zy-wcs-master.git

			@@ -1,17 +1,14 @@
			package com.zy.ai.service;

			import com.baomidou.mybatisplus.mapper.EntityWrapper;
			import com.baomidou.mybatisplus.core.conditions.query.QueryWrapper;
			import com.zy.ai.entity.ChatCompletionRequest;
			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.Comparator;
			import java.util.Collections;
			import java.util.Date;
			import java.util.HashMap;
			@@ -26,9 +23,18 @@
			private static final long CACHE_TTL_MS = 3000L;

			private final LlmRouteConfigService llmRouteConfigService;
			private final LlmSpringAiClientService llmSpringAiClientService;

			private volatile List<LlmRouteConfig> allRouteCache = Collections.emptyList();
			private volatile long cacheExpireAt = 0L;
			private static final Comparator<LlmRouteConfig> ROUTE_ORDER = (a, b) -> {
			int pa = a == null \|\| a.getPriority() == null ? Integer.MAX_VALUE : a.getPriority();
			int pb = b == null \|\| b.getPriority() == null ? Integer.MAX_VALUE : b.getPriority();
			if (pa != pb) return Integer.compare(pa, pb);
			long ia = a == null \|\| a.getId() == null ? Long.MAX_VALUE : a.getId();
			long ib = b == null \|\| b.getId() == null ? Long.MAX_VALUE : b.getId();
			return Long.compare(ia, ib);
			};

			public void evictCache() {
			cacheExpireAt = 0L;
			@@ -63,9 +69,11 @@
			}
			if (result.isEmpty() && !coolingRoutes.isEmpty()) {
			// 避免所有路由都处于冷却时系统完全不可用，降级允许使用冷却路由
			coolingRoutes.sort(ROUTE_ORDER);
			log.warn("LLM 路由均处于冷却，降级启用冷却路由。cooling={}, total={}", coolingRoutes.size(), total);
			return coolingRoutes;
			}
			result.sort(ROUTE_ORDER);
			if (result.isEmpty()) {
			log.warn("未找到可用 LLM 路由。total={}, disabled={}, invalid={}", total, disabled, invalid);
			}
			@@ -75,7 +83,7 @@
			public void markSuccess(Long routeId) {
			if (routeId == null) return;
			try {
			LlmRouteConfig db = llmRouteConfigService.selectById(routeId);
			LlmRouteConfig db = llmRouteConfigService.getById(routeId);
			if (db == null) return;
			db.setSuccessCount(nvl(db.getSuccessCount()) + 1);
			db.setConsecutiveFailCount(0);
			@@ -91,7 +99,7 @@
			public void markFailure(Long routeId, String errorText, boolean enterCooldown, Integer cooldownSeconds) {
			if (routeId == null) return;
			try {
			LlmRouteConfig db = llmRouteConfigService.selectById(routeId);
			LlmRouteConfig db = llmRouteConfigService.getById(routeId);
			if (db == null) return;
			Date now = new Date();
			db.setFailCount(nvl(db.getFailCount()) + 1);
			@@ -144,10 +152,15 @@
			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;
			QueryWrapper<LlmRouteConfig> wrapper = new QueryWrapper<>();
			wrapper.orderBy(true, true, "priority").orderBy(true, true, "id");
			List<LlmRouteConfig> list = llmRouteConfigService.list(wrapper);
			if (list == null) {
			allRouteCache = Collections.emptyList();
			} else {
			list.sort(ROUTE_ORDER);
			allRouteCache = list;
			}
			cacheExpireAt = System.currentTimeMillis() + CACHE_TTL_MS;
			return allRouteCache;
			}
			@@ -226,31 +239,31 @@
			}

			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");
			ChatCompletionRequest req = new ChatCompletionRequest();
			req.setModel(cfg.getModel());
			List<ChatCompletionRequest.Message> messages = new ArrayList<>();
			ChatCompletionRequest.Message msg = new ChatCompletionRequest.Message();
			msg.setRole("user");
			msg.setContent("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();
			req.setMessages(messages);
			req.setStream(false);
			req.setMax_tokens(8);
			req.setTemperature(0D);
			if (cfg.getThinking() != null && cfg.getThinking() == 1) {
			ChatCompletionRequest.Thinking thinking = new ChatCompletionRequest.Thinking();
			thinking.setType("enable");
			req.setThinking(thinking);
			}
			try {
			LlmSpringAiClientService.CompletionCallResult result =
			llmSpringAiClientService.callCompletion(cfg.getBaseUrl(), cfg.getApiKey(), req);
			return new TestHttpResult(result.getStatusCode(), result.getPayload());
			} catch (Throwable ex) {
			Integer statusCode = llmSpringAiClientService.statusCodeOf(ex);
			String body = llmSpringAiClientService.responseBodyOf(ex, 300);
			return new TestHttpResult(statusCode == null ? -1 : statusCode, safe(body != null ? body : ex.getMessage()));
			}
			}

			private String trimBody(String body) {