fix(sqlite): defer db connection until needed

ibis/backends/base/__init__.py

@@ -136,17 +136,64 @@ class BaseBackend(abc.ABC):
     database_class = Database
     table_class: type[ops.DatabaseTable] = ops.DatabaseTable
 
+    def __init__(self):
+        self._con_args: tuple[Any] = tuple()
+        self._con_kwargs: dict[str, Any] = dict()
+
+    def __getstate__(self):
+        return dict(
+            database_class=self.database_class,
+            table_class=self.table_class,
+            _con_args=self._con_args,
+            _con_kwargs=self._con_kwargs,
+        )
+
+    def __hash__(self):
+        return hash(self.datasource)
+
+    def __eq__(self, other):
+        return self.datasource == other.datasource
+
     @property
     @abc.abstractmethod
     def name(self) -> str:
         """
         Name of the backend, for example 'sqlite'.
         """
 
+    @property
+    def datasource(self) -> str:
+        """
+        Identity of the database.  Multiple connections to the same
+        database will have the same datasource.  Default implementation
+        assumes connection parameters uniquely specify the database.
+        """
+        return '_'.join(
+            str(x)
+            for x in (self.table_class, self._con_args, self._con_kwargs)
+        )
+
+    def connect(self, *args, **kwargs):
+        """
+        Return new client object with saved args/kwargs, having called
+        .reconnect() on it.
+        """
+        new_backend = self.__class__()
+        new_backend._con_args = args
+        new_backend._con_kwargs = kwargs
+        new_backend.reconnect()
+        return new_backend
+
+    def reconnect(self):
+        """
+        Reconnect to the target database already configured with connect().
+        """
+        self.do_connect(*self._con_args, **self._con_kwargs)
+
     @abc.abstractmethod
-    def connect(connection_string, **options):
+    def do_connect(self, *args, **kwargs):
         """
-        Connect to the underlying database and return a client object.
+        Connect to database specified by args and kwargs.
         """
 
     def database(self, name: str = None) -> Database:

ibis/backends/base/file/__init__.py

@@ -56,7 +56,7 @@ class BaseFileBackend(BaseBackend):
 
     database_class = FileDatabase
 
-    def connect(self, path):
+    def do_connect(self, path):
         """Create a Client for use with Ibis
 
         Parameters
@@ -67,10 +67,8 @@ def connect(self, path):
         -------
         Backend
         """
-        new_backend = self.__class__()
-        new_backend.path = new_backend.root = Path(path)
-        new_backend.dictionary = {}
-        return new_backend
+        self.path = self.root = Path(path)
+        self.dictionary = {}
 
     @property
     def version(self) -> str:

ibis/backends/base/sql/alchemy/__init__.py

@@ -101,13 +101,11 @@ def _build_alchemy_url(
             database=database,
         )
 
-    def connect(self, con: sqlalchemy.engine.Engine):
-        new_backend = self.__class__()
-        new_backend.con = con
-        new_backend.meta = sqlalchemy.MetaData(bind=con)
-        new_backend._inspector = sqlalchemy.inspect(con)
-        new_backend._schemas: Dict[str, sch.Schema] = {}
-        return new_backend
+    def do_connect(self, con):
+        self.con = con
+        self._inspector = sqlalchemy.inspect(self.con)
+        self.meta = sqlalchemy.MetaData(bind=self.con)
+        self._schemas: Dict[str, sch.Schema] = {}
 
     @property
     def version(self):

ibis/backends/base/sql/alchemy/database.py

@@ -21,3 +21,8 @@ def __init__(self, table, source, schema=None):
         schema = sch.infer(table, schema=schema)
         super().__init__(table.name, schema, source)
         self.sqla_table = table
+
+    def __getstate__(self):
+        d = super().__getstate__()
+        d['sqla_table'] = self.sqla_table
+        return d

ibis/backends/clickhouse/__init__.py

@@ -29,7 +29,7 @@ class Backend(BaseSQLBackend):
     table_expr_class = ClickhouseTable
     compiler = ClickhouseCompiler
 
