 负载策略: 根据客户端IP匹配最近/16子网节点 - 节点注册端点 (POST /api/nodes/register): 边缘节点自动注册 - 节点心跳端点 (POST /api/nodes/heartbeat): 上报CPU/RAM/带宽/连接数 - 健康汇总 (GET /api/nodes/health): 服务端汇总所有节点状态 - 配置节点新增subnet/is_origin字段 - 前端NodeManager: 实时健康状态/CPU/RAM进度条/带宽/连接数 - X-Forwarded-For/X-Real-IP透传支持客户端IP
This commit is contained in:
admin committed
1 parent
5f19fc6aa5
commit
980b91f41f
7 files changed
+319
-87
No files matched your search
@@ -1,43 +1,140 @@
|
||||
import json
|
||||
import asyncio
|
||||
import subprocess
|
||||
import ipaddress
|
||||
import time
|
||||
from config import settings
|
||||
|
||||
|
||||
class LoadBalancer:
|
||||
def __init__(self):
|
||||
self._rr_index = 0
|
||||
self._conn_counts: dict[str, int] = {n["id"]: 0 for n in settings.edge_nodes}
|
||||
self._conn_counts: dict[str, int] = {}
|
||||
self._node_health: dict[str, dict] = {}
|
||||
|
||||
def get_enabled_nodes(self):
|
||||
return [n for n in settings.edge_nodes if n.get("enabled", True)]
|
||||
|
||||
def select_nodes(self, strategy: str | None = None):
|
||||
nodes = self.get_enabled_nodes()
|
||||
strategy = strategy or settings.load_balance_strategy
|
||||
def get_node_by_id(self, node_id: str):
|
||||
for n in settings.edge_nodes:
|
||||
if n["id"] == node_id:
|
||||
return n
|
||||
return None
|
||||
|
||||
def select_best_node(self, client_ip: str | None = None):
|
||||
nodes = self.get_enabled_nodes()
|
||||
if not nodes:
|
||||
return None
|
||||
|
||||
if client_ip:
|
||||
best = self._find_subnet_match(client_ip, nodes)
|
||||
if best:
|
||||
return best
|
||||
|
||||
strategy = settings.load_balance_strategy
|
||||
if strategy == "round-robin":
|
||||
if not nodes:
|
||||
return []
|
||||
chosen = nodes[self._rr_index % len(nodes)]
|
||||
online = [n for n in nodes if self.is_online(n["id"])]
|
||||
if not online:
|
||||
online = nodes
|
||||
chosen = online[self._rr_index % len(online)]
|
||||
self._rr_index += 1
|
||||
return [chosen]
|
||||
return chosen
|
||||
elif strategy == "weighted":
|
||||
total = sum(n["weight"] for n in nodes)
|
||||
if total == 0:
|
||||
return nodes
|
||||
import random
|
||||
r = random.uniform(0, total)
|
||||
cumulative = 0
|
||||
for n in nodes:
|
||||
cumulative += n["weight"]
|
||||
if r <= cumulative:
|
||||
return [n]
|
||||
return [nodes[-1]]
|
||||
return self._weighted_select(nodes)
|
||||
elif strategy == "least-conn":
|
||||
sorted_nodes = sorted(nodes, key=lambda n: self._conn_counts.get(n["id"], 0))
|
||||
return [sorted_nodes[0]]
|
||||
return nodes
|
||||
return self._least_conn_select(nodes)
|
||||
else:
|
||||
return self._weighted_select(nodes)
|
||||
|
||||
def _find_subnet_match(self, client_ip: str, nodes: list[dict]):
|
||||
try:
|
||||
ip = ipaddress.ip_address(client_ip)
|
||||
except ValueError:
|
||||
return None
|
||||
|
||||
best = None
|
||||
best_prefix = -1
|
||||
|
||||
for node in nodes:
|
||||
if not self.is_online(node["id"]):
|
||||
continue
|
||||
|
||||
subnet_str = node.get("subnet", "0.0.0.0/0")
|
||||
if not subnet_str:
|
||||
continue
|
||||
|
||||
try:
|
||||
subnet = ipaddress.ip_network(subnet_str, strict=False)
|
||||
except ValueError:
|
||||
continue
|
||||
|
||||
if ip in subnet:
|
||||
if subnet.prefixlen > best_prefix:
|
||||
best_prefix = subnet.prefixlen
|
||||
best = node
|
||||
|
||||
return best
|
||||
|
||||
def _weighted_select(self, nodes: list[dict]):
|
||||
online = [n for n in nodes if self.is_online(n["id"])]
|
||||
if not online:
|
||||
online = nodes
|
||||
total = sum(n["weight"] for n in online)
|
||||
if total == 0:
|
||||
return online[0] if online else None
|
||||
import random
|
||||
r = random.uniform(0, total)
|
||||
cumulative = 0
|
||||
for n in online:
|
||||
cumulative += n["weight"]
|
||||
if r <= cumulative:
|
||||
return n
|
||||
return online[-1]
|
||||
|
||||
def _least_conn_select(self, nodes: list[dict]):
|
||||
online = [n for n in nodes if self.is_online(n["id"])]
|
||||
if not online:
|
||||
online = nodes
|
||||
sorted_nodes = sorted(online, key=lambda n: self._conn_counts.get(n["id"], 0))
|
||||
return sorted_nodes[0] if sorted_nodes else None
|
||||
|
||||
def report_health(self, node_id: str, metrics: dict):
|
||||
self._node_health[node_id] = {
|
||||
"last_seen": time.time(),
|
||||
"online": True,
|
||||
**metrics,
|
||||
}
|
||||
|
||||
def is_online(self, node_id: str) -> bool:
|
||||
h = self._node_health.get(node_id)
|
||||
if h is None:
|
||||
if any(n["id"] == node_id and n.get("is_origin") for n in settings.edge_nodes):
|
||||
return True
|
||||
return False
|
||||
if time.time() - h.get("last_seen", 0) > 30:
|
||||
return False
|
||||
return h.get("online", False)
|
||||
|
||||
def get_health_summary(self):
|
||||
result = {}
|
||||
for node in settings.edge_nodes:
|
||||
nid = node["id"]
|
||||
h = self._node_health.get(nid)
|
||||
result[nid] = {
|
||||
"name": node["name"],
|
||||
"host": node["host"],
|
||||
"online": self.is_online(nid),
|
||||
"weight": node["weight"],
|
||||
"enabled": node.get("enabled", True),
|
||||
"subnet": node.get("subnet"),
|
||||
"is_origin": node.get("is_origin", False),
|
||||
"last_seen": h.get("last_seen") if h else None,
|
||||
"metrics": {
|
||||
"cpu_percent": h.get("cpu_percent"),
|
||||
"memory_percent": h.get("memory_percent"),
|
||||
"bandwidth_mbps": h.get("bandwidth_mbps"),
|
||||
"active_connections": h.get("active_connections"),
|
||||
"hls_readers": h.get("hls_readers"),
|
||||
} if h else None,
|
||||
}
|
||||
return result
|
||||
|
||||
def increment_connection(self, node_id: str):
|
||||
self._conn_counts[node_id] = self._conn_counts.get(node_id, 0) + 1
|
||||
|
||||
Reference in new issue
Block a user