[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>

1 files changed, 16 insertions, 7 deletions

diff --git a/searx/sqlitedb.py b/searx/sqlitedb.py
index a4f498339..da8147166 100644
--- a/searx/sqlitedb.py
+++ b/searx/sqlitedb.py
@@ -121,8 +121,8 @@ class SQLiteAppl(abc.ABC):
 
     .. _WAL: https://sqlite.org/wal.html
     """
-    SQLITE_CONNECT_ARGS: dict[str,str|int|bool|None] = {
-        # "timeout": 5.0,
+    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
@@ -195,6 +195,7 @@ class SQLiteAppl(abc.ABC):
         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)
 
@@ -209,7 +210,7 @@ class SQLiteAppl(abc.ABC):
     def _compatibility(self):
 
         if self.SQLITE_THREADING_MODE == "serialized":
-            self._DB: sqlite3.Connection | None = None
+            self._DB = None
         else:
             msg = (
                 f"SQLite library is compiled with {self.SQLITE_THREADING_MODE} mode,"
@@ -228,7 +229,13 @@ class SQLiteAppl(abc.ABC):
 
     def _connect(self) -> sqlite3.Connection:
         conn = sqlite3.Connection(self.db_url, **self.SQLITE_CONNECT_ARGS)  # type: ignore
-        conn.execute(f"PRAGMA journal_mode={self.SQLITE_JOURNAL_MODE}")
+        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
 
@@ -312,7 +319,8 @@ class SQLiteAppl(abc.ABC):
         # 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.
-        self.init(conn)
+        with conn:
+            self.init(conn)
 
         return conn
 
@@ -330,7 +338,8 @@ class SQLiteAppl(abc.ABC):
         self._init_done = True
 
         logger.debug("init DB: %s", self.db_url)
-        self.properties.init(conn)
+        with conn:
+            self.properties.init(conn)
 
         ver = self.properties("DB_SCHEMA")
         if ver is None:
@@ -409,7 +418,7 @@ CREATE TABLE IF NOT EXISTS properties (
         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 be created
+        if res.fetchone() is None:  # DB schema needs to be created
             self.create_schema(conn)
         return True