Sliding sync: Store the per-connection state in the database.

changelog.d/17599.misc

@@ -0,0 +1 @@
+Store sliding sync per-connection state in the database.

changelog.d/17600.misc

@@ -0,0 +1 @@
+Make the sliding sync `PerConnectionState` class immutable.

scripts-dev/mypy_synapse_plugin.py

@@ -38,6 +38,7 @@
     NoneType,
     TupleType,
     TypeAliasType,
+    TypeVarType,
     UninhabitedType,
     UnionType,
 )
@@ -233,6 +234,7 @@ def check_is_cacheable(
     "synapse.synapse_rust.push.FilteredPushRules",
     # This is technically not immutable, but close enough.
     "signedjson.types.VerifyKey",
+    "synapse.types.StrCollection",
 }
 
 # Immutable containers only if the values are also immutable.
@@ -298,7 +300,7 @@ def is_cacheable(
 
         elif rt.type.fullname in MUTABLE_CONTAINER_TYPES:
             # Mutable containers are mutable regardless of their underlying type.
-            return False, None
+            return False, f"container {rt.type.fullname} is mutable"
 
         elif "attrs" in rt.type.metadata:
             # attrs classes are only cachable iff it is frozen (immutable itself)
@@ -318,6 +320,9 @@ def is_cacheable(
             else:
                 return False, "non-frozen attrs class"
 
+        elif rt.type.is_enum:
+            # We assume Enum values are immutable
+            return True, None
         else:
             # Ensure we fail for unknown types, these generally means that the
             # above code is not complete.
@@ -326,6 +331,18 @@ def is_cacheable(
                 f"Don't know how to handle {rt.type.fullname} return type instance",
             )
 
+    elif isinstance(rt, TypeVarType):
+        # We consider TypeVars immutable if they are bound to a set of immutable
+        # types.
+        if rt.values:
+            for value in rt.values:
+                ok, note = is_cacheable(value, signature, verbose)
+                if not ok:
+                    return False, f"TypeVar bound not cacheable {value}"
+            return True, None
+
+        return False, "TypeVar is unbound"
+
     elif isinstance(rt, NoneType):
         # None is cachable.
         return True, None

synapse/app/generic_worker.py

@@ -98,6 +98,7 @@
 from synapse.storage.databases.main.search import SearchStore
 from synapse.storage.databases.main.session import SessionStore
 from synapse.storage.databases.main.signatures import SignatureWorkerStore
+from synapse.storage.databases.main.sliding_sync import SlidingSyncStore
 from synapse.storage.databases.main.state import StateGroupWorkerStore
 from synapse.storage.databases.main.stats import StatsStore
 from synapse.storage.databases.main.stream import StreamWorkerStore
@@ -159,6 +160,7 @@ class GenericWorkerStore(
     SessionStore,
     TaskSchedulerWorkerStore,
     ExperimentalFeaturesStore,
+    SlidingSyncStore,
 ):
     # Properties that multiple storage classes define. Tell mypy what the
     # expected type is.

synapse/handlers/sliding_sync/__init__.py

@@ -45,13 +45,6 @@
 from synapse.handlers.relations import BundledAggregations
 from synapse.handlers.sliding_sync.extensions import SlidingSyncExtensionHandler
 from synapse.handlers.sliding_sync.store import SlidingSyncConnectionStore
-from synapse.handlers.sliding_sync.types import (
-    HaveSentRoomFlag,
-    MutablePerConnectionState,
-    PerConnectionState,
-    RoomSyncConfig,
-    StateValues,
-)
 from synapse.logging.opentracing import (
     SynapseTags,
     log_kv,
@@ -83,7 +76,16 @@
     StreamToken,
     UserID,
 )
-from synapse.types.handlers import OperationType, SlidingSyncConfig, SlidingSyncResult
+from synapse.types.handlers.sliding_sync import (
+    HaveSentRoomFlag,
+    MutablePerConnectionState,
+    OperationType,
+    PerConnectionState,
+    RoomSyncConfig,
+    SlidingSyncConfig,
+    SlidingSyncResult,
+    StateValues,
+)
 from synapse.types.state import StateFilter
 from synapse.util.async_helpers import concurrently_execute
 from synapse.visibility import filter_events_for_client
@@ -206,7 +208,7 @@ def __init__(self, hs: "HomeServer"):
         self.rooms_to_exclude_globally = hs.config.server.rooms_to_exclude_from_sync
         self.is_mine_id = hs.is_mine_id
 
