From 6d29b4cb573525c1092a67ef37aacf7ef2233723 Mon Sep 17 00:00:00 2001
From: Junjie <fallin.jie@qq.com>
Date: 星期二, 03 三月 2026 11:43:07 +0800
Subject: [PATCH] #
---
src/main/java/com/zy/ai/entity/LlmRouteConfig.java | 249 +++++++
src/main/java/com/zy/ai/mapper/LlmRouteConfigMapper.java | 11
src/main/resources/sql/20260303_add_ai_config_menu.sql | 46 +
src/main/resources/sql/20260303_create_sys_llm_route.sql | 32 +
src/main/java/com/zy/ai/controller/WcsDiagnosisController.java | 4
src/main/java/com/zy/ai/service/impl/LlmRouteConfigServiceImpl.java | 11
src/main/webapp/views/ai/llm_config.html | 421 +++++++++++++
/dev/null | 55 -
src/main/java/com/zy/ai/service/LlmChatService.java | 566 +++++++++-------
src/main/java/com/zy/ai/service/LlmRouteConfigService.java | 7
src/main/java/com/zy/ai/service/LlmRoutingService.java | 270 ++++++++
src/main/java/com/zy/ai/controller/LlmRouteConfigController.java | 131 ++++
src/main/java/com/zy/ai/service/WcsDiagnosisService.java | 19
src/main/resources/mapper/LlmRouteConfigMapper.xml | 29
src/main/resources/application.yml | 28
15 files changed, 1,538 insertions(+), 341 deletions(-)
diff --git a/src/main/java/com/zy/ai/config/LlmConfig.java b/src/main/java/com/zy/ai/config/LlmConfig.java
deleted file mode 100644
index 385286d..0000000
--- a/src/main/java/com/zy/ai/config/LlmConfig.java
+++ /dev/null
@@ -1,20 +0,0 @@
-package com.zy.ai.config;
-
-import org.springframework.beans.factory.annotation.Value;
-import org.springframework.context.annotation.Bean;
-import org.springframework.context.annotation.Configuration;
-import org.springframework.web.reactive.function.client.WebClient;
-
-@Configuration
-public class LlmConfig {
-
- @Value("${llm.base-url}")
- private String baseUrl;
-
- @Bean
- public WebClient llmWebClient() {
- return WebClient.builder()
- .baseUrl(baseUrl)
- .build();
- }
-}
\ No newline at end of file
diff --git a/src/main/java/com/zy/ai/controller/LlmRouteConfigController.java b/src/main/java/com/zy/ai/controller/LlmRouteConfigController.java
new file mode 100644
index 0000000..f43f408
--- /dev/null
+++ b/src/main/java/com/zy/ai/controller/LlmRouteConfigController.java
@@ -0,0 +1,131 @@
+package com.zy.ai.controller;
+
+import com.baomidou.mybatisplus.mapper.EntityWrapper;
+import com.core.annotations.ManagerAuth;
+import com.core.common.R;
+import com.zy.ai.entity.LlmRouteConfig;
+import com.zy.ai.service.LlmRouteConfigService;
+import com.zy.ai.service.LlmRoutingService;
+import com.zy.common.web.BaseController;
+import lombok.RequiredArgsConstructor;
+import org.springframework.web.bind.annotation.*;
+
+import java.util.Date;
+import java.util.List;
+import java.util.Map;
+
+@RestController
+@RequestMapping("/ai/llm/config")
+@RequiredArgsConstructor
+public class LlmRouteConfigController extends BaseController {
+
+ private final LlmRouteConfigService llmRouteConfigService;
+ private final LlmRoutingService llmRoutingService;
+
+ @GetMapping("/list/auth")
+ @ManagerAuth
+ public R list() {
+ EntityWrapper<LlmRouteConfig> wrapper = new EntityWrapper<>();
+ wrapper.orderBy("priority", true).orderBy("id", true);
+ List<LlmRouteConfig> list = llmRouteConfigService.selectList(wrapper);
+ return R.ok(list);
+ }
+
+ @PostMapping("/save/auth")
+ @ManagerAuth
+ public R save(@RequestBody LlmRouteConfig config) {
+ if (config == null) {
+ return R.error("鍙傛暟涓嶈兘涓虹┖");
+ }
+
+ if (isBlank(config.getBaseUrl()) || isBlank(config.getApiKey()) || isBlank(config.getModel())) {
+ return R.error("蹇呴』濉啓 baseUrl/apiKey/model");
+ }
+
+ if (config.getId() == null) {
+ llmRoutingService.fillAndNormalize(config, true);
+ llmRouteConfigService.insert(config);
+ } else {
+ LlmRouteConfig db = llmRouteConfigService.selectById(config.getId());
+ if (db == null) {
+ return R.error("閰嶇疆涓嶅瓨鍦�");
+ }
+ // 淇濈暀缁熻瀛楁锛岄伩鍏嶅墠绔瑕嗙洊
+ Integer failCount = db.getFailCount();
+ Integer successCount = db.getSuccessCount();
+ Integer consecutiveFailCount = db.getConsecutiveFailCount();
+ Date lastFailTime = db.getLastFailTime();
+ Date lastUsedTime = db.getLastUsedTime();
+ String lastError = db.getLastError();
+
+ llmRoutingService.fillAndNormalize(config, false);
+ config.setFailCount(failCount);
+ config.setSuccessCount(successCount);
+ config.setConsecutiveFailCount(consecutiveFailCount);
+ config.setLastFailTime(lastFailTime);
+ config.setLastUsedTime(lastUsedTime);
+ config.setLastError(lastError);
+ config.setCreateTime(db.getCreateTime());
+ llmRouteConfigService.updateById(config);
+ }
+
+ llmRoutingService.evictCache();
+ return R.ok(config);
+ }
+
+ @PostMapping("/delete/auth")
+ @ManagerAuth
+ public R delete(@RequestParam("id") Long id) {
+ if (id == null) {
+ return R.error("id涓嶈兘涓虹┖");
+ }
+ llmRouteConfigService.deleteById(id);
+ llmRoutingService.evictCache();
+ return R.ok();
+ }
+
+ @PostMapping("/clearCooldown/auth")
+ @ManagerAuth
+ public R clearCooldown(@RequestParam("id") Long id) {
+ if (id == null) {
+ return R.error("id涓嶈兘涓虹┖");
+ }
+ LlmRouteConfig cfg = llmRouteConfigService.selectById(id);
+ if (cfg == null) {
+ return R.error("閰嶇疆涓嶅瓨鍦�");
+ }
+ cfg.setCooldownUntil(null);
+ cfg.setConsecutiveFailCount(0);
+ cfg.setUpdateTime(new Date());
+ llmRouteConfigService.updateById(cfg);
+ llmRoutingService.evictCache();
+ return R.ok();
+ }
+
+ @PostMapping("/test/auth")
+ @ManagerAuth
+ public R test(@RequestBody LlmRouteConfig config) {
+ if (config == null) {
+ return R.error("鍙傛暟涓嶈兘涓虹┖");
+ }
+ if (isBlank(config.getBaseUrl()) || isBlank(config.getApiKey()) || isBlank(config.getModel())) {
+ return R.error("娴嬭瘯澶辫触锛氬繀椤诲～鍐� baseUrl/apiKey/model");
+ }
+ Map<String, Object> data = llmRoutingService.testRoute(config);
+ if (Boolean.TRUE.equals(data.get("ok")) && config.getId() != null) {
+ LlmRouteConfig db = llmRouteConfigService.selectById(config.getId());
+ if (db != null) {
+ db.setCooldownUntil(null);
+ db.setConsecutiveFailCount(0);
+ db.setUpdateTime(new Date());
+ llmRouteConfigService.updateById(db);
+ llmRoutingService.evictCache();
+ }
+ }
+ return R.ok(data);
+ }
+
+ private boolean isBlank(String s) {
+ return s == null || s.trim().isEmpty();
+ }
+}
diff --git a/src/main/java/com/zy/ai/controller/WcsDiagnosisController.java b/src/main/java/com/zy/ai/controller/WcsDiagnosisController.java
index 15857a0..127c9ed 100644
--- a/src/main/java/com/zy/ai/controller/WcsDiagnosisController.java
+++ b/src/main/java/com/zy/ai/controller/WcsDiagnosisController.java
@@ -33,7 +33,7 @@
WcsDiagnosisRequest request = aiUtils.makeAiRequest(1000, "瀵瑰綋鍓嶇郴缁熻繘琛屽贰妫�锛屽鏋滄湁寮傚父鎯呭喌灏辫繘琛岃缁嗙殑鍒嗘瀽锛屽鏋滄病鏈夊紓甯告儏鍐靛垯褰撴垚涓�娆℃鏌n\n");
wcsDiagnosisService.diagnoseStream(request, emitter);
} catch (Exception e) {
- emitter.completeWithError(e);
+ try { emitter.complete(); } catch (Exception ignore) {}
}
}).start();
@@ -50,7 +50,7 @@
WcsDiagnosisRequest request = aiUtils.makeAiRequest(100, null);
wcsDiagnosisService.askStream(request, prompt, chatId, reset, emitter);
} catch (Exception e) {
- emitter.completeWithError(e);
+ try { emitter.complete(); } catch (Exception ignore) {}
}
}).start();
return emitter;
diff --git a/src/main/java/com/zy/ai/entity/LlmRouteConfig.java b/src/main/java/com/zy/ai/entity/LlmRouteConfig.java
new file mode 100644
index 0000000..418ccb9
--- /dev/null
+++ b/src/main/java/com/zy/ai/entity/LlmRouteConfig.java
@@ -0,0 +1,249 @@
+package com.zy.ai.entity;
+
+import com.baomidou.mybatisplus.annotations.TableField;
+import com.baomidou.mybatisplus.annotations.TableId;
+import com.baomidou.mybatisplus.annotations.TableName;
+import com.baomidou.mybatisplus.enums.IdType;
+
+import java.io.Serializable;
+import java.util.Date;
+
+@TableName("sys_llm_route")
+public class LlmRouteConfig implements Serializable {
+
+ private static final long serialVersionUID = 1L;
+
+ @TableId(value = "id", type = IdType.AUTO)
+ private Long id;
+
+ private String name;
+
+ @TableField("base_url")
+ private String baseUrl;
+
+ @TableField("api_key")
+ private String apiKey;
+
+ private String model;
+
+ /**
+ * 1 寮�鍚繁搴︽�濊�� 0 鍏抽棴
+ */
+ private Short thinking;
+
+ /**
+ * 鏁板瓧瓒婂皬浼樺厛绾ц秺楂�
+ */
+ private Integer priority;
+
+ /**
+ * 1 鍚敤 0 绂佺敤
+ */
+ private Short status;
+
+ @TableField("switch_on_quota")
+ private Short switchOnQuota;
+
+ @TableField("switch_on_error")
+ private Short switchOnError;
+
+ @TableField("cooldown_seconds")
+ private Integer cooldownSeconds;
+
+ @TableField("cooldown_until")
+ private Date cooldownUntil;
+
+ @TableField("fail_count")
+ private Integer failCount;
+
+ @TableField("success_count")
+ private Integer successCount;
+
