diff --git a/tools/schema_generator.py b/tools/schema_generator.py
index 12e03bf86..8202ab241 100755
--- a/tools/schema_generator.py
+++ b/tools/schema_generator.py
@@ -17,6 +17,7 @@
 import contextlib
 import re
 
+import sqlalchemy as sa
 import tabulate
 
 from taskflow.persistence.backends import impl_sqlalchemy
@@ -53,29 +54,30 @@ def main():
         with contextlib.closing(backend.get_connection()) as conn:
             conn.upgrade()
         # Now make a prettier version of that schema...
-        tables = backend.engine.execute(TABLE_QUERY)
-        table_names = [r[0] for r in tables]
-        for i, table_name in enumerate(table_names):
-            pretty_name = NAME_MAPPING.get(table_name, table_name)
-            print("*" + pretty_name + "*")
-            # http://www.sqlite.org/faq.html#q24
-            table_name = table_name.replace("\"", "\"\"")
-            rows = []
-            for r in backend.engine.execute(SCHEMA_QUERY % table_name):
-                # Cut out the numbers from things like VARCHAR(12) since
-                # this is not very useful to show users who just want to
-                # see the basic schema...
-                row_type = re.sub(r"\(.*?\)", "", r['type']).strip()
-                if not row_type:
-                    raise ValueError("Row %s of table '%s' was empty after"
-                                     " cleaning" % (r['cid'], table_name))
-                rows.append([r['name'], row_type, to_bool_string(r['pk'])])
-            contents = tabulate.tabulate(
-                rows, headers=['Name', 'Type', 'Primary Key'],
-                tablefmt="rst")
-            print("\n%s" % contents.strip())
-            if i + 1 != len(table_names):
-                print("")
+        with backend.engine.connect() as conn, conn.begin():
+            tables = conn.execute(sa.text(TABLE_QUERY))
+            table_names = [r[0] for r in tables]
+            for i, table_name in enumerate(table_names):
+                pretty_name = NAME_MAPPING.get(table_name, table_name)
+                print("*" + pretty_name + "*")
+                # http://www.sqlite.org/faq.html#q24
+                table_name = table_name.replace("\"", "\"\"")
+                rows = []
+                for r in conn.execute(sa.text(SCHEMA_QUERY % table_name)):
+                    # Cut out the numbers from things like VARCHAR(12) since
+                    # this is not very useful to show users who just want to
+                    # see the basic schema...
+                    row_type = re.sub(r"\(.*?\)", "", r['type']).strip()
+                    if not row_type:
+                        raise ValueError("Row %s of table '%s' was empty after"
+                                         " cleaning" % (r['cid'], table_name))
+                    rows.append([r['name'], row_type, to_bool_string(r['pk'])])
+                contents = tabulate.tabulate(
+                    rows, headers=['Name', 'Type', 'Primary Key'],
+                    tablefmt="rst")
+                print("\n%s" % contents.strip())
+                if i + 1 != len(table_names):
+                    print("")
 
 
 if __name__ == '__main__':