summary refs log tree commit diff
path: root/synapse/storage
diff options
context:
space:
mode:
Diffstat (limited to 'synapse/storage')
-rw-r--r--synapse/storage/__init__.py5
-rw-r--r--synapse/storage/_base.py109
-rw-r--r--synapse/storage/state.py353
-rw-r--r--synapse/storage/stream.py7
4 files changed, 391 insertions, 83 deletions
diff --git a/synapse/storage/__init__.py b/synapse/storage/__init__.py
index 1a6a8a3762..c6ce65b4cc 100644
--- a/synapse/storage/__init__.py
+++ b/synapse/storage/__init__.py
@@ -354,6 +354,11 @@ def _upgrade_existing_database(cur, current_version, applied_delta_files,
                     )
                 logger.debug("Running script %s", relative_path)
                 module.run_upgrade(cur, database_engine)
+            elif ext == ".pyc":
+                # Sometimes .pyc files turn up anyway even though we've
+                # disabled their generation; e.g. from distribution package
+                # installers. Silently skip it
+                pass
             elif ext == ".sql":
                 # A plain old .sql file, just read and execute it
                 logger.debug("Applying schema %s", relative_path)
diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py
index 1e2d436660..a7590145ac 100644
--- a/synapse/storage/_base.py
+++ b/synapse/storage/_base.py
@@ -16,9 +16,11 @@ import logging
 
 from synapse.api.errors import StoreError
 from synapse.util.async import ObservableDeferred
+from synapse.util import unwrapFirstError
 from synapse.util.logutils import log_function
 from synapse.util.logcontext import preserve_context_over_fn, LoggingContext
 from synapse.util.lrucache import LruCache
+from synapse.util.dictionary_cache import DictionaryCache
 import synapse.metrics
 
 from util.id_generators import IdGenerator, StreamIdGenerator
@@ -231,6 +233,101 @@ class CacheDescriptor(object):
         return wrapped
 
 
+class CacheListDescriptor(object):
+    def __init__(self, orig, cache, list_name, num_args=1, inlineCallbacks=False):
+        self.orig = orig
+
+        if inlineCallbacks:
+            self.function_to_call = defer.inlineCallbacks(orig)
+        else:
+            self.function_to_call = orig
+
+        self.num_args = num_args
+        self.list_name = list_name
+
+        self.arg_names = inspect.getargspec(orig).args[1:num_args+1]
+        self.list_pos = self.arg_names.index(self.list_name)
+
+        self.cache = cache
+
+        self.sentinel = object()
+
+        if len(self.arg_names) < self.num_args:
+            raise Exception(
+                "Not enough explicit positional arguments to key off of for %r."
+                " (@cached cannot key off of *args or **kwars)"
+                % (orig.__name__,)
+            )
+
+        if self.list_name not in self.arg_names:
+            raise Exception(
+                "Couldn't see arguments %r for %r."
+                % (self.list_name, cache.name,)
+            )
+
+    def __get__(self, obj, objtype=None):
+
+        @functools.wraps(self.orig)
+        def wrapped(*args, **kwargs):
+            arg_dict = inspect.getcallargs(self.orig, obj, *args, **kwargs)
+            keyargs = [arg_dict[arg_nm] for arg_nm in self.arg_names]
+            list_args = arg_dict[self.list_name]
+
+            cached = {}
+            missing = []
+            for arg in list_args:
+                key = list(keyargs)
+                key[self.list_pos] = arg
+
+                try:
+                    res = self.cache.get(tuple(key)).observe()
+                    res.addCallback(lambda r, arg: (arg, r), arg)
+                    cached[arg] = res
+                except KeyError:
+                    missing.append(arg)
+
+            if missing:
+                sequence = self.cache.sequence
+                args_to_call = dict(arg_dict)
+                args_to_call[self.list_name] = missing
+
+                ret_d = defer.maybeDeferred(
+                    self.function_to_call,
+                    **args_to_call
+                )
+
+                ret_d = ObservableDeferred(ret_d)
+
+                for arg in missing:
+                    observer = ret_d.observe()
+                    observer.addCallback(lambda r, arg: r[arg], arg)
+
+                    observer = ObservableDeferred(observer)
+
+                    key = list(keyargs)
+                    key[self.list_pos] = arg
+                    self.cache.update(sequence, tuple(key), observer)
+
+                    def invalidate(f, key):
+                        self.cache.invalidate(key)
+                        return f
+                    observer.addErrback(invalidate, tuple(key))
+
+                    res = observer.observe()
+                    res.addCallback(lambda r, arg: (arg, r), arg)
+
+                    cached[arg] = res
+
+            return defer.gatherResults(
+                cached.values(),
+                consumeErrors=True,
+            ).addErrback(unwrapFirstError).addCallback(lambda res: dict(res))
+
+        obj.__dict__[self.orig.__name__] = wrapped
+
+        return wrapped
+
+
 def cached(max_entries=1000, num_args=1, lru=True):
     return lambda orig: CacheDescriptor(
         orig,
@@ -250,6 +347,16 @@ def cachedInlineCallbacks(max_entries=1000, num_args=1, lru=False):
     )
 
