Add configurable retries for getInfo calls and document retry tuning

docs/faq.md

@@ -13,6 +13,37 @@ Use `grid_shape` when a fixed pixel width/height is required (e.g., ML model inp
 ## I get 429 quota errors. What do I do?
 Reduce parallelism (fewer Dask workers), narrow the AOI or time range, combine server-side operations before opening, or switch to the standard endpoint for computed collections.
 
+Xee also exposes retry controls for both pixel fetches and metadata `getInfo()`
+calls. For example:
+
+```python
+import xarray as xr
+
+ds = xr.open_dataset(
+	'ee://ECMWF/ERA5_LAND/MONTHLY_AGGR',
+	engine='ee',
+	crs='EPSG:4326',
+	crs_transform=(0.25, 0, -180, 0, -0.25, 90),
+	shape_2d=(1440, 720),
+	getitem_kwargs={
+		'max_retries': 8,
+		'initial_delay': 500,
+	},
+	getinfo_kwargs={
+		'max_retries': 8,
+		'initial_delay': 1000,
+	},
+)
+```
+
+Defaults are:
+
+- `getitem_kwargs`: `max_retries=6`, `initial_delay=500` ms
+- `getinfo_kwargs`: `max_retries=6`, `initial_delay=1000` ms
+
+`helpers.extract_grid_params(...)` also supports `getinfo_kwargs` if metadata
+calls need tuning under quota pressure.
+
 ## Can I open a computed `ee.ImageCollection`?
 Yes. Build the collection with filtering / mapping functions, then pass the resulting collection object directly to `xr.open_dataset(..., engine='ee')` with grid parameters.
 

