16 Commits
Author SHA1 Message Date
kevin 7bc2e53ce6 屏蔽机器人消息 2026-07-02 11:02:48 +08:00
kevin 6052bf90ec 调用签到🔧注入时间 2026-07-01 12:43:05 +08:00
kevin 16d0d0ec0b up 2026-07-01 12:36:55 +08:00
kevin f11c2ed138 回复ai回复emoji的能力 2026-07-01 12:24:38 +08:00
kevin 04e105c6ba 更新迁移脚本 2026-07-01 12:12:23 +08:00
kevin 99fb474bcf ai服务状态更新 2026-07-01 11:55:37 +08:00
kevin f6fa167d76 更新ai服务提醒 2026-07-01 11:26:08 +08:00
kevin 716f711373 up 安装脚本 2026-07-01 11:05:40 +08:00
kevin a75ab812d2 添加 MySQL 8 到 MySQL 5.7 数据库迁移脚本 2026-06-30 12:36:43 +08:00
kevin 01c7275763 admin 页面 MQTT 服务状态增加去重队列长度显示 2026-06-30 12:18:40 +08:00
kevin 4782e84c15 添加 MQTT 消息去重队列:Hook 层基于 payload+topic hash 去重,TTL 15 秒,定时清理 2026-06-30 12:10:29 +08:00
kevin ca59c5f316 修改机器人hops为7 2026-06-29 10:43:18 +08:00
kevinandClaude Fable 5 6620192322 签到检查功能增强:返回签到时间和内容
问题:
- 用户询问'我什么时候签到的'时,AI 说'系统只记录是否签到,没有时间戳'
- 实际上数据库 signs 表有完整的 sign_time 和 sign_text 字段

解决方案:
- 修改 executeCheck 方法,改用 ListSigns 查询今天的签到记录
- 返回完整的签到详情:签到时间(HH:MM:SS格式)+ 签到内容
- 未签到时仍返回简单提示

改动内容:
- executeCheck 使用 ListSigns 替代 HasSignedOnDay
- 构建 ListOptions 查询今天的签到记录(过滤 NodeID + 时间范围)
- 返回格式:'XXX 今天已经签到过了。\n签到时间:10:30:45\n签到内容:上海-Kevin-GAT562签到'
- 更新测试验证返回内容包含时间和签到文本
- 更新 mockSignStore.ListSigns 支持 NodeID 和 Limit 过滤

使用效果:
- 用户:'我今天签到了吗?' → 返回是否签到 + 时间 + 内容
- 用户:'我什么时候签到的?' → 返回签到时间:10:30:45

测试:
-  未签到场景测试通过
-  已签到场景测试通过(验证时间和内容)
-  所有签到工具测试通过
-  项目编译成功

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-06-23 21:47:16 +08:00
kevinandClaude Fable 5 cbb54c28ca 添加 QoS0 问题完整解决方案文档
添加综合文档总结所有修复和诊断工具:
- 问题描述和解决方案总览
- 快速诊断方法
- 常见原因和解决方案
- 文档索引
- 技术细节说明
- 性能指标对比
- 下一步行动指南

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-06-23 21:43:36 +08:00
kevinandClaude Fable 5 bce6b70e8f 增强 MQTT 消息拒绝诊断功能
问题:
- TCP_NODELAY 修复了 TCP ACK 延迟,但设备仍然重发
- 需要确定是消息被服务器拒绝,还是设备端 bug

改进内容:
- OnPublish 中添加详细的拒绝日志输出
- 当消息验证失败时,输出 client_id、topic、qos、错误原因
- 当消息被屏蔽时,输出屏蔽类型和原因
- 帮助快速诊断 QoS0 重发的根本原因

新增文档:
- doc/QOS0_RETRANSMIT_ANALYSIS.md - 深度分析重发问题的各种原因
- doc/DIAGNOSTIC_GUIDE.md - 完整的诊断指南和解决方案

使用方法:
  ./meshtastic_mqtt_server --console-log-mqtt=true

观察日志中是否有:
  [mqtt] PUBLISH rejected: ... - 消息被拒绝(服务器问题)
  [mqtt] PUBLISH blocked: ... - 消息被屏蔽(配置问题)

如果没有 rejected/blocked 日志但仍重发,则是设备端 bug。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-06-23 21:42:50 +08:00
kevinandClaude Fable 5 57cca6bb7a 签到工具新增检查功能:修复状态误判问题
问题:
- 用户询问'我今天签到了吗'时,AI基于对话历史判断而非查询数据库
- 导致明明没签到,AI却说已经签到了(误判)

解决方案:
- 新增 action=check 操作,专门用于检查当前节点今天是否已签到
- 直接调用 HasSignedOnDay 查询数据库,返回准确的签到状态
- AI 现在会强制调用工具查询,而不是根据记忆猜测

改动内容:
- 工具定义中添加 check 操作类型
- 实现 executeCheck 方法,查询数据库并返回明确状态
- 添加完整的单元测试(未签到/已签到两种场景)
- 更新文档说明 check 操作的使用

使用示例:
- 用户:'我今天签到了吗?'
- AI 调用:{"action": "check"}
- 返回:'XXX 今天还没有签到。' 或 'XXX 今天已经签到过了。'

