gh-134004: Dbm vacuuming

Doc/library/dbm.rst

@@ -15,10 +15,16 @@
 * :mod:`dbm.ndbm`
 
 If none of these modules are installed, the
-slow-but-simple implementation in module :mod:`dbm.dumb` will be used.  There
+slow-but-simple implementation in module :mod:`dbm.dumb` will be used. There
 is a `third party interface <https://www.jcea.es/programacion/pybsddb.htm>`_ to
 the Oracle Berkeley DB.
 
+.. note::
+   None of the underlying modules will automatically shrink the disk space used by
+   the database file. However, :mod:`dbm.sqlite3`, :mod:`dbm.gnu` and :mod:`dbm.dumb`
+   provide a :meth:`!reorganize` method that can be used for this purpose.
+
+
 .. exception:: error
 
    A tuple containing the exceptions that can be raised by each of the supported
@@ -186,6 +192,16 @@ or any other SQLite browser, including the SQLite CLI.
       The Unix file access mode of the file (default: octal ``0o666``),
       used only when the database has to be created.
 
+   .. method:: sqlite3.reorganize()
+
+      If you have carried out a lot of deletions and would like to shrink the space
+      used on disk, this method will reorganize the database; otherwise, deleted file
+      space will be kept and reused as new (key, value) pairs are added.
+
+   .. note::
+      While reorganizing, as much as twice the size of the original database is required
+      in free disk space.
+
 
 :mod:`dbm.gnu` --- GNU database manager
 ---------------------------------------
@@ -438,6 +454,9 @@ The :mod:`!dbm.dumb` module defines the following:
       with a sufficiently large/complex entry due to stack depth limitations in
       Python's AST compiler.
 
+   .. warning::
+      :mod:`dbm.dumb` does not support concurrent writes, which can corrupt the database.
+
    .. versionchanged:: 3.5
       :func:`~dbm.dumb.open` always creates a new database when *flag* is ``'n'``.
 
@@ -460,3 +479,9 @@ The :mod:`!dbm.dumb` module defines the following:
    .. method:: dumbdbm.close()
 
       Close the database.
+
+   .. method:: dumbdbm.reorganize()
+
+      If you have carried out a lot of deletions and would like to shrink the space
+      used on disk, this method will reorganize the database; otherwise, deleted file
+      space will not be reused.

Doc/library/shelve.rst

@@ -75,8 +75,13 @@ Two additional methods are supported:
 
    Write back all entries in the cache if the shelf was opened with *writeback*
    set to :const:`True`.  Also empty the cache and synchronize the persistent
-   dictionary on disk, if feasible.  This is called automatically when the shelf
-   is closed with :meth:`close`.
+   dictionary on disk, if feasible.  This is called automatically when
+   :meth:`reorganize` is called or the shelf is closed with :meth:`close`.
+
+.. method:: Shelf.reorganize()
+
+   Calls :meth:`sync` and attempts to shrink space used on disk by removing empty
+   space resulting from deletions.
 
 .. method:: Shelf.close()
 
@@ -116,6 +121,11 @@ Restrictions
 * On macOS :mod:`dbm.ndbm` can silently corrupt the database file on updates,
   which can cause hard crashes when trying to read from the database.
 
+* :meth:`Shelf.reorganize` may not be available for all database packages and
+  may temporarely increase resource usage (especially disk space) when called.
+  Additionally, it will never run automatically and instead needs to be called
+  explicitly.
+
 
 .. class:: Shelf(dict, protocol=None, writeback=False, keyencoding='utf-8')
 

Lib/dbm/dumb.py

@@ -8,17 +8,12 @@
 
 - seems to contain a bug when updating...
 
-- reclaim free space (currently, space once occupied by deleted or expanded
-items is never reused)
-
 - support concurrent access (currently, if two processes take turns making
 updates, they can mess up the index)
 
 - support efficient access to large databases (currently, the whole index
 is read when the database is opened, and some updates rewrite the whole index)
 
-- support opening for read-only (flag = 'm')
-
 """
 
 import ast as _ast
@@ -289,6 +284,34 @@ def __enter__(self):
     def __exit__(self, *args):
         self.close()
 
+    def reorganize(self):
+        if self._readonly:
+            raise error('The database is opened for reading only')
+        self._verify_open()
+        # Ensure all changes are committed before reorganizing.
+        self._commit()
+        # Open file in r+ to allow changing in-place.
+        with _io.open(self._datfile, 'rb+') as f:
+            reorganize_pos = 0
+
+            # Iterate over existing keys, sorted by starting byte.
+            for key in sorted(self._index.keys(), key = lambda k: self._index[k][0]):
+                pos, siz = self._index[key]
+                f.seek(pos)
+                val = f.read(siz)
+
+                f.seek(reorganize_pos)
+                f.write(val)
+                self._index[key] = (reorganize_pos, siz)
+
+                blocks_occupied = (siz + _BLOCKSIZE - 1) // _BLOCKSIZE
+                reorganize_pos += blocks_occupied * _BLOCKSIZE
+
+            f.truncate(reorganize_pos)
+        # Commit changes to index, which were not in-place.
+        self._commit()
+
+
 
 def open(file, flag='c', mode=0o666):
     """Open the database file, filename, and return corresponding object.

