feat(caching): tie lifetime of cached tables to python refs

conda/environment-arm64-flink.yml

@@ -5,7 +5,6 @@ dependencies:
   # runtime dependencies
   - python =3.10
   - atpublic >=2.3
-  - bidict >=0.22.1
   - black >=22.1.0,<25
   - clickhouse-connect >=0.5.23
   - dask >=2022.9.1

conda/environment-arm64.yml

@@ -5,7 +5,6 @@ dependencies:
   # runtime dependencies
   - python >=3.10
   - atpublic >=2.3
-  - bidict >=0.22.1
   - black >=22.1.0,<25
   - clickhouse-connect >=0.5.23
   - dask >=2022.9.1

conda/environment.yml

@@ -5,7 +5,6 @@ dependencies:
   # runtime dependencies
   - apache-flink
   - atpublic >=2.3
-  - bidict >=0.22.1
   - black >=22.1.0,<25
   - clickhouse-connect >=0.5.23
   - dask >=2022.9.1

docs/posts/run-on-snowflake/index.qmd

@@ -111,7 +111,6 @@ session.sproc.register(
             "snowflake-snowpark-python",
             "toolz",
             "atpublic",
-            "bidict",
             "pyarrow",
             "pandas",
             "numpy",

ibis/backends/__init__.py

@@ -1228,8 +1228,7 @@ def _cached(self, expr: ir.Table):
         """
         op = expr.op()
         if (result := self._query_cache.get(op)) is None:
-            self._query_cache.store(expr)
-            result = self._query_cache[op]
+            result = self._query_cache.store(expr)
         return ir.CachedTable(result)
 
     def _release_cached(self, expr: ir.CachedTable) -> None:
@@ -1241,12 +1240,12 @@ def _release_cached(self, expr: ir.CachedTable) -> None:
             Cached expression to release
 
         """
-        del self._query_cache[expr.op()]
+        self._query_cache.release(expr.op().name)
 
     def _load_into_cache(self, name, expr):
         raise NotImplementedError(self.name)
 
-    def _clean_up_cached_table(self, op):
+    def _clean_up_cached_table(self, name):
         raise NotImplementedError(self.name)
 
     def _transpile_sql(self, query: str, *, dialect: str | None = None) -> str:

ibis/backends/bigquery/__init__.py

@@ -1144,9 +1144,9 @@ def drop_view(
     def _load_into_cache(self, name, expr):
         self.create_table(name, expr, schema=expr.schema(), temp=True)
 
-    def _clean_up_cached_table(self, op):
+    def _clean_up_cached_table(self, name):
         self.drop_table(
-            op.name,
+            name,
             database=(self._session_dataset.project, self._session_dataset.dataset_id),
         )
 

ibis/backends/duckdb/__init__.py

@@ -281,8 +281,8 @@ def create_table(
     def _load_into_cache(self, name, expr):
         self.create_table(name, expr, schema=expr.schema(), temp=True)
 
-    def _clean_up_cached_table(self, op):
-        self.drop_table(op.name)
+    def _clean_up_cached_table(self, name):
+        self.drop_table(name)
 
     def table(
         self, name: str, schema: str | None = None, database: str | None = None

ibis/backends/oracle/__init__.py

@@ -627,5 +627,5 @@
             with contextlib.suppress(oracledb.DatabaseError):
                 bind.execute(f'DROP TABLE "{name}"')
 
-    def _clean_up_cached_table(self, op):
-        self._clean_up_tmp_table(op.name)
+    def _clean_up_cached_table(self, name):
+        self._clean_up_tmp_table(name)

ibis/backends/pandas/__init__.py

@@ -275,8 +275,8 @@
     def has_operation(cls, operation: type[ops.Value]) -> bool:
         return operation in cls._get_operations()
 
-    def _clean_up_cached_table(self, op):
-        del self.dictionary[op.name]
+    def _clean_up_cached_table(self, name):
+        del self.dictionary[name]
 
     def to_pyarrow(
         self,

ibis/backends/polars/__init__.py

@@ -560,8 +560,8 @@
     def _load_into_cache(self, name, expr):
         self.create_table(name, self.compile(expr).cache())
 
-    def _clean_up_cached_table(self, op):
-        self._remove_table(op.name)
+    def _clean_up_cached_table(self, name):
+        self._remove_table(name)
 
 
 @lazy_singledispatch

ibis/backends/pyspark/__init__.py

@@ -638,8 +638,7 @@ def _load_into_cache(self, name, expr):
         # asked to, instead of when the session ends
         self._cached_dataframes[name] = t
 
-    def _clean_up_cached_table(self, op):
-        name = op.name
+    def _clean_up_cached_table(self, name):
         self._session.catalog.dropTempView(name)
         t = self._cached_dataframes.pop(name)
         assert t.is_cached

ibis/backends/snowflake/tests/test_udf.py

@@ -276,7 +276,6 @@ def ibis_sproc(session):
             "snowflake-snowpark-python",
             "toolz",
             "atpublic",
-            "bidict",
             "pyarrow",
             "pandas",
             "numpy",

ibis/backends/sql/__init__.py

@@ -275,8 +275,8 @@ def drop_view(
     def _load_into_cache(self, name, expr):
         self.create_table(name, expr, schema=expr.schema(), temp=True)
 
-    def _clean_up_cached_table(self, op):
-        self.drop_table(op.name)
+    def _clean_up_cached_table(self, name):
+        self.drop_table(name)
 
     def execute(
         self,

ibis/backends/tests/test_client.py

@@ -1334,31 +1334,6 @@
     assert result == schema
 
 
-@mark.notimpl(["datafusion", "flink", "impala", "trino", "druid"])
-@mark.never(
-    ["mssql"],
-    reason="mssql supports support temporary tables through naming conventions",
-)
-@mark.notimpl(["exasol"], reason="Exasol does not support temporary tables")
-@pytest.mark.never(
-    ["risingwave"],
-    raises=com.UnsupportedOperationError,
-    reason="Feature is not yet implemented: CREATE TEMPORARY TABLE",
-)
-def test_persist_expression_ref_count(backend, con, alltypes):
-    non_persisted_table = alltypes.mutate(test_column=ibis.literal("calculation"))
-    persisted_table = non_persisted_table.cache()
-
-    op = non_persisted_table.op()
-
-    # ref count is unaffected without a context manager
-    assert con._query_cache.refs[op] == 1
-    backend.assert_frame_equal(
-        non_persisted_table.to_pandas(), persisted_table.to_pandas()
-    )
-    assert con._query_cache.refs[op] == 1
-
-
 @mark.notimpl(["datafusion", "flink", "impala", "trino", "druid"])
 @mark.never(
     ["mssql"],
@@ -1391,14 +1366,16 @@
     raises=com.UnsupportedOperationError,
     reason="Feature is not yet implemented: CREATE TEMPORARY TABLE",
 )
-def test_persist_expression_contextmanager(backend, alltypes):
+def test_persist_expression_contextmanager(backend, con, alltypes):
     non_cached_table = alltypes.mutate(
         test_column=ibis.literal("calculation"), other_column=ibis.literal("big calc")
     )
-    with non_cached_table.cache() as cached_table:
-        backend.assert_frame_equal(
-            non_cached_table.to_pandas(), cached_table.to_pandas()
-        )
+    with pytest.warns(DeprecationWarning):
+        with non_cached_table.cache() as cached_table:
+            backend.assert_frame_equal(
+                non_cached_table.to_pandas(), cached_table.to_pandas()
+            )
+        assert non_cached_table.op() not in con._query_cache.cache
 
