3人参与 • 2025-12-09 • Java
现在大模型开发特别火,deepseek 因为中文理解好、反应快、还便宜,不少 java 开发者都用它。但实际写代码的时候,好多人因为没搞懂 api 机制、java 并发这些东西,踩了不少坑:token 过期把服务搞挂、多线程调用数据乱了、长文本传进去直接报错…… 结合项目的经验,把最常踩的 8 个坑拆明白,从为啥会踩坑、怎么解决,到能直接用的代码,全给你说明白,帮你顺顺利利把 deepseek 集成到项目里。
deepseek 的 token 一般能用 30 天,好多人一开始就把 token 写在代码里,根本没考虑过期的事。一旦 token 过期,所有 api 调用全返回 401 鉴权失败,如果没做异常处理,就会直接导致依赖该接口的业务服务中断。更糟的是,有的代码连异常都不捕获,直接抛个运行时异常,把线程池堵死,最后服务都熔断了。
// 错误示例:硬编码token,无过期处理、无异常捕获
public class deepseekclient {
// 写死的 token,过期直接失效
private static final string deepseek_token = "sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx";
private static final string api_url = "https://api.deepseek.com/v1/chat/completions";
public string callapi(string prompt) {
// 直接拼 token,不管过没过期
httpheaders headers = new httpheaders();
headers.set("authorization", "bearer " + deepseek_token);
headers.setcontenttype(mediatype.application_json);
// 拼请求体
map<string, object> requestbody = new hashmap<>();
requestbody.put("model", "deepseek-chat");
requestbody.put("messages", collections.singletonlist(
map.of("role", "user", "content", prompt)
));
resttemplate resttemplate = new resttemplate();
// 鉴权失败直接抛异常,服务跟着崩
responseentity<string> response = resttemplate.postforentity(api_url, new httpentity<>(requestbody, headers), string.class);
return response.getbody();
}
}
解决方案:实现 token 自动刷新 + 异常兜底 核心思路:
读取配置的工具类
import java.io.ioexception;
import java.io.inputstream;
import java.util.properties;
// 读取配置文件的工具
public class propertiesutils {
private static final properties props = new properties();
static {
try (inputstream in = propertiesutils.class.getclassloader().getresourceasstream("deepseek.properties")) {
props.load(in);
} catch (ioexception e) {
throw new runtimeexception("加载deepseek配置文件失败", e);
}
}
public static string getproperty(string key) {
string value = props.getproperty(key);
if (value == null) {
throw new runtimeexception("配置项[" + key + "]不存在");
}
return value;
}
}
然后是 token 管理类和 api 调用客户端:
import org.springframework.http.*;
import org.springframework.web.client.httpclienterrorexception;
import org.springframework.web.client.resttemplate;
import java.util.hashmap;
import java.util.map;
import java.util.concurrent.locks.reentrantlock;
// token 管理工具,自动刷新,线程安全
public class deepseektokenmanager {
// 从配置文件读
private static final string refresh_token = propertiesutils.getproperty("deepseek.refresh.token");
private static final string token_refresh_url = "https://api.deepseek.com/v1/auth/refresh";
// 30天有效期,提前1天刷新,单位都是毫秒
private static final long token_valid_period = 30 * 24 * 60 * 60 * 1000l;
private static final long refresh_advance = 24 * 60 * 60 * 1000l;
// 当前能用的token、过期时间、刷新用的锁
private volatile string currenttoken;
private volatile long expiretime;
private final reentrantlock refreshlock = new reentrantlock();
private final resttemplate resttemplate = new resttemplate();
// 单例模式
private static class singletonholder {
private static final deepseektokenmanager instance = new deepseektokenmanager();
}
public static deepseektokenmanager getinstance() {
return singletonholder.instance;
}
// 初始化的时候就加载第一个token
private deepseektokenmanager() {
refreshtoken();
}
// 拿能用的token,自动判断要不要刷新
public string getvalidtoken() {
long now = system.currenttimemillis();
// 现在的时间加提前量,快到过期时间就刷新
