Add connection identity to the rollback-failure DEBUG log

The AsyncAdaptedConnection.close rollback-failure DEBUG log named
the exception class but not the connection. In a pool of 20 async
connections churning several pool events per second, that produces
a stream of identical lines with no way to tell which adapter or
which peer is misbehaving.

Include id(self) and the peer address from the wrapped dbapi
connection so operators can correlate log records to specific
adapter instances and cluster nodes. id(self) is chosen over
id(self._connection) because SQLAlchemy's pool tracks the outer
adapter — that's the natural correlation handle. Peer address is
looked up defensively via getattr(..., "address", None) so the
log never fails on an adapter constructed in a test harness.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: antoineleclair

src/sqlalchemydqlite/aio.py

@@ -243,9 +243,15 @@ def close(self) -> None:
         ) as exc:
             # Silent suppression used to hide e.g. "leader flip
             # mid-rollback" from operators — a DEBUG line preserves the
-            # diagnostic without masking or propagating.
+            # diagnostic without masking or propagating. Include both
+            # id(self) and the peer address so a noisy pool can be
+            # correlated to specific adapter instances and nodes.
+            peer = getattr(self._connection, "address", None)
             logger.debug(
-                "AsyncAdaptedConnection.close: rollback failed (%s); proceeding to close",
+                "AsyncAdaptedConnection.close (id=%s, peer=%s): "
+                "rollback failed (%s); proceeding to close",
+                id(self),
+                peer,
                 type(exc).__name__,
                 exc_info=True,
             )

tests/test_aio_close_rollback_log.py

@@ -67,6 +67,11 @@ def _fake_await_only(coro: object) -> object:
     assert matching[0].exc_info is not None
     assert isinstance(matching[0].exc_info[1], OperationalError)
     assert fake.close_calls == 1  # close still ran
+    # Correlation fields: id=<num> and peer=<addr-or-None> so operators
+    # can attribute the log record to a specific adapter instance / node.
+    msg = matching[0].getMessage()
+    assert f"id={id(adapter)}" in msg, msg
+    assert "peer=" in msg, msg
 
 
 def test_close_propagates_programming_bug() -> None: