super/crawler-plugin
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: 商标查询失败自动重试、货源结果图切回Excel内嵌防WPS #REF、外观专利标题模型改gemini-3.8-flash

Browse Source 
- brand-check: 单品牌商标查询失败(异常/空响应/query_faild_data 非空)自动重试，
  次数与间隔可配(retry-times 默认3含首次 / retry-interval-millis 1000)，命中侵权不重试；
  collectdata 品牌过滤与外观专利商标核验两侧同时受益
- similarasin: 货源结果Excel图片由WPS DISPIMG 切回 Excel 365 richData 单元格内嵌，
  修复 WPS 打开 DISPIMG 显示 #REF 的兼容回归；下载失败单元格回退为URL文本
- appearance-patent: 标题商标提取模型由 deepseek-v4-flash 切 gemini-3.8-flash
This commit is contained in:
黄自达
2026-09-08 19:01:08 +08:00
parent ba581ed095
commit 197643a8fa
10 changed files with 191 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend-java/src/main/java/com/nanri/aiimage/config/AppearancePatentProperties.java
+1 -1
View File
@@ -14,7 +14,7 @@ public class AppearancePatentProperties {
/**
* 商标关键词提取模型
*/
private String titleModel = "deepseek-v4-flash";
private String titleModel = "gemini-3.8-flash";
/**
* 外观检测模型（视觉）
*/
backend-java/src/main/java/com/nanri/aiimage/config/BrandCheckProperties.java
+4
View File
@@ -10,6 +10,10 @@ public class BrandCheckProperties {
private String path = "/brand_check";
private String token = "";
private String defaultStrategy = "Terms";
/** 单品牌商标查询的重试次数（含首次），连续失败超过该次数才判定为查询失败。 */
private int retryTimes = 3;
/** 每次查询失败后到下一次重试前的等待毫秒数。 */
private int retryIntervalMillis = 1000;
private int connectTimeoutMillis = 10000;
private int readTimeoutMillis = 60000;
}
backend-java/src/main/java/com/nanri/aiimage/modules/brand/client/BrandCheckClient.java
+41 -7
View File
@@ -105,15 +105,49 @@ public class BrandCheckClient {
}
private BrandCheckOutcome checkOneBrand(String brand, String strategy) {
int attempts = Math.max(1, properties.getRetryTimes());
BrandCheckResponse response = null;
Exception lastFailure = null;
for (int attempt = 1; attempt <= attempts; attempt++) {
try {
BrandCheckResponse response = check(brand, strategy);
if (response == null) {
return new BrandCheckOutcome(List.of(), List.of(brand));
}
return new BrandCheckOutcome(nullToEmpty(response.getFaildData()), nullToEmpty(response.getQueryFaildData()));
response = check(brand, strategy);
} catch (Exception ex) {
log.warn("[brand-check] request failed brand={} strategy={} err={}", brand, strategy, ex.getMessage());
return new BrandCheckOutcome(List.of(), List.of(brand));
lastFailure = ex;
response = null;
if (attempt < attempts) {
log.warn("[brand-check] 请求异常将重试 brand={} attempt={}/{} err={}",
brand, attempt, attempts, ex.getMessage());
sleepBeforeRetry();
}
continue;
}
if (response != null && nullToEmpty(response.getQueryFaildData()).isEmpty()) {
// 本次查询完成且无失败项：命中侵权(faild_data)属有效结论，不重试。
return new BrandCheckOutcome(nullToEmpty(response.getFaildData()), List.of());
}
if (attempt < attempts) {
log.warn("[brand-check] 服务端返回查询失败将重试 brand={} attempt={}/{}",
brand, attempt, attempts);
sleepBeforeRetry();
}
}
log.warn("[brand-check] 重试耗尽判定查询失败 brand={} attempts={} lastErr={}",
brand, attempts,
lastFailure == null ? "query_faild_data 持续非空" : lastFailure.getMessage());
return new BrandCheckOutcome(List.of(),
response == null ? List.of(brand) : nullToEmpty(response.getQueryFaildData()));
}
private void sleepBeforeRetry() {
long delayMillis = Math.max(0L, properties.getRetryIntervalMillis());
