fix(openai): fail over on chat transport errors

backend/internal/service/openai_gateway_chat_completions.go

@@ -241,18 +241,7 @@ func (s *OpenAIGatewayService) ForwardAsChatCompletions(
 	}
 	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
 	if err != nil {
-		safeErr := sanitizeUpstreamErrorMessage(err.Error())
-		setOpsUpstreamError(c, 0, safeErr, "")
-		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
-			Platform:           account.Platform,
-			AccountID:          account.ID,
-			AccountName:        account.Name,
-			UpstreamStatusCode: 0,
-			Kind:               "request_error",
-			Message:            safeErr,
-		})
-		writeChatCompletionsError(c, http.StatusBadGateway, "upstream_error", "Upstream request failed")
-		return nil, fmt.Errorf("upstream request failed: %s", safeErr)
+		return nil, s.handleOpenAIUpstreamTransportError(ctx, c, account, err, false)
 	}
 	defer func() { _ = resp.Body.Close() }()
 

backend/internal/service/openai_gateway_chat_completions_raw.go

@@ -171,18 +171,7 @@ func (s *OpenAIGatewayService) forwardAsRawChatCompletions(
 	}
 	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
 	if err != nil {
-		safeErr := sanitizeUpstreamErrorMessage(err.Error())
-		setOpsUpstreamError(c, 0, safeErr, "")
-		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
-			Platform:           account.Platform,
-			AccountID:          account.ID,
-			AccountName:        account.Name,
-			UpstreamStatusCode: 0,
-			Kind:               "request_error",
-			Message:            safeErr,
-		})
-		writeChatCompletionsError(c, http.StatusBadGateway, "upstream_error", "Upstream request failed")
-		return nil, fmt.Errorf("upstream request failed: %s", safeErr)
+		return nil, s.handleOpenAIUpstreamTransportError(ctx, c, account, err, false)
 	}
 	defer func() { _ = resp.Body.Close() }()
 

backend/internal/service/openai_upstream_transport_error_handle_test.go

@@ -11,6 +11,8 @@ import (
 	"testing"
 	"time"
 
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/tlsfingerprint"
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/require"
 )
@@ -37,6 +39,21 @@ func newOpenAITransportErrTestContext() (*gin.Context, *httptest.ResponseRecorde
 	return c, rec
 }
 
+type failingOpenAIHTTPUpstream struct {
+	err   error
+	calls int
+}
+
+func (u *failingOpenAIHTTPUpstream) Do(_ *http.Request, _ string, _ int64, _ int) (*http.Response, error) {
+	u.calls++
+	return nil, u.err
+}
+
+func (u *failingOpenAIHTTPUpstream) DoWithTLS(_ *http.Request, _ string, _ int64, _ int, _ *tlsfingerprint.Profile) (*http.Response, error) {
+	u.calls++
+	return nil, u.err
+}
+
 // A durable proxy/credential failure must (a) temporarily unschedule the account
 // so it stops being hammered, and (b) return a failover error so the handler
 // switches to a healthy account instead of writing a hard 502 itself.
@@ -159,3 +176,37 @@ func TestHandleOpenAIUpstreamTransportError_DeadlineExceeded_StillFailsOver(t *t
 	var fo *UpstreamFailoverError
 	require.True(t, errors.As(err, &fo), "context.DeadlineExceeded must still return *UpstreamFailoverError")
 }
+
+func TestForwardAsRawChatCompletions_TransportErrorFailsOver(t *testing.T) {
+	repo := &openaiTransportAccountRepoStub{}
+	upstream := &failingOpenAIHTTPUpstream{
+		err: errors.New(`Post "https://opencode.ai/zen/v1/chat/completions": EOF`),
+	}
+	svc := &OpenAIGatewayService{
+		accountRepo:  repo,
+		httpUpstream: upstream,
+		cfg: &config.Config{
+			Security: config.SecurityConfig{
+				URLAllowlist: config.URLAllowlistConfig{Enabled: false},
+			},
+		},
+	}
+	account := &Account{
+		ID:          81,
+		Name:        "oc-20053",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeAPIKey,
+		Credentials: map[string]any{"api_key": "sk-test", "base_url": "https://opencode.ai/zen/v1"},
+	}
+	c, rec := newOpenAITransportErrTestContext()
+	body := []byte(`{"model":"deepseek-v4-flash-free","messages":[{"role":"user","content":"hello"}]}`)
+
+	_, err := svc.forwardAsRawChatCompletions(context.Background(), c, account, body, "")
+
+	require.Equal(t, 1, upstream.calls)
+	var fo *UpstreamFailoverError
+	require.True(t, errors.As(err, &fo), "transport error must trigger account failover")
+	require.Equal(t, http.StatusBadGateway, fo.StatusCode)
+	require.Empty(t, repo.tempUnschedCalls, "plain EOF is transient: fail over but do not evict")
+	require.Equal(t, 0, rec.Body.Len(), "service must not write a hard 502 before handler can fail over")
+}