Split state groups into a separate data store (#6296)

13 files changed, 1263 insertions, 0 deletions

diff --git a/synapse/storage/data_stores/state/__init__.py b/synapse/storage/data_stores/state/__init__.py
new file mode 100644
index 0000000000..86e09f6229
--- /dev/null
+++ b/synapse/storage/data_stores/state/__init__.py
@@ -0,0 +1,16 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from synapse.storage.data_stores.state.store import StateGroupDataStore  # noqa: F401
diff --git a/synapse/storage/data_stores/state/bg_updates.py b/synapse/storage/data_stores/state/bg_updates.py
new file mode 100644
index 0000000000..e8edaf9f7b
--- /dev/null
+++ b/synapse/storage/data_stores/state/bg_updates.py
@@ -0,0 +1,374 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+
+from six import iteritems
+
+from twisted.internet import defer
+
+from synapse.storage._base import SQLBaseStore
+from synapse.storage.database import Database
+from synapse.storage.engines import PostgresEngine
+from synapse.storage.state import StateFilter
+
+logger = logging.getLogger(__name__)
+
+
+MAX_STATE_DELTA_HOPS = 100
+
+
+class StateGroupBackgroundUpdateStore(SQLBaseStore):
+    """Defines functions related to state groups needed to run the state backgroud
+    updates.
+    """
+
+    def _count_state_group_hops_txn(self, txn, state_group):
+        """Given a state group, count how many hops there are in the tree.
+
+        This is used to ensure the delta chains don't get too long.
+        """
+        if isinstance(self.database_engine, PostgresEngine):
+            sql = """
+                WITH RECURSIVE state(state_group) AS (
+                    VALUES(?::bigint)
+                    UNION ALL
+                    SELECT prev_state_group FROM state_group_edges e, state s
+                    WHERE s.state_group = e.state_group
+                )
+                SELECT count(*) FROM state;
+            """
+
+            txn.execute(sql, (state_group,))
+            row = txn.fetchone()
+            if row and row[0]:
+                return row[0]
+            else:
+                return 0
+        else:
+            # We don't use WITH RECURSIVE on sqlite3 as there are distributions
+            # that ship with an sqlite3 version that doesn't support it (e.g. wheezy)
+            next_group = state_group
+            count = 0
+
+            while next_group:
+                next_group = self.db.simple_select_one_onecol_txn(
+                    txn,
+                    table="state_group_edges",
+                    keyvalues={"state_group": next_group},
+                    retcol="prev_state_group",
+                    allow_none=True,
+                )
+                if next_group:
+                    count += 1
+
+            return count
+
+    def _get_state_groups_from_groups_txn(
+        self, txn, groups, state_filter=StateFilter.all()
+    ):
+        results = {group: {} for group in groups}
+
+        where_clause, where_args = state_filter.make_sql_filter_clause()
+
+        # Unless the filter clause is empty, we're going to append it after an
+        # existing where clause
+        if where_clause:
+            where_clause = " AND (%s)" % (where_clause,)
+
+        if isinstance(self.database_engine, PostgresEngine):
+            # Temporarily disable sequential scans in this transaction. This is
+            # a temporary hack until we can add the right indices in
+            txn.execute("SET LOCAL enable_seqscan=off")
+
+            # The below query walks the state_group tree so that the "state"
+            # table includes all state_groups in the tree. It then joins
+            # against `state_groups_state` to fetch the latest state.
+            # It assumes that previous state groups are always numerically
+            # lesser.
+            # The PARTITION is used to get the event_id in the greatest state
+            # group for the given type, state_key.
+            # This may return multiple rows per (type, state_key), but last_value
+            # should be the same.