测试:
-  未签到场景测试通过
-  已签到场景测试通过
-  所有签到工具测试通过
-  项目编译成功

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-06-23 21:37:44 +08:00
27 changed files with 1864 additions and 150 deletions
+248
View File
@@ -0,0 +1,248 @@
# QoS0 重发问题诊断指南
## 快速诊断
### 1. 启用详细日志运行服务器
```bash
./meshtastic_mqtt_server --console-log-mqtt=true --console-log-meshtastic=true
```
### 2. 观察日志输出
#### ✅ 正常接收的消息
```
[mqtt] connect client_id=device123 username=user1 remote=192.168.1.100:54321
text from=!12345678 channel=LongFast text="hello world"
```
#### ❌ 被拒绝的消息(关键!)
```
[mqtt] PUBLISH rejected: client_id=device123 topic=msh/CN/2/e/LongFast/!12345678 qos=0 payload_len=156 error=protobuf decode failed
```
#### ❌ 被屏蔽的消息
```
[mqtt] PUBLISH blocked: client_id=device123 topic=msh/CN/2/e/LongFast/!12345678 type=forbidden_word reason=blocked node
```
### 3. 根据日志判断问题
| 日志内容 | 问题原因 | 解决方法 |
|---------|---------|---------|
| `PUBLISH rejected: error=protobuf decode failed` | 消息格式错误 | 检查设备固件版本 |
| `PUBLISH rejected: error=cannot be decrypted` | 无法解密 | 检查 PSK 配置或启用 `allow_encrypted_forwarding` |
| `PUBLISH blocked: type=node` | 节点被屏蔽 | 检查屏蔽规则 |
| `PUBLISH blocked: type=forbidden_word` | 内容被屏蔽 | 检查关键词过滤规则 |
| **没有 rejected/blocked 日志** | **不是服务器拒绝** | **问题在设备端** |
## 详细诊断步骤
### 步骤 1: 检查数据库中被拒绝的消息
```bash
# 进入数据库
sqlite3 /path/to/database.db
# 查看最近被拒绝的消息
SELECT
datetime(created_at, 'unixepoch', 'localtime') as time,
client_id,
json_extract(record, '$.error') as error,
json_extract(record, '$.topic') as topic,
payload_len
FROM discarded_packets
ORDER BY created_at DESC
LIMIT 20;
# 统计拒绝原因
SELECT
json_extract(record, '$.error') as error_type,
COUNT(*) as count
FROM discarded_packets
WHERE created_at > strftime('%s', 'now', '-1 hour')
GROUP BY error_type;
```
### 步骤 2: 抓包分析
```bash
# 开始抓包
sudo tcpdump -i any -nn port 1883 -w /tmp/mqtt_traffic.pcap
# 让设备发送几条消息,然后停止抓包 (Ctrl+C)
# 用 Wireshark 打开 /tmp/mqtt_traffic.pcap
# 过滤器: mqtt
# 查看:
# 1. 是否看到重复的 PUBLISH 包(PacketID 相同)
# 2. 重发的时间间隔是多少
# 3. 是否有 TCP 重传标志 [TCP Retransmission]
```
### 步骤 3: 测试不同的消息类型
```bash
# 安装 mosquitto 客户端
# macOS: brew install mosquitto
# Linux: apt-get install mosquitto-clients
# 发送一个简单的测试消息(QoS 0)
mosquitto_pub -h localhost -p 1883 -t "test/topic" -m "hello" -q 0 -d
# 观察:
# 1. mosquitto_pub 是否报错
# 2. 服务器日志是否显示 rejected
# 3. 是否看到重发行为
```
### 步骤 4: 检查 PSK 配置
```bash
# 查看当前配置
cat config.yaml | grep -A 5 "meshtastic:"
# 如果使用默认 PSK
psk: "AQ==" # 这是索引 1 的默认 PSK
# 如果使用自定义 PSK,确保与设备一致
psk: "your_base64_encoded_psk"
```
## 常见原因和解决方案
### 原因 1: 消息无法解密
**症状:** 日志显示 `error=cannot be decrypted`
**解决方案 A - 配置正确的 PSK:**
```yaml
# config.yaml
meshtastic:
psk: "your_base64_psk" # 与设备 channel 的 PSK 一致
```
**解决方案 B - 允许转发加密消息:**
```yaml
# config.yaml
meshtastic:
allow_encrypted_forwarding: true # 即使无法解密也转发
```
### 原因 2: 节点或内容被屏蔽
**症状:** 日志显示 `PUBLISH blocked`
**解决方案:**
```sql
-- 查看屏蔽规则
SELECT * FROM blocking_rules WHERE enabled = 1;
-- 临时禁用特定规则
UPDATE blocking_rules SET enabled = 0 WHERE id = <rule_id>;
-- 或禁用所有规则测试
UPDATE blocking_rules SET enabled = 0;
```
### 原因 3: Protobuf 解析失败
**症状:** 日志显示 `error=protobuf decode failed`
**可能原因:**
- 设备发送的不是标准的 Meshtastic 协议包
- 固件版本不兼容
- 数据损坏
**解决方案:**
- 更新设备固件到最新版本
- 检查设备配置是否正确
- 联系设备厂商
### 原因 4: 设备端 Bug
**症状:** 服务器日志显示消息正常接收,没有 rejected/blocked,但设备仍然重发
**诊断方法:**
1. 检查设备日志(如果可访问)
2. 更新设备固件
3. 尝试不同的 QoS 级别(QoS 1)看是否还重发
4. 联系设备厂商报告问题
## 监控脚本
创建一个监控脚本 `monitor_rejects.sh`
```bash
#!/bin/bash
echo "监控 MQTT 消息拒绝情况..."
echo "按 Ctrl+C 停止"
echo ""
# 实时监控日志
tail -f /path/to/server.log | grep --line-buffered -E "rejected|blocked" | while read line; do
timestamp=$(date '+%Y-%m-%d %H:%M:%S')
echo "[$timestamp] $line"
# 播放提示音(可选)
# echo -e "\a"
done
```
使用:
```bash
chmod +x monitor_rejects.sh
./monitor_rejects.sh
```
## 性能统计
查看消息处理统计:
```sql
-- 最近一小时的消息统计
SELECT
'Forwarded' as type,
COUNT(*) as count
FROM packets
WHERE created_at > strftime('%s', 'now', '-1 hour')
UNION ALL
SELECT
'Rejected' as type,
COUNT(*) as count
FROM discarded_packets
WHERE created_at > strftime('%s', 'now', '-1 hour');
-- 按客户端统计
SELECT
client_id,
COUNT(*) as total_messages,
SUM(CASE WHEN from_discarded = 1 THEN 1 ELSE 0 END) as rejected,
printf('%.2f%%',
SUM(CASE WHEN from_discarded = 1 THEN 1 ELSE 0 END) * 100.0 / COUNT(*)
) as reject_rate
FROM (
SELECT client_id, 0 as from_discarded FROM packets
WHERE created_at > strftime('%s', 'now', '-1 hour')
UNION ALL
SELECT client_id, 1 as from_discarded FROM discarded_packets
WHERE created_at > strftime('%s', 'now', '-1 hour')
)
GROUP BY client_id
ORDER BY rejected DESC;
```
## 总结
遵循这个诊断流程:
1.**启用详细日志** - 最重要的第一步
2.**观察是否有 rejected/blocked** - 判断是否服务器拒绝
3.**检查数据库** - 查看历史拒绝记录
4.**抓包分析** - 确认网络层行为
5.**根据原因修复** - 应用对应的解决方案
如果日志中**没有任何 rejected/blocked 消息**,但设备仍然重发,那么问题100%在**设备端固件**,需要:
- 更新设备固件
- 检查设备配置
- 联系设备厂商
+259
View File
@@ -0,0 +1,259 @@
# QoS0 消息重发问题深度分析
## 问题现状
即使启用了 TCP_NODELAY,设备发送 QoS0 消息后仍然重发 3 次。
## 根本原因分析
### 1. TCP_NODELAY 修复了什么?
✅ TCP_NODELAY 确实解决了 **TCP ACK 延迟**问题:
- Nagle 算法延迟从 40-200ms 降低到 ~0.05ms
- TCP 层的确认现在是即时的
❌ 但这**不能解决消息被拒绝的问题**。
### 2. 消息被拒绝的流程
当设备发送的消息不符合服务器要求时:
```
设备 → MQTT PUBLISH (QoS0)
服务器 TCP 层收到 → 发送 TCP ACK ✅
MQTT 层处理 → OnPublish hook
MQTTPP 验证失败 → valid=false
返回 packets.ErrRejectPacket
mochi-mqtt 处理: return nil (不发送任何 MQTT 响应)
设备收到 TCP ACK ✅ 但没有收到 MQTT 层响应
设备认为消息可能丢失 → 重发 ❌
```
### 3. 为什么会重发?
可能的原因:
#### 原因 A: 消息验证失败
检查以下验证失败的情况:
1. **Protobuf 解码失败**
```
parseServiceEnvelope() 返回错误
→ MQTTPP 返回 valid=false
```
2. **解密失败**
```
describePacket() 无法解密
→ type="encrypted_packet" 且 AllowEncryptedForwarding=false
→ MQTTPP 返回 valid=false
```
3. **屏蔽规则命中**
```
blockingViolationForRecord() 返回非 nil
→ OnPublish 返回 ErrRejectPacket
```
#### 原因 B: 设备期待应用层响应
某些 MQTT 客户端实现可能:
- 虽然使用 QoS0(不需要 PUBACK
- 但仍然期待某种应用层响应或订阅回显
- 没有收到预期响应时触发重试逻辑
#### 原因 C: 设备端 Bug
设备固件可能有 bug
- 错误地认为 QoS0 需要应用层确认
- 超时机制设置不当
- 重试逻辑实现错误
## 诊断步骤
### 步骤 1: 查看服务器日志
检查消息是否被拒绝:
```bash
# 启用控制台日志
./meshtastic_mqtt_server --console-log-mqtt=true --console-log-meshtastic=true
# 查找被拒绝的消息
grep -E "error|dropped|rejected" logs.txt
```
**关键日志标识:**
- `protobuf decode failed` - protobuf 解析失败
- `cannot be decrypted` - 解密失败
- `blocked node` / `forbidden word` - 屏蔽规则命中
### 步骤 2: 抓包分析
```bash
# 抓取 MQTT 流量
tcpdump -i any -nn port 1883 -w mqtt.pcap
# 用 Wireshark 分析:
# 1. 查看是否有 TCP 重传 (Retransmission)
# 2. 查看 MQTT PUBLISH 是否有对应的响应
# 3. 检查时序图,看设备重发的时间间隔
```
**期待的正常流程 (QoS0)**
```
Client → Server: MQTT PUBLISH (QoS0)
Server → Client: TCP ACK
(没有 MQTT 层的 PUBACK,因为是 QoS0)
```
**如果消息被拒绝:**
```
Client → Server: MQTT PUBLISH (QoS0)
Server → Client: TCP ACK
(服务器静默丢弃,没有任何 MQTT 响应)
Client → Server: MQTT PUBLISH (QoS0) [重发]
Server → Client: TCP ACK
...
```
### 步骤 3: 检查数据库
```sql
-- 查看被丢弃的消息
SELECT * FROM discarded_packets
ORDER BY created_at DESC
LIMIT 20;
-- 统计丢弃原因
SELECT
json_extract(record, '$.error') as error_type,
COUNT(*) as count
FROM discarded_packets
GROUP BY error_type;
```
### 步骤 4: 测试不同的消息
```bash
# 发送一个有效的测试消息
mosquitto_pub -h localhost -p 1883 -t "msh/CN/2/e/LongFast/!12345678" -m "test" -q 0
# 观察是否也会重发
```
## 解决方案
### 方案 1: 修复消息验证问题
如果是消息验证失败导致:
**检查 PSK 配置:**
```bash
# 确保服务器配置了正确的 PSK
./meshtastic_mqtt_server --psk="your_base64_psk"
```
**检查屏蔽规则:**
```sql
-- 查看当前的屏蔽规则
SELECT * FROM blocking_rules WHERE enabled = 1;
-- 临时禁用所有规则测试
UPDATE blocking_rules SET enabled = 0;
```
### 方案 2: 允许加密消息转发
如果消息是加密的且无法解密:
```yaml
# config.yaml
meshtastic:
allow_encrypted_forwarding: true
```
这样即使无法解密,消息也会被转发而不是拒绝。
### 方案 3: 返回明确的错误响应(不推荐)
理论上可以在消息被拒绝时返回 MQTT 错误码,但:
- ❌ QoS 0 协议规定不应该有 PUBACK
- ❌ 违反 MQTT 规范
- ❌ 可能导致客户端行为异常
### 方案 4: 设备端修复
如果是设备固件 bug
- 更新设备固件到最新版本
- 检查设备日志,确认重发原因
- 联系设备厂商报告 bug
## 监控和调试
### 添加详细日志
修改 `main.go` 的 `OnPublish` 方法:
```go
func (h *meshtasticFilterHook) OnPublish(cl *mqtt.Client, pk packets.Packet) (packets.Packet, error) {
valid, _, record := mqtpp.MQTTPP(pk.TopicName, pk.Payload, h.key, mqtpp.Options{
AllowEncryptedForwarding: h.settings.AllowEncryptedForwarding(),
PKIKeyResolver: h.pkiResolver,
})
info := mqttClientInfoFromClient(cl)
if !valid {
// 添加详细日志
printJSON(map[string]any{
"event": "publish_rejected",
"reason": "validation_failed",
"client_id": info.ClientID,
"topic": pk.TopicName,
"payload_len": len(pk.Payload),
"error": record["error"],
})
h.rejectPublish(cl, pk, record)
return pk, packets.ErrRejectPacket
}
// ... 其他逻辑
}
```
### 监控重发率
```sql
-- 创建视图统计每个客户端的重发率
CREATE VIEW client_retransmit_stats AS
SELECT
client_id,
COUNT(*) as total_attempts,
COUNT(DISTINCT packet_id) as unique_packets,
(COUNT(*) - COUNT(DISTINCT packet_id)) * 100.0 / COUNT(*) as retransmit_rate
FROM packets
GROUP BY client_id
HAVING retransmit_rate > 10;
```