-        self.connection_store = SlidingSyncConnectionStore()
+        self.connection_store = SlidingSyncConnectionStore(self.store)
         self.extensions = SlidingSyncExtensionHandler(hs)
 
     async def wait_for_sync_for_user(
@@ -331,11 +333,6 @@ async def current_sync_for_user(
             )
         )
 
-        await self.connection_store.mark_token_seen(
-            sync_config=sync_config,
-            from_token=from_token,
-        )
-
         # Get all of the room IDs that the user should be able to see in the sync
         # response
         has_lists = sync_config.lists is not None and len(sync_config.lists) > 0
@@ -430,15 +427,11 @@ async def current_sync_for_user(
                                     room_id
                                 )
                                 if existing_room_sync_config is not None:
-                                    existing_room_sync_config.combine_room_sync_config(
+                                    room_sync_config = existing_room_sync_config.combine_room_sync_config(
                                         room_sync_config
                                     )
-                                else:
-                                    # Make a copy so if we modify it later, it doesn't
-                                    # affect all references.
-                                    relevant_room_map[room_id] = (
-                                        room_sync_config.deep_copy()
-                                    )
+
+                                relevant_room_map[room_id] = room_sync_config
 
                                 room_ids_in_list.append(room_id)
 
@@ -503,11 +496,13 @@ async def current_sync_for_user(
                     # and need to fetch more info about.
                     existing_room_sync_config = relevant_room_map.get(room_id)
                     if existing_room_sync_config is not None:
-                        existing_room_sync_config.combine_room_sync_config(
-                            room_sync_config
+                        room_sync_config = (
+                            existing_room_sync_config.combine_room_sync_config(
+                                room_sync_config
+                            )
                         )
-                    else:
-                        relevant_room_map[room_id] = room_sync_config
+
+                    relevant_room_map[room_id] = room_sync_config
 
         # Fetch room data
         rooms: Dict[str, SlidingSyncResult.RoomResult] = {}

synapse/handlers/sliding_sync/extensions.py

@@ -19,11 +19,6 @@
 
 from synapse.api.constants import AccountDataTypes
 from synapse.handlers.receipts import ReceiptEventSource
-from synapse.handlers.sliding_sync.types import (
-    HaveSentRoomFlag,
-    MutablePerConnectionState,
-    PerConnectionState,
-)
 from synapse.logging.opentracing import trace
 from synapse.types import (
     DeviceListUpdates,
@@ -32,7 +27,14 @@
     SlidingSyncStreamToken,
     StreamToken,
 )
-from synapse.types.handlers import OperationType, SlidingSyncConfig, SlidingSyncResult
+from synapse.types.handlers.sliding_sync import (
+    HaveSentRoomFlag,
+    MutablePerConnectionState,
+    OperationType,
+    PerConnectionState,
+    SlidingSyncConfig,
+    SlidingSyncResult,
+)
 
 if TYPE_CHECKING:
     from synapse.server import HomeServer

synapse/handlers/sliding_sync/store.py

@@ -13,18 +13,18 @@
 #
 
 import logging
-from typing import TYPE_CHECKING, Dict, Optional, Tuple
+from typing import TYPE_CHECKING, Optional
 
 import attr
 
-from synapse.api.errors import SlidingSyncUnknownPosition
-from synapse.handlers.sliding_sync.types import (
+from synapse.logging.opentracing import trace
+from synapse.storage.databases.main import DataStore
+from synapse.types import SlidingSyncStreamToken
+from synapse.types.handlers.sliding_sync import (
     MutablePerConnectionState,
     PerConnectionState,
+    SlidingSyncConfig,
 )
-from synapse.logging.opentracing import trace
-from synapse.types import SlidingSyncStreamToken
-from synapse.types.handlers import SlidingSyncConfig
 
 if TYPE_CHECKING:
     pass
@@ -61,20 +61,7 @@ class SlidingSyncConnectionStore:
             to mapping of room ID to `HaveSentRoom`.
     """
 
-    # `(user_id, conn_id)` -> `connection_position` -> `PerConnectionState`
-    _connections: Dict[Tuple[str, str], Dict[int, PerConnectionState]] = attr.Factory(
-        dict
-    )
-
-    async def is_valid_token(
-        self, sync_config: SlidingSyncConfig, connection_token: int
-    ) -> bool:
-        """Return whether the connection token is valid/recognized"""
-        if connection_token == 0:
-            return True
-
-        conn_key = self._get_connection_key(sync_config)
-        return connection_token in self._connections.get(conn_key, {})
+    store: "DataStore"
 
     async def get_per_connection_state(
         self,
@@ -86,23 +73,20 @@ async def get_per_connection_state(
         Raises:
             SlidingSyncUnknownPosition if the connection_token is unknown
         """
-        if from_token is None:
+        if from_token is None or from_token.connection_position == 0:
             return PerConnectionState()
 
