CoW: Use weakref callbacks to track dead references

Co-authored-by: José Lucas Silva Mayer <[email protected]>

Author: wangwillian0

pandas/_libs/internals.pyx

@@ -890,29 +890,33 @@ cdef class BlockValuesRefs:
     """
     cdef:
         public list referenced_blocks
-        public int clear_counter
+        public int dead_counter
+        object __weakref__
+        object _weakref_cb
 
     def __cinit__(self, blk: Block | None = None) -> None:
+        def _weakref_cb(
+            item: weakref.ref,
+            selfref: weakref.ref = weakref.ref(self)
+        ) -> None:
+            self = selfref()
+            if self is not None:
+                self.dead_counter += 1
+                if self.dead_counter > 256:
+                    if self.dead_counter > len(self.referenced_blocks) // 2:
+                        self._clear_dead_references()
+        self._weakref_cb = _weakref_cb
         if blk is not None:
-            self.referenced_blocks = [weakref.ref(blk)]
+            self.referenced_blocks = [weakref.ref(blk, self._weakref_cb)]
         else:
             self.referenced_blocks = []
-        self.clear_counter = 500  # set reasonably high
-
-    def _clear_dead_references(self, force=False) -> None:
-        # Use exponential backoff to decide when we want to clear references
-        # if force=False. Clearing for every insertion causes slowdowns if
-        # all these objects stay alive, e.g. df.items() for wide DataFrames
-        # see GH#55245 and GH#55008
-        if force or len(self.referenced_blocks) > self.clear_counter:
-            self.referenced_blocks = [
-                ref for ref in self.referenced_blocks if ref() is not None
-            ]
-            nr_of_refs = len(self.referenced_blocks)
-            if nr_of_refs < self.clear_counter // 2:
-                self.clear_counter = max(self.clear_counter // 2, 500)
-            elif nr_of_refs > self.clear_counter:
-                self.clear_counter = max(self.clear_counter * 2, nr_of_refs)
+
+    def _clear_dead_references(self) -> None:
+        old_len = len(self.referenced_blocks)
+        self.referenced_blocks = [
+            ref for ref in self.referenced_blocks if ref() is not None
+        ]
+        self.dead_counter = self.dead_counter - (old_len - len(self.referenced_blocks))
 
     def add_reference(self, blk: Block) -> None:
         """Adds a new reference to our reference collection.
@@ -922,8 +926,7 @@ cdef class BlockValuesRefs:
         blk : Block
             The block that the new references should point to.
         """
-        self._clear_dead_references()
-        self.referenced_blocks.append(weakref.ref(blk))
+        self.referenced_blocks.append(weakref.ref(blk, self._weakref_cb))
 
     def add_index_reference(self, index: object) -> None:
         """Adds a new reference to our reference collection when creating an index.
@@ -933,8 +936,7 @@ cdef class BlockValuesRefs:
         index : Index
             The index that the new reference should point to.
         """
-        self._clear_dead_references()
-        self.referenced_blocks.append(weakref.ref(index))
+        self.referenced_blocks.append(weakref.ref(index, self._weakref_cb))
 
     def has_reference(self) -> bool:
         """Checks if block has foreign references.
@@ -946,6 +948,5 @@ cdef class BlockValuesRefs:
         -------
         bool
         """
-        self._clear_dead_references(force=True)
         # Checking for more references than block pointing to itself
-        return len(self.referenced_blocks) > 1
+        return len(self.referenced_blocks) - self.dead_counter > 1

pandas/tests/copy_view/test_internals.py

@@ -121,31 +121,44 @@ def test_iset_splits_blocks_inplace(using_copy_on_write, locs, arr, dtype):
             assert not np.shares_memory(get_array(df, col), get_array(df2, col))
 
 
-def test_exponential_backoff():
-    # GH#55518
+def test_clear_dead_references():
+    # GH#55539
     df = DataFrame({"a": [1, 2, 3]})
     for i in range(490):
         df.copy(deep=False)
 
-    assert len(df._mgr.blocks[0].refs.referenced_blocks) == 491
+    assert (
+        len(df._mgr.blocks[0].refs.referenced_blocks)
+        - df._mgr.blocks[0].refs.dead_counter
+        == 1
+    )
 
     df = DataFrame({"a": [1, 2, 3]})
     dfs = [df.copy(deep=False) for i in range(510)]
 
     for i in range(20):
         df.copy(deep=False)
-    assert len(df._mgr.blocks[0].refs.referenced_blocks) == 531
-    assert df._mgr.blocks[0].refs.clear_counter == 1000
+    assert (
+        len(df._mgr.blocks[0].refs.referenced_blocks)
+        - df._mgr.blocks[0].refs.dead_counter
+        == 511
+    )
 
     for i in range(500):
         df.copy(deep=False)
 
-    # Don't reduce since we still have over 500 objects alive
-    assert df._mgr.blocks[0].refs.clear_counter == 1000
+    assert (
+        len(df._mgr.blocks[0].refs.referenced_blocks)
+        - df._mgr.blocks[0].refs.dead_counter
+        == 511
+    )
 
     dfs = dfs[:300]
     for i in range(500):
         df.copy(deep=False)
 
-    # Reduce since there are less than 500 objects alive
-    assert df._mgr.blocks[0].refs.clear_counter == 500
+    assert (
+        len(df._mgr.blocks[0].refs.referenced_blocks)
+        - df._mgr.blocks[0].refs.dead_counter
+        == 301
+    )