fix: Google Cloud API translation key environment and fallback logic

docker-compose.yml

@@ -105,6 +105,7 @@ services:
       - STRIPE_PRO_PRICE_ID=${STRIPE_PRO_PRICE_ID:-}
       - STRIPE_BUSINESS_PRICE_ID=${STRIPE_BUSINESS_PRICE_ID:-}
       - GOOGLE_CLIENT_ID=${GOOGLE_CLIENT_ID:-}
+      - GOOGLE_CLOUD_API_KEY=${GOOGLE_CLOUD_API_KEY:-}
     volumes:
       - uploads_data:/app/uploads
       - outputs_data:/app/outputs

routes/translate_routes.py

@@ -836,6 +836,47 @@ def _estimate_pages(file_path: Path, file_extension: str) -> int:
     return 1
 
 
+class _GoogleCloudWithFallback:
+    """Tries Google Cloud API first, falls back to deep_translator on error.
+
+    This avoids a hard crash when the Cloud API key is invalid, quota is
+    exceeded, or the network is unreachable.  The legacy GoogleTranslator
+    (deep_translator, free, no key) is used as a best-effort fallback.
+    """
+
+    def __init__(self, cloud_adapter, legacy_provider):
+        self.cloud = cloud_adapter
+        self.legacy = legacy_provider
+        self.provider_name = "google_cloud_with_fallback"
+
+    def translate(self, text, target_language, source_language="auto"):
+        try:
+            return self.cloud.translate(text, target_language, source_language)
+        except Exception as e:
+            logger.warning(
+                "google_cloud_failed_fallback_to_legacy",
+                error=str(e)[:200],
+            )
+            return self.legacy.translate(text, target_language, source_language)
+
+    def translate_batch(
+        self, texts, target_language, source_language="auto", **kwargs
+    ):
+        try:
+            return self.cloud.translate_batch(
+                texts, target_language, source_language
+            )
+        except Exception as e:
+            logger.warning(
+                "google_cloud_batch_failed_fallback_to_legacy",
+                error=str(e)[:200],
+                text_count=len(texts),
+            )
+            return self.legacy.translate_batch(
+                texts, target_language, source_language
+            )
+
+
 async def _run_translation_job(
     job_id: str,
     input_path: Path,
@@ -953,6 +994,9 @@ async def _run_translation_job(
 
         # "google" (default classic mode): use Google Cloud API key if available,
         # otherwise fall back to deep_translator (legacy, no key).
+        # When the Cloud API key IS present, wrap it in a fallback so that
+        # quota / network errors don't kill the job — deep_translator is tried
+        # as a best-effort second chance.
         if _p == "google":
             gc_key = _cfg(
                 getattr(_admin_cfg.google_cloud, "api_key", None),
@@ -960,8 +1004,13 @@ async def _run_translation_job(
             )
             if gc_key:
                 from services.providers.google_cloud_provider import LegacyGoogleCloudAdapter
-                translation_provider = LegacyGoogleCloudAdapter(gc_key)
-                logger.info("google_provider_using_cloud_api", job_id=job_id)
+                from services.translation_service import GoogleTranslationProvider
+                cloud_adapter = LegacyGoogleCloudAdapter(gc_key)
+                legacy_fallback = GoogleTranslationProvider()
+                translation_provider = _GoogleCloudWithFallback(
+                    cloud_adapter, legacy_fallback
+                )
+                logger.info("google_provider_using_cloud_api_with_fallback", job_id=job_id)
             else:
                 logger.info("google_provider_no_cloud_key_using_legacy", job_id=job_id)
 
@@ -1142,7 +1191,7 @@ async def _run_translation_job(
         if attempted > 0:
             ratio = changed / attempted
             logger.info(f"Job {job_id}: translation stats — {changed}/{attempted} texts changed ({ratio:.0%})")
-            if ratio < 0.15 and changed == 0:
+            if changed == 0:
                 error_msg = (
                     f"Translation failed: 0 out of {attempted} texts were translated. "
                     f"The provider ({provider}) may be unavailable or misconfigured. "
@@ -1151,6 +1200,12 @@ async def _run_translation_job(
                 logger.error(f"Job {job_id}: {error_msg}")
                 tracker.set_error(error_msg)
                 return
+            elif ratio < 0.05:
+                # Very suspicious — likely partial failure, warn but don't block
+                logger.warning(
+                    f"Job {job_id}: suspiciously low translation rate: "
+                    f"{changed}/{attempted} ({ratio:.1%})"
+                )
 
         if user_id:
             await tier_quota_service.increment_on_success(user_id)