 
+def cachedList(cache, list_name, num_args=1, inlineCallbacks=False):
+    return lambda orig: CacheListDescriptor(
+        orig,
+        cache=cache,
+        list_name=list_name,
+        num_args=num_args,
+        inlineCallbacks=inlineCallbacks,
+    )
+
+
 class LoggingTransaction(object):
     """An object that almost-transparently proxies for the 'txn' object
     passed to the constructor. Adds logging and metrics to the .execute()
@@ -370,6 +477,8 @@ class SQLBaseStore(object):
         self._get_event_cache = Cache("*getEvent*", keylen=3, lru=True,
                                       max_entries=hs.config.event_cache_size)
 
+        self._state_group_cache = DictionaryCache("*stateGroupCache*", 100000)
+
         self._event_fetch_lock = threading.Condition()
         self._event_fetch_list = []
         self._event_fetch_ongoing = 0
diff --git a/synapse/storage/state.py b/synapse/storage/state.py
index 7ce51b9bdc..06f0ab3ff6 100644
--- a/synapse/storage/state.py
+++ b/synapse/storage/state.py
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from ._base import SQLBaseStore, cachedInlineCallbacks
+from ._base import SQLBaseStore, cached, cachedInlineCallbacks, cachedList
 
 from twisted.internet import defer
 
@@ -44,52 +44,26 @@ class StateStore(SQLBaseStore):
     """
 
     @defer.inlineCallbacks
-    def get_state_groups(self, event_ids):
+    def get_state_groups(self, room_id, event_ids):
         """ Get the state groups for the given list of event_ids
 
         The return value is a dict mapping group names to lists of events.
         """
+        if not event_ids:
+            defer.returnValue({})
 
-        def f(txn):
-            groups = set()
-            for event_id in event_ids:
-                group = self._simple_select_one_onecol_txn(
-                    txn,
-                    table="event_to_state_groups",
-                    keyvalues={"event_id": event_id},
-                    retcol="state_group",
-                    allow_none=True,
-                )
-                if group:
-                    groups.add(group)
-
-            res = {}
-            for group in groups:
-                state_ids = self._simple_select_onecol_txn(
-                    txn,
-                    table="state_groups_state",
-                    keyvalues={"state_group": group},
-                    retcol="event_id",
-                )
-
-                res[group] = state_ids
-
-            return res
-
-        states = yield self.runInteraction(
-            "get_state_groups",
-            f,
+        event_to_groups = yield self._get_state_group_for_events(
+            room_id, event_ids,
         )
 
-        state_list = yield defer.gatherResults(
-            [
-                self._fetch_events_for_group(group, vals)
-                for group, vals in states.items()
-            ],
-            consumeErrors=True,
-        )
+        groups = set(event_to_groups.values())
 
-        defer.returnValue(dict(state_list))
+        group_to_state = yield self._get_state_for_groups(groups)
+
+        defer.returnValue({
+            group: state_map.values()
+            for group, state_map in group_to_state.items()
+        })
 
     def _fetch_events_for_group(self, key, events):
         return self._get_events(
@@ -204,64 +178,285 @@ class StateStore(SQLBaseStore):
         events = yield self._get_events(event_ids, get_prev_content=False)
         defer.returnValue(events)
 