 
 @mark.notimpl(["datafusion", "flink", "impala", "trino", "druid"])
@@ -1417,12 +1394,12 @@
         test_column=ibis.literal("calculation"), other_column=ibis.literal("big calc 2")
     )
     op = non_cached_table.op()
-    with non_cached_table.cache() as cached_table:
-        backend.assert_frame_equal(
-            non_cached_table.to_pandas(), cached_table.to_pandas()
-        )
-        assert con._query_cache.refs[op] == 1
-    assert con._query_cache.refs[op] == 0
+    cached_table = non_cached_table.cache()
+    backend.assert_frame_equal(non_cached_table.to_pandas(), cached_table.to_pandas())
+
+    assert op in con._query_cache.cache
+    del cached_table
+    assert op not in con._query_cache.cache
 
 
 @mark.notimpl(["datafusion", "flink", "impala", "trino", "druid"])
@@ -1441,29 +1418,28 @@
         test_column=ibis.literal("calculation"), other_column=ibis.literal("big calc 2")
     )
     op = non_cached_table.op()
-    with non_cached_table.cache() as cached_table:
-        backend.assert_frame_equal(
-            non_cached_table.to_pandas(), cached_table.to_pandas()
-        )
+    cached_table = non_cached_table.cache()
 
-        name1 = cached_table.op().name
+    backend.assert_frame_equal(non_cached_table.to_pandas(), cached_table.to_pandas())
 
-        with non_cached_table.cache() as nested_cached_table:
-            name2 = nested_cached_table.op().name
-            assert not nested_cached_table.to_pandas().empty
+    name = cached_table.op().name
+    nested_cached_table = non_cached_table.cache()
 
-            # there are two refs to the uncached expression
-            assert con._query_cache.refs[op] == 2
+    # cached tables are identical and reusing the same op
+    assert cached_table.op() is nested_cached_table.op()
+    # table is cached
+    assert op in con._query_cache.cache
 
-        # one ref to the uncached expression was removed by the context manager
-        assert con._query_cache.refs[op] == 1
+    # deleting the first reference, leaves table in cache
+    del nested_cached_table
+    assert op in con._query_cache.cache
 
-    # no refs left after the outer context manager exits
-    assert con._query_cache.refs[op] == 0
+    # deleting the last reference, releases table from cache
+    del cached_table
+    assert op not in con._query_cache.cache
 
-    # assert that tables have been dropped
-    assert name1 not in con.list_tables()
-    assert name2 not in con.list_tables()
+    # assert that table has been dropped
+    assert name not in con.list_tables()
 
 
 @mark.notimpl(["datafusion", "flink", "impala", "trino", "druid"])
@@ -1477,13 +1453,19 @@
     raises=com.UnsupportedOperationError,
     reason="Feature is not yet implemented: CREATE TEMPORARY TABLE",
 )
-def test_persist_expression_repeated_cache(alltypes):
+def test_persist_expression_repeated_cache(alltypes, con):
     non_cached_table = alltypes.mutate(
         test_column=ibis.literal("calculation"), other_column=ibis.literal("big calc 2")
     )
-    with non_cached_table.cache() as cached_table:
-        with cached_table.cache() as nested_cached_table:
-            assert not nested_cached_table.to_pandas().empty
+    cached_table = non_cached_table.cache()
+    nested_cached_table = cached_table.cache()
+    name = cached_table.op().name
+
+    assert not nested_cached_table.to_pandas().empty
+
+    del nested_cached_table, cached_table
+
+    assert name not in con.list_tables()
 
 
 @mark.notimpl(["datafusion", "flink", "impala", "trino", "druid"])
@@ -1502,12 +1484,17 @@
         test_column=ibis.literal("calculation"), other_column=ibis.literal("big calc 3")
     )
     cached_table = non_cached_table.cache()
-    cached_table.release()
-    assert con._query_cache.refs[non_cached_table.op()] == 0
+    with pytest.warns(DeprecationWarning):
+        cached_table.release()
 
-    with pytest.raises(
-        com.IbisError,
-        match=r".+Did you call `\.release\(\)` twice on the same expression\?",
+    assert non_cached_table.op() not in con._query_cache.cache
+
+    with (
+        pytest.raises(
+            com.IbisError,
+            match=r".+Did you call `\.release\(\)` twice on the same expression\?",
+        ),
+        pytest.warns(DeprecationWarning),
     ):
         cached_table.release()