Add retry with exponential backoff for PVGIS API calls

src/relife_forecasting/main.py

@@ -19,6 +19,7 @@
 
 from relife_forecasting.config.logging import configure_logging
 from relife_forecasting.routes import health
+from relife_forecasting.utils.retry import retry_on_transient_error
 
 # Prefer your package imports; keep a small fallback to reduce friction during refactors.
 try:
@@ -308,11 +309,16 @@ async def simulate_building(
 
     # 3) Weather + ISO 52016
     if weather_source == "pvgis":
-        hourly_sim, annual_results_df = pybui.ISO52016.Temperature_and_Energy_needs_calculation(
-            bui_checked,
-            weather_source="pvgis",
-            sankey_graph=False,
-        )
+
+        @retry_on_transient_error()
+        def _run_pvgis_simulation():
+            return pybui.ISO52016.Temperature_and_Energy_needs_calculation(
+                bui_checked,
+                weather_source="pvgis",
+                sankey_graph=False,
+            )
+
+        hourly_sim, annual_results_df = _run_pvgis_simulation()
 
     elif weather_source == "epw":
         if epw_file is None:
@@ -987,7 +993,12 @@ def _spec_to_elements(spec: Dict[str, Any]) -> Set[str]:
         # 5) Baseline (optional)
         if include_baseline:
             if weather_source == "pvgis":
-                hourly_sim, annual_results_df = pybui.ISO52016.Temperature_and_Energy_needs_calculation(base_bui, weather_source="pvgis", sankey_graph=False,)
+
+                @retry_on_transient_error()
+                def _run_baseline_pvgis():
+                    return pybui.ISO52016.Temperature_and_Energy_needs_calculation(base_bui, weather_source="pvgis", sankey_graph=False,)
+
+                hourly_sim, annual_results_df = _run_baseline_pvgis()
             else:
                 hourly_sim, annual_results_df = pybui.ISO52016.Temperature_and_Energy_needs_calculation(
                     base_bui, weather_source="epw", path_weather_file=epw_path, sankey_graph=False, 
@@ -1031,7 +1042,12 @@ def _spec_to_elements(spec: Dict[str, Any]) -> Set[str]:
             )
 
             if weather_source == "pvgis":
-                hourly_sim, annual_results_df = pybui.ISO52016.Temperature_and_Energy_needs_calculation(bui_variant, weather_source="pvgis", sankey_graph=False,)
+
+                @retry_on_transient_error()
+                def _run_scenario_pvgis():
+                    return pybui.ISO52016.Temperature_and_Energy_needs_calculation(bui_variant, weather_source="pvgis", sankey_graph=False,)
+
+                hourly_sim, annual_results_df = _run_scenario_pvgis()
             else:
                 hourly_sim, annual_results_df = pybui.ISO52016.Temperature_and_Energy_needs_calculation(
                     bui_variant, weather_source="epw", path_weather_file=epw_path, sankey_graph=False,

src/relife_forecasting/routes/forecasting_service_functions.py

@@ -12,6 +12,8 @@
 from fastapi import HTTPException
 from pydantic import BaseModel, Field
 
+from relife_forecasting.utils.retry import retry_on_transient_error
+
 
 # =============================================================================
 # JSON / Data utilities
@@ -240,10 +242,15 @@ def simulate_building_worker(building_name: str, bui: dict, system: dict) -> Dic
       }
     """
     try:
-        hourly_sim, annual_results_df = pybui.ISO52016.Temperature_and_Energy_needs_calculation(
-            bui,
-            weather_source="pvgis",
-        )
+
+        @retry_on_transient_error()
+        def _run_worker_pvgis():
+            return pybui.ISO52016.Temperature_and_Energy_needs_calculation(
+                bui,
+                weather_source="pvgis",
+            )
+
+        hourly_sim, annual_results_df = _run_worker_pvgis()
 
         calc = pybui.HeatingSystemCalculator(system)
         calc.load_csv_data(hourly_sim)

src/relife_forecasting/utils/retry.py

@@ -0,0 +1,65 @@
+"""Retry utilities for resilient external API calls (e.g., PVGIS)."""
+
+import logging
+import time
+from functools import wraps
+from typing import TypeVar, Callable
+
+logger = logging.getLogger(__name__)
+
+T = TypeVar("T")
+
+# Exceptions that indicate a transient network issue worth retrying.
+TRANSIENT_EXCEPTIONS = (ConnectionError, TimeoutError, OSError)
+
+DEFAULT_MAX_RETRIES = 3
+DEFAULT_INITIAL_DELAY = 1.0
+DEFAULT_BACKOFF_FACTOR = 2.0
+
+
+def retry_on_transient_error(
+    max_retries: int = DEFAULT_MAX_RETRIES,
+    initial_delay: float = DEFAULT_INITIAL_DELAY,
+    backoff_factor: float = DEFAULT_BACKOFF_FACTOR,
+):
+    """Decorator that retries a function on transient network errors.
+
+    Uses exponential backoff between attempts.  Only retries on exceptions
+    that are subclasses of ``TRANSIENT_EXCEPTIONS``; all other exceptions
+    propagate immediately.
+    """
+
+    def decorator(fn: Callable[..., T]) -> Callable[..., T]:
+        @wraps(fn)
+        def wrapper(*args, **kwargs) -> T:
+            delay = initial_delay
+            last_exc: BaseException | None = None
+            for attempt in range(1, max_retries + 1):
+                try:
+                    return fn(*args, **kwargs)
+                except TRANSIENT_EXCEPTIONS as exc:
+                    last_exc = exc
+                    if attempt < max_retries:
+                        logger.warning(
+                            "Transient error in %s (attempt %d/%d): %s — retrying in %.1fs",
+                            fn.__name__,
+                            attempt,
+                            max_retries,
+                            exc,
+                            delay,
+                        )
+                        time.sleep(delay)
+                        delay *= backoff_factor
+                    else:
+                        logger.error(
+                            "Transient error in %s (attempt %d/%d): %s — no retries left",
+                            fn.__name__,
+                            attempt,
+                            max_retries,
+                            exc,
+                        )
+            raise last_exc  # type: ignore[misc]
+
+        return wrapper
+
+    return decorator