+            sql = """
+                WITH RECURSIVE state(state_group) AS (
+                    VALUES(?::bigint)
+                    UNION ALL
+                    SELECT prev_state_group FROM state_group_edges e, state s
+                    WHERE s.state_group = e.state_group
+                )
+                SELECT DISTINCT type, state_key, last_value(event_id) OVER (
+                    PARTITION BY type, state_key ORDER BY state_group ASC
+                    ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
+                ) AS event_id FROM state_groups_state
+                WHERE state_group IN (
+                    SELECT state_group FROM state
+                )
+            """
+
+            for group in groups:
+                args = [group]
+                args.extend(where_args)
+
+                txn.execute(sql + where_clause, args)
+                for row in txn:
+                    typ, state_key, event_id = row
+                    key = (typ, state_key)
+                    results[group][key] = event_id
+        else:
+            max_entries_returned = state_filter.max_entries_returned()
+
+            # We don't use WITH RECURSIVE on sqlite3 as there are distributions
+            # that ship with an sqlite3 version that doesn't support it (e.g. wheezy)
+            for group in groups:
+                next_group = group
+
+                while next_group:
+                    # We did this before by getting the list of group ids, and
+                    # then passing that list to sqlite to get latest event for
+                    # each (type, state_key). However, that was terribly slow
+                    # without the right indices (which we can't add until
+                    # after we finish deduping state, which requires this func)
+                    args = [next_group]
+                    args.extend(where_args)
+
+                    txn.execute(
+                        "SELECT type, state_key, event_id FROM state_groups_state"
+                        " WHERE state_group = ? " + where_clause,
+                        args,
+                    )
+                    results[group].update(
+                        ((typ, state_key), event_id)
+                        for typ, state_key, event_id in txn
+                        if (typ, state_key) not in results[group]
+                    )
+
+                    # If the number of entries in the (type,state_key)->event_id dict
+                    # matches the number of (type,state_keys) types we were searching
+                    # for, then we must have found them all, so no need to go walk
+                    # further down the tree... UNLESS our types filter contained
+                    # wildcards (i.e. Nones) in which case we have to do an exhaustive
+                    # search
+                    if (
+                        max_entries_returned is not None
+                        and len(results[group]) == max_entries_returned
+                    ):
+                        break
+
+                    next_group = self.db.simple_select_one_onecol_txn(
+                        txn,
+                        table="state_group_edges",
+                        keyvalues={"state_group": next_group},
+                        retcol="prev_state_group",
+                        allow_none=True,
+                    )
+
+        return results
+
+
+class StateBackgroundUpdateStore(StateGroupBackgroundUpdateStore):
+
+    STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication"
+    STATE_GROUP_INDEX_UPDATE_NAME = "state_group_state_type_index"
+    STATE_GROUPS_ROOM_INDEX_UPDATE_NAME = "state_groups_room_id_idx"
+
+    def __init__(self, database: Database, db_conn, hs):
+        super(StateBackgroundUpdateStore, self).__init__(database, db_conn, hs)
+        self.db.updates.register_background_update_handler(
+            self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME,
+            self._background_deduplicate_state,
+        )
+        self.db.updates.register_background_update_handler(
+            self.STATE_GROUP_INDEX_UPDATE_NAME, self._background_index_state
+        )
+        self.db.updates.register_background_index_update(
+            self.STATE_GROUPS_ROOM_INDEX_UPDATE_NAME,
+            index_name="state_groups_room_id_idx",
+            table="state_groups",
+            columns=["room_id"],
+        )
+
+    @defer.inlineCallbacks
+    def _background_deduplicate_state(self, progress, batch_size):
+        """This background update will slowly deduplicate state by reencoding
+        them as deltas.
+        """
+        last_state_group = progress.get("last_state_group", 0)
+        rows_inserted = progress.get("rows_inserted", 0)
+        max_group = progress.get("max_group", None)
+
+        BATCH_SIZE_SCALE_FACTOR = 100
+
+        batch_size = max(1, int(batch_size / BATCH_SIZE_SCALE_FACTOR))
+
+        if max_group is None:
+            rows = yield self.db.execute(
+                "_background_deduplicate_state",
+                None,
+                "SELECT coalesce(max(id), 0) FROM state_groups",
+            )
+            max_group = rows[0][0]
+
+        def reindex_txn(txn):
+            new_last_state_group = last_state_group
+            for count in range(batch_size):
+                txn.execute(
+                    "SELECT id, room_id FROM state_groups"
+                    " WHERE ? < id AND id <= ?"
+                    " ORDER BY id ASC"
+                    " LIMIT 1",
+                    (new_last_state_group, max_group),
+                )
+                row = txn.fetchone()
+                if row:
+                    state_group, room_id = row
+
+                if not row or not state_group:
+                    return True, count
+
+                txn.execute(
+                    "SELECT state_group FROM state_group_edges"
+                    " WHERE state_group = ?",
+                    (state_group,),
+                )
+
+                # If we reach a point where we've already started inserting
+                # edges we should stop.
+                if txn.fetchall():
+                    return True, count
+
+                txn.execute(
+                    "SELECT coalesce(max(id), 0) FROM state_groups"
+                    " WHERE id < ? AND room_id = ?",
+                    (state_group, room_id),
+                )
+                (prev_group,) = txn.fetchone()
+                new_last_state_group = state_group
+
+                if prev_group:
+                    potential_hops = self._count_state_group_hops_txn(txn, prev_group)
+                    if potential_hops >= MAX_STATE_DELTA_HOPS:
+                        # We want to ensure chains are at most this long,#
+                        # otherwise read performance degrades.