if (now + refresh_advance >= expiretime) {
refreshtoken();
}
return currenttoken;
}
// 刷新token
private void refreshtoken() {
if (system.currenttimemillis() + refresh_advance < expiretime) {
return;
}
// 加锁后再检查一遍,防止刚释放锁别人又进来了
refreshlock.lock();
try {
if (system.currenttimemillis() + refresh_advance < expiretime) {
return;
}
// 拼刷新token的请求
httpheaders headers = new httpheaders();
headers.setcontenttype(mediatype.application_json);
map<string, string> requestbody = new hashmap<>();
requestbody.put("refresh_token", refresh_token);
responseentity<map> response = resttemplate.postforentity(
token_refresh_url,
new httpentity<>(requestbody, headers),
map.class
);
if (response.getstatuscode() == httpstatus.ok) {
map<string, object> resbody = response.getbody();
this.currenttoken = (string) resbody.get("access_token");
this.expiretime = system.currenttimemillis() + token_valid_period;
system.out.println("token刷新成功,新的过期时间:" + expiretime);
} else {
throw new runtimeexception("token刷新失败,响应码:" + response.getstatuscode());
}
} catch (exception e) {
system.err.println("token刷新出问题了:" + e.getmessage());
// 临时延长10分钟,给运维留时间处理
this.expiretime = system.currenttimemillis() + 10 * 60 * 1000l;
throw new runtimeexception("token刷新失败,已临时续命10分钟,赶紧查!", e);
} finally {
refreshlock.unlock();
}
}
}
// deepseek api 调用客户端(集成token管理)
public class deepseekapiclient {
private static final string api_url = "https://api.deepseek.com/v1/chat/completions";
private final resttemplate resttemplate = new resttemplate();
private final deepseektokenmanager tokenmanager = deepseektokenmanager.getinstance();
public string calldeepseek(string prompt) {
// 获取有效token
string token = tokenmanager.getvalidtoken();
httpheaders headers = new httpheaders();
headers.set("authorization", "bearer " + token);
headers.setcontenttype(mediatype.application_json);
// 构建请求体
map<string, object> requestbody = new hashmap<>();
requestbody.put("model", "deepseek-chat");
requestbody.put("messages", collections.singletonlist(
map.of("role", "user", "content", prompt)
));
requestbody.put("timeout", 30000); // 30秒超时
httpentity<map<string, object>> request = new httpentity<>(requestbody, headers);
try {
responseentity<string> response = resttemplate.postforentity(api_url, request, string.class);
return response.getbody();
} catch (httpclienterrorexception.unauthorized e) {
system.err.println("鉴权失败,强制刷新token重试:" + e.getmessage());
tokenmanager.refreshtoken();
// 重试请求
headers.set("authorization", "bearer " + tokenmanager.getvalidtoken());
httpentity<map<string, object>> retryreq = new httpentity<>(requestbody, headers);
return resttemplate.postforentity(api_url, retryreq, string.class).getbody();
} catch (exception e) {
system.err.println("api调用出错:" + e.getmessage());
return "{\"choices\":[{\"message\":{\"content\":\"稍等一下,请求有点问题~\"}}]}";
}
}
}
好多人写代码的时候,resttemplate 不配置连接池,还把 httpheaders 这种东西做成全局共享的。多线程一并发就出问题:
// 错误示例:共享非线程安全对象,无连接池,无线程池隔离
public class unsafedeepseekclient {
// 错误:resttemplate未配置连接池,高并发下连接泄漏
private static final resttemplate resttemplate = new resttemplate();
// 错误:共享httpheaders,多线程下数据错乱
private static final httpheaders sharedheaders = new httpheaders();
static {
sharedheaders.set("authorization", "bearer sk-xxxxxxxx");
sharedheaders.setcontenttype(mediatype.application_json);
