Fix embedding timeout mismatch between future and SDK calls

Co-authored-by: Mte90 <403283+Mte90@users.noreply.github.com>

Author: Copilot

ai/analyzer.py

@@ -52,7 +52,9 @@
 # Increase batch size for parallel processing
 EMBEDDING_BATCH_SIZE = 16  # Process embeddings in batches for better throughput
 PROGRESS_LOG_INTERVAL = 10  # Log progress every N completed files
-EMBEDDING_TIMEOUT = 30  # Timeout in seconds for each embedding API call
+# Timeout for future.result() must account for retries: (max_retries + 1) × SDK_timeout + buffer
+# With SDK timeout of 15s and max_retries=2, this allows 3 × 15s = 45s + 15s buffer = 60s
+EMBEDDING_TIMEOUT = 60  # Timeout in seconds for each embedding API call (including retries)
 FILE_PROCESSING_TIMEOUT = 300  # Timeout in seconds for processing a single file (5 minutes)
 _THREADPOOL_WORKERS = max(16, EMBEDDING_CONCURRENCY + 8)
 _EXECUTOR = concurrent.futures.ThreadPoolExecutor(max_workers=_THREADPOOL_WORKERS)

ai/openai.py

@@ -122,7 +122,7 @@ def __init__(self,
                  api_url: Optional[str] = None,
                  api_key: Optional[str] = None,
                  model: Optional[str] = None,
-                 timeout: float = 30.0,
+                 timeout: float = 15.0,
                  max_retries: int = 2,
                  backoff: float = 1.5):
         self.api_url = api_url or CFG.get("api_url")