+ @TableField("consecutive_fail_count")
+ private Integer consecutiveFailCount;
+
+ @TableField("last_error")
+ private String lastError;
+
+ @TableField("last_used_time")
+ private Date lastUsedTime;
+
+ @TableField("last_fail_time")
+ private Date lastFailTime;
+
+ @TableField("create_time")
+ private Date createTime;
+
+ @TableField("update_time")
+ private Date updateTime;
+
+ private String memo;
+
+ public Long getId() {
+ return id;
+ }
+
+ public void setId(Long id) {
+ this.id = id;
+ }
+
+ public String getName() {
+ return name;
+ }
+
+ public void setName(String name) {
+ this.name = name;
+ }
+
+ public String getBaseUrl() {
+ return baseUrl;
+ }
+
+ public void setBaseUrl(String baseUrl) {
+ this.baseUrl = baseUrl;
+ }
+
+ public String getApiKey() {
+ return apiKey;
+ }
+
+ public void setApiKey(String apiKey) {
+ this.apiKey = apiKey;
+ }
+
+ public String getModel() {
+ return model;
+ }
+
+ public void setModel(String model) {
+ this.model = model;
+ }
+
+ public Short getThinking() {
+ return thinking;
+ }
+
+ public void setThinking(Short thinking) {
+ this.thinking = thinking;
+ }
+
+ public Integer getPriority() {
+ return priority;
+ }
+
+ public void setPriority(Integer priority) {
+ this.priority = priority;
+ }
+
+ public Short getStatus() {
+ return status;
+ }
+
+ public void setStatus(Short status) {
+ this.status = status;
+ }
+
+ public Short getSwitchOnQuota() {
+ return switchOnQuota;
+ }
+
+ public void setSwitchOnQuota(Short switchOnQuota) {
+ this.switchOnQuota = switchOnQuota;
+ }
+
+ public Short getSwitchOnError() {
+ return switchOnError;
+ }
+
+ public void setSwitchOnError(Short switchOnError) {
+ this.switchOnError = switchOnError;
+ }
+
+ public Integer getCooldownSeconds() {
+ return cooldownSeconds;
+ }
+
+ public void setCooldownSeconds(Integer cooldownSeconds) {
+ this.cooldownSeconds = cooldownSeconds;
+ }
+
+ public Date getCooldownUntil() {
+ return cooldownUntil;
+ }
+
+ public void setCooldownUntil(Date cooldownUntil) {
+ this.cooldownUntil = cooldownUntil;
+ }
+
+ public Integer getFailCount() {
+ return failCount;
+ }
+
+ public void setFailCount(Integer failCount) {
+ this.failCount = failCount;
+ }
+
+ public Integer getSuccessCount() {
+ return successCount;
+ }
+
+ public void setSuccessCount(Integer successCount) {
+ this.successCount = successCount;
+ }
+
+ public Integer getConsecutiveFailCount() {
+ return consecutiveFailCount;
+ }
+
+ public void setConsecutiveFailCount(Integer consecutiveFailCount) {
+ this.consecutiveFailCount = consecutiveFailCount;
+ }
+
+ public String getLastError() {
+ return lastError;
+ }
+
+ public void setLastError(String lastError) {
+ this.lastError = lastError;
+ }
+
+ public Date getLastUsedTime() {
+ return lastUsedTime;
+ }
+
+ public void setLastUsedTime(Date lastUsedTime) {
+ this.lastUsedTime = lastUsedTime;
+ }
+
+ public Date getLastFailTime() {
+ return lastFailTime;
+ }
+
+ public void setLastFailTime(Date lastFailTime) {
+ this.lastFailTime = lastFailTime;
+ }
+
+ public Date getCreateTime() {
+ return createTime;
+ }
+
+ public void setCreateTime(Date createTime) {
+ this.createTime = createTime;
+ }
+
+ public Date getUpdateTime() {
+ return updateTime;
+ }
+
+ public void setUpdateTime(Date updateTime) {
+ this.updateTime = updateTime;
+ }
+
+ public String getMemo() {
+ return memo;
+ }
+
+ public void setMemo(String memo) {
+ this.memo = memo;
+ }
+}
diff --git a/src/main/java/com/zy/ai/mapper/LlmRouteConfigMapper.java b/src/main/java/com/zy/ai/mapper/LlmRouteConfigMapper.java
new file mode 100644
index 0000000..7397da0
--- /dev/null
+++ b/src/main/java/com/zy/ai/mapper/LlmRouteConfigMapper.java
@@ -0,0 +1,11 @@
+package com.zy.ai.mapper;
+
+import com.baomidou.mybatisplus.mapper.BaseMapper;
+import com.zy.ai.entity.LlmRouteConfig;
+import org.apache.ibatis.annotations.Mapper;
+import org.springframework.stereotype.Repository;
+
+@Mapper
+@Repository
+public interface LlmRouteConfigMapper extends BaseMapper<LlmRouteConfig> {
+}
diff --git a/src/main/java/com/zy/ai/service/LlmChatService.java b/src/main/java/com/zy/ai/service/LlmChatService.java
index ddb333a..4e6bf19 100644
--- a/src/main/java/com/zy/ai/service/LlmChatService.java
+++ b/src/main/java/com/zy/ai/service/LlmChatService.java
@@ -1,7 +1,11 @@
package com.zy.ai.service;
+import com.alibaba.fastjson.JSON;
+import com.alibaba.fastjson.JSONArray;
+import com.alibaba.fastjson.JSONObject;
import com.zy.ai.entity.ChatCompletionRequest;
import com.zy.ai.entity.ChatCompletionResponse;
+import com.zy.ai.entity.LlmRouteConfig;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import org.springframework.beans.factory.annotation.Value;
@@ -9,37 +13,33 @@
import org.springframework.http.MediaType;
import org.springframework.stereotype.Service;
import org.springframework.web.reactive.function.client.WebClient;
-import reactor.core.publisher.Mono;
import reactor.core.publisher.Flux;
-import java.util.HashMap;
+import java.util.ArrayList;
import java.util.List;
-import java.util.function.Consumer;
import java.util.concurrent.LinkedBlockingQueue;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicBoolean;
-import com.alibaba.fastjson.JSON;
-import com.alibaba.fastjson.JSONArray;
-import com.alibaba.fastjson.JSONObject;
+import java.util.function.Consumer;
@Slf4j
@Service
@RequiredArgsConstructor
public class LlmChatService {
- private final WebClient llmWebClient;
+ private final LlmRoutingService llmRoutingService;
- @Value("${llm.api-key}")
- private String apiKey;
+ @Value("${llm.base-url:}")
+ private String fallbackBaseUrl;
- @Value("${llm.model}")
- private String model;
+ @Value("${llm.api-key:}")
+ private String fallbackApiKey;
- @Value("${llm.pythonPlatformUrl}")
- private String pythonPlatformUrl;
+ @Value("${llm.model:}")
+ private String fallbackModel;
- @Value("${llm.thinking}")
- private String thinking;
+ @Value("${llm.thinking:false}")
+ private String fallbackThinking;
/**
* 閫氱敤瀵硅瘽鏂规硶锛氫紶鍏� messages锛岃繑鍥炲ぇ妯″瀷鏂囨湰鍥炲
@@ -49,27 +49,12 @@
Integer maxTokens) {
ChatCompletionRequest req = new ChatCompletionRequest();
- req.setModel(model);
req.setMessages(messages);
req.setTemperature(temperature != null ? temperature : 0.3);
req.setMax_tokens(maxTokens != null ? maxTokens : 1024);
req.setStream(false);
- ChatCompletionResponse response = llmWebClient.post()
- .uri("/chat/completions")
- .header(HttpHeaders.AUTHORIZATION, "Bearer " + apiKey)
- .contentType(MediaType.APPLICATION_JSON)
- .accept(MediaType.APPLICATION_JSON, MediaType.TEXT_EVENT_STREAM)
- .bodyValue(req)
- .exchangeToMono(resp -> resp.bodyToFlux(String.class)
- .collectList()
- .map(list -> {
- String payload = String.join("\n\n", list);
- return parseCompletion(payload);
- }))
- .doOnError(ex -> log.error("璋冪敤 LLM 澶辫触", ex))
- .onErrorResume(ex -> Mono.empty())
- .block();
+ ChatCompletionResponse response = complete(req);
if (response == null ||
response.getChoices() == null ||
@@ -88,17 +73,10 @@
Integer maxTokens,
List<Object> tools) {
ChatCompletionRequest req = new ChatCompletionRequest();
- req.setModel(model);
req.setMessages(messages);
req.setTemperature(temperature != null ? temperature : 0.3);
req.setMax_tokens(maxTokens != null ? maxTokens : 1024);
req.setStream(false);
-
- if(thinking.equals("enable")) {
- ChatCompletionRequest.Thinking thinking = new ChatCompletionRequest.Thinking();
- thinking.setType("enable");
- req.setThinking(thinking);
- }
if (tools != null && !tools.isEmpty()) {
req.setTools(tools);
req.setTool_choice("auto");
@@ -107,26 +85,42 @@
}
public ChatCompletionResponse complete(ChatCompletionRequest req) {
- try {
- return llmWebClient.post()
- .uri("/chat/completions")
- .header(HttpHeaders.AUTHORIZATION, "Bearer " + apiKey)
- .contentType(MediaType.APPLICATION_JSON)
- .accept(MediaType.APPLICATION_JSON, MediaType.TEXT_EVENT_STREAM)
- .bodyValue(req)
- .exchangeToMono(resp -> resp.bodyToFlux(String.class)
- .collectList()
- .map(list -> {
- String payload = String.join("\n\n", list);
- return parseCompletion(payload);
- }))
- .doOnError(ex -> log.error("璋冪敤 LLM 澶辫触", ex))
- .onErrorResume(ex -> Mono.empty())
- .block();
- } catch (Exception e) {
- log.error("璋冪敤 LLM 澶辫触", e);
+ List<ResolvedRoute> routes = resolveRoutes();
+ if (routes.isEmpty()) {
+ log.error("璋冪敤 LLM 澶辫触: 鏈厤缃彲鐢� LLM 璺敱");
return null;
}
+
+ Throwable last = null;
+ for (int i = 0; i < routes.size(); i++) {
+ ResolvedRoute route = routes.get(i);
+ boolean hasNext = i < routes.size() - 1;
+ try {
