summary refs log tree commit diff
path: root/synapse/storage/state.py
diff options
context:
space:
mode:
authorDavid Baker <dave@matrix.org>2018-07-17 15:43:33 +0100
committerDavid Baker <dave@matrix.org>2018-07-17 15:43:33 +0100
commit1d11d9323d1d5eb419cbced9417f58d155c68255 (patch)
tree7b35de324a19d67f7fab5e6bc9c637c9c2253a71 /synapse/storage/state.py
parentMerge pull request #3502 from matrix-org/matthew/dinsic-tweak-display-names (diff)
parentchangelog (diff)
downloadsynapse-1d11d9323d1d5eb419cbced9417f58d155c68255.tar.xz
Merge remote-tracking branch 'origin/master' into dinsic
Diffstat (limited to 'synapse/storage/state.py')
-rw-r--r--synapse/storage/state.py66
1 files changed, 46 insertions, 20 deletions
diff --git a/synapse/storage/state.py b/synapse/storage/state.py

index bdee14a8eb..cd9821c270 100644 --- a/synapse/storage/state.py +++ b/synapse/storage/state.py
@@ -23,7 +23,7 @@ from twisted.internet import defer from synapse.storage.background_updates import BackgroundUpdateStore from synapse.storage.engines import PostgresEngine -from synapse.util.caches import intern_string, CACHE_SIZE_FACTOR +from synapse.util.caches import intern_string, get_cache_factor_for from synapse.util.caches.descriptors import cached, cachedList from synapse.util.caches.dictionary_cache import DictionaryCache from synapse.util.stringutils import to_ascii @@ -57,7 +57,7 @@ class StateGroupWorkerStore(SQLBaseStore): super(StateGroupWorkerStore, self).__init__(db_conn, hs) self._state_group_cache = DictionaryCache( - "*stateGroupCache*", 100000 * CACHE_SIZE_FACTOR + "*stateGroupCache*", 500000 * get_cache_factor_for("stateGroupCache") ) @cached(max_entries=100000, iterable=True) @@ -272,7 +272,7 @@ class StateGroupWorkerStore(SQLBaseStore): for typ in types: if typ[1] is None: where_clauses.append("(type = ?)") - where_args.extend(typ[0]) + where_args.append(typ[0]) wildcard_types = True else: where_clauses.append("(type = ? AND state_key = ?)") @@ -526,10 +526,23 @@ class StateGroupWorkerStore(SQLBaseStore): @defer.inlineCallbacks def _get_state_for_groups(self, groups, types=None): - """Given list of groups returns dict of group -> list of state events - with matching types. `types` is a list of `(type, state_key)`, where - a `state_key` of None matches all state_keys. If `types` is None then - all events are returned. + """Gets the state at each of a list of state groups, optionally + filtering by type/state_key + + Args: + groups (iterable[int]): list of state groups for which we want + to get the state. + types (None|iterable[(str, None|str)]): + indicates the state type/keys required. If None, the whole + state is fetched and returned. + + Otherwise, each entry should be a `(type, state_key)` tuple to + include in the response. A `state_key` of None is a wildcard + meaning that we require all state with that type. + + Returns: + Deferred[dict[int, dict[(type, state_key), EventBase]]] + a dictionary mapping from state group to state dictionary. """ if types: types = frozenset(types) @@ -538,7 +551,7 @@ class StateGroupWorkerStore(SQLBaseStore): if types is not None: for group in set(groups): state_dict_ids, _, got_all = self._get_some_state_from_cache( - group, types + group, types, ) results[group] = state_dict_ids @@ -559,26 +572,40 @@ class StateGroupWorkerStore(SQLBaseStore): # Okay, so we have some missing_types, lets fetch them. cache_seq_num = self._state_group_cache.sequence + # the DictionaryCache knows if it has *all* the state, but + # does not know if it has all of the keys of a particular type, + # which makes wildcard lookups expensive unless we have a complete + # cache. Hence, if we are doing a wildcard lookup, populate the + # cache fully so that we can do an efficient lookup next time. + + if types and any(k is None for (t, k) in types): + types_to_fetch = None + else: + types_to_fetch = types + group_to_state_dict = yield self._get_state_groups_from_groups( - missing_groups, types + missing_groups, types_to_fetch, ) - # Now we want to update the cache with all the things we fetched - # from the database. for group, group_state_dict in iteritems(group_to_state_dict): state_dict = results[group] - state_dict.update( - ((intern_string(k[0]), intern_string(k[1])), to_ascii(v)) - for k, v in iteritems(group_state_dict) - ) - + # update the result, filtering by `types`. + if types: + for k, v in iteritems(group_state_dict): + (typ, _) = k + if k in types or (typ, None) in types: + state_dict[k] = v + else: + state_dict.update(group_state_dict) + + # update the cache with all the things we fetched from the + # database. self._state_group_cache.update( cache_seq_num, key=group, - value=state_dict, - full=(types is None), - known_absent=types, + value=group_state_dict, + fetched_keys=types_to_fetch, ) defer.returnValue(results) @@ -685,7 +712,6 @@ class StateGroupWorkerStore(SQLBaseStore): self._state_group_cache.sequence, key=state_group, value=dict(current_state_ids), - full=True, ) return state_group