diff options
author | Richard van der Hoff <github@rvanderhoff.org.uk> | 2018-02-09 14:11:00 +0000 |
---|---|---|
committer | GitHub <noreply@github.com> | 2018-02-09 14:11:00 +0000 |
commit | 10b34dbb9abcc0800db84dae74a244caaea54e2d (patch) | |
tree | aedd171b7d6973ea2f4c76dfb6e8610737da9e4c /synapse/storage/events.py | |
parent | Merge pull request #2856 from matrix-org/erikj/remove_ratelimit (diff) | |
parent | purge: move room_depth update to end (diff) | |
download | synapse-10b34dbb9abcc0800db84dae74a244caaea54e2d.tar.xz |
Merge pull request #2858 from matrix-org/rav/purge_updates
delete_local_events for purge_room_history
Diffstat (limited to 'synapse/storage/events.py')
-rw-r--r-- | synapse/storage/events.py | 94 |
1 files changed, 61 insertions, 33 deletions
diff --git a/synapse/storage/events.py b/synapse/storage/events.py index af56f1ee57..86a7c5920d 100644 --- a/synapse/storage/events.py +++ b/synapse/storage/events.py @@ -2063,16 +2063,32 @@ class EventsStore(SQLBaseStore): ) return self.runInteraction("get_all_new_events", get_all_new_events_txn) - def delete_old_state(self, room_id, topological_ordering): - return self.runInteraction( - "delete_old_state", - self._delete_old_state_txn, room_id, topological_ordering - ) + def purge_history( + self, room_id, topological_ordering, delete_local_events, + ): + """Deletes room history before a certain point + + Args: + room_id (str): + + topological_ordering (int): + minimum topo ordering to preserve - def _delete_old_state_txn(self, txn, room_id, topological_ordering): - """Deletes old room state + delete_local_events (bool): + if True, we will delete local events as well as remote ones + (instead of just marking them as outliers and deleting their + state groups). """ + return self.runInteraction( + "purge_history", + self._purge_history_txn, room_id, topological_ordering, + delete_local_events, + ) + + def _purge_history_txn( + self, txn, room_id, topological_ordering, delete_local_events, + ): # Tables that should be pruned: # event_auth # event_backward_extremities @@ -2113,7 +2129,7 @@ class EventsStore(SQLBaseStore): 400, "topological_ordering is greater than forward extremeties" ) - logger.debug("[purge] looking for events to delete") + logger.info("[purge] looking for events to delete") txn.execute( "SELECT event_id, state_key FROM events" @@ -2125,16 +2141,16 @@ class EventsStore(SQLBaseStore): to_delete = [ (event_id,) for event_id, state_key in event_rows - if state_key is None and not self.hs.is_mine_id(event_id) + if state_key is None and ( + delete_local_events or not self.hs.is_mine_id(event_id) + ) ] logger.info( - "[purge] found %i events before cutoff, of which %i are remote" - " non-state events to delete", len(event_rows), len(to_delete)) - - for event_id, state_key in event_rows: - txn.call_after(self._get_state_group_for_event.invalidate, (event_id,)) + "[purge] found %i events before cutoff, of which %i can be deleted", + len(event_rows), len(to_delete), + ) - logger.debug("[purge] Finding new backward extremities") + logger.info("[purge] Finding new backward extremities") # We calculate the new entries for the backward extremeties by finding # all events that point to events that are to be purged @@ -2148,7 +2164,7 @@ class EventsStore(SQLBaseStore): ) new_backwards_extrems = txn.fetchall() - logger.debug("[purge] replacing backward extremities: %r", new_backwards_extrems) + logger.info("[purge] replacing backward extremities: %r", new_backwards_extrems) txn.execute( "DELETE FROM event_backward_extremities WHERE room_id = ?", @@ -2164,7 +2180,7 @@ class EventsStore(SQLBaseStore): ] ) - logger.debug("[purge] finding redundant state groups") + logger.info("[purge] finding redundant state groups") # Get all state groups that are only referenced by events that are # to be deleted. @@ -2181,15 +2197,15 @@ class EventsStore(SQLBaseStore): ) state_rows = txn.fetchall() - logger.debug("[purge] found %i redundant state groups", len(state_rows)) + logger.info("[purge] found %i redundant state groups", len(state_rows)) # make a set of the redundant state groups, so that we can look them up # efficiently state_groups_to_delete = set([sg for sg, in state_rows]) # Now we get all the state groups that rely on these state groups - logger.debug("[purge] finding state groups which depend on redundant" - " state groups") + logger.info("[purge] finding state groups which depend on redundant" + " state groups") remaining_state_groups = [] for i in xrange(0, len(state_rows), 100): chunk = [sg for sg, in state_rows[i:i + 100]] @@ -2214,7 +2230,7 @@ class EventsStore(SQLBaseStore): # Now we turn the state groups that reference to-be-deleted state # groups to non delta versions. for sg in remaining_state_groups: - logger.debug("[purge] de-delta-ing remaining state group %s", sg) + logger.info("[purge] de-delta-ing remaining state group %s", sg) curr_state = self._get_state_groups_from_groups_txn( txn, [sg], types=None ) @@ -2251,7 +2267,7 @@ class EventsStore(SQLBaseStore): ], ) - logger.debug("[purge] removing redundant state groups") + logger.info("[purge] removing redundant state groups") txn.executemany( "DELETE FROM state_groups_state WHERE state_group = ?", state_rows @@ -2261,18 +2277,15 @@ class EventsStore(SQLBaseStore): state_rows ) - # Delete all non-state - logger.debug("[purge] removing events from event_to_state_groups") + logger.info("[purge] removing events from event_to_state_groups") txn.executemany( "DELETE FROM event_to_state_groups WHERE event_id = ?", [(event_id,) for event_id, _ in event_rows] ) - - logger.debug("[purge] updating room_depth") - txn.execute( - "UPDATE room_depth SET min_depth = ? WHERE room_id = ?", - (topological_ordering, room_id,) - ) + for event_id, _ in event_rows: + txn.call_after(self._get_state_group_for_event.invalidate, ( + event_id, + )) # Delete all remote non-state events for table in ( @@ -2290,7 +2303,8 @@ class EventsStore(SQLBaseStore): "event_signatures", "rejections", ): - logger.debug("[purge] removing remote non-state events from %s", table) + logger.info("[purge] removing remote non-state events from %s", + table) txn.executemany( "DELETE FROM %s WHERE event_id = ?" % (table,), @@ -2298,16 +2312,30 @@ class EventsStore(SQLBaseStore): ) # Mark all state and own events as outliers - logger.debug("[purge] marking remaining events as outliers") + logger.info("[purge] marking remaining events as outliers") txn.executemany( "UPDATE events SET outlier = ?" " WHERE event_id = ?", [ (True, event_id,) for event_id, state_key in event_rows - if state_key is not None or self.hs.is_mine_id(event_id) + if state_key is not None or ( + not delete_local_events and self.hs.is_mine_id(event_id) + ) ] ) + # synapse tries to take out an exclusive lock on room_depth whenever it + # persists events (because upsert), and once we run this update, we + # will block that for the rest of our transaction. + # + # So, let's stick it at the end so that we don't block event + # persistence. + logger.info("[purge] updating room_depth") + txn.execute( + "UPDATE room_depth SET min_depth = ? WHERE room_id = ?", + (topological_ordering, room_id,) + ) + logger.info("[purge] done") @defer.inlineCallbacks |