+                        continue
+
+                    prev_state = self._get_state_groups_from_groups_txn(
+                        txn, [prev_group]
+                    )
+                    prev_state = prev_state[prev_group]
+
+                    curr_state = self._get_state_groups_from_groups_txn(
+                        txn, [state_group]
+                    )
+                    curr_state = curr_state[state_group]
+
+                    if not set(prev_state.keys()) - set(curr_state.keys()):
+                        # We can only do a delta if the current has a strict super set
+                        # of keys
+
+                        delta_state = {
+                            key: value
+                            for key, value in iteritems(curr_state)
+                            if prev_state.get(key, None) != value
+                        }
+
+                        self.db.simple_delete_txn(
+                            txn,
+                            table="state_group_edges",
+                            keyvalues={"state_group": state_group},
+                        )
+
+                        self.db.simple_insert_txn(
+                            txn,
+                            table="state_group_edges",
+                            values={
+                                "state_group": state_group,
+                                "prev_state_group": prev_group,
+                            },
+                        )
+
+                        self.db.simple_delete_txn(
+                            txn,
+                            table="state_groups_state",
+                            keyvalues={"state_group": state_group},
+                        )
+
+                        self.db.simple_insert_many_txn(
+                            txn,
+                            table="state_groups_state",
+                            values=[
+                                {
+                                    "state_group": state_group,
+                                    "room_id": room_id,
+                                    "type": key[0],
+                                    "state_key": key[1],
+                                    "event_id": state_id,
+                                }
+                                for key, state_id in iteritems(delta_state)
+                            ],
+                        )
+
+            progress = {
+                "last_state_group": state_group,
+                "rows_inserted": rows_inserted + batch_size,
+                "max_group": max_group,
+            }
+
+            self.db.updates._background_update_progress_txn(
+                txn, self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, progress
+            )
+
+            return False, batch_size
+
+        finished, result = yield self.db.runInteraction(
+            self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, reindex_txn
+        )
+
+        if finished:
+            yield self.db.updates._end_background_update(
+                self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME
+            )
+
+        return result * BATCH_SIZE_SCALE_FACTOR
+
+    @defer.inlineCallbacks
+    def _background_index_state(self, progress, batch_size):
+        def reindex_txn(conn):
+            conn.rollback()
+            if isinstance(self.database_engine, PostgresEngine):
+                # postgres insists on autocommit for the index
+                conn.set_session(autocommit=True)
+                try:
+                    txn = conn.cursor()
+                    txn.execute(
+                        "CREATE INDEX CONCURRENTLY state_groups_state_type_idx"
+                        " ON state_groups_state(state_group, type, state_key)"
+                    )
+                    txn.execute("DROP INDEX IF EXISTS state_groups_state_id")
+                finally:
+                    conn.set_session(autocommit=False)
+            else:
+                txn = conn.cursor()
+                txn.execute(
+                    "CREATE INDEX state_groups_state_type_idx"
+                    " ON state_groups_state(state_group, type, state_key)"
+                )
+                txn.execute("DROP INDEX IF EXISTS state_groups_state_id")
+
+        yield self.db.runWithConnection(reindex_txn)
+
+        yield self.db.updates._end_background_update(self.STATE_GROUP_INDEX_UPDATE_NAME)
+
+        return 1
diff --git a/synapse/storage/data_stores/state/schema/delta/23/drop_state_index.sql b/synapse/storage/data_stores/state/schema/delta/23/drop_state_index.sql
new file mode 100644
index 0000000000..ae09fa0065
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/delta/23/drop_state_index.sql
@@ -0,0 +1,16 @@
+/* Copyright 2015, 2016 OpenMarket Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+DROP INDEX IF EXISTS state_groups_state_tuple;
diff --git a/synapse/storage/data_stores/state/schema/delta/30/state_stream.sql b/synapse/storage/data_stores/state/schema/delta/30/state_stream.sql
new file mode 100644
index 0000000000..e85699e82e
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/delta/30/state_stream.sql
@@ -0,0 +1,33 @@
+/* Copyright 2016 OpenMarket Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+/* We used to create a table called current_state_resets, but this is no
+ * longer used and is removed in delta 54.
+ */
+
+/* The outlier events that have aquired a state group typically through
+ * backfill. This is tracked separately to the events table, as assigning a
+ * state group change the position of the existing event in the stream
+ * ordering.
+ * However since a stream_ordering is assigned in persist_event for the
+ * (event, state) pair, we can use that stream_ordering to identify when
+ * the new state was assigned for the event.