if (delayMillis <= 0L) {
return;
}
try {
Thread.sleep(delayMillis);
} catch (InterruptedException interruptedException) {
Thread.currentThread().interrupt();
throw new IllegalStateException("商标查询重试被中断", interruptedException);
}
}
backend-java/src/main/java/com/nanri/aiimage/modules/similarasin/service/SimilarAsinTaskService.java
+4 -4
View File
@@ -41,7 +41,7 @@ import com.nanri.aiimage.modules.similarasin.service.support.SimilarAsinExcelPar
import com.nanri.aiimage.modules.similarasin.service.support.SimilarAsinGroupingConverter;
import com.nanri.aiimage.modules.similarasin.service.support.SimilarAsinHistoryAssembler;
import com.nanri.aiimage.modules.similarasin.util.BoundedImageCache;
import com.nanri.aiimage.modules.similarasin.util.WpsCellImageWriter;
import com.nanri.aiimage.modules.similarasin.util.ExcelCellImageWriter;
import com.nanri.aiimage.modules.similarasin.util.SimilarAsinImageEmbedder;
import com.nanri.aiimage.modules.similarasin.util.SimilarAsinLogSupport;
import com.nanri.aiimage.modules.file.service.LocalFileStorageService;
@@ -4553,8 +4553,8 @@ public class SimilarAsinTaskService {
Map<String, SimilarAsinResultRowDto> resultMap,
Map<String, SimilarAsinImageEmbedder.ResizedImage> taskImageCache) {
ensureResultAssemblyNotInterrupted();
// 单元格内嵌图片：写完 workbook 后 patch WPS DISPIMG 结构（WPS/钉钉/飞书/Excel365 均可显示）。
WpsCellImageWriter.Session cellImageSession = WpsCellImageWriter.createSession();
// 单元格内嵌图片：写完 workbook 后 patch Excel 365 richData 单元格图片结构（跨 Excel/WPS 打开均可见）。
ExcelCellImageWriter.Session cellImageSession = ExcelCellImageWriter.createSession();
SimilarAsinImageEmbedder.ImageSpool imageSpool;
try {
Path spoolDir = Files.createTempDirectory(xlsx.getParentFile().toPath(), "similar-asin-images-");
@@ -4670,7 +4670,7 @@ public class SimilarAsinTaskService {
if (!cellImageSession.isEmpty()) {
try {
ensureResultAssemblyNotInterrupted();
WpsCellImageWriter.patchXlsxFile(xlsx, cellImageSession);
ExcelCellImageWriter.patchXlsxFile(xlsx, cellImageSession);
} catch (IOException ex) {
String cause = ex.toString();
log.error("[similar-asin] excel cell image patch failed xlsx={} images={} err={}",
backend-java/src/main/java/com/nanri/aiimage/modules/similarasin/util/SimilarAsinImageEmbedder.java
+20 -19
View File
@@ -611,7 +611,7 @@ public class SimilarAsinImageEmbedder {
String url,
Row row,
Map<String, ResizedImage> taskImageCache,
WpsCellImageWriter.Session excelCellImageSession) {
ExcelCellImageWriter.Session excelCellImageSession) {
return embedAsExcelCellImage(rowIdx, colIdx, url, row, taskImageCache, excelCellImageSession, null);
}
@@ -620,39 +620,40 @@ public class SimilarAsinImageEmbedder {
String url,
Row row,
Map<String, ResizedImage> taskImageCache,
WpsCellImageWriter.Session excelCellImageSession,
ExcelCellImageWriter.Session excelCellImageSession,
ImageSpool imageSpool) {
if (url == null || url.isBlank() || excelCellImageSession == null) {
return null;
}
String trimmedUrl = url.trim();
SpoolImage spooled = imageSpool == null ? null : imageSpool.get(trimmedUrl);
String imageId = null;
ImageDim dim = null;
if (spooled != null) {
// 预取已落盘：按路径注册，patch 阶段再读文件，避免整份图片字节驻留堆内存。
imageId = excelCellImageSession.registerImageByPath(spooled.path());
