fix(临时存储): RustFS 上传失败改为直接失败，不再静默回落 local 指针

多实例/容器化部署下 local 指针只有写入它的实例能读（跨节点读直接报错、容器重建即丢），此前上传失败静默降级会把跨节点不可读的脏指针落库，故障延后到其它节点的合并/组装才爆。现改为在写失败点抛 BusinessException（中文原因透传调用方），并新增 fallback-to-local-on-error 开关（默认 false）供单机部署回退旧行为；跨实例读 local 指针的报错改中文并带 objectKey；超限回落策略不变。
2026-09-14 02:31:16 +08:00
parent 3f5a234c59
commit c448f49e30
4 changed files with 51 additions and 9 deletions
@@ -1,6 +1,7 @@
 package com.nanri.aiimage.modules.task.service;

 import com.fasterxml.jackson.databind.ObjectMapper;
+import com.nanri.aiimage.common.exception.BusinessException;
 import com.nanri.aiimage.config.InstanceMetadata;
 import com.nanri.aiimage.config.StorageProperties;
 import com.nanri.aiimage.config.TransientStorageProperties;
@@ -275,16 +276,39 @@ class TransientPayloadStorageServiceTest {
        assertThrows(IllegalStateException.class, () -> service.resolvePayload(pointer, "read failed"));
    }

+    @Test
+    void uploadFailureFailsFastByDefaultWithoutLocalPointer() {
+        // 默认（fallbackToLocalOnError=false）：上传失败直接抛业务异常，不写本地文件、
+        // 不置 fallback 标记、不记降级指标——避免产生跨实例不可读的 local 脏指针。
+        RustfsObjectStorageService rustfs = mock(RustfsObjectStorageService.class);
+        when(rustfs.isConfigured()).thenReturn(true);
+        when(rustfs.uploadBytes(anyString(), any(byte[].class), anyBoolean()))
+                .thenThrow(new IllegalStateException("rustfs upload down"));
+        TransientPayloadStorageService service = newService(rustfs, propertiesWithLimit(1024, 4096, true));
+
+        BusinessException ex = assertThrows(BusinessException.class,
+                () -> service.storeChunkPayload("TEST", 9L, "scope", 1, "{\"a\":1}"));
+
+        assertTrue(ex.getMessage().contains("临时存储(RustFS)不可用"));
+        assertTrue(ex.getMessage().contains("objectKey=task-chunk/test/9/scope/chunk-1.json"));
+        assertFalse(service.wasLastStoreLocalFallback());
+        verify(rustfs, never()).recordLocalFallback();
+        assertFalse(java.nio.file.Files.exists(tempDir.resolve("transient-payload")),
+                "失败时不应落任何本地临时文件");
+    }
+
    @Test
    void test_task_064_payload_compression_dependency_failure_releases_resources() {
-        // 依赖失败：上传失败回落本地且标记 fallback；随后上传恢复重新走 rustfs，
-        // 失败期间无残留文件泄漏（本地文件可被清理），错误路径不产生重复对象。
+        // 依赖失败：开关显式打开（单机降级模式）时上传失败回落本地且标记 fallback；
+        // 随后上传恢复重新走 rustfs，失败期间无残留文件泄漏（本地文件可被清理），错误路径不产生重复对象。
        RustfsObjectStorageService rustfs = mock(RustfsObjectStorageService.class);
        when(rustfs.isConfigured()).thenReturn(true);
        when(rustfs.uploadBytes(anyString(), any(byte[].class), anyBoolean()))
                .thenThrow(new IllegalStateException("rustfs upload down"))
                .thenAnswer(invocation -> invocation.getArgument(0));
-        TransientPayloadStorageService service = newService(rustfs, propertiesWithLimit(1024, 4096, true));
+        TransientStorageProperties properties = propertiesWithLimit(1024, 4096, true);
+        properties.setFallbackToLocalOnError(true);
+        TransientPayloadStorageService service = newService(rustfs, properties);

        String fallbackPointer = service.storeParsedPayloadFast("TEST", 1L, "scope", "{\"a\":1}", false);
        assertTrue(fallbackPointer.startsWith("local:"));