Files
searxng/searx/sqlitedb.py
T
Markus Heiser 26801e92af [fix] sqlitedb: create DB Schema (DDL) during app initialization (hardening) (#6187)
The initialization of the DB schema ("base schema") has so far been done on
demand, which causes race conditions with competing threads and processes.

The DDL statements for creating the "base schema" are now executed as part of
the initialization of the app.

Further improvements were made to harden the database applications:

- Wikidata & Radio-Browser engine perform their initialization only once (so far
  the initialization was carried out in each thread/process).

- If multiple processes try to set DB's WAL mode when opening the DB at the same
  time, this usually leads to another race condition, which is now also caught.

Related:

- https://github.com/searxng/searxng/issues/6181#issuecomment-4586705

Closes: #6181

Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
2026-06-10 15:48:49 +02:00

478 lines
18 KiB
Python

# SPDX-License-Identifier: AGPL-3.0-or-later
"""Implementations to make access to SQLite databases a little more convenient.
:py:obj:`SQLiteAppl`
Abstract class with which DB applications can be implemented.
:py:obj:`SQLiteProperties`:
Class to manage properties stored in a database.
Examplarical implementations based on :py:obj:`SQLiteAppl`:
:py:obj:`searx.cache.ExpireCacheSQLite` :
Cache that manages key/value pairs in a SQLite DB, in which the key/value
pairs are deleted after an "expire" time. This type of cache is used, for
example, for the engines, see :py:obj:`searx.enginelib.EngineCache`.
:py:obj:`searx.favicons.cache.FaviconCacheSQLite` :
Favicon cache that manages the favicon BLOBs in a SQLite DB.
----
"""
import typing as t
import abc
import datetime
import re
import sqlite3
import sys
import threading
import uuid
from searx import logger
logger = logger.getChild("sqlitedb")
THREAD_LOCAL = threading.local()
class DBSession:
"""A *thead-local* DB session"""
@classmethod
def get_connect(cls, app: "SQLiteAppl") -> sqlite3.Connection:
"""Returns a thread local DB connection. The connection is only
established once per thread.
"""
if getattr(THREAD_LOCAL, "DBSession_map", None) is None:
url_to_session: dict[str, DBSession] = {}
THREAD_LOCAL.DBSession_map = url_to_session
session: DBSession | None = THREAD_LOCAL.DBSession_map.get(app.db_url)
if session is None:
session = cls(app)
return session.conn
def __init__(self, app: "SQLiteAppl"):
self.uuid: uuid.UUID = uuid.uuid4()
self.app: SQLiteAppl = app
self._conn: sqlite3.Connection | None = None
# self.__del__ will be called, when thread ends
if getattr(THREAD_LOCAL, "DBSession_map", None) is None:
url_to_session: dict[str, DBSession] = {}
THREAD_LOCAL.DBSession_map = url_to_session
THREAD_LOCAL.DBSession_map[self.app.db_url] = self
@property
def conn(self) -> sqlite3.Connection:
msg = f"[{threading.current_thread().ident}] DBSession: " f"{self.app.__class__.__name__}({self.app.db_url})"
if self._conn is None:
self._conn = self.app.connect()
logger.debug("%s --> created new connection", msg)
# else:
# logger.debug("%s --> already connected", msg)
return self._conn
def __del__(self):
try:
if self._conn is not None:
# HINT: Don't use Python's logging facility in a destructor, it
# will produce error reports when python aborts the process or
# thread, because at this point objects that the logging module
# needs, do not exist anymore.
# msg = f"DBSession: close [{self.uuid}] {self.app.__class__.__name__}({self.app.db_url})"
# logger.debug(msg)
self._conn.close()
except Exception: # pylint: disable=broad-exception-caught
pass
class SQLiteAppl(abc.ABC):
"""Abstract base class for implementing convenient DB access in SQLite
applications. In the constructor, a :py:obj:`SQLiteProperties` instance is
already aggregated under ``self.properties``."""
DDL_CREATE_TABLES: dict[str, str] = {}
DB_SCHEMA: int = 1
"""As soon as changes are made to the DB schema, the version number must be
increased. Changes to the version number require the DB to be recreated (or
migrated / if an migration path exists and is implemented)."""
SQLITE_THREADING_MODE: str = {
0: "single-thread",
1: "multi-thread",
3: "serialized"}[sqlite3.threadsafety] # fmt:skip
"""Threading mode of the SQLite library. Depends on the options used at
compile time and is different for different distributions and architectures.
Possible values are 0:``single-thread``, 1:``multi-thread``,
3:``serialized`` (see :py:obj:`sqlite3.threadsafety`). Pre- Python 3.11
this value was hard coded to 1.
Depending on this value, optimizations are made, e.g. in “serialized” mode
it is not necessary to create a separate DB connector for each thread.
"""
SQLITE_JOURNAL_MODE: str = "WAL"
"""``SQLiteAppl`` applications are optimized for WAL_ mode, its not recommend
to change the journal mode (see :py:obj:`SQLiteAppl.tear_down`).
.. _WAL: https://sqlite.org/wal.html
"""
SQLITE_CONNECT_ARGS: dict[str, str | float | int | bool | None] = {
"timeout": 3.0, # default is 5sec
# "detect_types": 0,
"check_same_thread": bool(SQLITE_THREADING_MODE != "serialized"),
"cached_statements": 0, # https://github.com/python/cpython/issues/118172
# "uri": False,
# "isolation_level": "",
# "autocommit": sqlite3.LEGACY_TRANSACTION_CONTROL,
} # fmt:skip
"""Connection arguments (:py:obj:`sqlite3.connect`)
``check_same_thread``: *bool*
Is disabled by default when :py:obj:`SQLITE_THREADING_MODE` is
`serialized`. The check is more of a hindrance when threadsafety_ is
`serialized` because it would prevent a DB connector from being used in
multiple threads.
Is enabled when threadsafety_ is ``single-thread`` or ``multi-thread``
(when threads cannot share a connection PEP-0249_).
``cached_statements``:
Is set to ``0`` by default. Note: Python 3.12+ fetch result are not
consistent in multi-threading application and causing an API misuse error.
The multithreading use in SQLiteAppl is intended and supported if
threadsafety is set to 3 (aka "serialized"). CPython supports “serialized”
from version 3.12 on, but unfortunately only with errors:
- https://github.com/python/cpython/issues/118172
- https://github.com/python/cpython/issues/123873
The workaround for SQLite3 multithreading cache inconsistency is to set
option ``cached_statements`` to ``0`` by default.
``isolation_level``: *unset*
If the connection attribute isolation_level_ is **not** ``None``, new
transactions are implicitly opened before ``execute()`` and
``executemany()`` executes SQL- INSERT, UPDATE, DELETE, or REPLACE
statements `[1]`_.
By default, the value is not set, which means the default from Python is
used: Python's default is ``""``, which is an alias for ``"DEFERRED"``.
``autocommit``: *unset*
Starting with Python 3.12 the DB connection has a ``autocommit`` attribute
and the recommended way of controlling transaction behaviour is through
this attribute `[2]`_.
By default, the value is not set, which means the default from Python is
used: Python's default is the constant LEGACY_TRANSACTION_CONTROL_:
Pre-Python 3.12 (non-PEP 249-compliant) transaction control, see
``isolation_level`` above for more details.
.. _PEP-0249:
https://peps.python.org/pep-0249/#threadsafety
.. _threadsafety:
https://docs.python.org/3/library/sqlite3.html#sqlite3.threadsafety
.. _isolation_level:
https://docs.python.org/3/library/sqlite3.html#sqlite3.Connection.isolation_level
.. _[1]:
https://docs.python.org/3/library/sqlite3.html#sqlite3-transaction-control-isolation-level
.. _autocommit:
https://docs.python.org/3/library/sqlite3.html#sqlite3.Connection.autocommit
.. _[2]:
https://docs.python.org/3/library/sqlite3.html#transaction-control-via-the-autocommit-attribute
.. _LEGACY_TRANSACTION_CONTROL:
https://docs.python.org/3/library/sqlite3.html#sqlite3.LEGACY_TRANSACTION_CONTROL
"""
def __init__(self, db_url: str):
self.db_url: str = db_url
self.properties: SQLiteProperties = SQLiteProperties(db_url)
self._init_done: bool = False
self._DB: sqlite3.Connection | None = None
self._compatibility()
# atexit.register(self.tear_down)
# def tear_down(self):
# """:ref:`Vacuuming the WALs` upon normal interpreter termination
# (:py:obj:`atexit.register`).
# .. _SQLite: Vacuuming the WALs: https://www.theunterminatedstring.com/sqlite-vacuuming/
# """
# self.DB.execute("PRAGMA wal_checkpoint(TRUNCATE)")
def _compatibility(self):
