fix #94 (#95)

src/scida/io/_base.py

@@ -281,7 +281,16 @@ def create_cachefile(
             hf.attrs["_cachingcomplete"] = True  # mark that caching complete
 
     def load_cachefile(self, location, token="", chunksize="auto", **kwargs):
-        self.file = h5py.File(location, "r")
+        try:
+            self.file = h5py.File(location, "r")
+        except (
+            OSError
+        ) as e:  # file potentially corrupted, raise InvalidCacheError and potentially recover
+            raise InvalidCacheError(
+                """Cache file '%s' might be invalid. An OSError '%s' was raised.
+                Delete file and try again."""
+                % (location, str(e))
+            )
         hf = self.file
         cache_valid = False
         if "_cachingcomplete" in hf.attrs:

tests/test_convenience.py

@@ -162,7 +162,6 @@ def signal_handler(signum, frame):
         with time_limit(dt_int):
             # signal just accepts integer seconds... so we work around it with sleep before
             time.sleep(dt_int - dt)
-
             load(testdatapath, **loadkwargs)
 
     # count total files by walking folders without counting those
@@ -191,3 +190,22 @@ def signal_handler(signum, frame):
     # attempt load and recover.
     load(testdatapath, **loadkwargs)
     assert "Invalid cache file, attempting to create new one." in caplog.text
+
+
+@require_testdata_path("interface", only=["TNG50-4_snapshot"])
+def test_load_cachefail_oserror(cachedir, testdatapath, caplog):
+    """Test recovery from failure during cache creation. Do so by keeping the file is kept open."""
+
+    # explicitly disable catalog discovery
+    loadkwargs = dict(catalog="none")
+    ds = load(testdatapath, **loadkwargs)
+    filename = ds.file.filename
+    ds.file.close()
+
+    # Let's destroy the hdf5 header with random data
+    with open(filename, "r+b") as f:
+        f.seek(0)
+        f.write(b"destroyheader")
+
+    ds = load(testdatapath, **loadkwargs)
+    print(ds.info())