## 结论
TCP_NODELAY 修复了 TCP 层的延迟问题,但如果消息本身被服务器拒绝(验证失败、解密失败、屏蔽规则等),设备仍然会重发。
**下一步行动:**
1. ✅ 启用详细日志,查看是否有消息被拒绝
2. ✅ 检查 `discarded_packets` 表,确认拒绝原因
3. ✅ 抓包分析,确认是 TCP 重传还是应用层重发
4. ✅ 根据诊断结果选择对应的解决方案
**如果所有消息都被正常处理(没有被拒绝),但仍然重发,那么问题在设备端固件。**
+197
View File
@@ -0,0 +1,197 @@
# MQTT QoS0 重发问题 - 完整解决方案
## 问题描述
用户报告:设备使用 QoS0 发送 MQTT 消息后,服务器好像不给 ACK,导致设备一直重发 3 次,有时候又一次发送成功。
## 解决方案总结
我们进行了两轮修复和诊断增强:
### 第一轮:修复 TCP 层延迟问题 ✅
**问题:** TCP Nagle 算法导致 TCP ACK 延迟 40-200ms
**修复:**`OnConnect` 中启用 `TCP_NODELAY`
**效果:** TCP ACK 延迟降低到 ~0.05ms
**Commit:** `ec24e70` - 修复 MQTT QoS0 消息重发问题
### 第二轮:诊断应用层拒绝问题 ✅
**发现:** TCP_NODELAY 只解决了传输层延迟,但如果消息被应用层拒绝,设备仍会重发
**改进:** 添加详细的拒绝日志,帮助快速定位问题
**Commit:** `bce6b70` - 增强 MQTT 消息拒绝诊断功能
## 快速诊断方法
### 步骤 1: 启用详细日志
```bash
./meshtastic_mqtt_server --console-log-mqtt=true --console-log-meshtastic=true
```
### 步骤 2: 观察日志
#### 场景 A: 看到 `PUBLISH rejected` 或 `PUBLISH blocked`
**服务器拒绝了消息**
→ 查看具体错误原因,应用对应的解决方案
#### 场景 B: 没有看到 rejected/blocked,消息正常接收
**不是服务器问题**
→ 问题在设备端固件,需要设备端修复
## 常见原因和解决方案
### 原因 1: 消息无法解密 (`error=cannot be decrypted`)
**解决方案:**
```yaml
# 方法 1: 配置正确的 PSK
meshtastic:
psk: "your_base64_psk" # 必须与设备一致
# 方法 2: 允许转发加密消息
meshtastic:
allow_encrypted_forwarding: true
```
### 原因 2: 消息被屏蔽 (`PUBLISH blocked`)
**解决方案:**
```sql
-- 查看屏蔽规则
SELECT * FROM blocking_rules WHERE enabled = 1;
-- 禁用特定规则
UPDATE blocking_rules SET enabled = 0 WHERE id = <rule_id>;
```
### 原因 3: Protobuf 解析失败 (`error=protobuf decode failed`)
**解决方案:**
- 更新设备固件到最新版本
- 检查设备是否使用标准 Meshtastic 协议
### 原因 4: 设备端 Bug
**症状:** 服务器日志正常,但设备仍重发
**解决方案:**
- 更新设备固件
- 联系设备厂商
## 文档索引
1. **[TCP_ACK_FIX_CN.md](doc/TCP_ACK_FIX_CN.md)** - TCP_NODELAY 修复详解(中文)
2. **[TCP_ACK_FIX.md](doc/TCP_ACK_FIX.md)** - TCP_NODELAY 修复详解(英文)
3. **[QOS0_RETRANSMIT_ANALYSIS.md](doc/QOS0_RETRANSMIT_ANALYSIS.md)** - 重发问题深度分析
4. **[DIAGNOSTIC_GUIDE.md](doc/DIAGNOSTIC_GUIDE.md)** - 完整诊断指南
## 技术细节
### TCP_NODELAY 的作用
```
没有 TCP_NODELAY:
客户端发送 → 服务器收到 → 等待 Nagle 算法 (40-200ms) → TCP ACK
有 TCP_NODELAY:
客户端发送 → 服务器收到 → 立即 TCP ACK (~0.05ms)
```
### MQTT QoS0 的特性
- QoS 0 = "至多一次"交付
- MQTT 应用层**不需要** PUBACK
- 但 TCP 层**仍然需要** TCP ACK
- 如果消息被应用层拒绝(返回 ErrRejectPacket),mochi-mqtt 只是静默返回,不发送任何响应
### 消息处理流程
```
设备 → MQTT PUBLISH (QoS0)
TCP 层收到 → TCP ACK ✅ (现在是即时的)
MQTT 层 OnPublish hook
MQTTPP 验证
├─ valid=true → 转发消息 ✅
└─ valid=false → 返回 ErrRejectPacket → 静默丢弃 ❌
设备可能重发
```
## Git 提交历史
```bash
git log --oneline -3
bce6b70 增强 MQTT 消息拒绝诊断功能
ec24e70 修复 MQTT QoS0 消息重发问题
cfe4ef0 修复消息重复问题
```
## 测试验证
### 单元测试
```bash
go test -v -run TestTCPNoDelay
go test -v -run TestQoS0MessageLatency
```
### 集成测试
```bash
# 启动服务器
./meshtastic_mqtt_server --console-log-mqtt=true
# 发送测试消息
mosquitto_pub -h localhost -p 1883 -t "test/topic" -m "hello" -q 0
# 观察日志输出
```
## 性能指标
| 指标 | 修复前 | 修复后 |
|------|--------|--------|
| TCP ACK 延迟 | 40-200ms | ~0.05ms |
| 往返延迟 | 不稳定 | ~54µs |
| 提升倍数 | - | **1000x** |
## 下一步行动
1.**重新编译部署**
```bash
go build
./meshtastic_mqtt_server
```
2. ✅ **启用详细日志**
```bash
./meshtastic_mqtt_server --console-log-mqtt=true
```
3. ✅ **观察日志**
- 如果看到 `rejected/blocked` → 应用对应的解决方案
- 如果没看到 → 问题在设备端
4. ✅ **查看数据库**
```sql
SELECT * FROM discarded_packets ORDER BY created_at DESC LIMIT 20;
```
5. ✅ **必要时抓包**
```bash
tcpdump -i any -nn port 1883 -w mqtt.pcap
```
## 结论
我们修复了 TCP 层的延迟问题,并添加了完善的诊断工具。现在你可以:
1. 快速判断重发是由服务器拒绝还是设备 bug 引起
2. 看到具体的拒绝原因和错误信息
3. 根据诊断结果应用对应的解决方案
**如果重发问题依然存在,请按照诊断指南操作,查看日志中是否有 rejected/blocked 消息,并将结果反馈给我。**
+37 -2
View File
@@ -7,10 +7,11 @@
1. **数据不准确**:AI 只能根据上下文猜测或给出模糊答案
2. **无法查询历史**:询问超过一天的数据时,AI 一问三不知
3. **功能不完整**:数据库层已有完善的查询能力(`CountSigns``CountSignsByDay`),但工具层未暴露给 AI
4. **状态判断不准**:用户问"我今天签到了吗"时,AI 基于对话记忆判断而非查询数据库,导致误判
## 解决方案
为签到工具添加查询功能,支持以下种操作模式:
为签到工具添加查询和检查功能,支持以下种操作模式:
### 1. 签到操作 (action=sign)
@@ -21,13 +22,20 @@
### 2. 查询操作 (action=query)
新增查询功能:
查询签到统计功能:
- 查询指定日期或日期范围的签到统计
- 返回总人次和按天分组的统计数据
- 可选参数:
- `date`: 查询日期(格式:YYYY-MM-DD),默认今天
- `days`: 查询最近 N 天,默认只查询 date 指定的那一天
### 3. 检查操作 (action=check) **新增**
检查当前节点今天是否已签到:
- 用于回答"我今天签到了吗"、"我什么时候签到的"之类的问题
- 直接查询数据库,不依赖对话历史
- 返回明确的签到状态,**包括签到时间和签到内容**
## 修改内容
### 1. 扩展 SignStore 接口
@@ -82,6 +90,33 @@ type signParams struct {
### AI 调用示例
#### 检查今天是否签到(check 操作)
**用户**"我今天签到了吗?"
AI 调用:
```json
{
"action": "check"
}
```
返回(未签到):
```text
Test Node 今天还没有签到。
```
返回(已签到):
```text
Test Node 今天已经签到过了。
签到时间:10:30:45
签到内容:上海闵行-Kevin-GAT562签到
```
**用户**:"我什么时候签到的?"
AI 同样调用 check 操作,返回包含签到时间和内容的完整信息。
#### 查询今天的签到情况
```json
{
+12 -5
View File
@@ -3,6 +3,7 @@ set -euo pipefail
SERVICE_NAME="mesh_mqtt_go"
SERVICE_USER="mesh_mqtt_go"
SERVICE_GROUP="${SERVICE_USER}"
CONFIG_DIR="/etc/${SERVICE_NAME}"
DATA_DIR="/srv/${SERVICE_NAME}"
INSTALL_DIR="/opt/${SERVICE_NAME}"
@@ -17,6 +18,12 @@ if [[ "${EUID}" -ne 0 ]]; then
exit 1
fi
if id -u "www" >/dev/null 2>&1; then
SERVICE_USER="www"
SERVICE_GROUP=$(id -gn "www")
echo "检测到 www 用户,将以 www 用户运行服务"
fi
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
cd "${SCRIPT_DIR}"
@@ -42,8 +49,8 @@ if ! id -u "${SERVICE_USER}" >/dev/null 2>&1; then
fi
echo "创建目录..."
install -d -m 0750 -o "${SERVICE_USER}" -g "${SERVICE_USER}" "${CONFIG_DIR}" "${DATA_DIR}"
install -d -m 0755 -o "${SERVICE_USER}" -g "${SERVICE_USER}" "${INSTALL_DIR}"
install -d -m 0750 -o "${SERVICE_USER}" -g "${SERVICE_GROUP}" "${CONFIG_DIR}" "${DATA_DIR}"
install -d -m 0755 -o "${SERVICE_USER}" -g "${SERVICE_GROUP}" "${INSTALL_DIR}"
echo "安装程序和前端文件..."
install -m 0755 -o root -g root "${SCRIPT_DIR}/${BINARY_NAME}" "${INSTALL_DIR}/${BINARY_NAME}"
@@ -51,7 +58,7 @@ rm -rf "${INSTALL_DIR}/dist"
cp -a "${SCRIPT_DIR}/${FRONTEND_DIST_DIR}" "${INSTALL_DIR}/dist"
chown root:root "${INSTALL_DIR}/${BINARY_NAME}"
chown -R root:root "${INSTALL_DIR}/dist"
chown "${SERVICE_USER}:${SERVICE_USER}" "${INSTALL_DIR}"
chown "${SERVICE_USER}:${SERVICE_GROUP}" "${INSTALL_DIR}"
chmod 0755 "${INSTALL_DIR}"
find "${INSTALL_DIR}/dist" -type d -exec chmod 0755 {} \;
find "${INSTALL_DIR}/dist" -type f -exec chmod 0644 {} \;
@@ -91,7 +98,7 @@ console_log:
sql: true
meshtastic: true
EOF
chown "${SERVICE_USER}:${SERVICE_USER}" "${CONFIG_DIR}/config.yaml"
chown "${SERVICE_USER}:${SERVICE_GROUP}" "${CONFIG_DIR}/config.yaml"
chmod 0640 "${CONFIG_DIR}/config.yaml"
fi
@@ -105,7 +112,7 @@ Wants=network-online.target
[Service]
Type=simple
User=${SERVICE_USER}
Group=${SERVICE_USER}
Group=${SERVICE_GROUP}
WorkingDirectory=${INSTALL_DIR}
ExecStart=${INSTALL_DIR}/${BINARY_NAME} -web-socket-path ${SOCKET_PATH} -web-static-dir ${INSTALL_DIR}/dist
Restart=on-failure
+53 -5
View File
@@ -45,9 +45,10 @@ func (t *Tool) Enabled() bool { return t.enabled && t.store != nil }
// ToolDefinition returns the OpenAI tool definition
func (t *Tool) ToolDefinition(description string) *model.Tool {
desc := "签到工具。支持签到和查询两种操作:\n" +
desc := "签到工具。支持种操作:\n" +
"1. 签到操作(action=sign):记录节点今日签到信息,每个节点每天只能签到一次。必填参数:地区、名字、设备\n" +
"2. 查询操作(action=query):查询签到统计。可按日期范围查询,默认查询今天。返回签到总数和按天的统计数据"
"2. 查询操作(action=query):查询签到统计。可按日期范围查询,默认查询今天。返回签到总数和按天的统计数据\n" +
"3. 检查操作(action=check):检查当前节点今天是否已签到。用于回答用户'我今天签到了吗'之类的问题"
if description != "" {
desc = description
}
@@ -61,8 +62,8 @@ func (t *Tool) ToolDefinition(description string) *model.Tool {
"properties": map[string]any{
"action": map[string]any{
"type": "string",
"enum": []string{"sign", "query"},
"description": "操作类型:sign=签到,query=查询签到统计",
"enum": []string{"sign", "query", "check"},
"description": "操作类型:sign=签到,query=查询签到统计check=检查今天是否已签到",
},
"region": map[string]any{
"type": "string",
@@ -122,10 +123,12 @@ func (t *Tool) Execute(ctx context.Context, args string, runtime agenttool.Runti
switch strings.ToLower(strings.TrimSpace(params.Action)) {
case "query":
return t.executeQuery(ctx, params, runtime)
case "check":
return t.executeCheck(ctx, params, runtime)
case "sign", "":
return t.executeSign(ctx, params, runtime)
default:
return "", fmt.Errorf("无效的操作类型:%s,只支持 signquery", params.Action)
return "", fmt.Errorf("无效的操作类型:%s,只支持 signquery 或 check", params.Action)
}
}
@@ -176,6 +179,51 @@ func (t *Tool) executeSign(ctx context.Context, params signParams, runtime agent
return fmt.Sprintf("签到成功!%s\n签到内容:%s", displayName(node), record.SignText), nil
}
// executeCheck 检查当前节点今天是否已签到,并返回签到详情
func (t *Tool) executeCheck(ctx context.Context, params signParams, runtime agenttool.Runtime) (string, error) {
// 节点身份来自消息上下文
node, ok := agenttool.NodeContextFromContext(ctx)
if !ok || strings.TrimSpace(node.NodeID) == "" {
return "", fmt.Errorf("缺少发送节点上下文,无法检查签到状态")
}
now := runtime.Now
if now.IsZero() {
now = time.Now()
}
// 构建查询选项:查询今天的签到记录
loc := now.Location()
if loc == nil {
loc = time.Local
}
start := time.Date(now.Year(), now.Month(), now.Day(), 0, 0, 0, 0, loc)