}
// 错误:无线程池,直接同步调用,超时阻塞主线程
public string concurrentcall(string prompt) {
map<string, object> requestbody = new hashmap<>(); // hashmap也不是线程安全的
requestbody.put("model", "deepseek-chat");
requestbody.put("messages", collections.singletonlist(
map.of("role", "user", "content", prompt)
));
// 多线程下sharedheaders可能被篡改
httpentity<map<string, object>> request = new httpentity<>(requestbody, sharedheaders);
// 无超时配置,调用超时阻塞线程
responseentity<string> response = resttemplate.postforentity(api_url, request, string.class);
return response.getbody();
}
}
核心思路:
配置 resttemplate 连接池,限制最大连接数、超时时间,避免连接泄漏;
import org.apache.http.client.config.requestconfig;
import org.apache.http.impl.client.closeablehttpclient;
import org.apache.http.impl.client.httpclientbuilder;
import org.apache.http.impl.conn.poolinghttpclientconnectionmanager;
import org.springframework.http.client.httpcomponentsclienthttprequestfactory;
import org.springframework.web.client.resttemplate;
import java.util.hashmap;
import java.util.map;
import java.util.concurrent.*;
// 线程安全的deepseek api客户端(含连接池、线程池配置)
public class threadsafedeepseekclient {
// 带连接池的resttemplate
private static final resttemplate resttemplate;
// 配置线程池:隔离deepseek api调用,避免影响核心业务
private static final executorservice deepseekexecutor;
static {
// 1. 先配http连接池,控制最大连接数
poolinghttpclientconnectionmanager connmanager = new poolinghttpclientconnectionmanager();
connmanager.setmaxtotal(100); // 总共最多100个连接
connmanager.setdefaultmaxperroute(50); // 同一个地址最多50个连接
// 超时配置,别卡太久
requestconfig requestconfig = requestconfig.custom()
.setconnecttimeout(5000) // 连服务器5秒超时
.setconnectionrequesttimeout(3000) // 从连接池拿连接3秒超时
.setsockettimeout(30000) // 读数据30秒超时
.build();
closeablehttpclient httpclient = httpclientbuilder.create()
.setconnectionmanager(connmanager)
.setdefaultrequestconfig(requestconfig)
.build();
// 把连接池配置给resttemplate
httpcomponentsclienthttprequestfactory requestfactory = new httpcomponentsclienthttprequestfactory(httpclient);
resttemplate = new resttemplate(requestfactory);
// 2. 再配个专用线程池
deepseekexecutor = new threadpoolexecutor(
10, // 平时保持10个活线程
50, // 最多扩到50个线程
60l, timeunit.seconds, // 空闲线程60秒没人用就关掉
new linkedblockingqueue<>(1000), // 任务排队最多1000个
new threadfactory() {
private int count = 0;
@override
public thread newthread(runnable r) {
thread thread = new thread(r);
thread.setname("deepseek-api-thread-" + count++);
thread.setdaemon(true); // 守护线程,程序关的时候不用等它
return thread;
}
},
new threadpoolexecutor.callerrunspolicy() // 任务满了就让调用者自己执行,别丢任务
);
}
private final deepseektokenmanager tokenmanager = deepseektokenmanager.getinstance();
// 异步调用deepseek api(线程安全)
public completablefuture<string> asynccall(string prompt) {
// 把任务丢到专用线程池里
return completablefuture.supplyasync(() -> {
// 关键:每个请求自己建请求头,别共享
httpheaders headers = new httpheaders();
headers.set("authorization", "bearer " + tokenmanager.getvalidtoken());
headers.setcontenttype(mediatype.application_json);
// 关键:每个请求独立创建请求体,避免hashmap线程不安全问题
map<string, object> requestbody = new hashmap<>();
requestbody.put("model", "deepseek-chat");
requestbody.put("messages", collections.singletonlist(
map.of("role", "user", "content", prompt)
));
requestbody.put("temperature", 0.7); // 随机性调中等
requestbody.put("max_tokens", 2000); // 最多返回2000个token
httpentity<map<string, object>> request = new httpentity<>(requestbody, headers);
try {
responseentity<string> response = resttemplate.postforentity(api_url, request, string.class);
return response.getbody();
} catch (exception e) {
system.err.println("线程" + thread.currentthread().getname() + "调用出错:" + e.getmessage());
return "{\"choices\":[{\"message\":{\"content\":\"请求失败,稍后再试~\"}}]}";
}
}, deepseekexecutor);
}
// 关闭线程池(应用关闭时调用)
public void shutdownexecutor() {
deepseekexecutor.shutdown();
try {
// 等10秒,还没关完就强制关
if (!deepseekexecutor.awaittermination(10, timeunit.seconds)) {
deepseekexecutor.shutdownnow();
}
} catch (interruptedexception e) {
deepseekexecutor.shutdownnow();
}
}
}
deepseek 的模型都有 token 限制,比如 deepseek-chat 单轮最多大概 8192 个 token。好多人不管文本多长都直接传,要么返回 400 说“超出长度”,要么自己瞎截断把关键信息切没了,模型回复得乱七八糟。更坑的是,有人按字数算长度,不知道中文和英文占的 token 不一样,切完还是超。
// 错误示例:直接传入超长文本,无分块、无token计算
public class longtexterrorclient {
public string calllongtext(string longcontent) {
map<string, object> requestbody = new hashmap<>();
requestbody.put("model", "deepseek-chat");
// 直接把超长文本丢进去,不管长度
requestbody.put("messages", collections.singletonlist(
map.of("role", "user", "content", longcontent)
));
httpheaders headers = new httpheaders();
headers.set("authorization", "bearer " + deepseektokenmanager.getinstance().getvalidtoken());
httpentity<map<string, object>> request = new httpentity<>(requestbody, headers);
resttemplate resttemplate = new resttemplate();
return resttemplate.postforentity(api_url, request, string.class).getbody();
}
}
核心思路:
先加依赖(maven),这个工具能精准算 token:
<dependency>
<groupid>com.knuddels</groupid>
<artifactid>jtokkit</artifactid>
<version>1.0.0</version>
</dependency>
然后是长文本处理工具:
import com.knuddels.jtokkit.encodings;
import com.knuddels.jtokkit.api.encoding;
import com.knuddels.jtokkit.api.encodingregistry;
import com.knuddels.jtokkit.api.modeltype;
import java.util.arraylist;
import java.util.list;
import java.util.concurrent.completablefuture;
import java.util.stream.collectors;
// 长文本处理工具,算token、分块、拼结果一条龙
public class longtextprocessor {
// 初始化token计算器,deepseek用的编码和gpt-4一样
private static final encodingregistry registry = encodings.newdefaultencodingregistry();
private static final encoding encoding = registry.getencodingformodel(modeltype.gpt_4);
// 单轮最大token数8192,留2000给模型回复,所以请求最多6192个token
private static final int max_request_tokens = 8192 - 2000;
private final threadsafedeepseekclient deepseekclient = new threadsafedeepseekclient();
// 算文本的token数,比按字数准多了
public int counttokens(string text) {
return encoding.counttokens(text);
}
// 超长文本分块,尽量不切句子
public list<string> splitlongtext(string longtext) {
list<string> chunks = new arraylist<>();
int totaltokens = counttokens(longtext);
// 没超限制就直接返回
if (totaltokens <= max_request_tokens) {
chunks.add(longtext);
return chunks;
}
// 按中文和英文的句号分割句子,保持意思完整
string[] sentences = longtext.split("(?<=[。!?.?!])");
stringbuilder currentchunk = new stringbuilder();
int currenttokens = 0;
for (string sentence : sentences) {
int sentencetokens = counttokens(sentence);
// 极端情况:一个句子就超了,那就按token硬切
if (sentencetokens > max_request_tokens) {
chunks.addall(splitoverlengthsentence(sentence, max_request_tokens));
continue;
}
// 加当前句子会超的话,先把之前的存起来
if (currenttokens + sentencetokens > max_request_tokens) {
