Merge pull request #6473 from ShahabAhmed01/fix/scheduler-research-settings-guard-5049

fix(scheduler): fall back to defaults for non-numeric research settings

src/task_scheduler.py

@@ -355,6 +355,36 @@ def _normalize_chat_endpoint(url: str) -> str:
         return url
 
 
+def _resolve_research_settings(get_setting_fn):
+    """Resolve the scheduled-research settings with crash-proof fallbacks.
+
+    A hand-edited or agent-written data/settings.json can hold non-numeric
+    values for the research_* keys; the bare int() reads previously raised
+    inside TaskScheduler._execute_research_task and killed every scheduled
+    research run until settings.json was fixed by hand (#5049).
+    src/research_handler.py already guards the same reads defensively —
+    this mirrors that for the scheduler. Valid values pass through
+    unchanged (numeric strings like "4096" still parse).
+    """
+    try:
+        max_tokens = int(get_setting_fn("research_max_tokens", 8192))
+    except (TypeError, ValueError, OverflowError):
+        max_tokens = 8192
+    try:
+        extraction_timeout = int(
+            get_setting_fn("research_extraction_timeout_seconds", 90) or 90
+        )
+    except (TypeError, ValueError, OverflowError):
+        extraction_timeout = 90
+    try:
+        extraction_concurrency = int(
+            get_setting_fn("research_extraction_concurrency", 3) or 3
+        )
+    except (TypeError, ValueError, OverflowError):
+        extraction_concurrency = 3
+    return max_tokens, extraction_timeout, extraction_concurrency
+
+
 class TaskScheduler:
     def __init__(self, session_manager):
         self._session_manager = session_manager
@@ -2187,9 +2217,9 @@ class TaskScheduler:
             pass
         endpoint_url = _normalize_chat_endpoint(endpoint_url)
 
-        max_tokens = int(get_setting("research_max_tokens", 8192))
-        extraction_timeout = int(get_setting("research_extraction_timeout_seconds", 90) or 90)
-        extraction_concurrency = int(get_setting("research_extraction_concurrency", 3) or 3)
+        max_tokens, extraction_timeout, extraction_concurrency = (
+            _resolve_research_settings(get_setting)
+        )
 
         researcher = DeepResearcher(
             llm_endpoint=endpoint_url,

tests/test_scheduler_research_settings_guard.py

@@ -0,0 +1,102 @@
+"""Guarded resolution of scheduled-research settings (#5049).
+
+The scheduler reads ``research_max_tokens`` /
+``research_extraction_timeout_seconds`` / ``research_extraction_concurrency``
+from settings.json with bare ``int()`` casts. A hand-edited value like
+``{"research_max_tokens": "invalid"}`` used to raise inside
+``TaskScheduler._execute_research_task`` and kill every scheduled research
+run until settings.json was fixed by hand.
+
+``_resolve_research_settings`` must fall back to the documented defaults
+(8192 / 90 / 3) for non-numeric input while preserving valid-value behavior
+(numeric strings still parse). ``src/research_handler.py`` already guards the
+same reads defensively; this is the scheduler's equivalent.
+"""
+
+from __future__ import annotations
+
+import json
+
+from src.task_scheduler import _resolve_research_settings
+
+
+def _fake_get_setting(values):
+    """A get_setting stand-in backed by a plain dict."""
+
+    def get(key, default=None):
+        return values.get(key, default)
+
+    return get
+
+
+def test_non_numeric_values_fall_back_to_documented_defaults():
+    get = _fake_get_setting(
+        {
+            "research_max_tokens": "invalid",
+            "research_extraction_timeout_seconds": "bogus",
+            "research_extraction_concurrency": "not-a-number",
+        }
+    )
+
+    assert _resolve_research_settings(get) == (8192, 90, 3)
+
+
+def test_none_values_fall_back_to_documented_defaults():
+    get = _fake_get_setting(
+        {
+            "research_max_tokens": None,
+            "research_extraction_timeout_seconds": None,
+            "research_extraction_concurrency": None,
+        }
+    )
+
+    assert _resolve_research_settings(get) == (8192, 90, 3)
+
+
+def test_numeric_strings_still_parse():
+    get = _fake_get_setting(
+        {
+            "research_max_tokens": "4096",
+            "research_extraction_timeout_seconds": "45",
+            "research_extraction_concurrency": "2",
+        }
+    )
+
+    assert _resolve_research_settings(get) == (4096, 45, 2)
+
+
+def test_missing_keys_keep_documented_defaults():
+    assert _resolve_research_settings(_fake_get_setting({})) == (8192, 90, 3)
+
+
+def test_valid_ints_pass_through():
+    get = _fake_get_setting(
+        {
+            "research_max_tokens": 2048,
+            "research_extraction_timeout_seconds": 120,
+            "research_extraction_concurrency": 6,
+        }
+    )
+
+    assert _resolve_research_settings(get) == (2048, 120, 6)
+
+
+def test_non_finite_json_numbers_fall_back_individually():
+    """JSON exponent overflow produces infinity, which int() cannot convert."""
+    keys = (
+        "research_max_tokens",
+        "research_extraction_timeout_seconds",
+        "research_extraction_concurrency",
+    )
+    valid = (4096, 45, 2)
+    defaults = (8192, 90, 3)
+
+    for non_finite in (json.loads("1e309"), json.loads("-1e309")):
+        for index, key in enumerate(keys):
+            values = dict(zip(keys, valid))
+            values[key] = non_finite
+            expected = list(valid)
+            expected[index] = defaults[index]
+            assert _resolve_research_settings(
+                _fake_get_setting(values)
+            ) == tuple(expected)