-    def connect(
+    def do_connect(
         self,
         host='localhost',
         port=9000,
@@ -80,8 +80,7 @@ def connect(
         -------
         ClickhouseClient
         """
-        new_backend = self.__class__()
-        new_backend.con = _DriverClient(
+        self.con = _DriverClient(
             host=host,
             port=port,
             database=database,
@@ -90,7 +89,6 @@ def connect(
             client_name=client_name,
             compression=compression,
         )
-        return new_backend
 
     def register_options(self):
         ibis.config.register_option(

ibis/backends/dask/__init__.py

@@ -27,11 +27,11 @@ class Backend(BasePandasBackend):
     database_class = DaskDatabase
     table_class = DaskTable
 
-    def connect(self, dictionary):
+    def do_connect(self, dictionary):
         # register dispatchers
         from . import udf  # noqa: F401
 
-        return super().connect(dictionary)
+        super().do_connect(dictionary)
 
     @property
     def version(self):

ibis/backends/impala/__init__.py

@@ -170,8 +170,8 @@ class Backend(BaseSQLBackend):
     def hdfs_connect(self, *args, **kwargs):
         return hdfs_connect(*args, **kwargs)
 
-    def connect(
-        self,
+    def do_connect(
+        new_backend,
         host='localhost',
         port=21050,
         database='default',
@@ -240,7 +240,6 @@ def connect(
         """
         import hdfs
 
-        new_backend = self.__class__()
         new_backend._kudu = None
         new_backend._temp_objects = set()
 
@@ -267,8 +266,6 @@ def connect(
 
         new_backend._ensure_temp_db_exists()
 
-        return new_backend
-
     @property
     def version(self):
         cursor = self.raw_sql('select version()')

ibis/backends/mysql/__init__.py

@@ -14,7 +14,7 @@ class Backend(BaseAlchemyBackend):
     name = 'mysql'
     compiler = MySQLCompiler
 
-    def connect(
+    def do_connect(
         self,
         host='localhost',
         user=None,
@@ -96,10 +96,8 @@ def connect(
             driver=f'mysql+{driver}',
         )
 
-        new_backend = super().connect(sqlalchemy.create_engine(alchemy_url))
-        new_backend.database_name = alchemy_url.database
-
-        return new_backend
+        self.database_name = alchemy_url.database
+        super().do_connect(sqlalchemy.create_engine(alchemy_url))
 
     @contextlib.contextmanager
     def begin(self):

ibis/backends/pandas/__init__.py

@@ -15,7 +15,7 @@ class BasePandasBackend(BaseBackend):
     Base class for backends based on pandas.
     """
 
-    def connect(self, dictionary):
+    def do_connect(self, dictionary):
         """Construct a client from a dictionary of DataFrames.
 
         Parameters
@@ -30,9 +30,7 @@ def connect(self, dictionary):
         from . import execution  # noqa F401
         from . import udf  # noqa F401
 
-        new_backend = self.__class__()
-        new_backend.dictionary = dictionary
-        return new_backend
+        self.dictionary = dictionary
 
     def from_dataframe(self, df, name='df', client=None):
         """

ibis/backends/postgres/__init__.py

@@ -13,7 +13,7 @@ class Backend(BaseAlchemyBackend):
     name = 'postgres'
     compiler = PostgreSQLCompiler
 
-    def connect(
+    def do_connect(
         self,
         host='localhost',
         user=None,
@@ -93,9 +93,8 @@ def connect(
             database=database,
             driver=f'postgresql+{driver}',
         )
-        new_backend = super().connect(sqlalchemy.create_engine(alchemy_url))
-        new_backend.database_name = alchemy_url.database
-        return new_backend
+        self.database_name = alchemy_url.database
+        super().do_connect(sqlalchemy.create_engine(alchemy_url))
 
     def list_databases(self, like=None):
         # http://dba.stackexchange.com/a/1304/58517

ibis/backends/pyspark/__init__.py

@@ -81,26 +81,23 @@ class Backend(BaseSQLBackend):
     table_class = PySparkDatabaseTable
     table_expr_class = PySparkTable
 
-    def connect(self, session):
+    def do_connect(self, session):
         """
         Create a pyspark `Backend` for use with Ibis.
 