chunks.add(currentchunk.tostring().trim());
currentchunk = new stringbuilder();
currenttokens = 0;
}
currentchunk.append(sentence);
currenttokens += sentencetokens;
}
// 把最后一块加进去
if (currentchunk.length() > 0) {
chunks.add(currentchunk.tostring().trim());
}
return chunks;
}
// 单个句子超token限制,按token硬切
private list<string> splitoverlengthsentence(string sentence, int maxtokens) {
list<string> subchunks = new arraylist<>();
char[] chars = sentence.tochararray();
stringbuilder subchunk = new stringbuilder();
int currenttokens = 0;
for (char c : chars) {
string charstr = string.valueof(c);
int chartokens = counttokens(charstr);
if (currenttokens + chartokens > maxtokens) {
subchunks.add(subchunk.tostring().trim());
subchunk = new stringbuilder();
currenttokens = 0;
}
subchunk.append(c);
currenttokens += chartokens;
}
if (subchunk.length() > 0) {
subchunks.add(subchunk.tostring().trim());
}
return subchunks;
}
// 分块调用api,最后拼结果
public string processlongtext(string longtext) {
list<string> chunks = splitlongtext(longtext);
// 就一块的话直接调
if (chunks.size() == 1) {
return deepseekclient.asynccall(chunks.get(0)).join();
}
// 多块的话异步调用,效率高
list<completablefuture<string>> futures = chunks.stream()
.map(chunk -> {
// 告诉模型这是第几块,总共多少块,让它有上下文
string prompt = "请处理以下文本片段(一共" + chunks.size() + "段,这是第" + (chunks.indexof(chunk) + 1) + "段):\n" + chunk;
return deepseekclient.asynccall(prompt);
})
.collect(collectors.tolist());
// 等所有调用都完成
completablefuture<void> alldone = completablefuture.allof(futures.toarray(new completablefuture[0]));
alldone.join();
// 拼结果
stringbuilder finalresult = new stringbuilder();
finalresult.append("超长文本处理结果(按片段拼接):\n");
for (completablefuture<string> future : futures) {
try {
string chunkres = future.get();
// 提取回复内容,实际项目里用fastjson或者jackson解析更靠谱
string content = extractcontent(chunkres);
finalresult.append(content).append("\n");
} catch (exception e) {
finalresult.append("【这段处理失败】:").append(e.getmessage()).append("\n");
}
}
return finalresult.tostring();
}
// 从api响应里把回复内容抠出来(简化版,实际用json库)
private string extractcontent(string response) {
int contentstart = response.indexof("\"content\":\"") + 10;
int contentend = response.indexof("\"}", contentstart);
if (contentstart > 0 && contentend > contentstart) {
return response.substring(contentstart, contentend);
}
return response;
}
}
不同模型的名称规范不同,若将其他模型的名称直接套用在 deepseek 上,会返回 404 错误(模型不存在)。
搞个枚举类存deepseek的模型名:
// deepseek模型名枚举,直接拿过来用
public class deepseekmodelenum {
// 通用对话、代码生成、推理增强,常用的就这三个
public static final string deepseek_chat = "deepseek-chat";
public static final string deepseek_coder = "deepseek-coder";
public static final string deepseek_r1 = "deepseek-r1";
// 如果你之前用别的模型,用这个方法转成deepseek的
public static string convertfromothermodel(string othermodelname) {
switch (othermodelname.tolowercase()) {
// 之前用其他ai的对话场景,转成deepseek-chat
case "gpt-3.5-turbo":
case "ernie-bot":
return deepseek_chat;
// 之前用代码模型的,转成deepseek-coder
case "code-davinci-002":
return deepseek_coder;
// 其他情况默认用通用对话模型
default:
return deepseek_chat;
}
}
}
// 调用示例
public class modelclient {
public string callwithrightmodel(string prompt) {
map<string, object> requestbody = new hashmap<>();
// 直接用枚举里的模型名,肯定不会错
requestbody.put("model", deepseekmodelenum.deepseek_chat);
// 其他参数...