+ ChatCompletionRequest routeReq = applyRoute(cloneRequest(req), route, false);
+ ChatCompletionResponse resp = callCompletion(route, routeReq);
+ if (!isValidCompletion(resp)) {
+ throw new RuntimeException("LLM 鍝嶅簲涓虹┖");
+ }
+ markSuccess(route);
+ return resp;
+ } catch (Throwable ex) {
+ last = ex;
+ boolean quota = isQuotaExhausted(ex);
+ boolean canSwitch = shouldSwitch(route, quota);
+ markFailure(route, ex, canSwitch);
+ if (hasNext && canSwitch) {
+ log.warn("LLM 鍒囨崲鍒颁笅涓�璺敱, current={}, reason={}", route.tag(), errorText(ex));
+ continue;
+ }
+ log.error("璋冪敤 LLM 澶辫触, route={}", route.tag(), ex);
+ break;
+ }
+ }
+
+ if (last != null) {
+ log.error("璋冪敤 LLM 鍏ㄩ儴璺敱澶辫触: {}", errorText(last));
+ }
+ return null;
}
public void chatStream(List<ChatCompletionRequest.Message> messages,
@@ -137,92 +131,12 @@
Consumer<Throwable> onError) {
ChatCompletionRequest req = new ChatCompletionRequest();
- req.setModel(model);
req.setMessages(messages);
req.setTemperature(temperature != null ? temperature : 0.3);
req.setMax_tokens(maxTokens != null ? maxTokens : 1024);
req.setStream(true);
-
- Flux<String> flux = llmWebClient.post()
- .uri("/chat/completions")
- .header(HttpHeaders.AUTHORIZATION, "Bearer " + apiKey)
- .contentType(MediaType.APPLICATION_JSON)
- .accept(MediaType.TEXT_EVENT_STREAM)
- .bodyValue(req)
- .retrieve()
- .bodyToFlux(String.class)
- .doOnError(ex -> log.error("璋冪敤 LLM 娴佸紡澶辫触", ex));
-
- AtomicBoolean doneSeen = new AtomicBoolean(false);
- AtomicBoolean errorSeen = new AtomicBoolean(false);
- LinkedBlockingQueue<String> queue = new LinkedBlockingQueue<>();
-
- Thread drain = new Thread(() -> {
- try {
- while (true) {
- String s = queue.poll(2, TimeUnit.SECONDS);
- if (s != null) {
- try { onChunk.accept(s); } catch (Exception ignore) {}
- }
- if (doneSeen.get() && queue.isEmpty()) {
- if (!errorSeen.get()) {
- try { if (onComplete != null) onComplete.run(); } catch (Exception ignore) {}
- }
- break;
- }
- }
- } catch (InterruptedException ignore) {
- ignore.printStackTrace();
- }
- });
- drain.setDaemon(true);
- drain.start();
-
- flux.subscribe(payload -> {
- if (payload == null || payload.isEmpty()) return;
- String[] events = payload.split("\\r?\\n\\r?\\n");
- for (String part : events) {
- String s = part;
- if (s == null || s.isEmpty()) continue;
- if (s.startsWith("data:")) {
- s = s.substring(5);
- if (s.startsWith(" ")) s = s.substring(1);
- }
- if ("[DONE]".equals(s.trim())) {
- doneSeen.set(true);
- continue;
- }
- try {
- JSONObject obj = JSON.parseObject(s);
- JSONArray choices = obj.getJSONArray("choices");
- if (choices != null && !choices.isEmpty()) {
- JSONObject c0 = choices.getJSONObject(0);
- JSONObject delta = c0.getJSONObject("delta");
- if (delta != null) {
- String content = delta.getString("content");
- if (content != null) {
- try { queue.offer(content); } catch (Exception ignore) {}
- }
- }
- }
- } catch (Exception e) {
- e.printStackTrace();
- }
- }
- }, err -> {
- errorSeen.set(true);
- doneSeen.set(true);
- if (onError != null) onError.accept(err);
- }, () -> {
- if (!doneSeen.get()) {
- errorSeen.set(true);
- doneSeen.set(true);
- if (onError != null) onError.accept(new RuntimeException("LLM 娴佹剰澶栧畬鎴�"));
- } else {
- doneSeen.set(true);
- }
- });
+ streamWithFailover(req, onChunk, onComplete, onError);
}
public void chatStreamWithTools(List<ChatCompletionRequest.Message> messages,
@@ -233,120 +147,46 @@
Runnable onComplete,
Consumer<Throwable> onError) {
ChatCompletionRequest req = new ChatCompletionRequest();
- req.setModel(model);
req.setMessages(messages);
req.setTemperature(temperature != null ? temperature : 0.3);
req.setMax_tokens(maxTokens != null ? maxTokens : 1024);
req.setStream(true);
- if(thinking.equals("enable")) {
- ChatCompletionRequest.Thinking thinking = new ChatCompletionRequest.Thinking();
- thinking.setType("enable");
- req.setThinking(thinking);
- }
if (tools != null && !tools.isEmpty()) {
req.setTools(tools);
req.setTool_choice("auto");
}
- Flux<String> flux = llmWebClient.post()
- .uri("/chat/completions")
- .header(HttpHeaders.AUTHORIZATION, "Bearer " + apiKey)
- .contentType(MediaType.APPLICATION_JSON)
- .accept(MediaType.TEXT_EVENT_STREAM)
- .bodyValue(req)
- .retrieve()
- .bodyToFlux(String.class)
- .doOnError(ex -> log.error("璋冪敤 LLM 娴佸紡澶辫触", ex));
-
- AtomicBoolean doneSeen = new AtomicBoolean(false);
- AtomicBoolean errorSeen = new AtomicBoolean(false);
- LinkedBlockingQueue<String> queue = new LinkedBlockingQueue<>();
-
- Thread drain = new Thread(() -> {
- try {
- while (true) {
- String s = queue.poll(5, TimeUnit.SECONDS);
- if (s != null) {
- try { onChunk.accept(s); } catch (Exception ignore) {}
- }
- if (doneSeen.get() && queue.isEmpty()) {
- if (!errorSeen.get()) {
- try { if (onComplete != null) onComplete.run(); } catch (Exception ignore) {}
- }
- break;
- }
- }
- } catch (InterruptedException ignore) {
- ignore.printStackTrace();
- }
- });
- drain.setDaemon(true);
- drain.start();
-
- flux.subscribe(payload -> {
- if (payload == null || payload.isEmpty()) return;
- String[] events = payload.split("\\r?\\n\\r?\\n");
- for (String part : events) {
- String s = part;
- if (s == null || s.isEmpty()) continue;
- if (s.startsWith("data:")) {
- s = s.substring(5);
- if (s.startsWith(" ")) s = s.substring(1);
- }
- if ("[DONE]".equals(s.trim())) {
- doneSeen.set(true);
- continue;
- }
- try {
- JSONObject obj = JSON.parseObject(s);
- JSONArray choices = obj.getJSONArray("choices");
- if (choices != null && !choices.isEmpty()) {
- JSONObject c0 = choices.getJSONObject(0);
- JSONObject delta = c0.getJSONObject("delta");
- if (delta != null) {
- String content = delta.getString("content");
- if (content != null) {
- try { queue.offer(content); } catch (Exception ignore) {}
- }
- }
- }
- } catch (Exception e) {
- e.printStackTrace();
- }
- }
- }, err -> {
- errorSeen.set(true);
- doneSeen.set(true);
- if (onError != null) onError.accept(err);
- }, () -> {
- if (!doneSeen.get()) {
- errorSeen.set(true);
- doneSeen.set(true);
- if (onError != null) onError.accept(new RuntimeException("LLM 娴佹剰澶栧畬鎴�"));
- } else {
- doneSeen.set(true);
- }
- });
+ streamWithFailover(req, onChunk, onComplete, onError);
}
- public void chatStreamRunPython(String prompt, String chatId, Consumer<String> onChunk,
+ private void streamWithFailover(ChatCompletionRequest req,
+ Consumer<String> onChunk,
Runnable onComplete,
Consumer<Throwable> onError) {
- HashMap<String, Object> req = new HashMap<>();
- req.put("prompt", prompt);
- req.put("chatId", chatId);
+ List<ResolvedRoute> routes = resolveRoutes();
+ if (routes.isEmpty()) {
+ if (onError != null) onError.accept(new RuntimeException("鏈厤缃彲鐢� LLM 璺敱"));
+ return;
+ }
+ attemptStream(routes, 0, req, onChunk, onComplete, onError);
+ }
- Flux<String> flux = llmWebClient.post()
- .uri(pythonPlatformUrl)
- .header(HttpHeaders.AUTHORIZATION, "Bearer " + apiKey)
- .contentType(MediaType.APPLICATION_JSON)
- .accept(MediaType.TEXT_EVENT_STREAM)
- .bodyValue(req)
- .retrieve()
- .bodyToFlux(String.class)
- .doOnError(ex -> log.error("璋冪敤 LLM 娴佸紡澶辫触", ex));
+ private void attemptStream(List<ResolvedRoute> routes,
+ int index,
+ ChatCompletionRequest req,
+ Consumer<String> onChunk,
+ Runnable onComplete,
+ Consumer<Throwable> onError) {
+ if (index >= routes.size()) {
+ if (onError != null) onError.accept(new RuntimeException("LLM 璺敱鍏ㄩ儴澶辫触"));
+ return;
+ }
+
+ ResolvedRoute route = routes.get(index);
+ ChatCompletionRequest routeReq = applyRoute(cloneRequest(req), route, true);
AtomicBoolean doneSeen = new AtomicBoolean(false);
AtomicBoolean errorSeen = new AtomicBoolean(false);
+ AtomicBoolean emitted = new AtomicBoolean(false);
LinkedBlockingQueue<String> queue = new LinkedBlockingQueue<>();
Thread drain = new Thread(() -> {
@@ -354,6 +194,7 @@
while (true) {
String s = queue.poll(2, TimeUnit.SECONDS);
if (s != null) {
+ emitted.set(true);
try {
onChunk.accept(s);
} catch (Exception ignore) {
@@ -370,13 +211,12 @@
}
}
} catch (InterruptedException ignore) {
- ignore.printStackTrace();
}
});
drain.setDaemon(true);
drain.start();
- flux.subscribe(payload -> {
+ streamFlux(route, routeReq).subscribe(payload -> {
if (payload == null || payload.isEmpty()) return;
String[] events = payload.split("\\r?\\n\\r?\\n");
for (String part : events) {
@@ -390,10 +230,6 @@
doneSeen.set(true);
continue;
}
- if("<think>".equals(s.trim()) || "</think>".equals(s.trim())) {
- queue.offer(s.trim());
- continue;
- }
try {
JSONObject obj = JSON.parseObject(s);
JSONArray choices = obj.getJSONArray("choices");
@@ -403,30 +239,190 @@
