fix(leonardo): 连续 3 次鉴权失败才判死，get-session 补齐浏览器头并保护 session_data

backend/internal/provider/leonardo/client.go

@@ -30,6 +30,9 @@ const (
 	graphqlURL    = "https://api.leonardo.ai/v1/graphql"
 	schemaVersion = "1.255.2"
 	userAgent     = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/133.0.0.0 Safari/537.36"
+	// sec-ch-ua must agree with userAgent's major version — a mismatch is itself a
+	// bot signal.
+	secChUA = `"Not(A:Brand";v="99", "Google Chrome";v="133", "Chromium";v="133"`
 )
 
 var (
@@ -137,6 +140,14 @@ func mergeCookies(cookie string, setCookies []string) string {
 	return strings.Join(out, "; ")
 }
 
+// keepsSession reports whether a merged cookie still carries BOTH components
+// get-session needs: the session token and better-auth's session_data cache.
+// A merge that loses either one (a Set-Cookie clearing a cache chunk) must be
+// discarded — sending it would answer 200 null, i.e. look like a dead account.
+func keepsSession(cookie string) bool {
+	return strings.Contains(cookie, "__Secure-better-auth.session_token") && HasSessionData(cookie)
+}
+
 // Session is the result of /api/auth/get-session: the short-lived bearer plus the
 // ids the GraphQL API needs (cognitoSub for the quota query, userId for the feed
 // and the CDN image path) and the human-facing account fields.
@@ -187,19 +198,30 @@ func (c *Client) GetSession(ctx context.Context, cookie string) (*Session, error
 		return nil, err
 	}
 	req = req.WithContext(ctx)
+	// Header set/order copied from a real browser's get-session call (HAR): a
+	// same-origin GET carries NO origin header and DOES carry the ua client hints
+	// + priority — sending origin while omitting the hints is exactly the shape
+	// Vercel's checkpoint 429s.
 	req.Header = http.Header{
 		"accept":             {"*/*"},
 		"accept-language":    {"en-US,en;q=0.9"},
+		"cache-control":      {"no-cache"},
 		"cookie":             {send},
-		"origin":          {appBase},
+		"pragma":             {"no-cache"},
+		"priority":           {"u=1, i"},
 		"referer":            {appBase + "/"},
-		"user-agent":      {userAgent},
+		"sec-ch-ua":          {secChUA},
+		"sec-ch-ua-mobile":   {"?0"},
+		"sec-ch-ua-platform": {`"Windows"`},
 		"sec-fetch-dest":     {"empty"},
 		"sec-fetch-mode":     {"cors"},
 		"sec-fetch-site":     {"same-origin"},
+		"user-agent":         {userAgent},
 		http.HeaderOrderKey: {
-			"accept", "accept-language", "cookie", "origin", "referer",
-			"user-agent", "sec-fetch-dest", "sec-fetch-mode", "sec-fetch-site",
+			"accept", "accept-language", "cache-control", "cookie", "pragma",
+			"priority", "referer", "sec-ch-ua", "sec-ch-ua-mobile",
+			"sec-ch-ua-platform", "sec-fetch-dest", "sec-fetch-mode",
+			"sec-fetch-site", "user-agent",
 		},
 	}
 	resp, err := client.Do(req)
@@ -208,14 +230,20 @@ func (c *Client) GetSession(ctx context.Context, cookie string) (*Session, error
 	}
 	defer resp.Body.Close()
 	body, _ := io.ReadAll(resp.Body)
-	if merged := mergeCookies(send, resp.Header["Set-Cookie"]); merged != send {
+	// Only a real app answer may rotate the stored cookie. The 403/429 人机校验 页
+	// also sends Set-Cookie (often CLEARING better-auth cookies), and persisting
+	// that would strip the session_data cache — after which get-session answers
+	// 200 null and a perfectly healthy account looks dead.
+	if resp.StatusCode == 200 {
+		if merged := mergeCookies(send, resp.Header["Set-Cookie"]); merged != send && keepsSession(merged) {
 			c.mu.Lock()
 			c.rotated[cookie] = merged
 			c.mu.Unlock()
 			send = merged
 		}
+	}
 	if resp.StatusCode == 401 {
-		return nil, ErrAuth
+		return nil, fmt.Errorf("%w: get-session http 401: %s", ErrAuth, clip(body, 160))
 	}
 	if resp.StatusCode != 200 {
 		// 403 / 429 here is the Vercel / Cloudflare 人机校验 页，不是 cookie 失效 —
@@ -240,8 +268,10 @@ func (c *Client) GetSession(ctx context.Context, cookie string) (*Session, error
 		return nil, fmt.Errorf("%w: get-session non-json", ErrTemporaryUpstream)
 	}
 	if strings.TrimSpace(raw.Session.AccessToken) == "" {
-		// No bearer despite 200 → the cookie no longer authenticates.
-		return nil, ErrAuth
+		// No bearer despite 200 → the cookie no longer authenticates. Carry the body
+		// so the log says WHICH shape it was (null session vs a session without a
+		// token) instead of a bare "auth failed".
+		return nil, fmt.Errorf("%w: get-session 200 without accessToken: %s", ErrAuth, clip(body, 160))
 	}
 	uid := raw.Session.UserID
 	if uid == "" {
@@ -278,6 +308,14 @@ func (c *Client) session(ctx context.Context, cookie string, force bool) (*Sessi
 	return c.GetSession(ctx, cookie)
 }
 