+ */
+CREATE TABLE IF NOT EXISTS ex_outlier_stream(
+    event_stream_ordering BIGINT PRIMARY KEY NOT NULL,
+    event_id TEXT NOT NULL,
+    state_group BIGINT NOT NULL
+);
diff --git a/synapse/storage/data_stores/state/schema/delta/32/remove_state_indices.sql b/synapse/storage/data_stores/state/schema/delta/32/remove_state_indices.sql
new file mode 100644
index 0000000000..1450313bfa
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/delta/32/remove_state_indices.sql
@@ -0,0 +1,19 @@
+/* Copyright 2016 OpenMarket Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+-- The following indices are redundant, other indices are equivalent or
+-- supersets
+DROP INDEX IF EXISTS state_groups_id; -- Duplicate of PRIMARY KEY
diff --git a/synapse/storage/data_stores/state/schema/delta/35/add_state_index.sql b/synapse/storage/data_stores/state/schema/delta/35/add_state_index.sql
new file mode 100644
index 0000000000..33980d02f0
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/delta/35/add_state_index.sql
@@ -0,0 +1,17 @@
+/* Copyright 2016 OpenMarket Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+INSERT into background_updates (update_name, progress_json, depends_on)
+    VALUES ('state_group_state_type_index', '{}', 'state_group_state_deduplication');
diff --git a/synapse/storage/data_stores/state/schema/delta/35/state.sql b/synapse/storage/data_stores/state/schema/delta/35/state.sql
new file mode 100644
index 0000000000..0f1fa68a89
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/delta/35/state.sql
@@ -0,0 +1,22 @@
+/* Copyright 2016 OpenMarket Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+CREATE TABLE state_group_edges(
+    state_group BIGINT NOT NULL,
+    prev_state_group BIGINT NOT NULL
+);
+
+CREATE INDEX state_group_edges_idx ON state_group_edges(state_group);
+CREATE INDEX state_group_edges_prev_idx ON state_group_edges(prev_state_group);
diff --git a/synapse/storage/data_stores/state/schema/delta/35/state_dedupe.sql b/synapse/storage/data_stores/state/schema/delta/35/state_dedupe.sql
new file mode 100644
index 0000000000..97e5067ef4
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/delta/35/state_dedupe.sql
@@ -0,0 +1,17 @@
+/* Copyright 2016 OpenMarket Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+INSERT into background_updates (update_name, progress_json)
+    VALUES ('state_group_state_deduplication', '{}');
diff --git a/synapse/storage/data_stores/state/schema/delta/47/state_group_seq.py b/synapse/storage/data_stores/state/schema/delta/47/state_group_seq.py
new file mode 100644
index 0000000000..9fd1ccf6f7
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/delta/47/state_group_seq.py
@@ -0,0 +1,34 @@
+# Copyright 2018 New Vector Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from synapse.storage.engines import PostgresEngine
+
+
+def run_create(cur, database_engine, *args, **kwargs):
+    if isinstance(database_engine, PostgresEngine):
+        # if we already have some state groups, we want to start making new
+        # ones with a higher id.
+        cur.execute("SELECT max(id) FROM state_groups")
+        row = cur.fetchone()
+
+        if row[0] is None:
+            start_val = 1
+        else:
+            start_val = row[0] + 1
+
+        cur.execute("CREATE SEQUENCE state_group_id_seq START WITH %s", (start_val,))
+
+
+def run_upgrade(*args, **kwargs):
+    pass
diff --git a/synapse/storage/data_stores/state/schema/delta/56/state_group_room_idx.sql b/synapse/storage/data_stores/state/schema/delta/56/state_group_room_idx.sql
new file mode 100644
index 0000000000..7916ef18b2
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/delta/56/state_group_room_idx.sql
@@ -0,0 +1,17 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+INSERT INTO background_updates (update_name, progress_json) VALUES
+    ('state_groups_room_id_idx', '{}');
diff --git a/synapse/storage/data_stores/state/schema/full_schemas/54/full.sql b/synapse/storage/data_stores/state/schema/full_schemas/54/full.sql
new file mode 100644
index 0000000000..35f97d6b3d
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/full_schemas/54/full.sql
@@ -0,0 +1,37 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+CREATE TABLE state_groups (
+    id BIGINT PRIMARY KEY,
+    room_id TEXT NOT NULL,
+    event_id TEXT NOT NULL
+);
+
+CREATE TABLE state_groups_state (
+    state_group BIGINT NOT NULL,
+    room_id TEXT NOT NULL,
+    type TEXT NOT NULL,
+    state_key TEXT NOT NULL,
+    event_id TEXT NOT NULL
+);
+
+CREATE TABLE state_group_edges (
+    state_group BIGINT NOT NULL,
+    prev_state_group BIGINT NOT NULL
+);
+
+CREATE INDEX state_group_edges_idx ON state_group_edges (state_group);
+CREATE INDEX state_group_edges_prev_idx ON state_group_edges (prev_state_group);
+CREATE INDEX state_groups_state_type_idx ON state_groups_state (state_group, type, state_key);
diff --git a/synapse/storage/data_stores/state/schema/full_schemas/54/sequence.sql.postgres b/synapse/storage/data_stores/state/schema/full_schemas/54/sequence.sql.postgres
new file mode 100644
index 0000000000..fcd926c9fb
--- /dev/null
+++ b/synapse/storage/data_stores/state/schema/full_schemas/54/sequence.sql.postgres
@@ -0,0 +1,21 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+CREATE SEQUENCE state_group_id_seq
+    START WITH 1
+    INCREMENT BY 1
+    NO MINVALUE
+    NO MAXVALUE
+    CACHE 1;
diff --git a/synapse/storage/data_stores/state/store.py b/synapse/storage/data_stores/state/store.py
new file mode 100644
index 0000000000..d53695f238
--- /dev/null
+++ b/synapse/storage/data_stores/state/store.py
@@ -0,0 +1,640 @@
+# -*- coding: utf-8 -*-
+# Copyright 2014-2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+from collections import namedtuple
+
+from six import iteritems
+from six.moves import range
+
+from twisted.internet import defer
+
+from synapse.api.constants import EventTypes
+from synapse.storage._base import SQLBaseStore
+from synapse.storage.data_stores.state.bg_updates import StateBackgroundUpdateStore
+from synapse.storage.database import Database
+from synapse.storage.state import StateFilter
+from synapse.util.caches import get_cache_factor_for
+from synapse.util.caches.descriptors import cached
+from synapse.util.caches.dictionary_cache import DictionaryCache
+
+logger = logging.getLogger(__name__)
+
+
+MAX_STATE_DELTA_HOPS = 100
+
+
+class _GetStateGroupDelta(
+    namedtuple("_GetStateGroupDelta", ("prev_group", "delta_ids"))
+):
+    """Return type of get_state_group_delta that implements __len__, which lets
+    us use the itrable flag when caching
+    """
+
+    __slots__ = []
+
+    def __len__(self):
+        return len(self.delta_ids) if self.delta_ids else 0
+
+
+class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
+    """A data store for fetching/storing state groups.
