atlasbot: enforce per-call timeout across retries

atlasbot/llm/client.py

@@ -1,5 +1,6 @@
 import json
 import logging
+import time
 from typing import Any
 
 import httpx
@@ -42,9 +43,16 @@ class LLMClient:
             "stream": False,
         }
         timeout = timeout_sec if timeout_sec is not None else self._timeout
+        deadline = (time.monotonic() + timeout) if timeout_sec is not None else None
         for attempt in range(max(1, self._settings.ollama_retries + 1)):
+            call_timeout = timeout
+            if deadline is not None:
+                remaining = deadline - time.monotonic()
+                if remaining <= 0:
+                    raise LLMError("timeout")
+                call_timeout = min(call_timeout, remaining)
             try:
-                async with httpx.AsyncClient(timeout=timeout) as client:
+                async with httpx.AsyncClient(timeout=call_timeout) as client:
                     resp = await client.post(self._endpoint(), json=payload, headers=self._headers)
                 if resp.status_code == FALLBACK_STATUS_CODE and self._settings.ollama_fallback_model:
                     payload["model"] = self._settings.ollama_fallback_model
@@ -61,6 +69,8 @@ class LLMClient:
                 return str(content)
             except Exception as exc:
                 log.warning("ollama call failed", extra={"extra": {"attempt": attempt + 1, "error": str(exc)}})
+                if deadline is not None and (deadline - time.monotonic()) <= 0:
+                    raise LLMError("timeout") from exc
                 if attempt + 1 >= max(1, self._settings.ollama_retries + 1):
                     raise LLMError(str(exc)) from exc
         raise LLMError("ollama retries exhausted")