+// ProbeSession force-mints a session from the cookie, bypassing the cached
+// bearer. Callers use it to double-check an auth failure before killing an
+// account: a rejected bearer (rotation race / expired token) still yields a
+// working cookie here, only a genuinely dead cookie returns ErrAuth.
+func (c *Client) ProbeSession(ctx context.Context, cookie string) (*Session, error) {
+	return c.session(ctx, cookie, true)
+}
+
 // callGraphQL runs one GraphQL call for an account cookie. The bearer only lives
 // ~1h, so a rejected token (401/403 or a JWTExpired GraphQL error) is re-minted
 // from the cookie and the call retried once. Only a cookie that itself stops
@@ -411,7 +449,11 @@ func (c *Client) graphqlP(ctx context.Context, accessToken string, payload []byt
 		"accept":               {"*/*"},
 		"accept-language":      {"en-US,en;q=0.9"},
 		"origin":               {appBase},
+		"priority":             {"u=1, i"},
 		"referer":              {appBase + "/"},
+		"sec-ch-ua":            {secChUA},
+		"sec-ch-ua-mobile":     {"?0"},
+		"sec-ch-ua-platform":   {`"Windows"`},
 		"user-agent":           {userAgent},
 		"authorization":        {"Bearer " + accessToken},
 		"x-leo-schema-version": {schemaVersion},
@@ -419,7 +461,8 @@ func (c *Client) graphqlP(ctx context.Context, accessToken string, payload []byt
 		"sec-fetch-mode":       {"cors"},
 		"sec-fetch-site":       {"same-site"},
 		http.HeaderOrderKey: {
-			"content-type", "accept", "accept-language", "origin", "referer",
+			"content-type", "accept", "accept-language", "origin", "priority",
+			"referer", "sec-ch-ua", "sec-ch-ua-mobile", "sec-ch-ua-platform",
 			"user-agent", "authorization", "x-leo-schema-version",
 			"sec-fetch-dest", "sec-fetch-mode", "sec-fetch-site",
 		},

backend/internal/service/tokens.go

@@ -1114,6 +1114,24 @@ func (s *TokenService) ImportCustomAccount(ctx context.Context, baseURL, apiKey,
 	return item, nil
 }
 
+// leonardoAuthStrikeLimit 是 leonardo 号被判死前允许的连续鉴权失败次数。上游偶发
+// 返回 200 null / 401(cookie 轮换竞态、人机校验)时一次就判死会误杀健康号，所以要
+// 连续失败到这个次数才判死；任何一次成功都会清零。
+const leonardoAuthStrikeLimit = 3
+
+// leonardoAuthStrike 记一次鉴权失败：返回要写回的 meta、当前连续失败次数,以及是否
+// 该判死。
+func leonardoAuthStrike(item *model.TokenAccount, reason string) (datatypes.JSONMap, int, bool) {
+	meta := cloneJSONMap(item.Meta)
+	strikes := 1
+	if n, ok := jsonMapInt(item.Meta, "auth_fails"); ok {
+		strikes = n + 1
+	}
+	meta["auth_fails"] = strikes
+	meta["last_auth_error"] = reason
+	return meta, strikes, strikes >= leonardoAuthStrikeLimit
+}
+
 // finishPending writes the terminal status/dead flag and clears the pending_check
 // marker (merging any cached quota) for a background import probe.
 func (s *TokenService) finishPending(ctx context.Context, pool, id, status string, dead bool, quotaMeta map[string]any) {
@@ -1469,17 +1487,23 @@ func (s *TokenService) Quota(ctx context.Context, pool, id string) (map[string]a
 		s.persistLeonardoCookie(ctx, item.ID, item.Value)
 		if err != nil {
 			if errors.Is(err, leonardo.ErrAuth) {
-				_, _ = s.tokens.Update(ctx, item.Pool, item.ID, map[string]any{
-					"status": "disabled",
-					"dead":   true,
-					"fails":  gorm.Expr("fails + 1"),
-				})
+				meta, strikes, kill := leonardoAuthStrike(item, "quota refresh: "+err.Error())
+				patch := map[string]any{"meta": meta, "fails": gorm.Expr("fails + 1")}
+				if kill {
+					patch["status"] = "disabled"
+					patch["dead"] = true
+					log.Printf("account leonardo/%s disabled after %d consecutive auth failures: %v", item.ID, strikes, err)
+				} else {
+					log.Printf("leonardo %s: auth failure %d/%d on quota refresh (%v) — kept active", item.ID, strikes, leonardoAuthStrikeLimit, err)
+				}
+				_, _ = s.tokens.Update(ctx, item.Pool, item.ID, patch)
 			}
 			return nil, err
 		}
 		patch := map[string]any{}
 		meta := cloneJSONMap(item.Meta)
 		meta["cached_quota_at"] = int(time.Now().Unix())
+		meta["auth_fails"] = 0 // cookie 还能换 token，连续失败计数清零
 		if remaining, ok := data["remaining"].(int); ok {
 			meta["cached_quota_remaining"] = remaining
 			// Below the per-generation floor → sink to "限额" so it stops being