+    """
+
+    def __init__(self, database: Database, db_conn, hs):
+        super(StateGroupDataStore, self).__init__(database, db_conn, hs)
+
+        # Originally the state store used a single DictionaryCache to cache the
+        # event IDs for the state types in a given state group to avoid hammering
+        # on the state_group* tables.
+        #
+        # The point of using a DictionaryCache is that it can cache a subset
+        # of the state events for a given state group (i.e. a subset of the keys for a
+        # given dict which is an entry in the cache for a given state group ID).
+        #
+        # However, this poses problems when performing complicated queries
+        # on the store - for instance: "give me all the state for this group, but
+        # limit members to this subset of users", as DictionaryCache's API isn't
+        # rich enough to say "please cache any of these fields, apart from this subset".
+        # This is problematic when lazy loading members, which requires this behaviour,
+        # as without it the cache has no choice but to speculatively load all
+        # state events for the group, which negates the efficiency being sought.
+        #
+        # Rather than overcomplicating DictionaryCache's API, we instead split the
+        # state_group_cache into two halves - one for tracking non-member events,
+        # and the other for tracking member_events.  This means that lazy loading
+        # queries can be made in a cache-friendly manner by querying both caches
+        # separately and then merging the result.  So for the example above, you
+        # would query the members cache for a specific subset of state keys
+        # (which DictionaryCache will handle efficiently and fine) and the non-members
+        # cache for all state (which DictionaryCache will similarly handle fine)
+        # and then just merge the results together.
+        #
+        # We size the non-members cache to be smaller than the members cache as the
+        # vast majority of state in Matrix (today) is member events.
+
+        self._state_group_cache = DictionaryCache(
+            "*stateGroupCache*",
+            # TODO: this hasn't been tuned yet
+            50000 * get_cache_factor_for("stateGroupCache"),
+        )
+        self._state_group_members_cache = DictionaryCache(
+            "*stateGroupMembersCache*",
+            500000 * get_cache_factor_for("stateGroupMembersCache"),
+        )
+
+    @cached(max_entries=10000, iterable=True)
+    def get_state_group_delta(self, state_group):
+        """Given a state group try to return a previous group and a delta between
+        the old and the new.
+
+        Returns:
+            (prev_group, delta_ids), where both may be None.
+        """
+
+        def _get_state_group_delta_txn(txn):
+            prev_group = self.db.simple_select_one_onecol_txn(
+                txn,
+                table="state_group_edges",
+                keyvalues={"state_group": state_group},
+                retcol="prev_state_group",
+                allow_none=True,
+            )
+
+            if not prev_group:
+                return _GetStateGroupDelta(None, None)
+
+            delta_ids = self.db.simple_select_list_txn(
+                txn,
+                table="state_groups_state",
+                keyvalues={"state_group": state_group},
+                retcols=("type", "state_key", "event_id"),
+            )
+
+            return _GetStateGroupDelta(
+                prev_group,
+                {(row["type"], row["state_key"]): row["event_id"] for row in delta_ids},
+            )
+
+        return self.db.runInteraction(
+            "get_state_group_delta", _get_state_group_delta_txn
+        )
+
+    @defer.inlineCallbacks
+    def _get_state_groups_from_groups(self, groups, state_filter):
+        """Returns the state groups for a given set of groups, filtering on
+        types of state events.
+
+        Args:
+            groups(list[int]): list of state group IDs to query
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+        Returns:
+            Deferred[dict[int, dict[tuple[str, str], str]]]:
+                dict of state_group_id -> (dict of (type, state_key) -> event id)
+        """
+        results = {}
+
+        chunks = [groups[i : i + 100] for i in range(0, len(groups), 100)]
+        for chunk in chunks:
+            res = yield self.db.runInteraction(
+                "_get_state_groups_from_groups",
+                self._get_state_groups_from_groups_txn,
+                chunk,
+                state_filter,
+            )
+            results.update(res)
+
+        return results
+
+    def _get_state_for_group_using_cache(self, cache, group, state_filter):
+        """Checks if group is in cache. See `_get_state_for_groups`
+
+        Args:
+            cache(DictionaryCache): the state group cache to use
+            group(int): The state group to lookup
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+
+        Returns 2-tuple (`state_dict`, `got_all`).
