Retry failed rewrite batches in smaller chunks

ai_daily_report/rewrite.py

@@ -74,15 +74,13 @@ def _is_transient_llm_error(exc: Exception) -> bool:
     return False
 
 
-def _apply_rewrite_batch(batch: list[NewsItem], llm_call: RewriteLlmCall) -> tuple[int, int]:
-    obj = parse_json_object(llm_call(_build_prompt(batch)))
-    rewrites = obj.get("rewrites", [])
-    if not isinstance(rewrites, list):
-        raise ValueError("rewrites is not a list")
+def _apply_rewrite_results(batch: list[NewsItem], rewrites: list[Any]) -> tuple[int, int]:
     by_id = {item.id: item for item in batch}
     seen_ids: set[str] = set()
     section_count = 0
     for entry in rewrites:
+        if not isinstance(entry, dict):
+            continue
         item_id = entry.get("id")
         title = str(entry.get("title") or "").strip()
         summary = str(entry.get("summary") or "").strip()
@@ -97,11 +95,20 @@ def _apply_rewrite_batch(batch: list[NewsItem], llm_call: RewriteLlmCall) -> tup
     return len(seen_ids), section_count
 
 
+def _apply_rewrite_batch(batch: list[NewsItem], llm_call: RewriteLlmCall) -> tuple[int, int]:
+    obj = parse_json_object(llm_call(_build_prompt(batch)))
+    rewrites = obj.get("rewrites", [])
+    if not isinstance(rewrites, list):
+        raise ValueError("rewrites is not a list")
+    return _apply_rewrite_results(batch, rewrites)
+
+
 def rewrite_items(
     items: list[NewsItem],
     *,
     llm_call: RewriteLlmCall,
     batch_size: int = 30,
+    retry_batch_size: int = 10,
     max_fallback_ratio: float = 0.2,
     retry_single_items: bool = False,
 ) -> tuple[list[NewsItem], dict[str, Any]]:
@@ -109,6 +116,7 @@ def rewrite_items(
     llm_section_count = 0
     fallback_count = 0
     missing_rewrite_count = 0
+    batch_retry_count = 0
     errors: list[str] = []
 
     for batch in _chunks(items, max(1, batch_size)):
@@ -129,6 +137,25 @@ def rewrite_items(
                     _fallback(item)
                     fallback_count += 1
                 continue
+            if len(batch) > max(1, retry_batch_size):
+                for retry_batch in _chunks(batch, max(1, retry_batch_size)):
+                    batch_retry_count += 1
+                    try:
+                        retry_rewritten_count, retry_section_count = _apply_rewrite_batch(retry_batch, llm_call)
+                        rewritten_count += retry_rewritten_count
+                        llm_section_count += retry_section_count
+                        for item in retry_batch:
+                            if item.title is None or item.summary is None:
+                                errors.append(f"missing_rewrite_for_item: {item.id}")
+                                _fallback(item)
+                                fallback_count += 1
+                                missing_rewrite_count += 1
+                    except Exception as retry_exc:
+                        errors.append(f"batch_retry:{type(retry_exc).__name__}: {retry_exc}")
+                        for item in retry_batch:
+                            _fallback(item)
+                            fallback_count += 1
+                continue
             if not retry_single_items:
                 for item in batch:
                     _fallback(item)
@@ -157,6 +184,7 @@ def rewrite_items(
         "missing_rewrite_count": missing_rewrite_count,
         "fallback_ratio": round(fallback_ratio, 4),
         "batch_count": len(_chunks(items, max(1, batch_size))),
+        "batch_retry_count": batch_retry_count,
         "errors": errors,
         "blocking_errors": blocking_errors,
         "quality_gate_failed": bool(blocking_errors),