docs/guide.md

@@ -154,6 +154,40 @@ temp_slice = ds['temperature_2m'].isel(time=0)
 temp_slice.plot()
 ```
 
+## Configure Retries
+
+Xee supports configurable retries for two paths:
+
+- Pixel reads via `getitem_kwargs`
+- Metadata `getInfo()` calls via `getinfo_kwargs`
+
+```python
+import ee
+import xarray as xr
+from xee import helpers
+
+ic = ee.ImageCollection('ECMWF/ERA5_LAND/MONTHLY_AGGR')
+
+# Optional: tune helper metadata fetch retries.
+grid_params = helpers.extract_grid_params(
+    ic,
+    getinfo_kwargs={'max_retries': 8, 'initial_delay': 1000},
+)
+
+ds = xr.open_dataset(
+    ic,
+    engine='ee',
+    **grid_params,
+    getitem_kwargs={'max_retries': 8, 'initial_delay': 500},
+    getinfo_kwargs={'max_retries': 8, 'initial_delay': 1000},
+)
+```
+
+Defaults:
+
+- `getitem_kwargs`: `max_retries=6`, `initial_delay=500` ms
+- `getinfo_kwargs`: `max_retries=6`, `initial_delay=1000` ms
+
 ## Further Resources
 
 - [Core Concepts](concepts.md)

docs/performance.md

@@ -26,6 +26,42 @@ Recommendations:
 3. Consolidate operations server-side (EE `.map`, `.select`, band math) before opening in Xee.
 4. Cache intermediate results in memory rather than re-opening repeatedly.
 
+## Retry Tuning
+
+Xee uses exponential backoff with jitter for:
+
+- Pixel requests (`getitem_kwargs`) used during array reads.
+- Metadata `getInfo()` requests (`getinfo_kwargs`) used during dataset setup and
+	helper metadata fetches.
+
+Defaults:
+
+- `getitem_kwargs`: `max_retries=6`, `initial_delay=500` ms
+- `getinfo_kwargs`: `max_retries=6`, `initial_delay=1000` ms
+
+`getinfo_kwargs` starts with a longer default delay to reduce setup-time retry bursts against EE metadata endpoints.
+
+You can tune these in `xr.open_dataset(...)`:
+
+```python
+ds = xr.open_dataset(
+		collection,
+		engine='ee',
+		crs='EPSG:4326',
+		crs_transform=(0.25, 0, -180, 0, -0.25, 90),
+		shape_2d=(1440, 720),
+		getitem_kwargs={'max_retries': 8, 'initial_delay': 500},
+		getinfo_kwargs={'max_retries': 8, 'initial_delay': 1000},
+)
+```
+
+Rule of thumb:
+
+1. If failures happen during dataset open / metadata fetch, tune
+	 `getinfo_kwargs` first.
+2. If failures happen during chunk reads / compute, tune `getitem_kwargs` first.
+3. Reduce Dask concurrency before increasing retries too aggressively.
+
 ## Chunk Size Considerations
 
 EE responses have an upper size limit (tens of MB). Xee's backend picks reasonable pixel window sizes automatically. If you see many small requests, consider choosing a coarser grid or limiting variable selection to needed bands.

xee/ext.py

@@ -41,6 +41,7 @@
 from xarray.backends import store as backends_store
 from xarray.core import indexing
 from xarray.core import utils
+from xee import retries
 from xee import types
 
 import ee
@@ -128,6 +129,11 @@ class EarthEngineStore(common.AbstractDataStore):
       'initial_delay': 500,
   }
 
+  GETINFO_KWARGS: dict[str, int] = {
+      'max_retries': 6,
+      'initial_delay': 1000,
+  }
+
   SCALE_UNITS: dict[str, int] = {
       'degree': 1,
       'metre': 10_000,
@@ -166,6 +172,7 @@ def open(
       ee_init_if_necessary: bool = False,
       executor_kwargs: dict[str, Any] | None = None,
       getitem_kwargs: dict[str, int] | None = None,
+      getinfo_kwargs: dict[str, int] | None = None,
       fast_time_slicing: bool = False,
   ) -> EarthEngineStore:
     if mode != 'r':
@@ -188,6 +195,7 @@ def open(
         ee_init_if_necessary=ee_init_if_necessary,
         executor_kwargs=executor_kwargs,
         getitem_kwargs=getitem_kwargs,
+        getinfo_kwargs=getinfo_kwargs,
         fast_time_slicing=fast_time_slicing,
     )
 
@@ -207,6 +215,7 @@ def __init__(
       ee_init_if_necessary: bool = False,
       executor_kwargs: dict[str, Any] | None = None,
       getitem_kwargs: dict[str, int] | None = None,
+      getinfo_kwargs: dict[str, int] | None = None,
       fast_time_slicing: bool = False,
   ):
     # Ensure crs_transform is a tuple and create the affine.Affine object.
@@ -236,6 +245,7 @@ def __init__(
     self.executor_kwargs = executor_kwargs
 
     self.getitem_kwargs = {**self.GETITEM_KWARGS, **(getitem_kwargs or {})}
+    self.getinfo_kwargs = {**self.GETINFO_KWARGS, **(getinfo_kwargs or {})}
 
     self.image_collection = image_collection
     if n_images != -1:
@@ -306,7 +316,11 @@ def get_info(self) -> dict[str, Any]:
         )
     )
 
-    info = ee.List([rpc for _, rpc in rpcs]).getInfo()
+    info = retries.robust_call(