+        `got_all` is a bool indicating if we successfully retrieved all
+        requests state from the cache, if False we need to query the DB for the
+        missing state.
+        """
+        is_all, known_absent, state_dict_ids = cache.get(group)
+
+        if is_all or state_filter.is_full():
+            # Either we have everything or want everything, either way
+            # `is_all` tells us whether we've gotten everything.
+            return state_filter.filter_state(state_dict_ids), is_all
+
+        # tracks whether any of our requested types are missing from the cache
+        missing_types = False
+
+        if state_filter.has_wildcards():
+            # We don't know if we fetched all the state keys for the types in
+            # the filter that are wildcards, so we have to assume that we may
+            # have missed some.
+            missing_types = True
+        else:
+            # There aren't any wild cards, so `concrete_types()` returns the
+            # complete list of event types we're wanting.
+            for key in state_filter.concrete_types():
+                if key not in state_dict_ids and key not in known_absent:
+                    missing_types = True
+                    break
+
+        return state_filter.filter_state(state_dict_ids), not missing_types
+
+    @defer.inlineCallbacks
+    def _get_state_for_groups(self, groups, state_filter=StateFilter.all()):
+        """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.
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+        Returns:
+            Deferred[dict[int, dict[tuple[str, str], str]]]:
+                dict of state_group_id -> (dict of (type, state_key) -> event id)
+        """
+
+        member_filter, non_member_filter = state_filter.get_member_split()
+
+        # Now we look them up in the member and non-member caches
+        (
+            non_member_state,
+            incomplete_groups_nm,
+        ) = yield self._get_state_for_groups_using_cache(
+            groups, self._state_group_cache, state_filter=non_member_filter
+        )
+
+        (
+            member_state,
+            incomplete_groups_m,
+        ) = yield self._get_state_for_groups_using_cache(
+            groups, self._state_group_members_cache, state_filter=member_filter
+        )
+
+        state = dict(non_member_state)
+        for group in groups:
+            state[group].update(member_state[group])
+
+        # Now fetch any missing groups from the database
+
+        incomplete_groups = incomplete_groups_m | incomplete_groups_nm
+
+        if not incomplete_groups:
+            return state
+
+        cache_sequence_nm = self._state_group_cache.sequence
+        cache_sequence_m = self._state_group_members_cache.sequence
+
+        # Help the cache hit ratio by expanding the filter a bit
+        db_state_filter = state_filter.return_expanded()
+
+        group_to_state_dict = yield self._get_state_groups_from_groups(
+            list(incomplete_groups), state_filter=db_state_filter
+        )
+
+        # Now lets update the caches
+        self._insert_into_cache(
+            group_to_state_dict,
+            db_state_filter,
+            cache_seq_num_members=cache_sequence_m,
+            cache_seq_num_non_members=cache_sequence_nm,
+        )
+
+        # And finally update the result dict, by filtering out any extra
+        # stuff we pulled out of the database.
+        for group, group_state_dict in iteritems(group_to_state_dict):
+            # We just replace any existing entries, as we will have loaded
+            # everything we need from the database anyway.
+            state[group] = state_filter.filter_state(group_state_dict)
+
+        return state
+
+    def _get_state_for_groups_using_cache(self, groups, cache, state_filter):
+        """Gets the state at each of a list of state groups, optionally
+        filtering by type/state_key, querying from a specific cache.
+
+        Args:
+            groups (iterable[int]): list of state groups for which we want
+                to get the state.
+            cache (DictionaryCache): the cache of group ids to state dicts which
+                we will pass through - either the normal state cache or the specific
+                members state cache.
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+
+        Returns:
+            tuple[dict[int, dict[tuple[str, str], str]], set[int]]: Tuple of
+            dict of state_group_id -> (dict of (type, state_key) -> event id)
+            of entries in the cache, and the state group ids either missing
+            from the cache or incomplete.
+        """
+        results = {}
+        incomplete_groups = set()
+        for group in set(groups):
+            state_dict_ids, got_all = self._get_state_for_group_using_cache(
+                cache, group, state_filter
+            )
+            results[group] = state_dict_ids
+
+            if not got_all:
+                incomplete_groups.add(group)
+
+        return results, incomplete_groups
+
+    def _insert_into_cache(
+        self,
+        group_to_state_dict,
+        state_filter,
+        cache_seq_num_members,
+        cache_seq_num_non_members,
+    ):
+        """Inserts results from querying the database into the relevant cache.
+
+        Args:
+            group_to_state_dict (dict): The new entries pulled from database.
+                Map from state group to state dict
+            state_filter (StateFilter): The state filter used to fetch state
+                from the database.
