Record failed trial context

src/aituner/llm.py

@@ -25,6 +25,8 @@ def build_prompt(
                 "best_request_rate": trial.best_request_rate,
                 "best_pass_rate": trial.best_pass_rate,
                 "diagnosis": trial.diagnosis,
+                "config_patch": trial.config_patch,
+                "failure_reason": trial.failure_reason,
             }
         )
     sections = [
@@ -34,6 +36,7 @@ def build_prompt(
         "expected_effects must be a JSON array of short strings, not an object.",
         "Only use allowed tunable env keys and allowed tunable flag keys.",
         "Do not wrap the JSON in markdown fences or any extra text.",
+        "Do not repeat a config that previously failed to launch unless the new patch explicitly removes the failing knob.",
         "",
         "Study stack:",
         json.dumps(

src/aituner/spec.py

@@ -446,6 +446,8 @@ class TrialSummary:
     best_pass_rate: float | None = None
     result_path: str | None = None
     diagnosis: str = ""
+    config_patch: dict[str, Any] | None = None
+    failure_reason: str = ""
 
 
 @dataclass

src/aituner/store.py

@@ -74,7 +74,12 @@ class StudyStore:
         self.write_json(trial_root / "trial_spec.json", to_jsonable(spec))
         next_state = replace(state, next_trial_index=state.next_trial_index + 1)
         next_state.trials.append(
-            TrialSummary(trial_id=trial_id, status="queued", diagnosis=proposal.diagnosis)
+            TrialSummary(
+                trial_id=trial_id,
+                status="queued",
+                diagnosis=proposal.diagnosis,
+                config_patch=to_jsonable(proposal.config_patch),
+            )
         )
         self.save_state(next_state)
         return spec, next_state
@@ -101,6 +106,7 @@ class StudyStore:
             summary.best_request_rate = payload.get("best_request_rate")
             summary.best_pass_rate = payload.get("best_pass_rate")
             summary.result_path = str(result_path)
+            summary.failure_reason = str(payload.get("failure_reason") or "").strip()
             if (
                 isinstance(summary.best_request_rate, (int, float))
                 and (best_rate is None or summary.best_request_rate > best_rate)

src/aituner/worker.py

@@ -183,6 +183,28 @@ def _replay_requests(
     return ordered, early_stopped, early_stop_reason
 
 
+def _wait_for_server_or_exit(
+    process: subprocess.Popen[str],
+    *,
+    base_url: str,
+    healthcheck_path: str,
+    ready_timeout_s: float,
+) -> None:
+    deadline = time.monotonic() + ready_timeout_s
+    last_error = "server_not_ready"
+    while time.monotonic() < deadline:
+        exit_code = process.poll()
+        if exit_code is not None:
+            raise RuntimeError(f"engine_process_exited_before_ready exit_code={exit_code}")
+        try:
+            wait_for_server(base_url, healthcheck_path, timeout_s=1.0)
+            return
+        except HttpClientError as exc:
+            last_error = str(exc)
+        time.sleep(1.0)
+    raise HttpClientError(f"Timed out waiting for {base_url}{healthcheck_path}: {last_error}")
+
+
 def run_trial(trial_spec_path: Path) -> dict[str, Any]:
     from .store import StudyStore
 
@@ -203,9 +225,14 @@ def run_trial(trial_spec_path: Path) -> dict[str, Any]:
             stderr=subprocess.STDOUT,
             text=True,
         )
+        probe_history: list[dict[str, Any]] = []
         try:
-            wait_for_server(recipe.base_url, recipe.healthcheck_path, recipe.ready_timeout_s)
-            probe_history: list[dict[str, Any]] = []
+            _wait_for_server_or_exit(
+                process,
+                base_url=recipe.base_url,
+                healthcheck_path=recipe.healthcheck_path,
+                ready_timeout_s=recipe.ready_timeout_s,
+            )
 
             def evaluator(threshold: float) -> ThresholdProbe[ProbePayload]:
                 selected = select_requests_for_threshold(requests, threshold=threshold)
@@ -297,10 +324,25 @@ def run_trial(trial_spec_path: Path) -> dict[str, Any]:
             }
             StudyStore.write_json(Path(trial.result_path), result)
             return result
+        except Exception as exc:  # noqa: BLE001
+            result = {
+                "study_id": trial.study_id,
+                "trial_id": trial.trial_id,
+                "status": "failed",
+                "best_sampling_u": None,
+                "best_request_rate": None,
+                "best_pass_rate": None,
+                "best_request_count": None,
+                "failure_reason": str(exc),
+                "probes": probe_history,
+            }
+            StudyStore.write_json(Path(trial.result_path), result)
+            return result
         finally:
-            process.terminate()
-            try:
-                process.wait(timeout=30)
-            except subprocess.TimeoutExpired:
-                process.kill()
-                process.wait(timeout=30)
+            if process.poll() is None:
+                process.terminate()
+                try:
+                    process.wait(timeout=30)
+                except subprocess.TimeoutExpired:
+                    process.kill()
+                    process.wait(timeout=30)