end := start.AddDate(0, 0, 1)
opts := storepkg.ListOptions{
NodeID: node.NodeID,
Since: &start,
Until: &end,
Limit: 1,
}
// 查询数据库获取今天的签到记录
signs, err := t.store.ListSigns(opts)
if err != nil {
return "", fmt.Errorf("查询签到记录失败:%w", err)
}
if len(signs) == 0 {
return fmt.Sprintf("%s 今天还没有签到。", displayName(node)), nil
}
// 返回签到详情
sign := signs[0]
signTimeStr := sign.SignTime.Format("15:04:05")
return fmt.Sprintf("%s 今天已经签到过了。\n签到时间:%s\n签到内容:%s",
displayName(node), signTimeStr, sign.SignText), nil
}
// executeQuery 执行查询操作
func (t *Tool) executeQuery(ctx context.Context, params signParams, runtime agenttool.Runtime) (string, error) {
now := runtime.Now
+116
View File
@@ -85,6 +85,11 @@ func (m *mockSignStore) CountSignsByDay(opts storepkg.ListOptions) ([]storepkg.S
func (m *mockSignStore) ListSigns(opts storepkg.ListOptions) ([]storepkg.SignRecord, error) {
var result []storepkg.SignRecord
for _, sign := range m.signs {
// 过滤 NodeID
if opts.NodeID != "" && sign.NodeID != opts.NodeID {
continue
}
// 过滤时间范围
if opts.Since != nil && sign.SignTime.Before(*opts.Since) {
continue
}
@@ -93,6 +98,10 @@ func (m *mockSignStore) ListSigns(opts storepkg.ListOptions) ([]storepkg.SignRec
}
result = append(result, sign)
}
// 应用 Limit
if opts.Limit > 0 && len(result) > opts.Limit {
result = result[:opts.Limit]
}
return result, nil
}
@@ -218,3 +227,110 @@ func TestSignTool_SignAction(t *testing.T) {
t.Errorf("Expected 1 sign record, got %d", len(store.signs))
}
}
func TestSignTool_CheckAction(t *testing.T) {
now := time.Date(2024, 6, 23, 12, 0, 0, 0, time.UTC)
// 测试场景1:今天未签到
t.Run("今天未签到", func(t *testing.T) {
store := &mockSignStore{
signs: []storepkg.SignRecord{},
nodeInfoMap: make(map[string]*storepkg.NodeInfoRecord),
}
tool := &Tool{
enabled: true,
store: store,
}
params := signParams{
Action: "check",
}
argsJSON, _ := json.Marshal(params)
nodeCtx := agenttool.NodeContext{
NodeID: "test_node_123",
LongName: "Test Node",
ShortName: "TN",
}
ctx := agenttool.WithNodeContext(context.Background(), nodeCtx)
runtime := agenttool.Runtime{Now: now}
result, err := tool.Execute(ctx, string(argsJSON), runtime)
if err != nil {
t.Errorf("Unexpected error: %v", err)
}
t.Logf("Check result (not signed): %s", result)
if !contains(result, "还没有签到") && !contains(result, "没有签到") {
t.Errorf("Expected result to indicate not signed yet")
}
})
// 测试场景2:今天已签到
t.Run("今天已签到", func(t *testing.T) {
signTime := time.Date(2024, 6, 23, 10, 30, 45, 0, time.UTC)
store := &mockSignStore{
signs: []storepkg.SignRecord{
{
NodeID: "test_node_123",
SignText: "上海-TestUser-TestDevice签到",
SignTime: signTime,
},
},
nodeInfoMap: make(map[string]*storepkg.NodeInfoRecord),
}
tool := &Tool{
enabled: true,
store: store,
}
params := signParams{
Action: "check",
}
argsJSON, _ := json.Marshal(params)
nodeCtx := agenttool.NodeContext{
NodeID: "test_node_123",
LongName: "Test Node",
ShortName: "TN",
}
ctx := agenttool.WithNodeContext(context.Background(), nodeCtx)
runtime := agenttool.Runtime{Now: now}
result, err := tool.Execute(ctx, string(argsJSON), runtime)
if err != nil {
t.Errorf("Unexpected error: %v", err)
}
t.Logf("Check result (already signed): %s", result)
if !contains(result, "已经签到") {
t.Errorf("Expected result to indicate already signed")
}
if !contains(result, "签到时间") {
t.Errorf("Expected result to contain sign time")
}
if !contains(result, "10:30:45") {
t.Errorf("Expected result to contain the exact sign time")
}
if !contains(result, "签到内容") {
t.Errorf("Expected result to contain sign text")
}
})
}
func contains(s, substr string) bool {
return len(s) > 0 && len(substr) > 0 && (s == substr || len(s) >= len(substr) && (s[:len(substr)] == substr || s[len(s)-len(substr):] == substr || containsMiddle(s, substr)))
}
func containsMiddle(s, substr string) bool {
for i := 0; i <= len(s)-len(substr); i++ {
if s[i:i+len(substr)] == substr {
return true
}
}
return false
}
+203
View File
@@ -0,0 +1,203 @@
package ai
import (
"context"
"fmt"
"sync"
"meshtastic_mqtt_server/internal/autoreply"
"meshtastic_mqtt_server/internal/llm"
storepkg "meshtastic_mqtt_server/internal/store"
"gorm.io/gorm"
)
// AIServiceStatus reports the current state of the AI service
type AIServiceStatus struct {
Running bool `json:"running"`
Enabled bool `json:"enabled"`
ProviderCount int `json:"provider_count"`
Message string `json:"message,omitempty"`
}
// AIManager manages the lifecycle of the AI service, supporting restart
type AIManager struct {
mu sync.Mutex
service *Service
cfg Config
db *gorm.DB
botSender autoreply.BotSender
ctx context.Context
store *storepkg.Store
}
// NewAIManager creates a new AIManager
func NewAIManager(cfg Config, db *gorm.DB, botSender autoreply.BotSender, ctx context.Context, store *storepkg.Store) *AIManager {
return &AIManager{
cfg: cfg,
db: db,
botSender: botSender,
ctx: ctx,
store: store,
}
}
// SetConfigEnabled sets the enabled flag on the config
func (m *AIManager) SetConfigEnabled(enabled bool) {
m.cfg.Enabled = enabled
}
// SetProviderConfigs sets the LLM provider configs
func (m *AIManager) SetProviderConfigs(configs []llm.ProviderConfig) {
m.cfg.LLMProviders = configs
}
// Init creates and starts the AI service
func (m *AIManager) Init() error {
m.mu.Lock()
defer m.mu.Unlock()
svc, err := NewService(m.cfg, m.db, m.botSender)
if err != nil {
return fmt.Errorf("failed to create AI service: %w", err)
}
if err := svc.Start(m.ctx); err != nil {
return fmt.Errorf("failed to start AI service: %w", err)
}
m.service = svc
return nil
}
// Stop stops the currently running AI service
func (m *AIManager) Stop() {
m.mu.Lock()
defer m.mu.Unlock()
if m.service != nil {
m.service.Stop()
m.service = nil
}
}
// Status returns the current AI service status
func (m *AIManager) Status() AIServiceStatus {
m.mu.Lock()
defer m.mu.Unlock()
providerCount := len(m.cfg.LLMProviders)
if m.service == nil {
msg := "AI 服务未运行,可在配置提供商后点击重启"
if providerCount == 0 {
msg = "尚未配置 AI 提供商,请先添加提供商配置"
}
return AIServiceStatus{
Running: false,
Enabled: m.cfg.Enabled,
ProviderCount: providerCount,
Message: msg,
}
}
enabled := m.service.Enabled()
if !enabled {
return AIServiceStatus{
Running: false,
Enabled: false,
ProviderCount: providerCount,
Message: "AI 服务未启用",
}
}
return AIServiceStatus{
Running: true,
Enabled: true,
ProviderCount: providerCount,
}
}
// Restart stops the current AI service and creates a new one from DB
func (m *AIManager) Restart() error {
m.mu.Lock()
defer m.mu.Unlock()
if m.service != nil {
m.service.Stop()
m.service = nil
}
providers, err := m.store.ListLLMProviders(true)
if err != nil {
return fmt.Errorf("加载 LLM 提供商列表失败: %w", err)
}
if len(providers) == 0 {
return fmt.Errorf("没有配置任何 LLM 提供商,请先添加配置")
}
providerConfigs := make([]llm.ProviderConfig, 0, len(providers))
for _, p := range providers {
providerConfigs = append(providerConfigs, llm.ProviderConfig{
Name: p.Name,
Active: p.Active,
APIKey: p.APIKey,
BaseURL: p.BaseURL,
Model: p.Model,
Timeout: p.Timeout,
ContextWindowTokens: p.ContextWindowTokens,
})
}
m.cfg.LLMProviders = providerConfigs
svc, err := NewService(m.cfg, m.db, m.botSender)
if err != nil {
return fmt.Errorf("创建 AI 服务失败: %w", err)
}
if err := svc.Start(m.ctx); err != nil {
return fmt.Errorf("启动 AI 服务失败: %w", err)
}
m.service = svc
return nil
}
// ReloadLLMProvider delegates to the current service
func (m *AIManager) ReloadLLMProvider(config interface{}) error {
m.mu.Lock()
svc := m.service
m.mu.Unlock()
if svc == nil {
return nil
}
return svc.ReloadLLMProvider(config)
}
// AddLLMProvider delegates to the current service
func (m *AIManager) AddLLMProvider(config interface{}) error {
m.mu.Lock()
svc := m.service
m.mu.Unlock()
if svc == nil {
return nil
}
return svc.AddLLMProvider(config)
}
// RemoveLLMProvider delegates to the current service
func (m *AIManager) RemoveLLMProvider(name string) error {
m.mu.Lock()
svc := m.service
m.mu.Unlock()
if svc == nil {
return nil
}
return svc.RemoveLLMProvider(name)
}
// AIServiceStatus returns the AI service status for the web interface
func (m *AIManager) AIServiceStatus() AIServiceStatus {
return m.Status()
}
// RestartAIService restarts the AI service for the web interface
func (m *AIManager) RestartAIService() error {
return m.Restart()
}
+9
View File
@@ -262,6 +262,9 @@ func (s *Service) Enabled() bool {
// ReloadLLMProvider reloads a specific LLM provider configuration
func (s *Service) ReloadLLMProvider(config interface{}) error {
if s == nil {
return nil
}
if !s.enabled || s.LLMState == nil {
return nil
}
@@ -274,6 +277,9 @@ func (s *Service) ReloadLLMProvider(config interface{}) error {