Lib/dbm/sqlite3.py

@@ -15,6 +15,7 @@
 STORE_KV = "REPLACE INTO Dict (key, value) VALUES (CAST(? AS BLOB), CAST(? AS BLOB))"
 DELETE_KEY = "DELETE FROM Dict WHERE key = CAST(? AS BLOB)"
 ITER_KEYS = "SELECT key FROM Dict"
+REORGANIZE = "VACUUM"
 
 
 class error(OSError):
@@ -122,6 +123,9 @@ def __enter__(self):
     def __exit__(self, *args):
         self.close()
 
+    def reorganize(self):
+        self._execute(REORGANIZE)
+
 
 def open(filename, /, flag="r", mode=0o666):
     """Open a dbm.sqlite3 database and return the dbm object.

Lib/shelve.py

@@ -171,6 +171,11 @@ def sync(self):
         if hasattr(self.dict, 'sync'):
             self.dict.sync()
 
+    def reorganize(self):
+        self.sync()
+        if hasattr(self.dict, 'reorganize'):
+            self.dict.reorganize()
+
 
 class BsdDbShelf(Shelf):
     """Shelf implementation using the "BSD" db interface.

Lib/test/test_dbm.py

@@ -135,6 +135,67 @@ def test_anydbm_access(self):
         assert(f[key] == b"Python:")
         f.close()
 
+    def test_anydbm_readonly_reorganize(self):
+        self.init_db()
+        with dbm.open(_fname, 'r') as d:
+            # Early stopping.
+            if not hasattr(d, 'reorganize'):
+                return
+
+            self.assertRaises(dbm.error, lambda: d.reorganize())
+
+    def test_anydbm_reorganize_not_changed_content(self):
+        self.init_db()
+        with dbm.open(_fname, 'c') as d:
+            # Early stopping.
+            if not hasattr(d, 'reorganize'):
+                return
+
+            keys_before = sorted(d.keys())
+            values_before = [d[k] for k in keys_before]
+            d.reorganize()
+            keys_after = sorted(d.keys())
+            values_after = [d[k] for k in keys_before]
+            self.assertEqual(keys_before, keys_after)
+            self.assertEqual(values_before, values_after)
+
+    def test_anydbm_reorganize_decreased_size(self):
+
+        def _calculate_db_size(db_path):
+            if os.path.isfile(db_path):
+                return os.path.getsize(db_path)
+            total_size = 0
+            for root, _, filenames in os.walk(db_path):
+                for filename in filenames:
+                    file_path = os.path.join(root, filename)
+                    total_size += os.path.getsize(file_path)
+            return total_size
+
+        # This test requires relatively large databases to reliably show difference in size before and after reorganizing.
+        with dbm.open(_fname, 'n') as f:
+            # Early stopping.
+            if not hasattr(f, 'reorganize'):
+                return
+
+            for k in self._dict:
+                f[k.encode('ascii')] = self._dict[k] * 100000
+            db_keys = list(f.keys())
+
+        # Make sure to calculate size of database only after file is closed to ensure file content are flushed to disk.
+        size_before = _calculate_db_size(os.path.dirname(_fname))
+
+        # Delete some elements from the start of the database.
+        keys_to_delete = db_keys[:len(db_keys) // 2]
+        with dbm.open(_fname, 'c') as f:
+            for k in keys_to_delete:
+                del f[k]
+            f.reorganize()
+
+        # Make sure to calculate size of database only after file is closed to ensure file content are flushed to disk.
+        size_after = _calculate_db_size(os.path.dirname(_fname))
+
+        self.assertLess(size_after, size_before)
+
     def test_open_with_bytes(self):
         dbm.open(os.fsencode(_fname), "c").close()
 

Misc/ACKS

@@ -1362,6 +1362,7 @@ Milan Oberkirch
 Pascal Oberndoerfer
 Géry Ogam
 Seonkyo Ok
+Andrea Oliveri
 Jeffrey Ollie
 Adam Olsen
 Bryan Olson

Misc/NEWS.d/next/Library/2025-05-15-00-27-09.gh-issue-134004.e8k4-R.rst

@@ -0,0 +1,2 @@
+:mod:`shelve` as well as underlying :mod:`!dbm.dumb` and :mod:`!dbm.sqlite` now have :meth:`!reorganize` methods to
+recover unused free space previously occupied by deleted entries.