if (delta != null) {
String content = delta.getString("content");
if (content != null) {
- try {
- queue.offer(content);
- } catch (Exception ignore) {
- }
+ queue.offer(content);
}
}
}
} catch (Exception e) {
- e.printStackTrace();
+ log.warn("瑙ｆ瀽 LLM stream 鐗囨澶辫触: {}", e.getMessage());
}
}
}, err -> {
errorSeen.set(true);
doneSeen.set(true);
+ boolean quota = isQuotaExhausted(err);
+ boolean canSwitch = shouldSwitch(route, quota);
+ markFailure(route, err, canSwitch);
+ if (!emitted.get() && canSwitch && index < routes.size() - 1) {
+ log.warn("LLM 璺敱澶辫触锛岃嚜鍔ㄥ垏鎹紝current={}, reason={}", route.tag(), errorText(err));
+ attemptStream(routes, index + 1, req, onChunk, onComplete, onError);
+ return;
+ }
if (onError != null) onError.accept(err);
}, () -> {
if (!doneSeen.get()) {
+ RuntimeException ex = new RuntimeException("LLM 娴佹剰澶栧畬鎴�");
errorSeen.set(true);
doneSeen.set(true);
- if (onError != null) onError.accept(new RuntimeException("LLM 娴佹剰澶栧畬鎴�"));
+ boolean canSwitch = shouldSwitch(route, false);
+ markFailure(route, ex, canSwitch);
+ if (!emitted.get() && canSwitch && index < routes.size() - 1) {
+ log.warn("LLM 璺敱娴佸紓甯稿畬鎴愶紝鑷姩鍒囨崲锛宑urrent={}", route.tag());
+ attemptStream(routes, index + 1, req, onChunk, onComplete, onError);
+ } else {
+ if (onError != null) onError.accept(ex);
+ }
} else {
+ markSuccess(route);
doneSeen.set(true);
}
});
+ }
+
+ private Flux<String> streamFlux(ResolvedRoute route, ChatCompletionRequest req) {
+ WebClient client = WebClient.builder().baseUrl(route.baseUrl).build();
+ return client.post()
+ .uri("/chat/completions")
+ .header(HttpHeaders.AUTHORIZATION, "Bearer " + route.apiKey)
+ .contentType(MediaType.APPLICATION_JSON)
+ .accept(MediaType.TEXT_EVENT_STREAM)
+ .bodyValue(req)
+ .exchangeToFlux(resp -> {
+ int status = resp.rawStatusCode();
+ if (status >= 200 && status < 300) {
+ return resp.bodyToFlux(String.class);
+ }
+ return resp.bodyToMono(String.class)
+ .defaultIfEmpty("")
+ .flatMapMany(body -> Flux.error(new LlmRouteException(status, body)));
+ })
+ .doOnError(ex -> log.error("璋冪敤 LLM 娴佸紡澶辫触, route={}", route.tag(), ex));
+ }
+
+ private ChatCompletionResponse callCompletion(ResolvedRoute route, ChatCompletionRequest req) {
+ WebClient client = WebClient.builder().baseUrl(route.baseUrl).build();
+ RawCompletionResult raw = client.post()
+ .uri("/chat/completions")
+ .header(HttpHeaders.AUTHORIZATION, "Bearer " + route.apiKey)
+ .contentType(MediaType.APPLICATION_JSON)
+ .accept(MediaType.APPLICATION_JSON, MediaType.TEXT_EVENT_STREAM)
+ .bodyValue(req)
+ .exchangeToMono(resp -> resp.bodyToFlux(String.class)
+ .collectList()
+ .map(list -> new RawCompletionResult(resp.rawStatusCode(), String.join("\\n\\n", list))))
+ .block();
+
+ if (raw == null) {
+ throw new RuntimeException("LLM 杩斿洖涓虹┖");
+ }
+ if (raw.statusCode < 200 || raw.statusCode >= 300) {
+ throw new LlmRouteException(raw.statusCode, raw.payload);
+ }
+ return parseCompletion(raw.payload);
+ }
+
+ private ChatCompletionRequest applyRoute(ChatCompletionRequest req, ResolvedRoute route, boolean stream) {
+ req.setModel(route.model);
+ req.setStream(stream);
+ if (route.thinkingEnabled) {
+ ChatCompletionRequest.Thinking t = new ChatCompletionRequest.Thinking();
+ t.setType("enable");
+ req.setThinking(t);
+ } else {
+ req.setThinking(null);
+ }
+ return req;
+ }
+
+ private ChatCompletionRequest cloneRequest(ChatCompletionRequest src) {
+ ChatCompletionRequest req = new ChatCompletionRequest();
+ req.setModel(src.getModel());
+ req.setMessages(src.getMessages());
+ req.setTemperature(src.getTemperature());
+ req.setMax_tokens(src.getMax_tokens());
+ req.setStream(src.getStream());
+ req.setTools(src.getTools());
+ req.setTool_choice(src.getTool_choice());
+ req.setThinking(src.getThinking());
+ return req;
+ }
+
+ private boolean isValidCompletion(ChatCompletionResponse response) {
+ if (response == null || response.getChoices() == null || response.getChoices().isEmpty()) {
+ return false;
+ }
+ ChatCompletionRequest.Message message = response.getChoices().get(0).getMessage();
+ if (message == null) {
+ return false;
+ }
+ if (!isBlank(message.getContent())) {
+ return true;
+ }
+ return message.getTool_calls() != null && !message.getTool_calls().isEmpty();
+ }
+
+ private boolean shouldSwitch(ResolvedRoute route, boolean quota) {
+ return quota ? route.switchOnQuota : route.switchOnError;
+ }
+
+ private void markSuccess(ResolvedRoute route) {
+ if (route.id != null) {
+ llmRoutingService.markSuccess(route.id);
+ }
+ }
+
+ private void markFailure(ResolvedRoute route, Throwable ex, boolean enterCooldown) {
+ if (route.id != null) {
+ llmRoutingService.markFailure(route.id, errorText(ex), enterCooldown, route.cooldownSeconds);
+ }
+ }
+
+ private String errorText(Throwable ex) {
+ if (ex == null) return "unknown";
+ if (ex instanceof LlmRouteException) {
+ LlmRouteException e = (LlmRouteException) ex;
+ String body = e.body == null ? "" : e.body;
+ if (body.length() > 240) {
+ body = body.substring(0, 240);
+ }
+ return "status=" + e.statusCode + ", body=" + body;
+ }
+ return ex.getMessage() == null ? ex.toString() : ex.getMessage();
+ }
+
+ private boolean isQuotaExhausted(Throwable ex) {
+ if (!(ex instanceof LlmRouteException)) return false;
+ LlmRouteException e = (LlmRouteException) ex;
+ if (e.statusCode == 429) return true;
+ String text = (e.body == null ? "" : e.body).toLowerCase();
+ return text.contains("insufficient_quota")
+ || text.contains("quota")
+ || text.contains("浣欓")
+ || text.contains("鐢ㄩ噺")
+ || text.contains("瓒呴檺")
+ || text.contains("rate limit");
+ }
+
+ private List<ResolvedRoute> resolveRoutes() {
+ List<ResolvedRoute> routes = new ArrayList<>();
+ List<LlmRouteConfig> dbRoutes = llmRoutingService.listAvailableRoutes();
+ for (LlmRouteConfig c : dbRoutes) {
+ routes.add(ResolvedRoute.fromDb(c));
+ }
+ // 鍏煎锛氭暟鎹簱涓虹┖鏃讹紝鍥為��鍒� yml
+ if (routes.isEmpty() && !isBlank(fallbackBaseUrl) && !isBlank(fallbackApiKey) && !isBlank(fallbackModel)) {
+ routes.add(ResolvedRoute.fromFallback(fallbackBaseUrl, fallbackApiKey, fallbackModel, isFallbackThinkingEnabled()));
+ }
+ return routes;
+ }
+
+ private boolean isFallbackThinkingEnabled() {
+ String x = fallbackThinking == null ? "" : fallbackThinking.trim().toLowerCase();
+ return "true".equals(x) || "1".equals(x) || "enable".equals(x);
+ }
+
+ private boolean isBlank(String s) {
+ return s == null || s.trim().isEmpty();
}
private ChatCompletionResponse mergeSseChunk(ChatCompletionResponse acc, String payload) {
@@ -452,7 +448,7 @@
ChatCompletionResponse.Choice choice = new ChatCompletionResponse.Choice();
ChatCompletionRequest.Message msg = new ChatCompletionRequest.Message();
choice.setMessage(msg);
- java.util.ArrayList<ChatCompletionResponse.Choice> list = new java.util.ArrayList<>();
+ ArrayList<ChatCompletionResponse.Choice> list = new ArrayList<>();
list.add(choice);
acc.setChoices(list);
}
@@ -490,7 +486,8 @@
if (created != null) acc.setCreated(created);
String object = obj.getString("object");
if (object != null && !object.isEmpty()) acc.setObjectName(object);
- } catch (Exception ignore) {}
+ } catch (Exception ignore) {
+ }
}
return acc;
}
@@ -502,7 +499,8 @@
if (r != null && r.getChoices() != null && !r.getChoices().isEmpty() && r.getChoices().get(0).getMessage() != null) {
return r;
}
- } catch (Exception ignore) {}
+ } catch (Exception ignore) {
+ }
ChatCompletionResponse sse = mergeSseChunk(new ChatCompletionResponse(), payload);
if (sse.getChoices() != null && !sse.getChoices().isEmpty() && sse.getChoices().get(0).getMessage() != null && sse.getChoices().get(0).getMessage().getContent() != null) {
return sse;
@@ -513,9 +511,75 @@
msg.setRole("assistant");
msg.setContent(payload);
choice.setMessage(msg);
- java.util.ArrayList<ChatCompletionResponse.Choice> list = new java.util.ArrayList<>();
+ ArrayList<ChatCompletionResponse.Choice> list = new ArrayList<>();
list.add(choice);
r.setChoices(list);
return r;
}
+
+ private static class RawCompletionResult {
+ private final int statusCode;
+ private final String payload;
+
+ private RawCompletionResult(int statusCode, String payload) {
+ this.statusCode = statusCode;
+ this.payload = payload;
+ }
+ }
+
+ private static class LlmRouteException extends RuntimeException {
+ private final int statusCode;
+ private final String body;
+
+ private LlmRouteException(int statusCode, String body) {
+ super("http status=" + statusCode);
+ this.statusCode = statusCode;
+ this.body = body;
+ }
+ }
+
+ private static class ResolvedRoute {