dim = new ImageDim(spooled.width(), spooled.height());
} else if (imageSpool != null) {
// 预取未命中说明该 URL 下载失败：不重试、不把 URL 文本写进单元格，留空并打日志。
log.warn("[similar-asin][image] cell-image-miss url={} colIdx={} (prefetch did not spool), leave blank",
trimmedUrl, colIdx);
row.createCell(colIdx).setCellValue("#VALUE!");
excelCellImageSession.registerImage(rowIdx, colIdx, spooled.path());
return new ImageDim(spooled.width(), spooled.height());
}
if (imageSpool != null) {
// 预取未命中说明该 URL 下载失败：不重试，单元格回退为 URL 文本便于人工核对。
row.createCell(colIdx).setCellValue(trimmedUrl);
return null;
} else {
}
ResizedImage thumb = downloadAndResize(trimmedUrl, taskImageCache);
if (thumb == null) {
log.warn("[similar-asin][image] cell-image-miss url={} colIdx={} (download failed), leave blank",
trimmedUrl, colIdx);
return null;
}
imageId = excelCellImageSession.registerImage(thumb.bytes());
try {
row.createCell(colIdx).setCellValue("#VALUE!");
excelCellImageSession.registerImage(rowIdx, colIdx, thumb.bytes());
taskImageCache.remove(trimmedUrl);
dim = new ImageDim(thumb.width(), thumb.height());
return new ImageDim(thumb.width(), thumb.height());
} catch (RuntimeException ex) {
taskImageCache.remove(trimmedUrl);
log.warn("[similar-asin][image] excel-cell-image-fail url={} colIdx={} err={}",
trimmedUrl, colIdx, ex.getMessage());
row.createCell(colIdx).setCellValue(trimmedUrl);
return null;
}
if (imageId != null) {
row.createCell(colIdx).setCellFormula(WpsCellImageWriter.dispImgFormula(imageId));
}
return dim;
}
private boolean hasLocalCachedThumb(String url) {
backend-java/src/main/resources/application-local.example.yml
+1 -1
View File
@@ -62,7 +62,7 @@ AIIMAGE_MODULE_CLEANUP_CRON=0 0 0 * * *
AIIMAGE_MODULE_CLEANUP_MODULE_TYPES=DEDUPE,SPLIT,CONVERT,DELETE_BRAND
AIIMAGE_APPEARANCE_PATENT_LLM_HOST=https://ai.t8star.org
AIIMAGE_APPEARANCE_PATENT_TITLE_MODEL=deepseek-v4-flash
AIIMAGE_APPEARANCE_PATENT_TITLE_MODEL=gemini-3.8-flash
AIIMAGE_APPEARANCE_PATENT_APPEARANCE_MODEL=gemini-3.8-flash
AIIMAGE_APPEARANCE_PATENT_LLM_MAX_TOKENS=64000
AIIMAGE_APPEARANCE_PATENT_LLM_BATCH_SIZE=10
backend-java/src/main/resources/application.yml
+3 -1
View File
@@ -237,11 +237,13 @@ aiimage:
path: ${AIIMAGE_BRAND_CHECK_PATH:/brand_check}
token: ${AIIMAGE_BRAND_CHECK_TOKEN:}
default-strategy: ${AIIMAGE_BRAND_CHECK_DEFAULT_STRATEGY:Terms}
retry-times: ${AIIMAGE_BRAND_CHECK_RETRY_TIMES:3}
retry-interval-millis: ${AIIMAGE_BRAND_CHECK_RETRY_INTERVAL_MILLIS:1000}
connect-timeout-millis: ${AIIMAGE_BRAND_CHECK_CONNECT_TIMEOUT_MILLIS:10000}
read-timeout-millis: ${AIIMAGE_BRAND_CHECK_READ_TIMEOUT_MILLIS:60000}
appearance-patent:
llm-host: ${AIIMAGE_APPEARANCE_PATENT_LLM_HOST:https://ai.t8star.org}
title-model: ${AIIMAGE_APPEARANCE_PATENT_TITLE_MODEL:deepseek-v4-flash}
title-model: ${AIIMAGE_APPEARANCE_PATENT_TITLE_MODEL:gemini-3.8-flash}
appearance-model: ${AIIMAGE_APPEARANCE_PATENT_APPEARANCE_MODEL:gemini-3.8-flash}
llm-max-tokens: ${AIIMAGE_APPEARANCE_PATENT_LLM_MAX_TOKENS:64000}
llm-connect-timeout-millis: ${AIIMAGE_APPEARANCE_PATENT_LLM_CONNECT_TIMEOUT_MILLIS:10000}
backend-java/src/test/java/com/nanri/aiimage/modules/appearancepatent/client/AppearancePatentLlmClientHttpTest.java
+14 -13
View File
@@ -63,22 +63,23 @@ class AppearancePatentLlmClientHttpTest {