if self.SQLITE_THREADING_MODE == "serialized":
self._DB = None
else:
msg = (
f"SQLite library is compiled with {self.SQLITE_THREADING_MODE} mode,"
" read https://docs.python.org/3/library/sqlite3.html#sqlite3.threadsafety"
)
if threading.active_count() > 1:
logger.error(msg)
else:
logger.warning(msg)
if sqlite3.sqlite_version_info <= (3, 35):
# See "Generalize UPSERT:" in https://sqlite.org/releaselog/3_35_0.html
logger.critical(
"SQLite runtime library version %s is not supported (require >= 3.35)", sqlite3.sqlite_version
)
def _connect(self) -> sqlite3.Connection:
conn = sqlite3.Connection(self.db_url, **self.SQLITE_CONNECT_ARGS) # type: ignore
try:
with conn:
conn.execute(f"PRAGMA journal_mode={self.SQLITE_JOURNAL_MODE}")
except sqlite3.OperationalError:
# when database is locked, the journal_mode is already set by
# different but concurrent process (no need to set it once more)
pass
self.register_functions(conn)
return conn
def connect(self) -> sqlite3.Connection:
"""Creates a new DB connection (:py:obj:`SQLITE_CONNECT_ARGS`). If not
already done, the DB schema is set up. The caller must take care of
closing the resource. Alternatively, :py:obj:`SQLiteAppl.DB` can also
be used (the resource behind `self.DB` is automatically closed when the
process or thread is terminated).
"""
if sys.version_info < (3, 12):
# Prior Python 3.12 there is no "autocommit" option
self.SQLITE_CONNECT_ARGS.pop("autocommit", None) # pyright: ignore[reportUnreachable]
msg = (
f"[{threading.current_thread().ident}] {self.__class__.__name__}({self.db_url})"
f" {self.SQLITE_CONNECT_ARGS} // {self.SQLITE_JOURNAL_MODE}"
)
logger.debug(msg)
with self._connect() as conn:
self.init(conn)
return conn
def register_functions(self, conn: sqlite3.Connection):
"""Create user-defined_ SQL functions.
``REGEXP(<pattern>, <field>)`` : 0 | 1
`re.search`_ returns (int) 1 for a match and 0 for none match of
``<pattern>`` in ``<field>``.
.. code:: sql
SELECT '12' AS field WHERE REGEXP('^[0-9][0-9]$', field)
-- 12
SELECT REGEXP('[0-9][0-9]', 'X12Y')
-- 1
SELECT REGEXP('[0-9][0-9]', 'X1Y')
-- 0
.. _user-defined: https://docs.python.org/3/library/sqlite3.html#sqlite3.Connection.create_function
.. _deterministic: https://sqlite.org/deterministic.html
.. _re.search: https://docs.python.org/3/library/re.html#re.search
"""
conn.create_function("regexp", 2, lambda x, y: 1 if re.search(x, y) else 0, deterministic=True) # type: ignore
@property
def DB(self) -> sqlite3.Connection:
"""Provides a DB connection. The connection is a *singleton* and
therefore well suited for read access. If
:py:obj:`SQLITE_THREADING_MODE` is ``serialized`` only one DB connection
is created for all threads.
.. note::
For dedicated `transaction control`_, it is recommended to create a
new connection (:py:obj:`SQLiteAppl.connect`).
.. _transaction control:
https://docs.python.org/3/library/sqlite3.html#sqlite3-controlling-transactions
"""
conn: sqlite3.Connection
if self.SQLITE_THREADING_MODE == "serialized":
# Theoretically it is possible to reuse the DB cursor across threads
# as of Python 3.12, in practice the threading of the cursor seems
# to me a little faulty that I prefer to establish one connection
# per thread.
#
# may we can activate this code one day ..
# if self._DB is None:
# self._DB = self.connect()
# conn = self._DB
conn = DBSession.get_connect(self)
else:
conn = DBSession.get_connect(self)
# Since more than one instance of SQLiteAppl share the same DB
# connection, we need to make sure that each SQLiteAppl instance has run
# its init method at least once.
with conn:
self.init(conn)
return conn
def init(self, conn: sqlite3.Connection) -> bool:
"""Initializes the DB schema and properties, is only executed once even
if called several times.
If the initialization has not yet taken place, it is carried out and a
`True` is returned to the caller at the end. If the initialization has
already been carried out in the past, `False` is returned.