+ private Long id;
+ private String name;
+ private String baseUrl;
+ private String apiKey;
+ private String model;
+ private boolean thinkingEnabled;
+ private boolean switchOnQuota;
+ private boolean switchOnError;
+ private Integer cooldownSeconds;
+
+ private static ResolvedRoute fromDb(LlmRouteConfig c) {
+ ResolvedRoute r = new ResolvedRoute();
+ r.id = c.getId();
+ r.name = c.getName();
+ r.baseUrl = c.getBaseUrl();
+ r.apiKey = c.getApiKey();
+ r.model = c.getModel();
+ r.thinkingEnabled = c.getThinking() != null && c.getThinking() == 1;
+ r.switchOnQuota = c.getSwitchOnQuota() == null || c.getSwitchOnQuota() == 1;
+ r.switchOnError = c.getSwitchOnError() == null || c.getSwitchOnError() == 1;
+ r.cooldownSeconds = c.getCooldownSeconds();
+ return r;
+ }
+
+ private static ResolvedRoute fromFallback(String baseUrl, String apiKey, String model, boolean thinkingEnabled) {
+ ResolvedRoute r = new ResolvedRoute();
+ r.name = "fallback-yml";
+ r.baseUrl = baseUrl;
+ r.apiKey = apiKey;
+ r.model = model;
+ r.thinkingEnabled = thinkingEnabled;
+ r.switchOnQuota = true;
+ r.switchOnError = true;
+ r.cooldownSeconds = 300;
+ return r;
+ }
+
+ private String tag() {
+ String showName = name == null ? "unnamed" : name;
+ String showModel = model == null ? "" : (" model=" + model);
+ return showName + showModel;
+ }
+ }
}
diff --git a/src/main/java/com/zy/ai/service/LlmRouteConfigService.java b/src/main/java/com/zy/ai/service/LlmRouteConfigService.java
new file mode 100644
index 0000000..4bf082e
--- /dev/null
+++ b/src/main/java/com/zy/ai/service/LlmRouteConfigService.java
@@ -0,0 +1,7 @@
+package com.zy.ai.service;
+
+import com.baomidou.mybatisplus.service.IService;
+import com.zy.ai.entity.LlmRouteConfig;
+
+public interface LlmRouteConfigService extends IService<LlmRouteConfig> {
+}
diff --git a/src/main/java/com/zy/ai/service/LlmRoutingService.java b/src/main/java/com/zy/ai/service/LlmRoutingService.java
new file mode 100644
index 0000000..4323d6a
--- /dev/null
+++ b/src/main/java/com/zy/ai/service/LlmRoutingService.java
@@ -0,0 +1,270 @@
+package com.zy.ai.service;
+
+import com.baomidou.mybatisplus.mapper.EntityWrapper;
+import com.zy.ai.entity.LlmRouteConfig;
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.http.HttpHeaders;
+import org.springframework.http.MediaType;
+import org.springframework.stereotype.Service;
+import org.springframework.web.reactive.function.client.WebClient;
+import reactor.core.publisher.Mono;
+
+import java.time.Duration;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.Date;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+@Slf4j
+@Service
+@RequiredArgsConstructor
+public class LlmRoutingService {
+
+ private static final long CACHE_TTL_MS = 3000L;
+
+ private final LlmRouteConfigService llmRouteConfigService;
+
+ private volatile List<LlmRouteConfig> allRouteCache = Collections.emptyList();
+ private volatile long cacheExpireAt = 0L;
+
+ public void evictCache() {
+ cacheExpireAt = 0L;
+ }
+
+ public List<LlmRouteConfig> listAllOrdered() {
+ return new ArrayList<>(loadAllRoutes());
+ }
+
+ public List<LlmRouteConfig> listAvailableRoutes() {
+ Date now = new Date();
+ List<LlmRouteConfig> result = new ArrayList<>();
+ List<LlmRouteConfig> coolingRoutes = new ArrayList<>();
+ int total = 0;
+ int disabled = 0;
+ int invalid = 0;
+ for (LlmRouteConfig c : loadAllRoutes()) {
+ total++;
+ if (!isEnabled(c)) {
+ disabled++;
+ continue;
+ }
+ if (isBlank(c.getBaseUrl()) || isBlank(c.getApiKey()) || isBlank(c.getModel())) {
+ invalid++;
+ continue;
+ }
+ if (isCooling(c, now)) {
+ coolingRoutes.add(c);
+ continue;
+ }
+ result.add(c);
+ }
+ if (result.isEmpty() && !coolingRoutes.isEmpty()) {
+ // 閬垮厤鎵�鏈夎矾鐢遍兘澶勪簬鍐峰嵈鏃剁郴缁熷畬鍏ㄤ笉鍙敤锛岄檷绾у厑璁镐娇鐢ㄥ喎鍗磋矾鐢�
+ log.warn("LLM 璺敱鍧囧浜庡喎鍗达紝闄嶇骇鍚敤鍐峰嵈璺敱銆俢ooling={}, total={}", coolingRoutes.size(), total);
+ return coolingRoutes;
+ }
+ if (result.isEmpty()) {
+ log.warn("鏈壘鍒板彲鐢� LLM 璺敱銆倀otal={}, disabled={}, invalid={}", total, disabled, invalid);
+ }
+ return result;
+ }
+
+ public void markSuccess(Long routeId) {
+ if (routeId == null) return;
+ try {
+ LlmRouteConfig db = llmRouteConfigService.selectById(routeId);
+ if (db == null) return;
+ db.setSuccessCount(nvl(db.getSuccessCount()) + 1);
+ db.setConsecutiveFailCount(0);
+ db.setLastUsedTime(new Date());
+ db.setUpdateTime(new Date());
+ llmRouteConfigService.updateById(db);
+ evictCache();
+ } catch (Exception e) {
+ log.warn("鏇存柊璺敱鎴愬姛鐘舵�佸け璐�, routeId={}", routeId, e);
+ }
+ }
+
+ public void markFailure(Long routeId, String errorText, boolean enterCooldown, Integer cooldownSeconds) {
+ if (routeId == null) return;
+ try {
+ LlmRouteConfig db = llmRouteConfigService.selectById(routeId);
+ if (db == null) return;
+ Date now = new Date();
+ db.setFailCount(nvl(db.getFailCount()) + 1);
+ db.setConsecutiveFailCount(nvl(db.getConsecutiveFailCount()) + 1);
+ db.setLastFailTime(now);
+ db.setLastError(trimError(errorText));
+ if (enterCooldown) {
+ int sec = cooldownSeconds != null && cooldownSeconds > 0
+ ? cooldownSeconds
+ : defaultCooldown(db.getCooldownSeconds());
+ db.setCooldownUntil(new Date(now.getTime() + sec * 1000L));
+ }
+ db.setUpdateTime(now);
+ llmRouteConfigService.updateById(db);
+ evictCache();
+ } catch (Exception e) {
+ log.warn("鏇存柊璺敱澶辫触鐘舵�佸け璐�, routeId={}", routeId, e);
+ }
+ }
+
+ private int defaultCooldown(Integer sec) {
+ return sec == null || sec <= 0 ? 300 : sec;
+ }
+
+ private String trimError(String err) {
+ if (err == null) return null;
+ String x = err.replace("\n", " ").replace("\r", " ");
+ return x.length() > 500 ? x.substring(0, 500) : x;
+ }
+
+ private Integer nvl(Integer x) {
+ return x == null ? 0 : x;
+ }
+
+ private boolean isEnabled(LlmRouteConfig c) {
+ return c != null && c.getStatus() != null && c.getStatus() == 1;
+ }
+
+ private boolean isCooling(LlmRouteConfig c, Date now) {
+ return c != null && c.getCooldownUntil() != null && c.getCooldownUntil().after(now);
+ }
+
+ private List<LlmRouteConfig> loadAllRoutes() {
+ long now = System.currentTimeMillis();
+ if (now < cacheExpireAt && allRouteCache != null) {
+ return allRouteCache;
+ }
+ synchronized (this) {
+ now = System.currentTimeMillis();
+ if (now < cacheExpireAt && allRouteCache != null) {
+ return allRouteCache;
+ }
+ EntityWrapper<LlmRouteConfig> wrapper = new EntityWrapper<>();
+ wrapper.orderBy("priority", true).orderBy("id", true);
+ List<LlmRouteConfig> list = llmRouteConfigService.selectList(wrapper);
+ allRouteCache = list == null ? Collections.emptyList() : list;
+ cacheExpireAt = System.currentTimeMillis() + CACHE_TTL_MS;
+ return allRouteCache;
+ }
+ }
+
+ private String safe(String s) {
+ return s == null ? "" : s.trim();
+ }
+
+ private boolean isBlank(String s) {
+ return s == null || s.trim().isEmpty();
+ }
+
+ public LlmRouteConfig fillAndNormalize(LlmRouteConfig cfg, boolean isCreate) {
+ Date now = new Date();
+ if (isBlank(cfg.getName())) {
+ cfg.setName("LLM_ROUTE_" + now.getTime());
+ }
+ if (cfg.getThinking() == null) {
+ cfg.setThinking((short) 0);
+ }
+ if (cfg.getPriority() == null) {
+ cfg.setPriority(100);
+ }
+ if (cfg.getStatus() == null) {
+ cfg.setStatus((short) 1);
+ }
+ if (cfg.getSwitchOnQuota() == null) {
+ cfg.setSwitchOnQuota((short) 1);
+ }
+ if (cfg.getSwitchOnError() == null) {
+ cfg.setSwitchOnError((short) 1);
+ }
+ if (cfg.getCooldownSeconds() == null || cfg.getCooldownSeconds() < 0) {
+ cfg.setCooldownSeconds(300);
+ }
+ if (cfg.getFailCount() == null) {
+ cfg.setFailCount(0);
+ }
+ if (cfg.getSuccessCount() == null) {
+ cfg.setSuccessCount(0);
+ }
+ if (cfg.getConsecutiveFailCount() == null) {
+ cfg.setConsecutiveFailCount(0);
+ }
+ if (isCreate) {
+ cfg.setCreateTime(now);
+ }
+ cfg.setUpdateTime(now);
+ return cfg;
+ }
+
+ public Map<String, Object> testRoute(LlmRouteConfig cfg) {
+ HashMap<String, Object> result = new HashMap<>();
+ long start = System.currentTimeMillis();
+ try {
+ TestHttpResult raw = testJavaRoute(cfg);
+ fillTestResult(result, raw, start);
+ } catch (Exception e) {
+ result.put("ok", false);
+ result.put("statusCode", -1);
+ result.put("latencyMs", System.currentTimeMillis() - start);
+ result.put("message", "娴嬭瘯寮傚父: " + safe(e.getMessage()));
+ result.put("responseSnippet", "");
+ }
+ return result;
+ }
+
+ private void fillTestResult(HashMap<String, Object> result, TestHttpResult raw, long start) {
+ boolean ok = raw.statusCode >= 200 && raw.statusCode < 300;
+ result.put("ok", ok);
+ result.put("statusCode", raw.statusCode);
+ result.put("latencyMs", System.currentTimeMillis() - start);