String model = String.valueOf(request.get("model"));
callCounts.computeIfAbsent(model, ignored -> new AtomicInteger()).incrementAndGet();
// 标题提取与外观检测现同用 gemini-3.8-flash，按请求是否带 image_url 区分两路：
// 外观检测消息带 image_url parts，标题商标提取仅纯文本。
String content;
if (model.contains("deepseek")) {
// 商标提取：按标题内容返回品牌词或"无"
String messages = String.valueOf(request.get("messages"));
if (messages.contains("Apple")) {
content = "Apple，Apple";
} else {
content = "无";
}
} else {
if (body.contains("\"type\":\"image_url\"")) {
// 外观检测：返回 JSON（带 ```json 包裹与换行，模拟脏输出）
if (messagesBodyContains(exchange, "原创个性杯")) {
if (body.contains("原创个性杯")) {
content = "```json\n{\"appearance_status\": \"侵权\", \"appearance_reason\": \"【视觉拆解】：特殊造型\\n【判定依据】：高度相似知名设计\"}\n```";
} else {
content = "{\"appearance_status\": \"无侵权\", \"appearance_reason\": \"【视觉拆解】：普通直筒杯。\\n【对比评估】：行业通用基础形状。\\n【判定依据】：无侵权。\"}";
}
} else {
// 商标提取：按标题内容返回品牌词或"无"
if (String.valueOf(request.get("messages")).contains("Apple")) {
content = "Apple，Apple";
} else {
content = "无";
}
}
String response = objectMapper.writeValueAsString(Map.of(
"model", model,
@@ -125,8 +126,8 @@ class AppearancePatentLlmClientHttpTest {
assertThat(result.getAppearanceRisk()).isEqualTo("无侵权");
assertThat(result.getAppearanceReason()).contains("视觉拆解");
assertThat(result.getTitleReason()).contains("Apple");
assertThat(callCounts.get("deepseek-v4-flash")).hasValue(1);
assertThat(callCounts.get("gemini-3.8-flash")).hasValue(1);
assertThat(callCounts.get("gemini-3.8-flash")).hasValue(2);
assertThat(callCounts.get("deepseek-v4-flash")).isNull();
}
@Test
@@ -182,7 +183,7 @@ class AppearancePatentLlmClientHttpTest {
client.inspectRows(List.of(row("6", "B006", "普通数据线", "", "https://img.example.com/6.jpg")), null, "test-key");
String appearanceBody = capturedBodies.stream()
.filter(b -> b.contains("gemini-3.8-flash"))
.filter(b -> b.contains("\"type\":\"image_url\""))
.findFirst()
.orElseThrow();
assertThat(appearanceBody).contains("https://img.example.com/6.jpg");
backend-java/src/test/java/com/nanri/aiimage/modules/brand/client/BrandCheckClientTest.java
+95
View File
@@ -1,9 +1,15 @@
package com.nanri.aiimage.modules.brand.client;
import com.nanri.aiimage.config.BrandCheckProperties;
import com.sun.net.httpserver.HttpServer;
import org.junit.jupiter.api.Test;
import java.io.IOException;
import java.io.OutputStream;
import java.net.InetSocketAddress;
import java.nio.charset.StandardCharsets;
import java.util.List;
import java.util.concurrent.atomic.AtomicInteger;
import static org.assertj.core.api.Assertions.assertThat;
@@ -26,4 +32,93 @@ class BrandCheckClientTest {
assertThat(brands).containsExactly("任天堂", "Sony");
}
@Test
void queryFailureRetriesUntilSuccess() throws Exception {
AtomicInteger attempts = new AtomicInteger();
HttpServer server = startBrandCheckServer(exchange -> {
int count = attempts.incrementAndGet();
String body = count < 3
? "{\"faild_data\":[],\"query_faild_data\":[\"Apple\"]}"
: "{\"faild_data\":[],\"query_faild_data\":[]}";
respond(exchange, body);
});
try {
BrandCheckClient client = newClient(server, 3);
BrandCheckClient.BrandCheckBatchResult result = client.checkAll(List.of("Apple"), "Terms");
assertThat(attempts.get()).as("前两次失败后应重试到成功").isEqualTo(3);