// AddLLMProvider adds a new LLM provider
func (s *Service) AddLLMProvider(config interface{}) error {
if s == nil {
return nil
}
if !s.enabled || s.LLMState == nil {
return nil
}
@@ -286,6 +292,9 @@ func (s *Service) AddLLMProvider(config interface{}) error {
// RemoveLLMProvider removes an LLM provider
func (s *Service) RemoveLLMProvider(name string) error {
if s == nil {
return nil
}
if !s.enabled || s.LLMState == nil {
return nil
}
+10
View File
@@ -509,6 +509,16 @@ func cleanReplyText(text string) string {
sb.WriteRune(r)
case r == 0x3002 || r == 0xFF1F || r == 0xFF01 || r == 0xFF0C || r == 0xFF1A: // Fullwidth punctuation
sb.WriteRune(r)
case r >= 0x1F600 && r <= 0x1F64F: // Emoticons
sb.WriteRune(r)
case r >= 0x1F300 && r <= 0x1F5FF: // Misc Symbols and Pictographs
sb.WriteRune(r)
case r >= 0x1F680 && r <= 0x1F6FF: // Transport and Map Symbols
sb.WriteRune(r)
case r >= 0x1F900 && r <= 0x1F9FF: // Supplemental Symbols and Pictographs
sb.WriteRune(r)
case r >= 0x2600 && r <= 0x27BF: // Misc Symbols + Dingbats
sb.WriteRune(r)
default:
continue // Skip all other characters
}
+114 -47
View File
@@ -9,6 +9,7 @@ import (
"github.com/gin-gonic/gin"
"gorm.io/gorm"
aipkg "meshtastic_mqtt_server/internal/ai"
storepkg "meshtastic_mqtt_server/internal/store"
"meshtastic_mqtt_server/internal/webutil"
)
@@ -18,6 +19,8 @@ type LLMProviderReloader interface {
ReloadLLMProvider(config interface{}) error
AddLLMProvider(config interface{}) error
RemoveLLMProvider(name string) error
AIServiceStatus() aipkg.AIServiceStatus
RestartAIService() error
}
func RegisterRoutes(r *gin.RouterGroup, store *storepkg.Store, aiService LLMProviderReloader) {
@@ -49,6 +52,10 @@ func RegisterRoutes(r *gin.RouterGroup, store *storepkg.Store, aiService LLMProv
// LLM Primary Config - 主 AI 回复配置
group.GET("/primary-config", handleGetLLMPrimaryConfig(store))
group.PUT("/primary-config", handleUpdateLLMPrimaryConfig(store))
// AI Service Status
group.GET("/status", handleGetAIServiceStatus(aiService))
group.POST("/restart", handleRestartAIService(aiService))
}
}
@@ -298,25 +305,30 @@ func handleCreateLLMProvider(store *storepkg.Store, aiService LLMProviderReloade
}
// Reload AI service with new provider
if aiService != nil {
providerConfig := map[string]interface{}{
"Name": record.Name,
"Active": record.Active,
"APIKey": record.APIKey,
"BaseURL": record.BaseURL,
"Model": record.Model,
"Timeout": record.Timeout,
"ContextWindowTokens": record.ContextWindowTokens,
}
if err := aiService.AddLLMProvider(providerConfig); err != nil {
// Log warning but don't fail the request - database is already updated
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"item": llmProviderDTO(*record),
"warning": "provider created but failed to reload AI service: " + err.Error(),
})
return
}
if aiService == nil {
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"item": llmProviderDTO(*record),
"warning": "AI 服务未运行,配置已保存但需重启服务后生效",
})
return
}
providerConfig := map[string]interface{}{
"Name": record.Name,
"Active": record.Active,
"APIKey": record.APIKey,
"BaseURL": record.BaseURL,
"Model": record.Model,
"Timeout": record.Timeout,
"ContextWindowTokens": record.ContextWindowTokens,
}
if err := aiService.AddLLMProvider(providerConfig); err != nil {
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"item": llmProviderDTO(*record),
"warning": "provider created but failed to reload AI service: " + err.Error(),
})
return
}
c.JSON(http.StatusOK, gin.H{"status": "ok", "item": llmProviderDTO(*record)})
@@ -385,25 +397,30 @@ func handleUpdateLLMProvider(store *storepkg.Store, aiService LLMProviderReloade
}
// Reload AI service with updated provider
if aiService != nil {
providerConfig := map[string]interface{}{
"Name": record.Name,
"Active": record.Active,
"APIKey": record.APIKey,
"BaseURL": record.BaseURL,
"Model": record.Model,
"Timeout": record.Timeout,
"ContextWindowTokens": record.ContextWindowTokens,
}
if err := aiService.ReloadLLMProvider(providerConfig); err != nil {
// Log warning but don't fail the request - database is already updated
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"item": llmProviderDTO(*record),
"warning": "provider updated but failed to reload AI service: " + err.Error(),
})
return
}
if aiService == nil {
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"item": llmProviderDTO(*record),
"warning": "AI 服务未运行,配置已保存但需重启服务后生效",
})
return
}
providerConfig := map[string]interface{}{
"Name": record.Name,
"Active": record.Active,
"APIKey": record.APIKey,
"BaseURL": record.BaseURL,
"Model": record.Model,
"Timeout": record.Timeout,
"ContextWindowTokens": record.ContextWindowTokens,
}
if err := aiService.ReloadLLMProvider(providerConfig); err != nil {
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"item": llmProviderDTO(*record),
"warning": "provider updated but failed to reload AI service: " + err.Error(),
})
return
}
c.JSON(http.StatusOK, gin.H{"status": "ok", "item": llmProviderDTO(*record)})
@@ -424,15 +441,19 @@ func handleDeleteLLMProvider(store *storepkg.Store, aiService LLMProviderReloade
}
// Remove provider from AI service
if aiService != nil {
if err := aiService.RemoveLLMProvider(name); err != nil {
// Log warning but don't fail the request - database is already updated
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"warning": "provider deleted but failed to reload AI service: " + err.Error(),
})
return
}
if aiService == nil {
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"warning": "AI 服务未运行,配置已删除但需重启服务后生效",
})
return
}
if err := aiService.RemoveLLMProvider(name); err != nil {
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"warning": "provider deleted but failed to reload AI service: " + err.Error(),
})
return
}
c.JSON(http.StatusOK, gin.H{"status": "ok"})
@@ -814,3 +835,49 @@ func llmPrimaryConfigDTO(row storepkg.LLMPrimaryConfigRecord) map[string]any {
"updated_at": row.UpdatedAt,
}
}
// ============================================
// AI Service Status & Restart Handlers
// ============================================
func handleGetAIServiceStatus(aiService LLMProviderReloader) gin.HandlerFunc {
return func(c *gin.Context) {
if aiService == nil {
c.JSON(http.StatusOK, gin.H{
"running": false,
"enabled": false,
"provider_count": 0,
"message": "AI 服务未初始化",
})
return
}
status := aiService.AIServiceStatus()
c.JSON(http.StatusOK, gin.H{
"running": status.Running,
"enabled": status.Enabled,
"provider_count": status.ProviderCount,
"message": status.Message,
})
}
}
func handleRestartAIService(aiService LLMProviderReloader) gin.HandlerFunc {
return func(c *gin.Context) {
if aiService == nil {
c.JSON(http.StatusInternalServerError, gin.H{"error": "AI 服务未初始化,无法重启"})
return
}
if err := aiService.RestartAIService(); err != nil {
c.JSON(http.StatusInternalServerError, gin.H{"error": "重启 AI 服务失败: " + err.Error()})
return
}
status := aiService.AIServiceStatus()
c.JSON(http.StatusOK, gin.H{
"status": "ok",
"running": status.Running,
"enabled": status.Enabled,
"provider_count": status.ProviderCount,
"message": "AI 服务已重启",
})
}
}
+9
View File
@@ -20,6 +20,7 @@ type PacketBuildOptions struct {
PSK []byte
Encrypt bool
ViaMQTT bool
HopLimit uint32
}
type TextMessageBuildOptions struct {
@@ -252,6 +253,14 @@ func buildMeshPacket(opts PacketBuildOptions, data []byte) ([]byte, error) {
out = protowire.AppendTag(out, 14, protowire.VarintType)
out = protowire.AppendVarint(out, 1)
}
hop := opts.HopLimit
if hop == 0 {
hop = 7
}
out = protowire.AppendTag(out, 9, protowire.VarintType)
out = protowire.AppendVarint(out, uint64(hop))
out = protowire.AppendTag(out, 15, protowire.VarintType)
out = protowire.AppendVarint(out, uint64(hop))
return out, nil
}
+78
View File
@@ -0,0 +1,78 @@
package mqttforward
import (
"crypto/sha256"
"encoding/hex"
"sync"
"time"
)
const dedupTTL = 15 * time.Second
type DedupQueue struct {
mu sync.Mutex
entries map[string]time.Time
stopCh chan struct{}
}
func NewDedupQueue() *DedupQueue {
return &DedupQueue{
entries: make(map[string]time.Time),
stopCh: make(chan struct{}),
}
}
func (dq *DedupQueue) TryForward(topic string, payload []byte) bool {
hash := dedupHash(topic, payload)
now := time.Now()
dq.mu.Lock()
defer dq.mu.Unlock()
if expiry, ok := dq.entries[hash]; ok && now.Before(expiry) {
return false
}
dq.entries[hash] = now.Add(dedupTTL)
return true
}
func (dq *DedupQueue) Start() {
go func() {
ticker := time.NewTicker(dedupTTL)
defer ticker.Stop()
for {
select {
case <-ticker.C:
dq.cleanup()
case <-dq.stopCh:
return
}
}
}()
}
func (dq *DedupQueue) Stop() {
close(dq.stopCh)
}