+ result.put("message", ok ? "娴嬭瘯鎴愬姛" : "娴嬭瘯澶辫触");
+ result.put("responseSnippet", trimBody(raw.body));
+ }
+
+ private TestHttpResult testJavaRoute(LlmRouteConfig cfg) {
+ HashMap<String, Object> req = new HashMap<>();
+ req.put("model", cfg.getModel());
+ List<Map<String, String>> messages = new ArrayList<>();
+ HashMap<String, String> msg = new HashMap<>();
+ msg.put("role", "user");
+ msg.put("content", "ping");
+ messages.add(msg);
+ req.put("messages", messages);
+ req.put("stream", false);
+ req.put("max_tokens", 8);
+ req.put("temperature", 0);
+
+ WebClient client = WebClient.builder().baseUrl(cfg.getBaseUrl()).build();
+ return client.post()
+ .uri("/chat/completions")
+ .header(HttpHeaders.AUTHORIZATION, "Bearer " + cfg.getApiKey())
+ .contentType(MediaType.APPLICATION_JSON)
+ .accept(MediaType.APPLICATION_JSON, MediaType.TEXT_EVENT_STREAM)
+ .bodyValue(req)
+ .exchangeToMono(resp -> resp.bodyToMono(String.class)
+ .defaultIfEmpty("")
+ .map(body -> new TestHttpResult(resp.rawStatusCode(), body)))
+ .timeout(Duration.ofSeconds(12))
+ .onErrorResume(ex -> Mono.just(new TestHttpResult(-1, safe(ex.getMessage()))))
+ .block();
+ }
+
+ private String trimBody(String body) {
+ String x = safe(body).replace("\r", " ").replace("\n", " ");
+ return x.length() > 300 ? x.substring(0, 300) : x;
+ }
+
+ private static class TestHttpResult {
+ private final int statusCode;
+ private final String body;
+
+ private TestHttpResult(int statusCode, String body) {
+ this.statusCode = statusCode;
+ this.body = body;
+ }
+ }
+}
diff --git a/src/main/java/com/zy/ai/service/PythonService.java b/src/main/java/com/zy/ai/service/PythonService.java
deleted file mode 100644
index adc712b..0000000
--- a/src/main/java/com/zy/ai/service/PythonService.java
+++ /dev/null
@@ -1,55 +0,0 @@
-package com.zy.ai.service;
-
-import lombok.extern.slf4j.Slf4j;
-import org.springframework.beans.factory.annotation.Autowired;
-import org.springframework.stereotype.Component;
-import org.springframework.web.servlet.mvc.method.annotation.SseEmitter;
-
-@Slf4j
-@Component
-public class PythonService {
-
- @Autowired
- private LlmChatService llmChatService;
-
- public boolean runPython(String prompt, String chatId, SseEmitter emitter) {
- try {
- llmChatService.chatStreamRunPython(prompt, chatId, s -> {
- try {
- String safe = s == null ? "" : s.replace("\r", "").replace("\n", "\\n");
- if (!safe.isEmpty()) {
- sse(emitter, safe);
- }
- } catch (Exception ignore) {
- }
- }, () -> {
- try {
- sse(emitter, "\\n\\n銆怉I銆戣繍琛屽凡鍋滄锛堟甯哥粨鏉燂級\\n\\n");
- log.info("AI MCP diagnose stopped: final end");
- emitter.complete();
- } catch (Exception ignore) {
- }
- }, e -> {
- sse(emitter, "\\n\\n銆怉I銆戝垎鏋愬嚭閿欙紝姝ｅ湪鍥為��...\\n\\n");
- });
- return true;
- } catch (Exception e) {
- try {
- sse(emitter, "\\n\\n銆怉I銆戣繍琛屽凡鍋滄锛堝紓甯革級\\n\\n");
- log.error("AI MCP diagnose stopped: error", e);
- emitter.completeWithError(e);
- } catch (Exception ignore) {}
- return true;
- }
- }
-
- private void sse(SseEmitter emitter, String data) {
- if (data == null) return;
- try {
- emitter.send(SseEmitter.event().data(data));
- } catch (Exception e) {
- log.warn("SSE send failed", e);
- }
- }
-
-}
diff --git a/src/main/java/com/zy/ai/service/WcsDiagnosisService.java b/src/main/java/com/zy/ai/service/WcsDiagnosisService.java
index 189e57d..45dea58 100644
--- a/src/main/java/com/zy/ai/service/WcsDiagnosisService.java
+++ b/src/main/java/com/zy/ai/service/WcsDiagnosisService.java
@@ -11,7 +11,6 @@
import com.zy.common.utils.RedisUtil;
import com.zy.core.enums.RedisKeyType;
import org.springframework.beans.factory.annotation.Autowired;
-import org.springframework.beans.factory.annotation.Value;
import org.springframework.web.servlet.mvc.method.annotation.SseEmitter;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
@@ -28,8 +27,6 @@
private static final long CHAT_TTL_SECONDS = 7L * 24 * 3600;
- @Value("${llm.platform}")
- private String platform;
@Autowired
private LlmChatService llmChatService;
@Autowired
@@ -40,8 +37,6 @@
private AiUtils aiUtils;
@Autowired(required = false)
private McpController mcpController;
- @Autowired
- private PythonService pythonService;
public void diagnoseStream(WcsDiagnosisRequest request, SseEmitter emitter) {
List<ChatCompletionRequest.Message> messages = new ArrayList<>();
@@ -85,7 +80,7 @@
try {
try { emitter.send(SseEmitter.event().data("銆怉I銆戣繍琛屽凡鍋滄锛堝紓甯革級")); } catch (Exception ignore) {}
log.error("AI diagnose stream stopped: error", e);
- emitter.completeWithError(e);
+ emitter.complete();
} catch (Exception ignore) {}
});
}
@@ -95,11 +90,6 @@
String chatId,
boolean reset,
SseEmitter emitter) {
- if (platform.equals("python")) {
- pythonService.runPython(prompt, chatId, emitter);
- return;
- }
-
List<ChatCompletionRequest.Message> messages = new ArrayList<>();
List<ChatCompletionRequest.Message> history = null;
@@ -187,7 +177,10 @@
emitter.complete();
} catch (Exception ignore) {}
}, e -> {
- try { emitter.completeWithError(e); } catch (Exception ignore) {}
+ try {
+ try { emitter.send(SseEmitter.event().data("銆怉I銆戣繍琛屽凡鍋滄锛堝紓甯革級")); } catch (Exception ignore) {}
+ emitter.complete();
+ } catch (Exception ignore) {}
});
}
@@ -380,7 +373,7 @@
try {
sse(emitter, "\\n\\n銆怉I銆戣繍琛屽凡鍋滄锛堝紓甯革級\\n\\n");
log.error("AI MCP diagnose stopped: error", e);
- emitter.completeWithError(e);
+ emitter.complete();
} catch (Exception ignore) {}
return true;
}
diff --git a/src/main/java/com/zy/ai/service/impl/LlmRouteConfigServiceImpl.java b/src/main/java/com/zy/ai/service/impl/LlmRouteConfigServiceImpl.java
new file mode 100644
index 0000000..0315c11
--- /dev/null
+++ b/src/main/java/com/zy/ai/service/impl/LlmRouteConfigServiceImpl.java
@@ -0,0 +1,11 @@
+package com.zy.ai.service.impl;
+
+import com.baomidou.mybatisplus.service.impl.ServiceImpl;
+import com.zy.ai.entity.LlmRouteConfig;
+import com.zy.ai.mapper.LlmRouteConfigMapper;
+import com.zy.ai.service.LlmRouteConfigService;
+import org.springframework.stereotype.Service;
+
+@Service("llmRouteConfigService")
+public class LlmRouteConfigServiceImpl extends ServiceImpl<LlmRouteConfigMapper, LlmRouteConfig> implements LlmRouteConfigService {
+}
diff --git a/src/main/resources/application.yml b/src/main/resources/application.yml
index fe77635..116f9b4 100644
--- a/src/main/resources/application.yml
+++ b/src/main/resources/application.yml
@@ -1,6 +1,6 @@
# 绯荤粺鐗堟湰淇℃伅
app:
- version: 1.0.4.2
+ version: 1.0.4.3
version-type: dev # prd 鎴� dev
server:
@@ -83,27 +83,15 @@
expireDays: 7
llm:
- platform: java
- pythonPlatformUrl: http://127.0.0.1:9000/ai/diagnose/askStream
- thinking: enable
-# base-url: https://api.siliconflow.cn/v1
-# api-key: sk-sxdtebtquwrugzrmaqqqkzdzmrgzhzmplwwuowysdasccent
-# model: deepseek-ai/DeepSeek-V3.2
-# base-url: http://47.76.147.249:9998/e/7g7kqxxt1ei2un71
-# api-key: app-mP0O6aY5WpbfaHs7BNnjVkli
-# model: deepseek-ai/DeepSeek-V3.2
-# base-url: http://34.2.134.223:3000/v1
-# api-key: sk-WabrmtOezCFwVo7XvVOrO3QkmfcKG7T7jy0BaVnmQTWm5GXh
-# model: gemini-3-pro-preview
-# base-url: http://127.0.0.1:8317/v1
-# api-key: WznOjAGJNVFKSe9kBZTr
-# model: gpt-5
- base-url: https://api.xiaomimimo.com/v1
- api-key: sk-cw7e4se9cal8cxdgjml8dmtn4pdmqtvfccg5fcermt0ddtys
- model: mimo-v2-flash
+ # 鐜板凡杩佺Щ鍒版暟鎹簱琛� sys_llm_route 缁存姢锛堟敮鎸佸API/澶氭ā鍨�/澶欿ey鑷姩鍒囨崲锛�
+ # 浠ヤ笅浠呬綔涓烘暟鎹簱涓虹┖鏃剁殑鍏煎鍥為��閰嶇疆
+ thinking: false
+ base-url:
+ api-key:
+ model:
perf:
methodTiming:
enabled: false
thresholdMs: 50
- sampleRate: 1.0
\ No newline at end of file
+ sampleRate: 1.0
diff --git a/src/main/resources/mapper/LlmRouteConfigMapper.xml b/src/main/resources/mapper/LlmRouteConfigMapper.xml
new file mode 100644
index 0000000..1930a0c
--- /dev/null
+++ b/src/main/resources/mapper/LlmRouteConfigMapper.xml
@@ -0,0 +1,29 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE mapper PUBLIC "-//mybatis.org//DTD Mapper 3.0//EN" "http://mybatis.org/dtd/mybatis-3-mapper.dtd">
+<mapper namespace="com.zy.ai.mapper.LlmRouteConfigMapper">
+
+ <resultMap id="BaseResultMap" type="com.zy.ai.entity.LlmRouteConfig">
+ <id column="id" property="id"/>
+ <result column="name" property="name"/>
+ <result column="base_url" property="baseUrl"/>
+ <result column="api_key" property="apiKey"/>
+ <result column="model" property="model"/>
+ <result column="thinking" property="thinking"/>
+ <result column="priority" property="priority"/>
+ <result column="status" property="status"/>
+ <result column="switch_on_quota" property="switchOnQuota"/>
+ <result column="switch_on_error" property="switchOnError"/>
