Add a background job to automatically delete stale devices (#12855)

Co-authored-by: Patrick Cloke <clokep@users.noreply.github.com>

3 files changed, 79 insertions, 1 deletions

diff --git a/synapse/config/server.py b/synapse/config/server.py
index f73d5e1f66..657322cb1f 100644
--- a/synapse/config/server.py
+++ b/synapse/config/server.py
@@ -679,6 +679,17 @@ class ServerConfig(Config):
             config.get("exclude_rooms_from_sync") or []
         )
 
+        delete_stale_devices_after: Optional[str] = (
+            config.get("delete_stale_devices_after") or None
+        )
+
+        if delete_stale_devices_after is not None:
+            self.delete_stale_devices_after: Optional[int] = self.parse_duration(
+                delete_stale_devices_after
+            )
+        else:
+            self.delete_stale_devices_after = None
+
     def has_tls_listener(self) -> bool:
         return any(listener.tls for listener in self.listeners)
 
diff --git a/synapse/handlers/device.py b/synapse/handlers/device.py
index 438a549339..2a56473dc6 100644
--- a/synapse/handlers/device.py
+++ b/synapse/handlers/device.py
@@ -61,6 +61,7 @@ if TYPE_CHECKING:
 logger = logging.getLogger(__name__)
 
 MAX_DEVICE_DISPLAY_NAME_LEN = 100
+DELETE_STALE_DEVICES_INTERVAL_MS = 24 * 60 * 60 * 1000
 
 
 class DeviceWorkerHandler:
@@ -295,6 +296,19 @@ class DeviceHandler(DeviceWorkerHandler):
         # On start up check if there are any updates pending.
         hs.get_reactor().callWhenRunning(self._handle_new_device_update_async)
 
+        self._delete_stale_devices_after = hs.config.server.delete_stale_devices_after
+
+        # Ideally we would run this on a worker and condition this on the
+        # "run_background_tasks_on" setting, but this would mean making the notification
+        # of device list changes over federation work on workers, which is nontrivial.
+        if self._delete_stale_devices_after is not None:
+            self.clock.looping_call(
+                run_as_background_process,
+                DELETE_STALE_DEVICES_INTERVAL_MS,
+                "delete_stale_devices",
+                self._delete_stale_devices,
+            )
+
     def _check_device_name_length(self, name: Optional[str]) -> None:
         """
         Checks whether a device name is longer than the maximum allowed length.
@@ -370,6 +384,19 @@ class DeviceHandler(DeviceWorkerHandler):
 
         raise errors.StoreError(500, "Couldn't generate a device ID.")
 
+    async def _delete_stale_devices(self) -> None:
+        """Background task that deletes devices which haven't been accessed for more than
+        a configured time period.
+        """
+        # We should only be running this job if the config option is defined.
+        assert self._delete_stale_devices_after is not None
+        now_ms = self.clock.time_msec()
+        since_ms = now_ms - self._delete_stale_devices_after
+        devices = await self.store.get_local_devices_not_accessed_since(since_ms)
+
+        for user_id, user_devices in devices.items():
+            await self.delete_devices(user_id, user_devices)
+
     @trace
     async def delete_device(self, user_id: str, device_id: str) -> None:
         """Delete the given device
@@ -692,7 +719,8 @@ class DeviceHandler(DeviceWorkerHandler):
                             )
                             # TODO: when called, this isn't in a logging context.
                             # This leads to log spam, sentry event spam, and massive
-                            # memory usage. See #12552.
+                            # memory usage.
+                            # See https://github.com/matrix-org/synapse/issues/12552.
                             # log_kv(
                             #     {"message": "sent device update to host", "host": host}
                             # )
diff --git a/synapse/storage/databases/main/devices.py b/synapse/storage/databases/main/devices.py
index dd43bae784..d900064c07 100644
--- a/synapse/storage/databases/main/devices.py
+++ b/synapse/storage/databases/main/devices.py
@@ -1154,6 +1154,45 @@ class DeviceWorkerStore(SQLBaseStore):
             _prune_txn,
         )
 
+    async def get_local_devices_not_accessed_since(
+        self, since_ms: int
+    ) -> Dict[str, List[str]]:
+        """Retrieves local devices that haven't been accessed since a given date.
+
+        Args:
+            since_ms: the timestamp to select on, every device with a last access date
+                from before that time is returned.
+
+        Returns:
+            A dictionary with an entry for each user with at least one device matching
+            the request, which value is a list of the device ID(s) for the corresponding
+            device(s).
+        """
+
+        def get_devices_not_accessed_since_txn(
+            txn: LoggingTransaction,
+        ) -> List[Dict[str, str]]:
+            sql = """
+                SELECT user_id, device_id
+                FROM devices WHERE last_seen < ? AND hidden = FALSE
+            """
+            txn.execute(sql, (since_ms,))
+            return self.db_pool.cursor_to_dict(txn)
+
+        rows = await self.db_pool.runInteraction(
+            "get_devices_not_accessed_since",
+            get_devices_not_accessed_since_txn,
+        )
+
+        devices: Dict[str, List[str]] = {}
+        for row in rows:
+            # Remote devices are never stale from our point of view.
+            if self.hs.is_mine_id(row["user_id"]):
+                user_devices = devices.setdefault(row["user_id"], [])
+                user_devices.append(row["device_id"])
+
+        return devices
+
 
 class DeviceBackgroundUpdateStore(SQLBaseStore):
     def __init__(