func (dq *DedupQueue) Len() int {
dq.mu.Lock()
defer dq.mu.Unlock()
return len(dq.entries)
}
func (dq *DedupQueue) cleanup() {
now := time.Now()
dq.mu.Lock()
defer dq.mu.Unlock()
for hash, expiry := range dq.entries {
if now.After(expiry) {
delete(dq.entries, hash)
}
}
}
func dedupHash(topic string, payload []byte) string {
h := sha256.New()
h.Write([]byte(topic))
h.Write(payload)
return hex.EncodeToString(h.Sum(nil))
}
+18 -21
View File
@@ -290,27 +290,24 @@ func insertInboundBotDirectMessage(s *Store, record map[string]any, clientInfo M
}
_ = clientInfo // mqtt 元数据已经记录在 content_json 里,这里保留参数以保持队列签名一致
// 同时将消息添加到 LLM 队列(忽略机器人自己发送的消息)
if peerNodeID != bot.NodeID {
longName := NullableString(record["long_name"])
shortName := NullableString(record["short_name"])
channelID := NullableString(record["channel_id"])
_, err = s.EnqueueLLMMessage(LLMMessageQueueInput{
BotID: bot.ID,
BotNodeID: bot.NodeID,
BotNodeNum: bot.NodeNum,
FromNodeID: peerNodeID,
FromNodeNum: int64(peerNum),
LongName: longName,
ShortName: shortName,
Text: text,
PacketID: int64(packetID),
ChannelID: channelID,
Topic: topic,
MessageType: "direct",
ContentJSON: contentPtr,
})
}
longName := NullableString(record["long_name"])
shortName := NullableString(record["short_name"])
channelID := NullableString(record["channel_id"])
_, err = s.EnqueueLLMMessage(LLMMessageQueueInput{
BotID: bot.ID,
BotNodeID: bot.NodeID,
BotNodeNum: bot.NodeNum,
FromNodeID: peerNodeID,
FromNodeNum: int64(peerNum),
LongName: longName,
ShortName: shortName,
Text: text,
PacketID: int64(packetID),
ChannelID: channelID,
Topic: topic,
MessageType: "direct",
ContentJSON: contentPtr,
})
if err != nil {
printJSON(map[string]any{
"event": "llm_queue_enqueue_failed",
+6
View File
@@ -66,6 +66,12 @@ func (s *Store) CountBotNodes(opts ListOptions) (int64, error) {
return total, s.db.Model(&BotNodeRecord{}).Count(&total).Error
}
func (s *Store) IsBotNodeID(nodeID string) bool {
var count int64
s.db.Model(&BotNodeRecord{}).Where("node_id = ?", nodeID).Count(&count)
return count > 0
}
func (s *Store) GetBotNode(id uint64) (*BotNodeRecord, error) {
var row BotNodeRecord
if err := s.db.Where("id = ?", id).Take(&row).Error; err != nil {
+5 -6
View File
@@ -306,8 +306,7 @@ func (s *Store) EnqueueLLMMessage(input LLMMessageQueueInput) (*LLMMessageQueueR
return nil, nil // 机器人的 LLM 队列未启用,静默返回
}
// 忽略机器人自己发送的消息,避免自循环
if input.FromNodeID == bot.NodeID {
if s.IsBotNodeID(input.FromNodeID) {
return nil, nil
}
@@ -526,6 +525,10 @@ func enqueueChannelMessageToLLM(s *Store, record map[string]any) error {
shortName = &sn
}
if s.IsBotNodeID(fromNodeID) {
return nil
}
var channelID *string
if cid, ok := record["channel_id"].(string); ok && cid != "" {
channelID = &cid
@@ -546,11 +549,7 @@ func enqueueChannelMessageToLLM(s *Store, record map[string]any) error {
return fmt.Errorf("query bots for channel message enqueue: %w", err)
}
// 为每个符合条件的机器人创建一条队列记录(忽略机器人自己发送的消息)
for _, bot := range bots {
if fromNodeID == bot.NodeID {
continue
}
_, err = s.EnqueueLLMMessage(LLMMessageQueueInput{
BotID: bot.ID,
BotNodeID: bot.NodeID,
+1
View File
@@ -88,6 +88,7 @@ func RunAgentToolLoop(ctx context.Context, state *State, profile *llm.Profile, s
if routerPrompt == "" {
routerPrompt = primaryPrompt
}
routerPrompt = routerPrompt + "\n当前日期:" + time.Now().Format("2006-01-02")
if primaryPrompt != "" {
primarySystemMessage := &model.ChatCompletionMessage{
Role: "system",
+1 -1
View File
@@ -48,7 +48,7 @@ func NewState(cfg *Config, ai *llm.State, options ...Option) (*State, error) {
Enabled: true,
Timeout: 30,
MaxTokens: 512,
SystemPrompt: "你可以按需直接调用可用工具来回答用户问题。\n每个工具的 description 描述了它的适用场景和调用条件。\n工具结果优先于模型内置知识;工具失败时必须如实说明,不要编造结果。\n只调用确实必要的工具。",
SystemPrompt: "你可以按需直接调用可用工具来回答用户问题。\n每个工具的 description 描述了它的适用场景和调用条件。\n工具结果优先于模型内置知识;工具失败时必须如实说明,不要编造结果。\n只调用确实必要的工具。\n\n重要规则:\n- 当用户问题包含\"今天\"、\"昨天\"、\"最近\"、\"本周\"、\"本月\"等相对时间词时,你必须先调用 time 工具获取当前准确日期,然后再用该日期调用其他工具。绝不可以使用模型内置知识猜测日期。\n- 你不知道\"今天\"是哪一天,必须通过 time 工具查询。",
}
}
if ai == nil {
+11 -1
View File
@@ -30,6 +30,7 @@ type MQTTRuntimeStatus struct {
Stats *mqttforwardpkg.Stats
ClientStats *mqttforwardpkg.ClientStats
DBQueue *storepkg.WriteQueue
DedupQueue *mqttforwardpkg.DedupQueue
}
// AdminMQTTStatus 是 admin 路由 GET /admin/mqtt-status 返回的 JSON 视图。
@@ -50,6 +51,7 @@ type AdminMQTTStatus struct {
MessagesSent int64 `json:"messages_sent"`
MessagesDropped int64 `json:"messages_dropped"`
DBWriteQueueLength int `json:"db_write_queue_length"`
DedupQueueLength int `json:"dedup_queue_len"`
Retained int64 `json:"retained"`
Inflight int64 `json:"inflight"`
InflightDropped int64 `json:"inflight_dropped"`
@@ -71,7 +73,7 @@ type AdminMQTTClient struct {
// Status 实现 MQTTStatusProvider。
func (m MQTTRuntimeStatus) Status() AdminMQTTStatus {
if m.Server == nil || m.Server.Info == nil {
return AdminMQTTStatus{Running: false, Address: m.Address, TLS: m.TLS, DBWriteQueueLength: m.DBQueue.Len()}
return AdminMQTTStatus{Running: false, Address: m.Address, TLS: m.TLS, DBWriteQueueLength: m.DBQueue.Len(), DedupQueueLength: m.dedupQueueLen()}
}
info := m.Server.Info.Clone()
status := AdminMQTTStatus{
@@ -91,6 +93,7 @@ func (m MQTTRuntimeStatus) Status() AdminMQTTStatus {
MessagesSent: m.Stats.Forwarded(),
MessagesDropped: m.Stats.Dropped(),
DBWriteQueueLength: m.DBQueue.Len(),
DedupQueueLength: m.dedupQueueLen(),
Retained: info.Retained,
Inflight: info.Inflight,
InflightDropped: info.InflightDropped,
@@ -136,6 +139,13 @@ func mqttClientInfo(c *mqtt.Client) mqttClientInfoView {
}
}
func (m MQTTRuntimeStatus) dedupQueueLen() int {
if m.DedupQueue == nil {
return 0
}
return m.DedupQueue.Len()
}
// DisconnectClient 实现 MQTTStatusProvider:发送 Disconnect 报文并关闭连接。
// 使用 ErrAdministrativeAction 作为断开理由,便于日志区分。
func (m MQTTRuntimeStatus) DisconnectClient(clientID string) bool {
+3
View File
@@ -13,6 +13,7 @@ import (
"github.com/gin-gonic/gin"
"gorm.io/gorm"
aipkg "meshtastic_mqtt_server/internal/ai"
"meshtastic_mqtt_server/internal/auth"
blockingpkg "meshtastic_mqtt_server/internal/blocking"
botpkg "meshtastic_mqtt_server/internal/bot"
@@ -32,6 +33,8 @@ type LLMProviderReloader interface {
ReloadLLMProvider(config interface{}) error
AddLLMProvider(config interface{}) error
RemoveLLMProvider(name string) error
AIServiceStatus() aipkg.AIServiceStatus
RestartAIService() error
}
func NewHTTPServer(cfg configpkg.WebConfig, consoleLog bool, store *storepkg.Store, sessions *auth.Manager, mqttStatus MQTTStatusProvider, blocking *blockingpkg.Cache, forwarder mqttforwardpkg.Reloader, settings *rspkg.Cache, botSender botpkg.TextSender, aiService LLMProviderReloader) *http.Server {
+68 -50
View File
@@ -57,8 +57,9 @@ type meshtasticFilterHook struct {
settings *rspkg.Cache
pkiResolver func(toNodeNum, fromNodeNum uint32) ([]byte, []byte, bool)
autoAcker func(record map[string]any)
consoleLog bool // 控制台是否打印 MQTT 连接/订阅事件
consoleLog bool // 控制台是否打印 MQTT 连接/订阅事件
packetConsoleLog bool // 控制台是否打印 Meshtastic 数据包
dedupQueue *mqttforwardpkg.DedupQueue
}
// ID 返回用于识别 Meshtastic payload 过滤器的 hook 名称。
@@ -203,6 +204,12 @@ func (h *meshtasticFilterHook) OnPublish(cl *mqtt.Client, pk packets.Packet) (pa
PKIKeyResolver: h.pkiResolver,
})
if !valid {
// 记录拒绝原因,帮助诊断 QoS0 重发问题
if h.consoleLog {
info := mqttClientInfoFromClient(cl)
fmt.Fprintf(os.Stderr, "[mqtt] PUBLISH rejected: client_id=%s topic=%s qos=%d payload_len=%d error=%v\n",
info.ClientID, pk.TopicName, pk.FixedHeader.Qos, len(pk.Payload), record["error"])
}
h.rejectPublish(cl, pk, record)
return pk, packets.ErrRejectPacket
}
@@ -210,9 +217,19 @@ func (h *meshtasticFilterHook) OnPublish(cl *mqtt.Client, pk packets.Packet) (pa
for key, value := range violation {
record[key] = value
}
// 记录屏蔽原因
if h.consoleLog {
info := mqttClientInfoFromClient(cl)
fmt.Fprintf(os.Stderr, "[mqtt] PUBLISH blocked: client_id=%s topic=%s type=%v reason=%v\n",
info.ClientID, pk.TopicName, violation["blocking_type"], violation["error"])
}
h.rejectPublish(cl, pk, record)
return pk, packets.ErrRejectPacket
}
if h.dedupQueue != nil && !h.dedupQueue.TryForward(pk.TopicName, pk.Payload) {
h.stats.IncDropped()
return pk, packets.ErrRejectPacket
}
h.stats.IncForwarded()
h.dbQueue.EnqueueRecord(record, mqttClientInfoFromClient(cl))
@@ -381,16 +398,48 @@ func run(cfg *configpkg.Config) error {
defer forwardManager.StopAll()
// Initialize AI Service
var aiService *ai.Service
// Create bot sender adapter - 支持频道消息和私聊消息两种发送方式
botSenderAdapter := autoreply.NewBotServiceAdapter(
// SendDirectText: 发送私聊消息
func(ctx context.Context, botID uint64, toNodeNum int64, text string) error {
_, err := botSender.SendText(ctx, botpkg.SendTextRequest{