-        connection_position = from_token.connection_position
-        if connection_position == 0:
-            # Initial sync (request without a `from_token`) starts at `0` so
-            # there is no existing per-connection state
-            return PerConnectionState()
-
-        conn_key = self._get_connection_key(sync_config)
-        sync_statuses = self._connections.get(conn_key, {})
-        connection_state = sync_statuses.get(connection_position)
+        conn_id = sync_config.conn_id or ""
 
-        if connection_state is None:
-            raise SlidingSyncUnknownPosition()
+        device_id = sync_config.requester.device_id
+        assert device_id is not None
 
-        return connection_state
+        return await self.store.get_per_connection_state(
+            sync_config.user.to_string(),
+            device_id,
+            conn_id,
+            from_token.connection_position,
+        )
 
     @trace
     async def record_new_state(
@@ -116,85 +100,24 @@ async def record_new_state(
         If there are no changes to the state this may return the same token as
         the existing per-connection state.
         """
-        prev_connection_token = 0
-        if from_token is not None:
-            prev_connection_token = from_token.connection_position
-
         if not new_connection_state.has_updates():
-            return prev_connection_token
-
-        conn_key = self._get_connection_key(sync_config)
-        sync_statuses = self._connections.setdefault(conn_key, {})
-
-        # Generate a new token, removing any existing entries in that token
-        # (which can happen if requests get resent).
-        new_store_token = prev_connection_token + 1
-        sync_statuses.pop(new_store_token, None)
+            if from_token is not None:
+                return from_token.connection_position
+            else:
+                return 0
 
-        # We copy the `MutablePerConnectionState` so that the inner `ChainMap`s
-        # don't grow forever.
-        sync_statuses[new_store_token] = new_connection_state.copy()
+        if from_token is not None and from_token.connection_position == 0:
+            from_token = None
 
-        return new_store_token
-
-    @trace
-    async def mark_token_seen(
-        self,
-        sync_config: SlidingSyncConfig,
-        from_token: Optional[SlidingSyncStreamToken],
-    ) -> None:
-        """We have received a request with the given token, so we can clear out
-        any other tokens associated with the connection.
-
-        If there is no from token then we have started afresh, and so we delete
-        all tokens associated with the device.
-        """
-        # Clear out any tokens for the connection that doesn't match the one
-        # from the request.
-
-        conn_key = self._get_connection_key(sync_config)
-        sync_statuses = self._connections.pop(conn_key, {})
-        if from_token is None:
-            return
-
-        sync_statuses = {
-            connection_token: room_statuses
-            for connection_token, room_statuses in sync_statuses.items()
-            if connection_token == from_token.connection_position
-        }
-        if sync_statuses:
-            self._connections[conn_key] = sync_statuses
-
-    @staticmethod
-    def _get_connection_key(sync_config: SlidingSyncConfig) -> Tuple[str, str]:
-        """Return a unique identifier for this connection.
-
-        The first part is simply the user ID.
-
-        The second part is generally a combination of device ID and conn_id.
-        However, both these two are optional (e.g. puppet access tokens don't
-        have device IDs), so this handles those edge cases.
-
-        We use this over the raw `conn_id` to avoid clashes between different
-        clients that use the same `conn_id`. Imagine a user uses a web client
-        that uses `conn_id: main_sync_loop` and an Android client that also has
-        a `conn_id: main_sync_loop`.
-        """
-
-        user_id = sync_config.user.to_string()
-
-        # Only one sliding sync connection is allowed per given conn_id (empty
-        # or not).
         conn_id = sync_config.conn_id or ""
 
-        if sync_config.requester.device_id:
-            return (user_id, f"D/{sync_config.requester.device_id}/{conn_id}")
-
-        if sync_config.requester.access_token_id:
-            # If we don't have a device, then the access token ID should be a
-            # stable ID.
-            return (user_id, f"A/{sync_config.requester.access_token_id}/{conn_id}")
+        device_id = sync_config.requester.device_id
+        assert device_id is not None
 
-        # If we have neither then its likely an AS or some weird token. Either
-        # way we can just fail here.
-        raise Exception("Cannot use sliding sync with access token type")
+        return await self.store.persist_per_connection_state(
+            sync_config.user.to_string(),
+            device_id,
+            conn_id,
+            from_token.connection_position if from_token else None,
+            new_connection_state,
+        )