"""
if self._init_done:
return False
self._init_done = True
logger.debug("init DB: %s", self.db_url)
with conn:
self.properties.init(conn)
ver = self.properties("DB_SCHEMA")
if ver is None:
with conn:
self.create_schema(conn)
else:
ver = int(ver)
if ver != self.DB_SCHEMA:
raise sqlite3.DatabaseError("Expected DB schema v%s, DB schema is v%s" % (self.DB_SCHEMA, ver))
logger.debug("DB_SCHEMA = %s", ver)
return True
def create_schema(self, conn: sqlite3.Connection):
logger.debug("create schema ..")
self.properties.set("DB_SCHEMA", self.DB_SCHEMA)
self.properties.set("LAST_MAINTENANCE", "")
with conn:
for table_name, sql in self.DDL_CREATE_TABLES.items():
conn.execute(sql)
self.properties.set(f"Table {table_name} created", table_name)
class SQLiteProperties(SQLiteAppl):
"""Simple class to manage properties of a DB application in the DB. The
object has its own DB connection and transaction area.
.. code:: sql
CREATE TABLE IF NOT EXISTS properties (
name TEXT,
value TEXT,
m_time INTEGER DEFAULT (strftime('%s', 'now')),
PRIMARY KEY (name))
"""
SQLITE_JOURNAL_MODE: str = "WAL"
DDL_PROPERTIES: str = """\
CREATE TABLE IF NOT EXISTS properties (
name TEXT,
value TEXT,
m_time INTEGER DEFAULT (strftime('%s', 'now')), -- last modified (unix epoch) time in sec.
PRIMARY KEY (name))"""
"""Table to store properties of the DB application"""
SQL_GET: str = "SELECT value FROM properties WHERE name = ?"
SQL_M_TIME: str = "SELECT m_time FROM properties WHERE name = ?"
SQL_SET: str = (
"INSERT INTO properties (name, value) VALUES (?, ?)"
" ON CONFLICT(name) DO UPDATE"
" SET value=excluded.value, m_time=strftime('%s', 'now')"
)
SQL_DELETE: str = "DELETE FROM properties WHERE name = ?"
SQL_TABLE_EXISTS: str = (
"SELECT name FROM sqlite_master"
" WHERE type='table' AND name='properties'"
) # fmt:skip
SQLITE_CONNECT_ARGS: dict[str, str | int | bool | None] = dict(SQLiteAppl.SQLITE_CONNECT_ARGS)
# pylint: disable=super-init-not-called
def __init__(self, db_url: str): # pyright: ignore[reportMissingSuperCall]
self.db_url: str = db_url
self._init_done: bool = False
self._compatibility()
def init(self, conn: sqlite3.Connection) -> bool:
"""Initializes DB schema of the properties in the DB."""
if self._init_done:
return False
self._init_done = True
logger.debug("init properties of DB: %s", self.db_url)
res = conn.execute(self.SQL_TABLE_EXISTS)
if res.fetchone() is None: # DB schema needs to be created
self.create_schema(conn)
return True
def __call__(self, name: str, default: t.Any = None) -> t.Any:
"""Returns the value of the property ``name`` or ``default`` if property
not exists in DB."""
res = self.DB.execute(self.SQL_GET, (name,)).fetchone()
if res is None:
return default
return res[0]
def set(self, name: str, value: str | int):
"""Set ``value`` of property ``name`` in DB. If property already
exists, update the ``m_time`` (and the value)."""
with self.DB:
self.DB.execute(self.SQL_SET, (name, value))
def delete(self, name: str) -> int:
"""Delete of property ``name`` from DB."""
with self.DB:
cur = self.DB.execute(self.SQL_DELETE, (name,))
return cur.rowcount
def row(self, name: str, default: t.Any = None):
"""Returns the DB row of property ``name`` or ``default`` if property
not exists in DB."""
res = self.DB.execute("SELECT * FROM properties WHERE name = ?", (name,))
row = res.fetchone()
if row is None:
return default
col_names = [column[0] for column in row.description]
return dict(zip(col_names, row))
def m_time(self, name: str, default: int = 0) -> int:
"""Last modification time of this property."""
res = self.DB.execute(self.SQL_M_TIME, (name,))
row = res.fetchone()
if row is None:
return default
return int(row[0])
def create_schema(self, conn: sqlite3.Connection):
with conn:
conn.execute(self.DDL_PROPERTIES)
def __str__(self) -> str:
lines: list[str] = []
for row in self.DB.execute("SELECT name, value, m_time FROM properties"):
name, value, m_time = row
m_time = datetime.datetime.fromtimestamp(m_time).strftime("%Y-%m-%d %H:%M:%S")
lines.append(f"[last modified: {m_time}] {name:20s}: {value}")
return "\n".join(lines)