BotID: botID,
MessageType: "direct",
ToNodeNum: &toNodeNum,
Text: text,
})
return err
},
// SendChannelText: 发送频道消息
func(ctx context.Context, botID uint64, channelID string, text string) error {
_, err := botSender.SendText(ctx, botpkg.SendTextRequest{
BotID: botID,
MessageType: "channel",
ChannelID: channelID,
Text: text,
})
return err
},
)
aiManager := ai.NewAIManager(ai.Config{
DataDir: cfg.AI.DataDir,
Enabled: cfg.AI.Enabled,
ConsoleLog: cfg.ConsoleLog.LLM,
ToolConfigStore: store,
ToolRouterStore: store,
TopicRouterStore: store,
Store: store,
}, store.DB(), botSenderAdapter, botCtx, store)
if cfg.AI.Enabled {
// Get LLM providers from database
llmProviders, err := store.ListLLMProviders(true)
aiManager.SetConfigEnabled(true)
providers, err := store.ListLLMProviders(true)
if err != nil {
fmt.Fprintf(os.Stderr, "Warning: failed to load LLM providers: %v\n", err)
} else if len(llmProviders) > 0 {
// Convert database records to provider configs
providerConfigs := make([]llm.ProviderConfig, 0, len(llmProviders))
for _, p := range llmProviders {
} else if len(providers) > 0 {
providerConfigs := make([]llm.ProviderConfig, 0, len(providers))
for _, p := range providers {
providerConfigs = append(providerConfigs, llm.ProviderConfig{
Name: p.Name,
Active: p.Active,
@@ -401,48 +450,11 @@ func run(cfg *configpkg.Config) error {
ContextWindowTokens: p.ContextWindowTokens,
})
}
// Create bot sender adapter - 支持频道消息和私聊消息两种发送方式
botSenderAdapter := autoreply.NewBotServiceAdapter(
// SendDirectText: 发送私聊消息
func(ctx context.Context, botID uint64, toNodeNum int64, text string) error {
_, err := botSender.SendText(ctx, botpkg.SendTextRequest{
BotID: botID,
MessageType: "direct",
ToNodeNum: &toNodeNum,
Text: text,
})
return err
},
// SendChannelText: 发送频道消息
func(ctx context.Context, botID uint64, channelID string, text string) error {
_, err := botSender.SendText(ctx, botpkg.SendTextRequest{
BotID: botID,
MessageType: "channel",
ChannelID: channelID,
Text: text,
})
return err
},
)
aiService, err = ai.NewService(ai.Config{
LLMProviders: providerConfigs,
DataDir: cfg.AI.DataDir,
Enabled: cfg.AI.Enabled,
ConsoleLog: cfg.ConsoleLog.LLM,
ToolConfigStore: store,
ToolRouterStore: store,
TopicRouterStore: store,
Store: store,
}, store.DB(), botSenderAdapter)
if err != nil {
aiManager.SetProviderConfigs(providerConfigs)
if err := aiManager.Init(); err != nil {
fmt.Fprintf(os.Stderr, "Warning: failed to initialize AI service: %v\n", err)
} else {
if err := aiService.Start(botCtx); err != nil {
fmt.Fprintf(os.Stderr, "Warning: failed to start AI service: %v\n", err)
}
defer aiService.Stop()
defer aiManager.Stop()
printJSON(map[string]any{"event": "ai_service_started", "providers": len(providerConfigs)})
}
} else {
@@ -457,8 +469,8 @@ func run(cfg *configpkg.Config) error {
if err != nil {
return err
}
mqttStatus := webpkg.MQTTRuntimeStatus{Server: server, Address: mqttAddr, TLS: cfg.MQTT.TLS.Enabled, Stats: messageStats, ClientStats: clientStats, DBQueue: dbQueue}
handler := webpkg.NewRouter(cfg.Web, cfg.ConsoleLog.Web, store, sessions, mqttStatus, blocking, forwardManager, settings, botSender, aiService)
mqttStatus := webpkg.MQTTRuntimeStatus{Server: server, Address: mqttAddr, TLS: cfg.MQTT.TLS.Enabled, Stats: messageStats, ClientStats: clientStats, DBQueue: dbQueue, DedupQueue: mqttHook.dedupQueue}
handler := webpkg.NewRouter(cfg.Web, cfg.ConsoleLog.Web, store, sessions, mqttStatus, blocking, forwardManager, settings, botSender, aiManager)
webAddresses := []string{}
if cfg.Web.PortEnabled {
httpServer := &http.Server{
@@ -509,6 +521,9 @@ func run(cfg *configpkg.Config) error {
if err := server.Close(); err != nil && runErr == nil {
runErr = err
}
if mqttHook.dedupQueue != nil {
mqttHook.dedupQueue.Stop()
}
return runErr
}
@@ -517,6 +532,8 @@ func startMQTTServer(cfg *configpkg.Config, store *storepkg.Store, dbQueue *stor
if err := server.AddHook(new(mqttauth.AllowHook), nil); err != nil {
return nil, nil, "", err
}
dedupQueue := mqttforwardpkg.NewDedupQueue()
dedupQueue.Start()
hook := &meshtasticFilterHook{
server: server,
key: cfg.Key,
@@ -528,6 +545,7 @@ func startMQTTServer(cfg *configpkg.Config, store *storepkg.Store, dbQueue *stor
pkiResolver: botpkg.NewPKIKeyResolver(store),
consoleLog: cfg.ConsoleLog.MQTT,
packetConsoleLog: cfg.ConsoleLog.Meshtastic,
dedupQueue: dedupQueue,
}
if err := server.AddHook(hook, nil); err != nil {
return nil, nil, "", err
+12 -2
View File
@@ -6,6 +6,7 @@ import type {
AdminRuntimeSettingsPayload,
AdminRuntimeSettingsResponse,
AdminUsersResponse,
AIServiceStatus,
BotMessage,
BotMessageMutationResponse,
BotNode,
@@ -515,8 +516,17 @@ export function updateLLMProvider(name: string, payload: Partial<LLMProviderPayl
return putJSON<LLMProviderResponse>(`/api/admin/llm/providers/${encodeURIComponent(name)}`, payload)
}
export function deleteLLMProvider(name: string): Promise<{ status: string }> {
return deleteJSON<{ status: string }>(`/api/admin/llm/providers/${encodeURIComponent(name)}`)
export function deleteLLMProvider(name: string): Promise<{ status: string; warning?: string }> {
return deleteJSON<{ status: string; warning?: string }>(`/api/admin/llm/providers/${encodeURIComponent(name)}`)
}
// AI Service Status API
export function getAIServiceStatus(): Promise<AIServiceStatus> {
return getJSON<AIServiceStatus>('/api/admin/llm/status')
}
export function restartAIService(): Promise<AIServiceStatus & { status: string; message?: string }> {
return postJSON<AIServiceStatus & { status: string; message?: string }>('/api/admin/llm/restart')
}
// LLM Tool Router API
@@ -200,6 +200,7 @@ onBeforeUnmount(() => {
<div><span>订阅数</span><strong>{{ status.subscriptions }}</strong></div>
<div><span>转发消息</span><strong>{{ status.messages_sent }}</strong></div>
<div><span>数据库队列</span><strong>{{ status.db_write_queue_length }}</strong></div>
<div><span>去重队列</span><strong>{{ status.dedup_queue_len }}</strong></div>
<a class="status-card-link" href="/admin/discard_details"><span>丢弃消息</span><strong>{{ status.messages_dropped }}</strong></a>
<div><span>收到包</span><strong>{{ status.packets_received }}</strong></div>
<div><span>发送包</span><strong>{{ status.packets_sent }}</strong></div>
+116 -9
View File
@@ -11,12 +11,19 @@ import {
updateLLMToolRouter,
updateLLMTopicConfig,
updateLLMPrimaryConfig,
getAIServiceStatus,
restartAIService,
} from '../api'
import type { LLMPlatformRouter, LLMTopicConfig, LLMProvider, LLMPrimaryConfig } from '../types'
import type { LLMPlatformRouter, LLMTopicConfig, LLMProvider, LLMPrimaryConfig, AIServiceStatus } from '../types'
const loading = ref(false)
const error = ref('')
const success = ref('')
const showWarning = ref(false)
// AI Service Status
const aiStatus = ref<AIServiceStatus>({ running: false, enabled: false, provider_count: 0 })
const restarting = ref(false)
// LLM Provider 相关
const providers = ref<LLMProvider[]>([])
@@ -169,12 +176,18 @@ async function saveProvider() {
}
try {
let response: any
if (isCreatingProvider.value) {
await createLLMProvider(providerForm.value)
success.value = '创建成功'
response = await createLLMProvider(providerForm.value)
} else if (editingProvider.value) {
await updateLLMProvider(editingProvider.value.name, providerForm.value)
success.value = '更新成功'
response = await updateLLMProvider(editingProvider.value.name, providerForm.value)
}
if (response.warning) {
success.value = response.warning
showWarning.value = true
} else {
success.value = isCreatingProvider.value ? '创建成功' : '更新成功'
showWarning.value = false
}
clearSuccess()
closeProviderForm()
@@ -189,8 +202,14 @@ async function confirmDeleteProvider(name: string) {
return
}
try {
await deleteLLMProvider(name)
success.value = '删除成功'
const response = await deleteLLMProvider(name)
if (response.warning) {
success.value = response.warning
showWarning.value = true
} else {
success.value = '删除成功'
showWarning.value = false
}
clearSuccess()
await loadProviders()
} catch (err) {
@@ -286,7 +305,31 @@ async function savePrimaryConfig() {
}
}
async function loadAIStatus() {
try {
aiStatus.value = await getAIServiceStatus()
} catch (err) {
console.warn('Failed to load AI service status', err)
}
}
async function handleRestartAI() {
if (!confirm('确定要重启 AI 服务吗?')) return
restarting.value = true
try {
const result = await restartAIService()
aiStatus.value = result
success.value = result.message || 'AI 服务已重启'
clearSuccess()
} catch (err) {
error.value = err instanceof Error ? err.message : String(err)
} finally {
restarting.value = false
}
}
onMounted(() => {
loadAIStatus()
