fix(acceptance): 阻断本地控制面漂移污染验收

本地 K3s 节点此前在 Kubelet 中登记为宿主机资源，负载可挤压 etcd 并在 server 重启后继续污染同一 Run。现在为三节点设置真实可调度预算、独立 etcd/数据卷和锁定依赖镜像，并持续核对 server 与 API/etcd 健康。\n\n每次验收生成独立 Run ID，报告使用独占或原子写入，负载错误记录具体阶段；数据库鉴权不可用返回带 Retry-After 的 503，避免基础设施故障被误报为 401。\n\n验证：Go 全量测试、go vet、gofmt、OpenAPI、bash -n、ShellCheck、报告测试和 k3d 配置解析均通过。
2026-07-31 23:07:19 +08:00
parent 015ff8ea6c
commit a95184b5b6
20 changed files with 970 additions and 85 deletions
@@ -59,15 +59,16 @@ type options struct {
 }

 type report struct {
-	SchemaVersion string        `json:"schemaVersion"`
-	RunID         string        `json:"runId"`
-	Profile       string        `json:"profile"`
-	StartedAt     time.Time     `json:"startedAt"`
-	FinishedAt    time.Time     `json:"finishedAt"`
-	Passed        bool          `json:"passed"`
-	Phases        []phaseReport `json:"phases"`
-	Failure       string        `json:"failure,omitempty"`
-	SecretSafe    bool          `json:"secretSafe"`
+	SchemaVersion    string        `json:"schemaVersion"`
+	RunID            string        `json:"runId"`
+	Profile          string        `json:"profile"`
+	StartedAt        time.Time     `json:"startedAt"`
+	FinishedAt       time.Time     `json:"finishedAt"`
+	Passed           bool          `json:"passed"`
+	Phases           []phaseReport `json:"phases"`
+	Failure          string        `json:"failure,omitempty"`
+	FailureOperation string        `json:"failureOperation,omitempty"`
+	SecretSafe       bool          `json:"secretSafe"`
 }

 type phaseReport struct {
@@ -101,6 +102,25 @@ type httpStatusError struct {
 	Body   string
 }

+type operationError struct {
+	Operation string
+	Err       error
+}
+
+func (e *operationError) Error() string { return e.Operation + ": " + e.Err.Error() }
+func (e *operationError) Unwrap() error { return e.Err }
+
+func withOperation(operation string, err error) error {
+	if err == nil {
+		return nil
+	}
+	var existing *operationError
+	if errors.As(err, &existing) {
+		return err
+	}
+	return &operationError{Operation: operation, Err: err}
+}
+
 func (e *httpStatusError) Error() string {
 	return fmt.Sprintf("HTTP %d: %s", e.Status, e.Body)
 }
@@ -137,11 +157,15 @@ func main() {
 	result.Passed = runErr == nil
 	if runErr != nil {
 		result.Failure = redactError(runErr.Error(), opts)
+		var operationErr *operationError
+		if errors.As(runErr, &operationErr) {
+			result.FailureOperation = operationErr.Operation
+		}
 	}
 	payload, _ := json.MarshalIndent(result, "", "  ")
 	fmt.Println(string(payload))
 	if opts.reportPath != "" {
-		if err := os.WriteFile(opts.reportPath, append(payload, '\n'), 0o600); err != nil {
+		if err := writeReportExclusive(opts.reportPath, payload); err != nil {
 			fmt.Fprintln(os.Stderr, "write acceptance report failed:", err)
 			os.Exit(1)
 		}
@@ -151,6 +175,18 @@ func main() {
 	}
 }