return "";
}
}
虽然deepseek响应格式和openai像,但有些字段不一样,比如finish_reason的取值、usage里的统计方式。有人直接抄其他ai的解析代码,结果要么字段拿不到,要么报解析异常。
搞个专门的解析工具,兼容这些差异:
import com.alibaba.fastjson.json;
import com.alibaba.fastjson.jsonobject;
// deepseek响应解析工具类(适配字段差异)
public class deepseekresponseparser {
// 提取回复内容
public static string getcontent(string responsejson) {
try {
jsonobject root = json.parseobject(responsejson);
// 兼容deepseek和openai的响应结构
if (root.containskey("choices") && !root.getjsonarray("choices").isempty()) {
jsonobject choice = root.getjsonarray("choices").getjsonobject(0);
// deepseek的message字段与openai一致,但需判空
if (choice.containskey("message")) {
return choice.getjsonobject("message").getstring("content");
}
}
// 要是有error字段,直接抛异常
if (root.containskey("error")) {
throw new runtimeexception("api报错:" + root.getjsonobject("error").getstring("message"));
}
return "没拿到有效回复";
} catch (exception e) {
throw new runtimeexception("解析响应失败:" + e.getmessage());
}
}
// 解析token使用量(适配deepseek的usage字段)
public static int getusedtokens(string responsejson) {
jsonobject root = json.parseobject(responsejson);
if (root.containskey("usage")) {
return root.getjsonobject("usage").getintvalue("total_tokens");
}
return 0;
}
}
// 调用示例
public class parserclient {
public void parseresponse(string response) {
// 直接拿回复内容,不用自己处理json
string content = deepseekresponseparser.getcontent(response);
// 看看用了多少token
int usedtokens = deepseekresponseparser.getusedtokens(response);
system.out.println("回复:" + content);
system.out.println("消耗token:" + usedtokens);
}
}
deepseek api 的响应速度与模型类型、文本长度相关(如 deepseek-coder 处理代码时响应较慢),若直接复用其他ai的超时配置(如 10 秒),会导致频繁超时;反之,超时配置过长会导致线程阻塞。
// 动态超时配置工具
public class timeoutconfig {
// 不同模型的基础超时时间(毫秒)
private static final int chat_timeout = 30000; // 对话模型30秒
private static final int coder_timeout = 60000; // 代码模型60秒
private static final int r1_timeout = 45000; // 推理模型45秒
// 按模型拿基础超时
public static int getbasetimeout(string modelname) {
switch (modelname) {
case deepseekmodelenum.deepseek_coder:
return coder_timeout;
case deepseekmodelenum.deepseek_r1:
return r1_timeout;
default:
return chat_timeout;
}
}
// 按文本长度加超时,长文本给更多时间
public static int getdynamictimeout(string modelname, string text) {
int basetimeout = getbasetimeout(modelname);
int textlen = text.length();
// 每1000字加5秒,最多不超过基础超时的2倍(别无限加)
int extratimeout = math.min((textlen / 1000) * 5000, basetimeout);
return basetimeout + extratimeout;
}
}
// 调用示例
public class timeoutclient {
public string callwithdynamictimeout(string prompt) {
string model = deepseekmodelenum.deepseek_coder;
// 按模型和文本长度算超时
int timeout = timeoutconfig.getdynamictimeout(model, prompt);
map<string, object> requestbody = new hashmap<>();
requestbody.put("model", model);
requestbody.put("messages", collections.singletonlist(map.of("role", "user", "content", prompt)));
requestbody.put("timeout", timeout); // 把算好的超时传进去
// 其他请求逻辑...
return "";
}
}
有些参数在别的模型里有用,但deepseek不支持。
import java.util.hashmap;
import java.util.list;
import java.util.map;
// 请求参数适配工具
public class paramsadapter {
// deepseek不支持的参数列表
private static final list<string> unsupported_params = list.of(
"frequency_penalty", "presence_penalty", "logit_bias"
);
// 适配参数,保证传过去的都能用
public static map<string, object> adapt(map<string, object> originalparams) {
map<string, object> adaptedparams = new hashmap<>(originalparams);
// 1. 把不支持的参数删掉
unsupported_params.foreach(adaptedparams::remove);
// 2. 修正temperature:只能0-1
if (adaptedparams.containskey("temperature")) {
double temp = (double) adaptedparams.get("temperature");
// 小于0取0,大于1取1,中间的不变
adaptedparams.put("temperature", math.min(math.max(temp, 0.0), 1.0));
}
// 3. 修正max_tokens:最少10,最多4096
if (adaptedparams.containskey("max_tokens")) {
int maxtokens = (int) adaptedparams.get("max_tokens");
adaptedparams.put("max_tokens", math.min(math.max(maxtokens, 10), 4096));
}
return adaptedparams;
}
}
// 调用示例
public class paramsclient {
public string callwithrightparams(string prompt) {
// 原来的参数,可能有无效的
map<string, object> originalparams = new hashmap<>();
originalparams.put("model", deepseekmodelenum.deepseek_chat);
originalparams.put("temperature", 1.5); // 超出deepseek的范围
originalparams.put("frequency_penalty", 0.5); // deepseek不支持
originalparams.put("messages", collections.singletonlist(map.of("role", "user", "content", prompt)));
// 适配后再传
map<string, object> adaptedparams = paramsadapter.adapt(originalparams);
// 其他请求逻辑...