loadProviders()
loadToolRouter()
loadTopicConfig()
@@ -298,8 +341,23 @@ onMounted(() => {
<div class="admin-llm-api">
<h2>LLM API 配置管理</h2>
<div class="ai-status-bar">
<span class="status-indicator" :class="{ running: aiStatus.running, stopped: !aiStatus.running }"></span>
<span class="status-text">
<template v-if="aiStatus.running">AI 服务运行中{{ aiStatus.provider_count }} 个提供商</template>
<template v-else>{{ aiStatus.message || 'AI 服务未运行' }}</template>
</span>
<button
class="admin-button admin-button-small"
:disabled="restarting"
@click="handleRestartAI"
>
{{ restarting ? '重启中...' : '重启 AI 服务' }}
</button>
</div>
<p v-if="error" class="error">{{ error }}</p>
<p v-if="success" class="success">{{ success }}</p>
<p v-if="success" :class="showWarning ? 'warning' : 'success'">{{ success }}</p>
<!-- LLM Provider 列表 -->
<div class="admin-section">
@@ -690,13 +748,49 @@ onMounted(() => {
}
.admin-llm-api h2 {
margin: 0 0 2rem;
margin: 0 0 1rem;
font-size: 1.75rem;
font-weight: 700;
color: #1e293b;
letter-spacing: -0.02em;
}
.ai-status-bar {
display: flex;
align-items: center;
gap: 0.75rem;
padding: 0.75rem 1.25rem;
background: white;
border-radius: 12px;
border: 1px solid #e2e8f0;
margin-bottom: 1.5rem;
box-shadow: 0 1px 3px rgba(0, 0, 0, 0.05);
}
.status-indicator {
width: 10px;
height: 10px;
border-radius: 50%;
flex-shrink: 0;
}
.status-indicator.running {
background: #22c55e;
box-shadow: 0 0 6px rgba(34, 197, 94, 0.5);
}
.status-indicator.stopped {
background: #ef4444;
box-shadow: 0 0 6px rgba(239, 68, 68, 0.4);
}
.status-text {
flex: 1;
font-size: 0.9rem;
color: #475569;
font-weight: 500;
}
.admin-section {
background: white;
padding: 1.75rem;
@@ -973,6 +1067,19 @@ onMounted(() => {
gap: 0.5rem;
}
.warning {
color: #92400e;
padding: 1rem 1.25rem;
background: linear-gradient(135deg, #fef3c7 0%, #fde68a 100%);
border-radius: 10px;
margin-bottom: 1.25rem;
border: 1px solid #fcd34d;
font-weight: 500;
display: flex;
align-items: center;
gap: 0.5rem;
}
.admin-loading {
padding: 3rem;
text-align: center;
+9
View File
@@ -364,6 +364,7 @@ export interface AdminMqttStatus {
messages_sent: number
messages_dropped: number
db_write_queue_length: number
dedup_queue_len: number
retained: number
inflight: number
inflight_dropped: number
@@ -643,6 +644,14 @@ export interface LLMProviderPayload {
export interface LLMProviderResponse {
item: LLMProvider
warning?: string
}
export interface AIServiceStatus {
running: boolean
enabled: boolean
provider_count: number
message?: string
}
// LLM Tool Router 相关类型
+2 -1
View File
@@ -1 +1,2 @@
__pycache__
__pycache__
db_config.py
+266
View File
@@ -0,0 +1,266 @@
#!/usr/bin/env python3
"""
MySQL 8 -> MySQL 5.7 数据库迁移脚本
依赖: pip install pymysql
使用方法:
python py/migrate_mysql8_to_mysql57.py
功能:
1. 从源 MySQL 8 读取表结构,修正 collation 后在目标 MySQL 5.7 建表
2. 逐表批量拷贝数据
3. 修正 auto-increment 值
注意:
- 目标库应为空库(无同名表)
- 运行前请确保源库和目标库都可连接
"""
from __future__ import annotations
import re
import sys
from typing import Any
import pymysql
import pymysql.cursors
from db_config import SOURCE_CONFIG, TARGET_CONFIG
BATCH_SIZE = 5000
# ============================================================
# 工具函数
# ============================================================
def _conn(config: dict[str, Any]) -> pymysql.Connection:
"""创建数据库连接,使用 DictCursor 方便按列名访问"""
return pymysql.connect(
host=config["host"],
port=config["port"],
user=config["user"],
password=config["password"],
database=config["database"],
charset=config["charset"],
cursorclass=pymysql.cursors.DictCursor,
)
def fix_collation(ddl: str) -> str:
"""将 MySQL 8 默认 collation 替换为 MySQL 5.7 兼容版本"""
ddl = ddl.replace("utf8mb4_0900_ai_ci", "utf8mb4_general_ci")
ddl = ddl.replace("utf8mb4_0900_as_ci", "utf8mb4_general_ci")
return ddl
def fix_text_blob_index(ddl: str) -> str:
"""为 TEXT/BLOB 列在索引中添加前缀长度 (255),兼容 MySQL 5.7"""
text_cols = set()
for m in re.finditer(r'`(\w+)`\s+(?:tinytext|text|mediumtext|longtext|tinyblob|blob|mediumblob|longblob)\b', ddl, re.IGNORECASE):
text_cols.add(m.group(1))
if not text_cols:
return ddl
def _add_prefix(match: re.Match) -> str:
col = match.group(1)
rest = match.group(2)
if col in text_cols and not rest.strip().startswith('('):
return f'`{col}`(255){rest}'
return match.group(0)
# 匹配 KEY 定义中的列引用: `col_name` 后面不跟 ( 的情况
ddl = re.sub(r'`(\w+)`(\s*[,\)])', _add_prefix, ddl)
return ddl
def get_all_tables(conn: pymysql.Connection) -> list[str]:
with conn.cursor() as cur:
cur.execute("SHOW TABLES")
key = list(cur.description[0])[0]
return [row[key] for row in cur.fetchall()]
def get_table_columns(conn: pymysql.Connection, table: str) -> list[str]:
"""返回表的所有列名(按顺序)"""
with conn.cursor() as cur:
cur.execute("SHOW COLUMNS FROM `%s`" % table)
return [row["Field"] for row in cur.fetchall()]
def get_auto_increment(
conn: pymysql.Connection, table: str
) -> int | None:
"""获取某张表当前的 AUTO_INCREMENT 值"""
with conn.cursor() as cur:
cur.execute(
"SELECT AUTO_INCREMENT FROM information_schema.TABLES "
"WHERE TABLE_SCHEMA = DATABASE() AND TABLE_NAME = %s",
(table,),
)
row = cur.fetchone()
if row and row["AUTO_INCREMENT"]:
return int(row["AUTO_INCREMENT"])
return None
def row_count(conn: pymysql.Connection, table: str) -> int:
with conn.cursor() as cur:
cur.execute("SELECT COUNT(*) AS cnt FROM `%s`" % table)
return cur.fetchone()["cnt"]
# ============================================================
# 主流程
# ============================================================
def migrate() -> int:
src = _conn(SOURCE_CONFIG)
tgt = _conn(TARGET_CONFIG)
print(f"[连接] 源 MySQL 8 @ {SOURCE_CONFIG['host']}:{SOURCE_CONFIG['port']}")
print(f"[连接] 目标 MySQL 5.7 @ {TARGET_CONFIG['host']}:{TARGET_CONFIG['port']}")
print()
# 1. 在目标库建库(如还不存在)
try:
with tgt.cursor() as cur:
cur.execute(
"CREATE DATABASE IF NOT EXISTS `%s` "
"CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci"
% TARGET_CONFIG["database"]
)
cur.execute("USE `%s`" % TARGET_CONFIG["database"])
tgt.select_db(TARGET_CONFIG["database"])
except Exception as e:
print(f"[错误] 创建目标数据库失败: {e}")
return 1
tables = get_all_tables(src)
print(f"[发现] 源库共 {len(tables)} 张表: {', '.join(tables)}")
print()
# 2. 逐表建表(修正 collation
print("=" * 60)
print("阶段 1: 在目标库创建表结构")
print("=" * 60)
for idx, table in enumerate(tables, 1):
with src.cursor() as cur:
cur.execute("SHOW CREATE TABLE `%s`" % table)
row = cur.fetchone()
ddl = row["Create Table"]
ddl = fix_collation(ddl)
ddl = fix_text_blob_index(ddl)
try:
with tgt.cursor() as cur:
cur.execute(ddl)
tgt.commit()
print(f" [{idx:2d}/{len(tables)}] OK `{table}`")
except Exception as e:
print(f" [{idx:2d}/{len(tables)}] 错误 `{table}`: {e}")
tgt.rollback()
return 1
print()
# 3. 逐表拷贝数据
print("=" * 60)
print("阶段 2: 拷贝表数据")
print("=" * 60)
total_rows_copied = 0
auto_increments: dict[str, int | None] = {}
for idx, table in enumerate(tables, 1):
columns = get_table_columns(src, table)
if not columns:
auto_increments[table] = None
print(f" [{idx:2d}/{len(tables)}] SKIP `{table}` (0 列)")
continue
col_quoted = ", ".join("`%s`" % c for c in columns)
placeholders = ", ".join(["%s"] * len(columns))
insert_sql = "INSERT INTO `%s` (%s) VALUES (%s)" % (
table,
col_quoted,
placeholders,
)
table_count = 0
with src.cursor() as read_cur:
read_cur.execute("SELECT * FROM `%s`" % table)
batch = read_cur.fetchmany(BATCH_SIZE)
while batch:
rows_values = [
[row.get(c) for c in columns] for row in batch
]
try:
with tgt.cursor() as write_cur:
write_cur.executemany(insert_sql, rows_values)
tgt.commit()
table_count += len(rows_values)
print(
f"\r [{idx:2d}/{len(tables)}] `{table}` -> {table_count}",
end="",
flush=True,
)
except Exception as e:
print()
print(f" [{idx:2d}/{len(tables)}] 错误 `{table}`: {e}")
tgt.rollback()
return 1
batch = read_cur.fetchmany(BATCH_SIZE)
# 修正 auto_increment
ai = get_auto_increment(src, table)
auto_increments[table] = ai
if ai is not None:
with tgt.cursor() as cur:
cur.execute(
"ALTER TABLE `%s` AUTO_INCREMENT = %s" % (table, ai)
)
tgt.commit()
total_rows_copied += table_count
print(
f"\r [{idx:2d}/{len(tables)}] `{table}` -> {table_count} 行 [OK]"
)
# 4. 验证
print()
print("=" * 60)
print("阶段 3: 验证行数")
print("=" * 60)
all_match = True
src_total = 0
tgt_total = 0
for table in tables:
s = row_count(src, table)
t = row_count(tgt, table)
src_total += s
tgt_total += t
status = "OK" if s == t else "不匹配!"
if s != t:
all_match = False
print(f" `{table}`: 源={s} 目标={t} [{status}]")
print()
print(f" 总计: 源={src_total} 目标={tgt_total}")
src.close()
tgt.close()
if all_match:
print()
print("迁移完成,所有表行数一致。")
return 0
else:
print()
print("警告: 部分表行数不一致,请检查。")
return 1
if __name__ == "__main__":
sys.exit(migrate())