+func writeReportExclusive(path string, payload []byte) error {
+	reportFile, err := os.OpenFile(path, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o600)
+	if err != nil {
+		return err
+	}
+	if _, err := reportFile.Write(append(payload, '\n')); err != nil {
+		_ = reportFile.Close()
+		return err
+	}
+	return reportFile.Close()
+}
+
 func parseOptions() (options, error) {
 	var profile string
 	var reportPath string
@@ -347,14 +383,15 @@ func runGemini(
 				response, err = client.Do(req)
 				if err == nil {
 					if response.StatusCode != http.StatusOK {
-						err = responseStatusError(response)
+						err = withOperation("gemini_submit", responseStatusError(response))
 					} else {
 						var size int64
 						var outputHash string
 						size, outputHash, err = streamGeminiImageHash(response.Body)
+						err = withOperation("gemini_response", err)
 						_ = response.Body.Close()
 						if err == nil && expectedOutputBytes > 0 && (size != int64(expectedOutputBytes) || outputHash != expectedHash) {
-							err = fmt.Errorf("Gemini output mismatch: bytes=%d hash=%s", size, outputHash)
+							err = withOperation("gemini_response", fmt.Errorf("Gemini output mismatch: bytes=%d hash=%s", size, outputHash))
 						}
 						if err == nil {
 							mu.Lock()
@@ -367,6 +404,7 @@ func runGemini(
 					}
 				}
 			}
+			err = withOperation("gemini_submit", err)
 			latency := time.Since(requestStarted)
 			mu.Lock()
 			latencies[index] = latency
@@ -503,7 +541,7 @@ func runVideo(
 				response, err = client.Do(req)
 				if err == nil {
 					if response.StatusCode != http.StatusAccepted {
-						err = responseStatusError(response)
+						err = withOperation("video_submit", responseStatusError(response))
 					} else {
 						var accepted struct {
 							TaskID string `json:"taskId"`
@@ -517,6 +555,7 @@ func runVideo(
 					}
 				}
 			}
+			err = withOperation("video_submit", err)
 			mu.Lock()
 			latencies[index] = time.Since(requestStarted)
 			if !realUpstream && logicalIndex%4 == 0 {
@@ -742,7 +781,7 @@ func pollVideoTask(ctx context.Context, client *http.Client, opts options, taskI
 		endpoint := opts.gateways[index%len(opts.gateways)] + "/api/v1/ai/result/" + url.PathEscape(taskID)
 		req, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint, nil)
 		if err != nil {
-			return err
+			return withOperation("video_poll", err)
 		}
 		opts.setHeaders(req, index, realUpstream)
 		response, err := client.Do(req)
@@ -758,20 +797,20 @@ func pollVideoTask(ctx context.Context, client *http.Client, opts options, taskI
 				case "succeeded", "success":
 					mediaURL := findMediaURL(payload)
 					if mediaURL == "" {
-						return fmt.Errorf("video task %s succeeded without a media URL", taskID)
+						return withOperation("video_poll", fmt.Errorf("video task %s succeeded without a media URL", taskID))
 					}
-					return validateVideoAsset(ctx, client, opts, mediaURL, index)
+					return withOperation("video_download", validateVideoAsset(ctx, client, opts, mediaURL, index))
 				case "failed", "cancelled", "canceled":
-					return fmt.Errorf("video task %s finished with status %s", taskID, status)
+					return withOperation("video_poll", fmt.Errorf("video task %s finished with status %s", taskID, status))
 				}
 			}
 		}
 		if err != nil {
-			return err
+			return withOperation("video_poll", err)
 		}
 		select {
 		case <-ctx.Done():
-			return ctx.Err()
+			return withOperation("video_poll", ctx.Err())
 		case <-ticker.C:
 		}
 	}
@@ -6,6 +6,7 @@ import (
 	"encoding/base64"
 	"encoding/hex"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"net/http"
 	"net/http/httptest"
@@ -212,6 +213,35 @@ func TestAcceptanceReportErrorRedactsSecretsAndURLs(t *testing.T) {
 	}
 }

+func TestAcceptanceReportCannotOverwriteExistingRunArtifact(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "load-report.json")
+	if err := writeReportExclusive(path, []byte(`{"runId":"first"}`)); err != nil {
+		t.Fatalf("write first report: %v", err)
+	}
+	if err := writeReportExclusive(path, []byte(`{"runId":"second"}`)); err == nil {
+		t.Fatal("second report unexpectedly overwrote the first report")
+	}
+	payload, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("read report: %v", err)
+	}
+	if string(payload) != "{\"runId\":\"first\"}\n" {
+		t.Fatalf("report changed after rejected overwrite: %s", payload)
+	}
+}
+
+func TestAcceptanceFailurePreservesOperationAndHTTPStatus(t *testing.T) {
+	err := withOperation("video_poll", &httpStatusError{Status: http.StatusUnauthorized, Body: "unauthorized"})
+	var operationErr *operationError
+	if !errors.As(err, &operationErr) || operationErr.Operation != "video_poll" {
+		t.Fatalf("operation error=%#v", operationErr)
+	}
+	var statusErr *httpStatusError
+	if !errors.As(err, &statusErr) || statusErr.Status != http.StatusUnauthorized {
+		t.Fatalf("HTTP status error=%#v", statusErr)
+	}
+}
+
 func TestAcceptanceGatewayTLSNameSetsHostHeader(t *testing.T) {
 	request := httptest.NewRequest(http.MethodGet, "https://127.0.0.1/api/v1/healthz", nil)
 	opts := options{