+        lambda: ee.List([rpc for _, rpc in rpcs]).getInfo(),
+        catch=ee.ee_exception.EEException,
+        **self.getinfo_kwargs,
+    )
 
     return dict(zip((name for name, _ in rpcs), info))
 
@@ -657,9 +671,20 @@ def _ee_bounds_to_bounds(bounds: dict[str, Any]) -> types.Bounds:
   return x_min, y_min, x_max, y_max
 
 
-def geometry_to_bounds(geom: ee.Geometry) -> types.Bounds:
+def geometry_to_bounds(
+    geom: ee.Geometry,
+    getinfo_kwargs: dict[str, int] | None = None,
+) -> types.Bounds:
   """Finds the CRS bounds from a ee.Geometry polygon."""
-  bounds = geom.bounds().getInfo()
+  getinfo_kwargs = {
+      **EarthEngineStore.GETINFO_KWARGS,
+      **(getinfo_kwargs or {}),
+  }
+  bounds = retries.robust_call(
+      lambda: geom.bounds().getInfo(),
+      catch=ee.ee_exception.EEException,
+      **getinfo_kwargs,
+  )
   return _ee_bounds_to_bounds(bounds)
 
 
@@ -920,6 +945,7 @@ def open_dataset(
       ee_init_kwargs: dict[str, Any] | None = None,
       executor_kwargs: dict[str, Any] | None = None,
       getitem_kwargs: dict[str, int] | None = None,
+      getinfo_kwargs: dict[str, int] | None = None,
       fast_time_slicing: bool = False,
   ) -> xarray.Dataset:  # type: ignore
     """Open an Earth Engine ImageCollection as an Xarray Dataset.
@@ -989,6 +1015,11 @@ def open_dataset(
         - 'max_retries', the maximum number of retry attempts. Defaults to 6.
         - 'initial_delay', the initial delay in milliseconds before the first
           retry. Defaults to 500.
+      getinfo_kwargs (optional): Exponential backoff kwargs applied to
+        Earth Engine `getInfo()` calls used by Xee metadata workflows.
+        - 'max_retries', the maximum number of retry attempts. Defaults to 6.
+        - 'initial_delay', the initial delay in milliseconds before the first
+          retry. Defaults to 1000.
       fast_time_slicing (optional): Whether to perform an optimization that
         makes slicing an ImageCollection across time faster. This optimization
         loads EE images in a slice by ID, so any modifications to images in a
@@ -1023,6 +1054,7 @@ def open_dataset(
         ee_init_if_necessary=ee_init_if_necessary,
         executor_kwargs=executor_kwargs,
         getitem_kwargs=getitem_kwargs,
+        getinfo_kwargs=getinfo_kwargs,
         fast_time_slicing=fast_time_slicing,
     )
 

xee/ext_integration_test.py

@@ -105,6 +105,14 @@ def setUp(self):
         n_images=64,
         getitem_kwargs={'max_retries': 9},
     )
+    self.getinfo_tuned_store = xee.EarthEngineStore(
+      ee.ImageCollection('LANDSAT/LC08/C02/T1').filterDate(
+        '2017-01-01', '2017-01-03'
+      ),
+      n_images=64,
+      getinfo_kwargs={'max_retries': 9, 'initial_delay': 1200},
+      **_TEST_GRID_PARAMS,
+    )
     self.all_img_store = xee.EarthEngineStore(
         ee.ImageCollection('LANDSAT/LC08/C02/T1').filterDate(
             '2017-01-01', '2017-01-03'
@@ -298,6 +306,15 @@ def test_getitem_kwargs(self):
     self.assertEqual(arr2.store.getitem_kwargs['initial_delay'], 500)
     self.assertEqual(arr2.store.getitem_kwargs['max_retries'], 9)
 
+  def test_getinfo_kwargs(self):
+    arr = xee.EarthEngineBackendArray('B4', self.getinfo_tuned_store)
+    self.assertEqual(arr.store.getinfo_kwargs['initial_delay'], 1200)
+    self.assertEqual(arr.store.getinfo_kwargs['max_retries'], 9)
+
+    arr1 = xee.EarthEngineBackendArray('longitude', self.lnglat_store)
+    self.assertEqual(arr1.store.getinfo_kwargs['initial_delay'], 1000)
+    self.assertEqual(arr1.store.getinfo_kwargs['max_retries'], 6)
+
 
 class EEBackendEntrypointTest(absltest.TestCase):
 
@@ -632,6 +649,15 @@ def test_extract_grid_params_from_image(self):
     self.assertEqual(grid_params['crs'], 'EPSG:32613')
     np.allclose(grid_params['crs_transform'], [30, 0, 643185, 0, -30, 4255815])
 
+  def test_extract_grid_params_from_image_with_getinfo_kwargs(self):
+    img = ee.Image('LANDSAT/LT05/C02/T1_TOA/LT05_031034_20110619')
+    grid_params = helpers.extract_grid_params(
+        img, getinfo_kwargs={'max_retries': 8, 'initial_delay': 1100}
+    )
+    self.assertEqual(grid_params['shape_2d'], (7881, 6981))
+    self.assertEqual(grid_params['crs'], 'EPSG:32613')
+    np.allclose(grid_params['crs_transform'], [30, 0, 643185, 0, -30, 4255815])
+
   def test_extract_grid_params_from_image_collection(self):
     dem = ee.ImageCollection('COPERNICUS/DEM/GLO30')
     grid_params = helpers.extract_grid_params(dem)