         Pipes `**kwargs` into Backend, which pipes them into SparkContext.
         See documentation for SparkContext:
         https://spark.apache.org/docs/latest/api/python/_modules/pyspark/context.html#SparkContext
         """
-        new_backend = self.__class__()
-        new_backend._context = session.sparkContext
-        new_backend._session = session
-        new_backend._catalog = session.catalog
+        self._context = session.sparkContext
+        self._session = session
+        self._catalog = session.catalog
 
         # Spark internally stores timestamps as UTC values, and timestamp data
         # that is brought in without a specified time zone is converted as
         # local time to UTC with microsecond resolution.
         # https://spark.apache.org/docs/latest/sql-pyspark-pandas-with-arrow.html#timestamp-with-time-zone-semantics
-        new_backend._session.conf.set('spark.sql.session.timeZone', 'UTC')
-
-        return new_backend
+        self._session.conf.set('spark.sql.session.timeZone', 'UTC')
 
     @property
     def version(self):

ibis/backends/sqlite/__init__.py

@@ -31,8 +31,42 @@ class Backend(BaseAlchemyBackend):
     database_class = Database
     compiler = SQLiteCompiler
 
-    def connect(self, path=None, create=False):
+    def __init__(self):
+        super().__init__()
+        self._con: sqlalchemy.engine.Engine = None
+        self._meta: sqlalchemy.MetaData = None
+
+    def __getstate__(self) -> dict:
+        r = super().__getstate__()
+        r.update(
+            dict(
+                compiler=self.compiler,
+                database_name=self.database_name,
+            )
+        )
+        return r
+
+    @property
+    def con(self) -> sqlalchemy.engine.Engine:
+        if self._con is None:
+            self.reconnect()
+        return self._con
+
+    @con.setter
+    def con(self, v: Optional[sqlalchemy.engine.Engine]):
+        self._con = v
+
+    @property
+    def meta(self) -> sqlalchemy.MetaData:
+        if self._meta is None:
+            self.reconnect()
+        return self._meta
+
+    @meta.setter
+    def meta(self, v: sqlalchemy.MetaData):
+        self._meta = v
 
+    def do_connect(self, path=None, create=False):
         """
         Create an Ibis client connected to a SQLite database.
 
@@ -47,15 +81,15 @@ def connect(self, path=None, create=False):
         create : boolean, default False
             If file does not exist, create it
         """
-        new_backend = super().connect(sqlalchemy.create_engine("sqlite://"))
-        new_backend.database_name = "base"
+        self.database_name = "base"
 
+        super().do_connect(sqlalchemy.create_engine("sqlite://"))
         if path is not None:
-            new_backend.attach(new_backend.database_name, path, create=create)
+            self.attach(self.database_name, path, create=create)
 
-        udf.register_all(new_backend.con)
+        udf.register_all(self.con)
 
-        return new_backend
+        self._meta = sqlalchemy.MetaData(bind=self.con)
 
     def list_tables(self, like=None, database=None):
         if database is None:

ibis/backends/sqlite/tests/test_client.py

@@ -57,7 +57,7 @@ def test_list_tables(con):
 
 
 def test_attach_file(dbpath):
-    client = Backend().connect()
+    client = Backend().connect(None)
 
     client.attach('foo', dbpath)
     client.attach('bar', dbpath)

ibis/expr/types.py

@@ -2,7 +2,7 @@
 import os
 import warnings
 import webbrowser
-from typing import TYPE_CHECKING, Optional
+from typing import TYPE_CHECKING, Dict, Optional
 
 import numpy as np
 
@@ -205,7 +205,7 @@ def _find_backends(self):
         """
         from ibis.backends.base import BaseBackend
 
-        seen_backends = set()
+        seen_backends: Dict[str, BaseBackend] = {}  # key is backend.datasource
 
         stack = [self.op()]
         seen = set()
@@ -218,11 +218,12 @@ def _find_backends(self):
 
                 for arg in node.flat_args():
                     if isinstance(arg, BaseBackend):
-                        seen_backends.add(arg)
+                        if arg.datasource not in seen_backends:
+                            seen_backends[arg.datasource] = arg
                     elif isinstance(arg, ir.Expr):
                         stack.append(arg.op())
 
-        return list(seen_backends)
+        return list(seen_backends.values())
 
     def _find_backend(self):
         backends = self._find_backends()

ibis/tests/expr/mocks.py

@@ -355,9 +355,10 @@ class MockBackend(BaseSQLBackend):
     current_database = 'mockdb'
 
     def __init__(self):
+        super().__init__()
         self.executed_queries = []
 
-    def connect(self):
+    def do_connect(self):
         pass
 
     def list_tables(self):