+            cache_seq_num_members (int): Sequence number of member cache since
+                last lookup in cache
+            cache_seq_num_non_members (int): Sequence number of member cache since
+                last lookup in cache
+        """
+
+        # We need to work out which types we've fetched from the DB for the
+        # member vs non-member caches. This should be as accurate as possible,
+        # but can be an underestimate (e.g. when we have wild cards)
+
+        member_filter, non_member_filter = state_filter.get_member_split()
+        if member_filter.is_full():
+            # We fetched all member events
+            member_types = None
+        else:
+            # `concrete_types()` will only return a subset when there are wild
+            # cards in the filter, but that's fine.
+            member_types = member_filter.concrete_types()
+
+        if non_member_filter.is_full():
+            # We fetched all non member events
+            non_member_types = None
+        else:
+            non_member_types = non_member_filter.concrete_types()
+
+        for group, group_state_dict in iteritems(group_to_state_dict):
+            state_dict_members = {}
+            state_dict_non_members = {}
+
+            for k, v in iteritems(group_state_dict):
+                if k[0] == EventTypes.Member:
+                    state_dict_members[k] = v
+                else:
+                    state_dict_non_members[k] = v
+
+            self._state_group_members_cache.update(
+                cache_seq_num_members,
+                key=group,
+                value=state_dict_members,
+                fetched_keys=member_types,
+            )
+
+            self._state_group_cache.update(
+                cache_seq_num_non_members,
+                key=group,
+                value=state_dict_non_members,
+                fetched_keys=non_member_types,
+            )
+
+    def store_state_group(
+        self, event_id, room_id, prev_group, delta_ids, current_state_ids
+    ):
+        """Store a new set of state, returning a newly assigned state group.
+
+        Args:
+            event_id (str): The event ID for which the state was calculated
+            room_id (str)
+            prev_group (int|None): A previous state group for the room, optional.
+            delta_ids (dict|None): The delta between state at `prev_group` and
+                `current_state_ids`, if `prev_group` was given. Same format as
+                `current_state_ids`.
+            current_state_ids (dict): The state to store. Map of (type, state_key)
+                to event_id.
+
+        Returns:
+            Deferred[int]: The state group ID
+        """
+
+        def _store_state_group_txn(txn):
+            if current_state_ids is None:
+                # AFAIK, this can never happen
+                raise Exception("current_state_ids cannot be None")
+
+            state_group = self.database_engine.get_next_state_group_id(txn)
+
+            self.db.simple_insert_txn(
+                txn,
+                table="state_groups",
+                values={"id": state_group, "room_id": room_id, "event_id": event_id},
+            )
+
+            # We persist as a delta if we can, while also ensuring the chain
+            # of deltas isn't tooo long, as otherwise read performance degrades.
+            if prev_group:
+                is_in_db = self.db.simple_select_one_onecol_txn(
+                    txn,
+                    table="state_groups",
+                    keyvalues={"id": prev_group},
+                    retcol="id",
+                    allow_none=True,
+                )
+                if not is_in_db:
+                    raise Exception(
+                        "Trying to persist state with unpersisted prev_group: %r"
+                        % (prev_group,)
+                    )
+
+                potential_hops = self._count_state_group_hops_txn(txn, prev_group)
+            if prev_group and potential_hops < MAX_STATE_DELTA_HOPS:
+                self.db.simple_insert_txn(
+                    txn,
+                    table="state_group_edges",
+                    values={"state_group": state_group, "prev_state_group": prev_group},
+                )
+
+                self.db.simple_insert_many_txn(
+                    txn,
+                    table="state_groups_state",
+                    values=[
+                        {
+                            "state_group": state_group,
+                            "room_id": room_id,
+                            "type": key[0],
+                            "state_key": key[1],
+                            "event_id": state_id,
+                        }
+                        for key, state_id in iteritems(delta_ids)
+                    ],
+                )
+            else:
+                self.db.simple_insert_many_txn(
+                    txn,
+                    table="state_groups_state",
+                    values=[
+                        {
+                            "state_group": state_group,
+                            "room_id": room_id,
+                            "type": key[0],
+                            "state_key": key[1],
+                            "event_id": state_id,
+                        }
+                        for key, state_id in iteritems(current_state_ids)
+                    ],
+                )
+
+            # Prefill the state group caches with this group.
+            # It's fine to use the sequence like this as the state group map
+            # is immutable. (If the map wasn't immutable then this prefill could
+            # race with another update)
+
+            current_member_state_ids = {
+                s: ev
+                for (s, ev) in iteritems(current_state_ids)
+                if s[0] == EventTypes.Member
+            }
+            txn.call_after(
+                self._state_group_members_cache.update,
+                self._state_group_members_cache.sequence,
+                key=state_group,
+                value=dict(current_member_state_ids),
+            )
+
+            current_non_member_state_ids = {
+                s: ev
+                for (s, ev) in iteritems(current_state_ids)
+                if s[0] != EventTypes.Member
+            }
+            txn.call_after(
+                self._state_group_cache.update,
+                self._state_group_cache.sequence,
+                key=state_group,
+                value=dict(current_non_member_state_ids),
+            )
+
+            return state_group
+
+        return self.db.runInteraction("store_state_group", _store_state_group_txn)
+
+    def purge_unreferenced_state_groups(
+        self, room_id: str, state_groups_to_delete
+    ) -> defer.Deferred:
+        """Deletes no longer referenced state groups and de-deltas any state
+        groups that reference them.