+ <result column="cooldown_seconds" property="cooldownSeconds"/>
+ <result column="cooldown_until" property="cooldownUntil"/>
+ <result column="fail_count" property="failCount"/>
+ <result column="success_count" property="successCount"/>
+ <result column="consecutive_fail_count" property="consecutiveFailCount"/>
+ <result column="last_error" property="lastError"/>
+ <result column="last_used_time" property="lastUsedTime"/>
+ <result column="last_fail_time" property="lastFailTime"/>
+ <result column="create_time" property="createTime"/>
+ <result column="update_time" property="updateTime"/>
+ <result column="memo" property="memo"/>
+ </resultMap>
+
+</mapper>
diff --git a/src/main/resources/sql/20260303_add_ai_config_menu.sql b/src/main/resources/sql/20260303_add_ai_config_menu.sql
new file mode 100644
index 0000000..58c9a97
--- /dev/null
+++ b/src/main/resources/sql/20260303_add_ai_config_menu.sql
@@ -0,0 +1,46 @@
+-- 灏� AI閰嶇疆 鑿滃崟鎸傝浇鍒帮細寮�鍙戜笓鐢� -> AI閰嶇疆
+-- 璇存槑锛氭湰绯荤粺鑿滃崟鏉ユ簮浜� sys_resource锛屾墽琛屾湰鑴氭湰鍚庤鍦ㄢ�滆鑹叉巿鏉冣�濋噷缁欏搴旇鑹插嬀閫夋柊鑿滃崟銆�
+
+-- 1) 瀹氫綅鈥滃紑鍙戜笓鐢ㄢ�濅竴绾ц彍鍗�
+SET @dev_parent_id := (
+ SELECT id
+ FROM sys_resource
+ WHERE name = '寮�鍙戜笓鐢�' AND level = 1
+ ORDER BY id
+ LIMIT 1
+);
+
+-- 2) 鏂板浜岀骇鑿滃崟锛欰I閰嶇疆锛堥〉闈級
+INSERT INTO sys_resource(code, name, resource_id, level, sort, status)
+SELECT 'ai/llm_config.html', 'AI閰嶇疆', @dev_parent_id, 2, 999, 1
+FROM dual
+WHERE @dev_parent_id IS NOT NULL
+ AND NOT EXISTS (
+ SELECT 1
+ FROM sys_resource
+ WHERE code = 'ai/llm_config.html' AND level = 2
+ );
+
+-- 3) 鏂板涓夌骇鎸夐挳鏉冮檺锛氭煡鐪嬶紙鐢ㄤ簬瑙掕壊缁嗙矑搴︽巿鏉冿級
+SET @ai_cfg_id := (
+ SELECT id
+ FROM sys_resource
+ WHERE code = 'ai/llm_config.html' AND level = 2
+ ORDER BY id
+ LIMIT 1
+);
+
+INSERT INTO sys_resource(code, name, resource_id, level, sort, status)
+SELECT 'ai/llm_config.html#view', '鏌ョ湅', @ai_cfg_id, 3, 1, 1
+FROM dual
+WHERE @ai_cfg_id IS NOT NULL
+ AND NOT EXISTS (
+ SELECT 1
+ FROM sys_resource
+ WHERE code = 'ai/llm_config.html#view' AND level = 3
+ );
+
+-- 鍙�夋鏌�
+SELECT id, code, name, resource_id, level, sort, status
+FROM sys_resource
+WHERE code IN ('ai/llm_config.html', 'ai/llm_config.html#view');
diff --git a/src/main/resources/sql/20260303_create_sys_llm_route.sql b/src/main/resources/sql/20260303_create_sys_llm_route.sql
new file mode 100644
index 0000000..47d67ff
--- /dev/null
+++ b/src/main/resources/sql/20260303_create_sys_llm_route.sql
@@ -0,0 +1,32 @@
+CREATE TABLE IF NOT EXISTS `sys_llm_route` (
+ `id` BIGINT NOT NULL AUTO_INCREMENT COMMENT '涓婚敭',
+ `name` VARCHAR(64) NOT NULL COMMENT '璺敱鍚嶇О',
+ `base_url` VARCHAR(255) DEFAULT NULL COMMENT 'LLM API Base URL',
+ `api_key` VARCHAR(512) DEFAULT NULL COMMENT 'API Key',
+ `model` VARCHAR(128) DEFAULT NULL COMMENT '妯″瀷鍚�',
+ `thinking` TINYINT NOT NULL DEFAULT 0 COMMENT '鏄惁寮�鍚繁搴︽�濊��:1鏄�0鍚�',
+ `priority` INT NOT NULL DEFAULT 100 COMMENT '浼樺厛绾�(瓒婂皬瓒婁紭鍏�)',
+ `status` TINYINT NOT NULL DEFAULT 1 COMMENT '鐘舵��:1鍚敤0绂佺敤',
+ `switch_on_quota` TINYINT NOT NULL DEFAULT 1 COMMENT '棰濆害鑰楀敖鏃舵槸鍚﹀垏鎹�',
+ `switch_on_error` TINYINT NOT NULL DEFAULT 1 COMMENT '鏁呴殰鏃舵槸鍚﹀垏鎹�',
+ `cooldown_seconds` INT NOT NULL DEFAULT 300 COMMENT '鏁呴殰鍚庡喎鍗寸鏁�',
+ `cooldown_until` DATETIME DEFAULT NULL COMMENT '鍐峰嵈鎴鏃堕棿',
+ `fail_count` INT NOT NULL DEFAULT 0 COMMENT '鎬诲け璐ユ鏁�',
+ `success_count` INT NOT NULL DEFAULT 0 COMMENT '鎬绘垚鍔熸鏁�',
+ `consecutive_fail_count` INT NOT NULL DEFAULT 0 COMMENT '杩炵画澶辫触娆℃暟',
+ `last_error` VARCHAR(512) DEFAULT NULL COMMENT '鏈�杩戦敊璇憳瑕�',
+ `last_used_time` DATETIME DEFAULT NULL COMMENT '鏈�杩戞垚鍔熻皟鐢ㄦ椂闂�',
+ `last_fail_time` DATETIME DEFAULT NULL COMMENT '鏈�杩戝け璐ユ椂闂�',
+ `create_time` DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '鍒涘缓鏃堕棿',
+ `update_time` DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP COMMENT '鏇存柊鏃堕棿',
+ `memo` VARCHAR(255) DEFAULT NULL COMMENT '澶囨敞',
+ PRIMARY KEY (`id`),
+ KEY `idx_sys_llm_route_status_priority` (`status`, `priority`),
+ KEY `idx_sys_llm_route_cooldown` (`cooldown_until`)
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COMMENT='LLM 璺敱閰嶇疆(澶欰PI/澶氭ā鍨�/澶欿ey鑷姩鍒囨崲)';
+
+-- 绀轰緥鏁版嵁锛堟寜闇�淇敼鍚庢墽琛岋級
+-- INSERT INTO sys_llm_route(name, base_url, api_key, model, thinking, priority, status, switch_on_quota, switch_on_error, cooldown_seconds)
+-- VALUES ('涓昏矾鐢�-gpt5', 'https://api.xiaomimimo.com/v1', 'sk-xxxx', 'gpt-5', 1, 10, 1, 1, 1, 300);
+-- INSERT INTO sys_llm_route(name, base_url, api_key, model, thinking, priority, status, switch_on_quota, switch_on_error, cooldown_seconds)
+-- VALUES ('澶囪矾鐢�-mimo', 'https://api.xiaomimimo.com/v1', 'sk-yyyy', 'mimo-v2-flash', 0, 20, 1, 1, 1, 300);
diff --git a/src/main/webapp/views/ai/llm_config.html b/src/main/webapp/views/ai/llm_config.html
new file mode 100644
index 0000000..dbbbc6a
--- /dev/null
+++ b/src/main/webapp/views/ai/llm_config.html
@@ -0,0 +1,421 @@
+<!DOCTYPE html>
+<html lang="zh-CN">
+<head>
+ <meta charset="UTF-8" />
+ <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+ <title>AI閰嶇疆</title>
+ <link rel="stylesheet" href="../../static/vue/element/element.css" />
+ <style>
+ body {
+ margin: 0;
+ background:
+ radial-gradient(1200px 500px at 10% -10%, rgba(26, 115, 232, 0.14), transparent 50%),
+ radial-gradient(900px 450px at 100% 0%, rgba(38, 166, 154, 0.11), transparent 55%),
+ #f4f7fb;
+ }
+ .container {
+ max-width: 1640px;
+ margin: 16px auto;
+ padding: 0 14px;
+ }
+ .hero {
+ background: linear-gradient(135deg, #0f4c81 0%, #1f6fb2 45%, #2aa198 100%);
+ color: #fff;
+ border-radius: 14px;
+ padding: 14px 16px;
+ margin-bottom: 10px;
+ box-shadow: 0 10px 28px rgba(23, 70, 110, 0.22);
+ }
+ .hero-top {
+ display: flex;
+ align-items: center;
+ justify-content: space-between;
+ gap: 10px;
+ }
+ .hero-title {
+ display: flex;
+ align-items: center;
+ gap: 10px;
+ }
+ .hero-title .main {
+ font-size: 16px;
+ font-weight: 700;
+ letter-spacing: 0.2px;
+ }
+ .hero-title .sub {
+ font-size: 12px;
+ opacity: 0.9;
+ }
+ .summary-grid {
+ margin-top: 10px;
+ display: grid;
+ grid-template-columns: repeat(5, minmax(0, 1fr));
+ gap: 8px;
+ }
+ .summary-card {
+ border-radius: 10px;
+ background: rgba(255, 255, 255, 0.16);
+ border: 1px solid rgba(255, 255, 255, 0.24);
+ padding: 8px 10px;
+ min-height: 56px;
+ backdrop-filter: blur(3px);
+ }
+ .summary-card .k {
+ font-size: 11px;
+ opacity: 0.88;
+ }
+ .summary-card .v {
+ margin-top: 4px;
+ font-size: 22px;
+ font-weight: 700;
+ line-height: 1.1;
+ }
+ .table-shell {
+ border-radius: 12px;
+ overflow: hidden;
+ box-shadow: 0 6px 22px rgba(15, 28, 48, 0.08);
+ border: 1px solid #e8edf5;
+ background: #fff;
+ }
+ .mono {
+ font-family: Menlo, Monaco, Consolas, "Liberation Mono", monospace;
+ font-size: 12px;
+ }
+ @media (max-width: 1280px) {
+ .summary-grid { grid-template-columns: repeat(2, minmax(0, 1fr)); }
+ }
+ </style>
+</head>
+<body>
+<div id="app" class="container">
+ <div class="hero">
+ <div class="hero-top">
+ <div class="hero-title">
+ <div v-html="headerIcon" style="display:flex;"></div>
+ <div>
+ <div class="main">AI閰嶇疆 - LLM璺敱</div>
+ <div class="sub">鏀寔澶欰PI銆佸妯″瀷銆佸Key锛岄搴﹁�楀敖鎴栨晠闅滆嚜鍔ㄥ垏鎹�</div>
+ </div>
+ </div>
+ <div>
+ <el-button type="primary" size="mini" @click="addRoute">鏂板璺敱</el-button>
+ <el-button size="mini" @click="loadRoutes">鍒锋柊</el-button>
+ </div>
+ </div>
+ <div class="summary-grid">
+ <div class="summary-card">
+ <div class="k">鎬昏矾鐢�</div>
+ <div class="v">{{ summary.total }}</div>
+ </div>
+ <div class="summary-card">
+ <div class="k">鍚敤</div>
+ <div class="v">{{ summary.enabled }}</div>
+ </div>
+ <div class="summary-card">
+ <div class="k">鏁呴殰鍒囨崲寮�鍚�</div>
+ <div class="v">{{ summary.errorSwitch }}</div>
+ </div>
+ <div class="summary-card">
+ <div class="k">棰濆害鍒囨崲寮�鍚�</div>
+ <div class="v">{{ summary.quotaSwitch }}</div>
+ </div>
+ <div class="summary-card">
+ <div class="k">鍐峰嵈涓�</div>
+ <div class="v">{{ summary.cooling }}</div>
+ </div>
+ </div>
+ </div>
+
+ <div class="table-shell">
+ <el-table :data="routes" stripe height="72vh" v-loading="loading" :header-cell-style="{background:'#f7f9fc', color:'#2e3a4d', fontWeight:600}">
+ <el-table-column label="鍚嶇О" width="170">
+ <template slot-scope="scope">
+ <el-input v-model="scope.row.name" size="mini"></el-input>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="Base URL" min-width="220">
+ <template slot-scope="scope">
+ <el-input v-model="scope.row.baseUrl" class="mono" size="mini" placeholder="蹇呭～锛屼緥濡�: https://api.deepseek.com"></el-input>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="妯″瀷" width="180">
+ <template slot-scope="scope">