backend/internal/service/v1.go

@@ -7,6 +7,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"log"
 	"net/http"
 	"path/filepath"
 	"sort"
@@ -3762,9 +3763,30 @@ func (s *V1Service) markTokenFailure(ctx context.Context, pool string, token mod
 		// grok is intentionally excluded: a grok sso can momentarily 401 while
 		// still valid (upstream blip / proxy / anti-bot), so an auth failure just
 		// fails over for this request without permanently killing the account.
-		if pool == "chatgpt" || pool == "runway" || pool == "leonardo" || pool == "krea" || pool == "imagine" {
+		disable := pool == "chatgpt" || pool == "runway" || pool == "leonardo" || pool == "krea" || pool == "imagine"
+		if disable && pool == "leonardo" {
+			// 两道保险：先重新 get-session 复核（单次失败常是 bearer 轮换竞态），复核
+			// 也不过就只记一次连续失败，连续到上限才判死。
+			if s.leonardoCookieAlive(ctx, token) {
+				log.Printf("leonardo %s: auth failure on %s but cookie still authenticates — kept active", token.ID, kind)
+				disable = false
+			} else {
+				meta, strikes, kill := leonardoAuthStrike(&token, "auth failure on "+kind)
+				patch["meta"] = meta
+				disable = kill
+				if kill {
+					log.Printf("account leonardo/%s disabled after %d consecutive auth failures: %s", token.ID, strikes, kind)
+				} else {
+					log.Printf("leonardo %s: auth failure %d/%d on %s — kept active", token.ID, strikes, leonardoAuthStrikeLimit, kind)
+				}
+			}
+		}
+		if disable {
 			patch["status"] = "disabled"
 			patch["dead"] = true
+			if pool != "leonardo" {
+				log.Printf("account %s/%s disabled: auth failure on %s", pool, token.ID, kind)
+			}
 		}
 	default:
 		// Neither pool is auto-disabled on generic (non-auth / non-quota) failures
@@ -3775,6 +3797,24 @@ func (s *V1Service) markTokenFailure(ctx context.Context, pool string, token mod
 	_, _ = s.tokens.Update(ctx, pool, token.ID, patch)
 }
 
+// leonardoCookieAlive re-checks a Leonardo cookie after an auth failure by
+// force-minting a session (bypassing the cached bearer). Only a cookie that
+// still fails to authenticate counts as dead; a temporary upstream answer
+// (403/429 人机校验) also keeps the account alive.
+func (s *V1Service) leonardoCookieAlive(ctx context.Context, token model.TokenAccount) bool {
+	if s.leonardo == nil || strings.TrimSpace(token.Value) == "" {
+		return false
+	}
+	probeCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), 30*time.Second)
+	defer cancel()
+	sess, err := s.leonardo.ProbeSession(probeCtx, token.Value)
+	if err == nil && sess != nil && strings.TrimSpace(sess.AccessToken) != "" {
+		s.leonardoPersistCookie(probeCtx, token.ID, token.Value)
+		return true
+	}
+	return !errors.Is(err, leonardo.ErrAuth)
+}
+
 // markTokenDead disables an account and marks it dead on a fatal upstream error
 // (a non-overload temporary Adobe failure that ops policy treats as account death).
 func (s *V1Service) markTokenDead(ctx context.Context, pool string, token model.TokenAccount, kind string) {