+
+        Args:
+            room_id: The room the state groups belong to (must all be in the
+                same room).
+            state_groups_to_delete (Collection[int]): Set of all state groups
+                to delete.
+        """
+
+        return self.db.runInteraction(
+            "purge_unreferenced_state_groups",
+            self._purge_unreferenced_state_groups,
+            room_id,
+            state_groups_to_delete,
+        )
+
+    def _purge_unreferenced_state_groups(self, txn, room_id, state_groups_to_delete):
+        logger.info(
+            "[purge] found %i state groups to delete", len(state_groups_to_delete)
+        )
+
+        rows = self.db.simple_select_many_txn(
+            txn,
+            table="state_group_edges",
+            column="prev_state_group",
+            iterable=state_groups_to_delete,
+            keyvalues={},
+            retcols=("state_group",),
+        )
+
+        remaining_state_groups = set(
+            row["state_group"]
+            for row in rows
+            if row["state_group"] not in state_groups_to_delete
+        )
+
+        logger.info(
+            "[purge] de-delta-ing %i remaining state groups",
+            len(remaining_state_groups),
+        )
+
+        # Now we turn the state groups that reference to-be-deleted state
+        # groups to non delta versions.
+        for sg in remaining_state_groups:
+            logger.info("[purge] de-delta-ing remaining state group %s", sg)
+            curr_state = self._get_state_groups_from_groups_txn(txn, [sg])
+            curr_state = curr_state[sg]
+
+            self.db.simple_delete_txn(
+                txn, table="state_groups_state", keyvalues={"state_group": sg}
+            )
+
+            self.db.simple_delete_txn(
+                txn, table="state_group_edges", keyvalues={"state_group": sg}
+            )
+
+            self.db.simple_insert_many_txn(
+                txn,
+                table="state_groups_state",
+                values=[
+                    {
+                        "state_group": sg,
+                        "room_id": room_id,
+                        "type": key[0],
+                        "state_key": key[1],
+                        "event_id": state_id,
+                    }
+                    for key, state_id in iteritems(curr_state)
+                ],
+            )
+
+        logger.info("[purge] removing redundant state groups")
+        txn.executemany(
+            "DELETE FROM state_groups_state WHERE state_group = ?",
+            ((sg,) for sg in state_groups_to_delete),
+        )
+        txn.executemany(
+            "DELETE FROM state_groups WHERE id = ?",
+            ((sg,) for sg in state_groups_to_delete),
+        )
+
+    @defer.inlineCallbacks
+    def get_previous_state_groups(self, state_groups):
+        """Fetch the previous groups of the given state groups.
+
+        Args:
+            state_groups (Iterable[int])
+
+        Returns:
+            Deferred[dict[int, int]]: mapping from state group to previous
+            state group.
+        """
+
+        rows = yield self.db.simple_select_many_batch(
+            table="state_group_edges",
+            column="prev_state_group",
+            iterable=state_groups,
+            keyvalues={},
+            retcols=("prev_state_group", "state_group"),
+            desc="get_previous_state_groups",
+        )
+
+        return {row["state_group"]: row["prev_state_group"] for row in rows}
+
+    def purge_room_state(self, room_id, state_groups_to_delete):
+        """Deletes all record of a room from state tables
+
+        Args:
+            room_id (str):
+            state_groups_to_delete (list[int]): State groups to delete
+        """
+
+        return self.db.runInteraction(
+            "purge_room_state",
+            self._purge_room_state_txn,
+            room_id,
+            state_groups_to_delete,
+        )
+
+    def _purge_room_state_txn(self, txn, room_id, state_groups_to_delete):
+        # first we have to delete the state groups states
+        logger.info("[purge] removing %s from state_groups_state", room_id)
+
+        self.db.simple_delete_many_txn(
+            txn,
+            table="state_groups_state",
+            column="state_group",
+            iterable=state_groups_to_delete,
+            keyvalues={},
+        )
+
+        # ... and the state group edges
+        logger.info("[purge] removing %s from state_group_edges", room_id)
+
+        self.db.simple_delete_many_txn(
+            txn,
+            table="state_group_edges",
+            column="state_group",
+            iterable=state_groups_to_delete,
+            keyvalues={},
+        )
+
+        # ... and the state groups
+        logger.info("[purge] removing %s from state_groups", room_id)
+
+        self.db.simple_delete_many_txn(
+            txn,
+            table="state_groups",
+            column="id",
+            iterable=state_groups_to_delete,
+            keyvalues={},
+        )