+    def _get_state_groups_from_group(self, group, types):
+        def f(txn):
+            if types is not None:
+                where_clause = "AND (%s)" % (
+                    " OR ".join(["(type = ? AND state_key = ?)"] * len(types)),
+                )
+            else:
+                where_clause = ""
+
+            sql = (
+                "SELECT event_id FROM state_groups_state WHERE"
+                " state_group = ? %s"
+            ) % (where_clause,)
+
+            args = [group]
+            if types is not None:
+                args.extend([i for typ in types for i in typ])
+
+            txn.execute(sql, args)
+
+            return [r[0] for r in txn.fetchall()]
+
+        return self.runInteraction(
+            "_get_state_groups_from_group",
+            f,
+        )
+
+    def _get_state_groups_from_groups(self, groups_and_types):
+        def f(txn):
+            results = {}
+            for group, types in groups_and_types:
+                if types is not None:
+                    where_clause = "AND (%s)" % (
+                        " OR ".join(["(type = ? AND state_key = ?)"] * len(types)),
+                    )
+                else:
+                    where_clause = ""
+
+                sql = (
+                    "SELECT event_id FROM state_groups_state WHERE"
+                    " state_group = ? %s"
+                ) % (where_clause,)
+
+                args = [group]
+                if types is not None:
+                    args.extend([i for typ in types for i in typ])
+
+                txn.execute(sql, args)
+
+                results[group] = [
+                    r[0]
+                    for r in txn.fetchall()
+                ]
+
+            return results
+
+        return self.runInteraction(
+            "_get_state_groups_from_groups",
+            f,
+        )
+
+    @cached(num_args=3, lru=True, max_entries=10000)
+    def _get_state_for_event_id(self, room_id, event_id, types):
+        def f(txn):
+            type_and_state_sql = " OR ".join([
+                "(type = ? AND state_key = ?)"
+                if typ[1] is not None
+                else "type = ?"
+                for typ in types
+            ])
+
+            sql = (
+                "SELECT e.event_id, sg.state_group, sg.event_id"
+                " FROM state_groups_state as sg"
+                " INNER JOIN event_to_state_groups as e"
+                " ON e.state_group = sg.state_group"
+                " WHERE e.event_id = ? AND (%s)"
+            ) % (type_and_state_sql,)
+
+            args = [event_id]
+            for typ, state_key in types:
+                args.extend(
+                    [typ, state_key] if state_key is not None else [typ]
+                )
+            txn.execute(sql, args)
+
+            return event_id, [
+                r[0]
+                for r in txn.fetchall()
+            ]
+
+        return self.runInteraction(
+            "_get_state_for_event_id",
+            f,
+        )
+
     @defer.inlineCallbacks
-    def get_state_for_events(self, room_id, event_ids):
+    def get_state_for_events(self, room_id, event_ids, types):
+        """Given a list of event_ids and type tuples, return a list of state
+        dicts for each event. The state dicts will only have the type/state_keys
+        that are in the `types` list.
+
+        Args:
+            room_id (str)
+            event_ids (list)
+            types (list): List of (type, state_key) tuples which are used to
+                filter the state fetched. `state_key` may be None, which matches
+                any `state_key`
+
+        Returns:
+            deferred: A list of dicts corresponding to the event_ids given.
+            The dicts are mappings from (type, state_key) -> state_events
+        """
+        event_to_groups = yield self._get_state_group_for_events(
+            room_id, event_ids,
+        )
+
+        groups = set(event_to_groups.values())
+
+        group_to_state = yield self._get_state_for_groups(
+            groups, types
+        )
+
+        event_to_state = {
+            event_id: group_to_state[group]
+            for event_id, group in event_to_groups.items()
+        }
+
+        defer.returnValue([
+            event_to_state[event]
+            for event in event_ids
+        ])
+
+    @cached(num_args=2, lru=True, max_entries=100000)
+    def _get_state_group_for_event(self, room_id, event_id):
+        return self._simple_select_one_onecol(
+            table="event_to_state_groups",
+            keyvalues={
+                "event_id": event_id,
+            },
+            retcol="state_group",
+            allow_none=True,
+            desc="_get_state_group_for_event",
+        )
+
+    @cachedList(cache=_get_state_group_for_event.cache, list_name="event_ids",
+                num_args=2)
+    def _get_state_group_for_events(self, room_id, event_ids):
         def f(txn):
-            groups = set()
-            event_to_group = {}
+            results = {}
             for event_id in event_ids:
-                # TODO: Remove this loop.
-                group = self._simple_select_one_onecol_txn(
+                results[event_id] = self._simple_select_one_onecol_txn(
                     txn,
                     table="event_to_state_groups",
-                    keyvalues={"event_id": event_id},
+                    keyvalues={
+                        "event_id": event_id,
+                    },
                     retcol="state_group",
                     allow_none=True,
                 )
-                if group:
-                    event_to_group[event_id] = group
-                    groups.add(group)
-
-            group_to_state_ids = {}
-            for group in groups:
-                state_ids = self._simple_select_onecol_txn(
-                    txn,
-                    table="state_groups_state",
-                    keyvalues={"state_group": group},
-                    retcol="event_id",
-                )
-
-                group_to_state_ids[group] = state_ids
 
-            return event_to_group, group_to_state_ids
+            return results
 
-        res = yield self.runInteraction(
-            "annotate_events_with_state_groups",
+        return self.runInteraction(
+            "_get_state_group_for_events",
             f,
         )
 