return "";
}
}
同样是429错误,deepseek表示“请求太频繁,触发限流了”,但文心一言可能表示“token用完了”;还有500错误,有人以为是自己代码的问题,其实是deepseek服务端的问题,白查半天。
import org.springframework.web.client.httpclienterrorexception;
// 错误码适配工具类
public class deepseekerrorhandler {
// deepseek常见错误码说明
public enum errorcode {
token_expired(401, "token过期或无效,该刷新了"),
rate_limit(429, "请求太频繁,歇会儿再试"),
too_long(400, "文本太长,超过token限制了"),
model_not_found(404, "模型名填错了"),
server_error(500, "deepseek服务端出问题了");
private final int code;
private final string desc;
errorcode(int code, string desc) {
this.code = code;
this.desc = desc;
}
public int getcode() {
return code;
}
public string getdesc() {
return desc;
}
}
// 处理deepseek错误,返回适配的重试策略
public static boolean needretry(exception e) {
if (e instanceof httpclienterrorexception) {
int statuscode = ((httpclienterrorexception) e).getstatuscode().value();
// 限流和服务端错误可以重试,其他的别瞎试
return statuscode == errorcode.rate_limit.getcode()
|| statuscode == errorcode.server_error.getcode();
}
// 网络超时、连接失败这些也可以重试
return e instanceof java.net.sockettimeoutexception
|| e instanceof java.net.connectexception;
}
// 获取错误描述,适配不同模型的错误码
public static string geterrormsg(exception e) {
if (e instanceof httpclienterrorexception) {
httpclienterrorexception ex = (httpclienterrorexception) e;
int statuscode = ex.getstatuscode().value();
// 匹配错误码
for (errorcode errorcode : errorcode.values()) {
if (errorcode.getcode() == statuscode) {
return errorcode.getdesc() + ",详情:" + ex.getresponsebodyasstring();
}
}
}
// 其他错误直接返回信息
return "未知错误:" + e.getmessage();
}
}
// 调用示例,带重试逻辑
public class retryclient {
private static final int max_retry = 3; // 最多重试3次
public string callwithretry(string prompt) {
int retrycount = 0;
while (retrycount < max_retry) {
try {
// 调用api的逻辑
deepseekapiclient client = new deepseekapiclient();
return client.calldeepseek(prompt);
} catch (exception e) {
retrycount++;
string errormsg = deepseekerrorhandler.geterrormsg(e);
system.err.println("第" + retrycount + "次调用失败:" + errormsg);
// 不该重试就直接退出
if (!deepseekerrorhandler.needretry(e)) {
break;
}
// 指数退避重试:第1次等2秒,第2次等4秒,第3次等8秒
try {
thread.sleep((long) (math.pow(2, retrycount) * 1000));
} catch (interruptedexception ie) {
thread.currentthread().interrupt();
break;
}
}
}
return "{\"choices\":[{\"message\":{\"content\":\"试了好几次都不行,稍后再试吧~\"}}]}";
}
}
本文拆解了 java 调用 deepseek api 的 8 个高频错误,从 token 管理、并发安全、超长文本处理到跨模型适配,核心避坑思路可总结为:
到此这篇关于java调用deepseek api的8个高频坑与解决方法的文章就介绍到这了,更多相关java调用deepseek 内容请搜索代码网以前的文章或继续浏览下面的相关文章希望大家以后多多支持代码网!
您想发表意见!!点此发布评论
版权声明:本文内容由互联网用户贡献,该文观点仅代表作者本人。本站仅提供信息存储服务,不拥有所有权,不承担相关法律责任。 如发现本站有涉嫌抄袭侵权/违法违规的内容, 请发送邮件至 2386932994@qq.com 举报,一经查实将立刻删除。
发表评论