assertThat(result.hasFailedData()).isFalse();
assertThat(result.hasQueryFailedData()).isFalse();
} finally {
server.stop(0);
}
}
@Test
void persistentQueryFailureIsReportedAfterExhaustingRetries() throws Exception {
AtomicInteger attempts = new AtomicInteger();
HttpServer server = startBrandCheckServer(exchange -> {
attempts.incrementAndGet();
respond(exchange, "{\"faild_data\":[],\"query_faild_data\":[\"Apple\"]}");
});
try {
BrandCheckClient client = newClient(server, 3);
BrandCheckClient.BrandCheckBatchResult result = client.checkAll(List.of("Apple"), "Terms");
assertThat(attempts.get()).as("持续失败应耗尽重试次数").isEqualTo(3);
assertThat(result.hasFailedData()).isFalse();
assertThat(result.hasQueryFailedData()).isTrue();
assertThat(result.queryFaildData()).contains("Apple");
} finally {
server.stop(0);
}
}
@Test
void confirmedInfringementIsNotRetried() throws Exception {
AtomicInteger attempts = new AtomicInteger();
HttpServer server = startBrandCheckServer(exchange -> {
attempts.incrementAndGet();
respond(exchange, "{\"faild_data\":[\"Apple\"],\"query_faild_data\":[]}");
});
try {
BrandCheckClient client = newClient(server, 3);
BrandCheckClient.BrandCheckBatchResult result = client.checkAll(List.of("Apple"), "Terms");
assertThat(attempts.get()).as("命中侵权属有效结论，不应重试").isEqualTo(1);
assertThat(result.hasFailedData()).isTrue();
assertThat(result.hasQueryFailedData()).isFalse();
} finally {
server.stop(0);
}
}
private BrandCheckClient newClient(HttpServer server, int retryTimes) {
BrandCheckProperties props = new BrandCheckProperties();
props.setBaseUrl("http://127.0.0.1:" + server.getAddress().getPort());
props.setPath("/brand_check");
props.setRetryTimes(retryTimes);
props.setRetryIntervalMillis(1);
return new BrandCheckClient(props, null);
}
private HttpServer startBrandCheckServer(com.sun.net.httpserver.HttpHandler handler) throws IOException {
HttpServer server = HttpServer.create(new InetSocketAddress("127.0.0.1", 0), 0);
server.createContext("/brand_check", handler);
server.start();
return server;
}
private void respond(com.sun.net.httpserver.HttpExchange exchange, String body) throws IOException {
byte[] bytes = body.getBytes(StandardCharsets.UTF_8);
exchange.getResponseHeaders().set("Content-Type", "application/json; charset=utf-8");
exchange.sendResponseHeaders(200, bytes.length);
try (OutputStream os = exchange.getResponseBody()) {
os.write(bytes);
}
}
}
backend-java/src/test/java/com/nanri/aiimage/modules/similarasin/util/SimilarAsinImageEmbedderTest.java
+3 -2
View File
@@ -229,12 +229,13 @@ class SimilarAsinImageEmbedderTest {
var row = workbook.createSheet().createRow(1);
SimilarAsinImageEmbedder.ImageDim result = failedEmbedder.embedAsExcelCellImage(
1, 9, url, row, new HashMap<>(), WpsCellImageWriter.createSession(), spool);
1, 9, url, row, new HashMap<>(), ExcelCellImageWriter.createSession(), spool);
assertEquals(SimilarAsinImageEmbedder.DOWNLOAD_MAX_RETRY + 1, callsAfterPrefetch);
assertEquals(callsAfterPrefetch, networkCalls.get());
assertNull(result);
assertNull(row.getCell(9), "预取失败不应把 URL 文本写进单元格，应留空");
assertTrue(row.getCell(9) != null, "预取失败应回退为 URL 文本便于人工核对");
assertEquals(url, row.getCell(9).getStringCellValue());
} finally {
failedEmbedder.shutdown();
}