diff --git a/docs/patterns/sqlite3.rst b/docs/patterns/sqlite3.rst
index bc471f66..0d02e465 100644
--- a/docs/patterns/sqlite3.rst
+++ b/docs/patterns/sqlite3.rst
@@ -24,7 +24,15 @@ So here is a simple example of how you can use SQLite 3 with Flask::
 
     @app.teardown_request
     def teardown_request(exception):
-        g.db.close()
+        if hasattr(g, 'db'):
+            g.db.close()
+
+.. note::
+
+   Please keep in mind that the teardown request functions are always
+   executed, even if a before-request handler failed or was never
+   executed.  Because of this we have to make sure here that the database
+   is there before we close it.
 
 Connect on Demand
 -----------------
diff --git a/docs/reqcontext.rst b/docs/reqcontext.rst
index 3b49e1d5..0249b88e 100644
--- a/docs/reqcontext.rst
+++ b/docs/reqcontext.rst
@@ -131,7 +131,9 @@ understand what is actually happening.  The new behavior is quite simple:
 
 4.  At the end of the request the :meth:`~flask.Flask.teardown_request`
     functions are executed.  This always happens, even in case of an
-    unhandled exception down the road.
+    unhandled exception down the road or if a before-request handler was
+    not executed yet or at all (for example in test environments sometimes
+    you might want to not execute before-request callbacks).
 
 Now what happens on errors?  In production mode if an exception is not
 caught, the 500 internal server handler is called.  In development mode
@@ -183,6 +185,12 @@ It's easy to see the behavior from the command line:
 this runs after request
 >>>
 
+Keep in mind that teardown callbacks are always executed, even if
+before-request callbacks were not executed yet but an exception happened.
+Certain parts of the test system might also temporarily create a request
+context without calling the before-request handlers.  Make sure to write
+your teardown-request handlers in a way that they will never fail.
+
 .. _notes-on-proxies:
 
 Notes On Proxies
diff --git a/docs/upgrading.rst b/docs/upgrading.rst
index 13b5be71..b318292c 100644
--- a/docs/upgrading.rst
+++ b/docs/upgrading.rst
@@ -142,7 +142,8 @@ You are now encouraged to use this instead::
 
     @app.teardown_request
     def after_request(exception):
-        g.db.close()
+        if hasattr(g, 'db'):
+            g.db.close()
 
 On the upside this change greatly improves the internal code flow and
 makes it easier to customize the dispatching and error handling.  This
diff --git a/examples/flaskr/flaskr.py b/examples/flaskr/flaskr.py
index 361c1aee..6f9b06fc 100644
--- a/examples/flaskr/flaskr.py
+++ b/examples/flaskr/flaskr.py
@@ -50,7 +50,8 @@ def before_request():
 @app.teardown_request
 def teardown_request(exception):
     """Closes the database again at the end of the request."""
-    g.db.close()
+    if hasattr(g, 'db'):
+        g.db.close()
 
 
 @app.route('/')
diff --git a/examples/minitwit/minitwit.py b/examples/minitwit/minitwit.py
index f7c700d3..fee023fd 100644
--- a/examples/minitwit/minitwit.py
+++ b/examples/minitwit/minitwit.py
@@ -85,7 +85,8 @@ def before_request():
 @app.teardown_request
 def teardown_request(exception):
     """Closes the database again at the end of the request."""
-    g.db.close()
+    if hasattr(g, 'db'):
+        g.db.close()
 
 
 @app.route('/')