-        event_to_group, group_to_state_ids = res
+    def _get_state_for_group_from_cache(self, group, types=None):
+        is_all, state_dict = self._state_group_cache.get(group)
+
+        type_to_key = {}
+        missing_types = set()
+        if types is not None:
+            for typ, state_key in types:
+                if state_key is None:
+                    type_to_key[typ] = None
+                    missing_types.add((typ, state_key))
+                else:
+                    if type_to_key.get(typ, object()) is not None:
+                        type_to_key.setdefault(typ, set()).add(state_key)
+
+                    if (typ, state_key) not in state_dict:
+                        missing_types.add((typ, state_key))
+
+        if is_all and types is None:
+            return state_dict, missing_types
+
+        if is_all or (types is not None and not missing_types):
+            sentinel = object()
+
+            def include(typ, state_key):
+                valid_state_keys = type_to_key.get(typ, sentinel)
+                if valid_state_keys is sentinel:
+                    return False
+                if valid_state_keys is None:
+                    return True
+                if state_key in valid_state_keys:
+                    return True
+                return False
+
+            return {
+                k: v
+                for k, v in state_dict.items()
+                if v and include(k[0], k[1])
+            }, missing_types
+
+        return {}, missing_types
 
-        state_list = yield defer.gatherResults(
-            [
-                self._fetch_events_for_group(group, vals)
-                for group, vals in group_to_state_ids.items()
-            ],
-            consumeErrors=True,
+    @defer.inlineCallbacks
+    def _get_state_for_groups(self, groups, types=None):
+        results = {}
+        missing_groups_and_types = []
+        for group in groups:
+            state_dict, missing_types = self._get_state_for_group_from_cache(
+                group, types
+            )
+
+            if types is not None and not missing_types:
+                results[group] = {
+                    key: value
+                    for key, value in state_dict.items()
+                    if value
+                }
+            else:
+                missing_groups_and_types.append((
+                    group,
+                    missing_types if types else None
+                ))
+
+        if not missing_groups_and_types:
+            defer.returnValue(results)
+
+        # Okay, so we have some missing_types, lets fetch them.
+        cache_seq_num = self._state_group_cache.sequence
+
+        group_state_dict = yield self._get_state_groups_from_groups(
+            missing_groups_and_types
         )
 
-        state_dict = {
-            group: {
-                (ev.type, ev.state_key): ev
-                for ev in state
-            }
-            for group, state in state_list
+        state_events = yield self._get_events(
+            [e_id for l in group_state_dict.values() for e_id in l],
+            get_prev_content=False
+        )
+
+        state_events = {
+            e.event_id: e
+            for e in state_events
         }
 
-        defer.returnValue([
-            state_dict.get(event_to_group.get(event, None), None)
-            for event in event_ids
-        ])
+        for group, state_ids in group_state_dict.items():
+            state_dict = {
+                key: None
+                for key in missing_types
+            }
+            evs = [
+                state_events[e_id] for e_id in state_ids
+                if e_id in state_events  # This can happen if event is rejected.
+            ]
+            state_dict.update({
+                (e.type, e.state_key): e
+                for e in evs
+            })
+
+            # Update the cache
+            self._state_group_cache.update(
+                cache_seq_num,
+                key=group,
+                value=state_dict,
+                full=(types is None),
+            )
+
+            results[group] = {
+                key: value
+                for key, value in state_dict.items()
+                if value
+            }
+
+        defer.returnValue(results)
 
 
 def _make_group_id(clock):
diff --git a/synapse/storage/stream.py b/synapse/storage/stream.py
index af45fc5619..b59fe81004 100644
--- a/synapse/storage/stream.py
+++ b/synapse/storage/stream.py
@@ -35,7 +35,7 @@ what sort order was used:
 
 from twisted.internet import defer
 
-from ._base import SQLBaseStore
+from ._base import SQLBaseStore, cachedInlineCallbacks
 from synapse.api.constants import EventTypes
 from synapse.types import RoomStreamToken
 from synapse.util.logutils import log_function
@@ -299,9 +299,8 @@ class StreamStore(SQLBaseStore):
 
         defer.returnValue((events, token))
 
-    @defer.inlineCallbacks
-    def get_recent_events_for_room(self, room_id, limit, end_token,
-                                   with_feedback=False, from_token=None):
+    @cachedInlineCallbacks(num_args=4)
+    def get_recent_events_for_room(self, room_id, limit, end_token, from_token=None):
         # TODO (erikj): Handle compressed feedback
 
         end_token = RoomStreamToken.parse_stream_token(end_token)