+ <el-input v-model="scope.row.model" class="mono" size="mini" placeholder="蹇呭～锛屼緥濡�: deepseek-chat"></el-input>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="API Key" min-width="220">
+ <template slot-scope="scope">
+ <el-input v-model="scope.row.apiKey" class="mono" type="password" size="mini" placeholder="蹇呭～"></el-input>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="浼樺厛绾�" width="90">
+ <template slot-scope="scope">
+ <el-input-number v-model="scope.row.priority" size="mini" :min="0" :max="99999" :controls="false" style="width:80px;"></el-input-number>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="鐘舵��" width="70">
+ <template slot-scope="scope">
+ <el-switch v-model="scope.row.status" :active-value="1" :inactive-value="0"></el-switch>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="鎬濊��" width="70">
+ <template slot-scope="scope">
+ <el-switch v-model="scope.row.thinking" :active-value="1" :inactive-value="0"></el-switch>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="棰濆害鍒囨崲" width="90">
+ <template slot-scope="scope">
+ <el-switch v-model="scope.row.switchOnQuota" :active-value="1" :inactive-value="0"></el-switch>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="鏁呴殰鍒囨崲" width="90">
+ <template slot-scope="scope">
+ <el-switch v-model="scope.row.switchOnError" :active-value="1" :inactive-value="0"></el-switch>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="鍐峰嵈绉掓暟" width="100">
+ <template slot-scope="scope">
+ <el-input-number v-model="scope.row.cooldownSeconds" size="mini" :min="0" :max="86400" :controls="false" style="width:90px;"></el-input-number>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="缁熻" min-width="220">
+ <template slot-scope="scope">
+ <div>鎴愬姛: {{ scope.row.successCount || 0 }} / 澶辫触: {{ scope.row.failCount || 0 }} / 杩炵画澶辫触: {{ scope.row.consecutiveFailCount || 0 }}</div>
+ <div style="color:#909399;">鍐峰嵈鍒�: {{ scope.row.cooldownUntil || '-' }}</div>
+ <div style="color:#909399;">鏈�杩戦敊璇�: {{ scope.row.lastError || '-' }}</div>
+ </template>
+ </el-table-column>
+
+ <el-table-column label="鎿嶄綔" width="120" fixed="right" align="center">
+ <template slot-scope="scope">
+ <el-dropdown trigger="click" @command="function(cmd){ handleRouteCommand(cmd, scope.row, scope.$index); }">
+ <el-button size="mini" type="primary" plain>
+ 鎿嶄綔<i class="el-icon-arrow-down el-icon--right"></i>
+ </el-button>
+ <el-dropdown-menu slot="dropdown">
+ <el-dropdown-item command="test" :disabled="scope.row.__testing === true">
+ {{ scope.row.__testing === true ? '娴嬭瘯涓�...' : '娴嬭瘯' }}
+ </el-dropdown-item>
+ <el-dropdown-item command="save">淇濆瓨</el-dropdown-item>
+ <el-dropdown-item command="cooldown">娓呭喎鍗�</el-dropdown-item>
+ <el-dropdown-item command="delete" divided>鍒犻櫎</el-dropdown-item>
+ </el-dropdown-menu>
+ </el-dropdown>
+ </template>
+ </el-table-column>
+ </el-table>
+ </div>
+</div>
+
+<script type="text/javascript" src="../../static/vue/js/vue.min.js"></script>
+<script type="text/javascript" src="../../static/vue/element/element.js"></script>
+<script type="text/javascript" src="../../static/js/common.js" charset="utf-8"></script>
+<script>
+ new Vue({
+ el: '#app',
+ data: function() {
+ return {
+ headerIcon: getAiIconHtml(34, 34),
+ loading: false,
+ routes: []
+ };
+ },
+ computed: {
+ summary: function() {
+ var now = Date.now();
+ var total = this.routes.length;
+ var enabled = 0, quotaSwitch = 0, errorSwitch = 0, cooling = 0;
+ for (var i = 0; i < this.routes.length; i++) {
+ var x = this.routes[i];
+ if (x.status === 1) enabled++;
+ if (x.switchOnQuota === 1) quotaSwitch++;
+ if (x.switchOnError === 1) errorSwitch++;
+ if (x.cooldownUntil && new Date(x.cooldownUntil).getTime() > now) cooling++;
+ }
+ return { total: total, enabled: enabled, quotaSwitch: quotaSwitch, errorSwitch: errorSwitch, cooling: cooling };
+ }
+ },
+ methods: {
+ authHeaders: function() {
+ return { 'token': localStorage.getItem('token') };
+ },
+ handleRouteCommand: function(command, route, idx) {
+ if (command === 'test') return this.testRoute(route);
+ if (command === 'save') return this.saveRoute(route);
+ if (command === 'cooldown') return this.clearCooldown(route);
+ if (command === 'delete') return this.deleteRoute(route, idx);
+ },
+ loadRoutes: function() {
+ var self = this;
+ self.loading = true;
+ fetch(baseUrl + '/ai/llm/config/list/auth', { headers: self.authHeaders() })
+ .then(function(r){ return r.json(); })
+ .then(function(res){
+ self.loading = false;
+ if (res && res.code === 200) {
+ self.routes = Array.isArray(res.data) ? res.data : [];
+ } else {
+ self.$message.error((res && res.msg) ? res.msg : '鍔犺浇澶辫触');
+ }
+ })
+ .catch(function(){
+ self.loading = false;
+ self.$message.error('鍔犺浇澶辫触');
+ });
+ },
+ addRoute: function() {
+ this.routes.unshift({
+ id: null,
+ name: '',
+ baseUrl: '',
+ apiKey: '',
+ model: '',
+ thinking: 0,
+ priority: 100,
+ status: 1,
+ switchOnQuota: 1,
+ switchOnError: 1,
+ cooldownSeconds: 300,
+ successCount: 0,
+ failCount: 0,
+ consecutiveFailCount: 0,
+ cooldownUntil: null,
+ lastError: null
+ });
+ },
+ buildPayload: function(route) {
+ return {
+ id: route.id,
+ name: route.name,
+ baseUrl: route.baseUrl,
+ apiKey: route.apiKey,
+ model: route.model,
+ thinking: route.thinking,
+ priority: route.priority,
+ status: route.status,
+ switchOnQuota: route.switchOnQuota,
+ switchOnError: route.switchOnError,
+ cooldownSeconds: route.cooldownSeconds,
+ memo: route.memo
+ };
+ },
+ saveRoute: function(route) {
+ var self = this;
+ fetch(baseUrl + '/ai/llm/config/save/auth', {
+ method: 'POST',
+ headers: Object.assign({ 'Content-Type': 'application/json' }, self.authHeaders()),
+ body: JSON.stringify(self.buildPayload(route))
+ })
+ .then(function(r){ return r.json(); })
+ .then(function(res){
+ if (res && res.code === 200) {
+ self.$message.success('淇濆瓨鎴愬姛');
+ self.loadRoutes();
+ } else {
+ self.$message.error((res && res.msg) ? res.msg : '淇濆瓨澶辫触');
+ }
+ })
+ .catch(function(){
+ self.$message.error('淇濆瓨澶辫触');
+ });
+ },
+ deleteRoute: function(route, idx) {
+ var self = this;
+ if (!route.id) {
+ self.routes.splice(idx, 1);
+ return;
+ }
+ self.$confirm('纭畾鍒犻櫎璇ヨ矾鐢卞悧锛�', '鎻愮ず', { type: 'warning' }).then(function() {
+ fetch(baseUrl + '/ai/llm/config/delete/auth?id=' + encodeURIComponent(route.id), {
+ method: 'POST',
+ headers: self.authHeaders()
+ })
+ .then(function(r){ return r.json(); })
+ .then(function(res){
+ if (res && res.code === 200) {
+ self.$message.success('鍒犻櫎鎴愬姛');
+ self.loadRoutes();
+ } else {
+ self.$message.error((res && res.msg) ? res.msg : '鍒犻櫎澶辫触');
+ }
+ })
+ .catch(function(){
+ self.$message.error('鍒犻櫎澶辫触');
+ });
+ }).catch(function(){});
+ },
+ clearCooldown: function(route) {
+ var self = this;
+ if (!route.id) return;
+ fetch(baseUrl + '/ai/llm/config/clearCooldown/auth?id=' + encodeURIComponent(route.id), {
+ method: 'POST',
+ headers: self.authHeaders()
+ })
+ .then(function(r){ return r.json(); })
+ .then(function(res){
+ if (res && res.code === 200) {
+ self.$message.success('宸叉竻闄ゅ喎鍗�');
+ self.loadRoutes();
+ } else {
+ self.$message.error((res && res.msg) ? res.msg : '鎿嶄綔澶辫触');
+ }
+ })
+ .catch(function(){
+ self.$message.error('鎿嶄綔澶辫触');
+ });
+ },
+ testRoute: function(route) {
+ var self = this;
+ if (route.__testing === true) return;
+ if (!route.id) {
+ self.$message.warning('褰撳墠鏄湭淇濆瓨閰嶇疆锛屾祴璇曢�氳繃鍚庝粛闇�鍏堜繚瀛樻墠浼氱敓鏁�');
+ }
+ self.$set(route, '__testing', true);
+ fetch(baseUrl + '/ai/llm/config/test/auth', {
+ method: 'POST',
+ headers: Object.assign({ 'Content-Type': 'application/json' }, self.authHeaders()),
+ body: JSON.stringify(self.buildPayload(route))
+ })
+ .then(function(r){ return r.json(); })
+ .then(function(res){
+ if (!res || res.code !== 200) {
+ self.$message.error((res && res.msg) ? res.msg : '娴嬭瘯澶辫触');
+ return;
+ }
+ var data = res.data || {};
+ var ok = data.ok === true;
+ var title = ok ? '娴嬭瘯鎴愬姛' : '娴嬭瘯澶辫触';
+ var msg = ''
+ + '璺敱: ' + (route.name || '-') + '\n'
+ + 'Base URL: ' + (route.baseUrl || '-') + '\n'
+ + '鐘舵�佺爜: ' + (data.statusCode != null ? data.statusCode : '-') + '\n'
+ + '鑰楁椂: ' + (data.latencyMs != null ? data.latencyMs : '-') + ' ms\n'
+ + '缁撴灉: ' + (data.message || '-') + '\n'
+ + '杩斿洖鐗囨: ' + (data.responseSnippet || '-');
+ self.$alert(msg, title, { confirmButtonText: '纭畾', type: ok ? 'success' : 'error' });
+ })
+ .catch(function(){
+ self.$message.error('娴嬭瘯澶辫触');
+ })
+ .finally(function(){
+ self.$set(route, '__testing', false);
+ });
+ }
+ },
+ mounted: function() {
+ this.loadRoutes();
+ }
+ });
+</script>
+</body>
+</html>
--
Gitblit v1.9.1