synapse-old/synapse/storage/search.py

491 lines
17 KiB
Python
Raw Normal View History

2015-10-09 08:48:31 -06:00
# -*- coding: utf-8 -*-
2016-01-06 21:26:29 -07:00
# Copyright 2015, 2016 OpenMarket Ltd
2015-10-09 08:48:31 -06:00
#
# 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 twisted.internet import defer
from .background_updates import BackgroundUpdateStore
2015-11-05 09:10:54 -07:00
from synapse.api.errors import SynapseError
2015-10-16 07:37:14 -06:00
from synapse.storage.engines import PostgresEngine, Sqlite3Engine
2015-10-09 08:48:31 -06:00
2015-11-04 10:57:44 -07:00
import logging
import re
2015-11-04 10:57:44 -07:00
logger = logging.getLogger(__name__)
class SearchStore(BackgroundUpdateStore):
EVENT_SEARCH_UPDATE_NAME = "event_search"
def __init__(self, hs):
super(SearchStore, self).__init__(hs)
self.register_background_update_handler(
self.EVENT_SEARCH_UPDATE_NAME, self._background_reindex_search
)
@defer.inlineCallbacks
def _background_reindex_search(self, progress, batch_size):
target_min_stream_id = progress["target_min_stream_id_inclusive"]
max_stream_id = progress["max_stream_id_exclusive"]
rows_inserted = progress.get("rows_inserted", 0)
INSERT_CLUMP_SIZE = 1000
TYPES = ["m.room.name", "m.room.message", "m.room.topic"]
def reindex_search_txn(txn):
sql = (
2015-11-11 06:59:40 -07:00
"SELECT stream_ordering, event_id FROM events"
" WHERE ? <= stream_ordering AND stream_ordering < ?"
" AND (%s)"
" ORDER BY stream_ordering DESC"
" LIMIT ?"
2015-11-11 06:59:40 -07:00
) % (" OR ".join("type = '%s'" % (t,) for t in TYPES),)
2015-11-11 06:59:40 -07:00
txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size))
2015-11-11 06:59:40 -07:00
rows = txn.fetchall()
if not rows:
2015-11-11 06:59:40 -07:00
return 0
min_stream_id = rows[-1][0]
event_ids = [row[1] for row in rows]
events = self._get_events_txn(txn, event_ids)
event_search_rows = []
for event in events:
try:
event_id = event.event_id
room_id = event.room_id
content = event.content
if event.type == "m.room.message":
key = "content.body"
value = content["body"]
elif event.type == "m.room.topic":
key = "content.topic"
value = content["topic"]
elif event.type == "m.room.name":
key = "content.name"
value = content["name"]
except (KeyError, AttributeError):
# If the event is missing a necessary field then
# skip over it.
continue
if not isinstance(value, basestring):
# If the event body, name or topic isn't a string
# then skip over it
continue
event_search_rows.append((event_id, room_id, key, value))
if isinstance(self.database_engine, PostgresEngine):
sql = (
"INSERT INTO event_search (event_id, room_id, key, vector)"
" VALUES (?,?,?,to_tsvector('english', ?))"
)
elif isinstance(self.database_engine, Sqlite3Engine):
sql = (
"INSERT INTO event_search (event_id, room_id, key, value)"
" VALUES (?,?,?,?)"
)
else:
# This should be unreachable.
raise Exception("Unrecognized database engine")
for index in range(0, len(event_search_rows), INSERT_CLUMP_SIZE):
clump = event_search_rows[index:index + INSERT_CLUMP_SIZE]
2015-11-11 06:59:40 -07:00
txn.executemany(sql, clump)
progress = {
"target_min_stream_id_inclusive": target_min_stream_id,
"max_stream_id_exclusive": min_stream_id,
"rows_inserted": rows_inserted + len(event_search_rows)
}
self._background_update_progress_txn(
txn, self.EVENT_SEARCH_UPDATE_NAME, progress
)
return len(event_search_rows)
2015-11-11 06:59:40 -07:00
result = yield self.runInteraction(
self.EVENT_SEARCH_UPDATE_NAME, reindex_search_txn
)
2015-11-11 06:59:40 -07:00
if not result:
yield self._end_background_update(self.EVENT_SEARCH_UPDATE_NAME)
defer.returnValue(result)
2015-10-09 08:48:31 -06:00
@defer.inlineCallbacks
def search_msgs(self, room_ids, search_term, keys):
2015-10-16 09:46:48 -06:00
"""Performs a full text search over events with given keys.
2015-10-16 04:28:12 -06:00
Args:
room_ids (list): List of room ids to search in
search_term (str): Search term to search for
keys (list): List of keys to search in, currently supports
2015-10-16 09:46:48 -06:00
"content.body", "content.name", "content.topic"
2015-10-16 04:28:12 -06:00
Returns:
list of dicts
2015-10-16 04:28:12 -06:00
"""
2015-10-09 08:48:31 -06:00
clauses = []
2015-12-02 06:28:13 -07:00
search_query = search_query = _parse_query(self.database_engine, search_term)
2015-12-11 04:12:57 -07:00
args = []
2015-10-09 08:48:31 -06:00
2015-10-22 06:25:22 -06:00
# Make sure we don't explode because the person is in too many rooms.
2015-10-22 09:54:56 -06:00
# We filter the results below regardless.
2015-10-22 09:18:35 -06:00
if len(room_ids) < 500:
2015-10-22 06:25:22 -06:00
clauses.append(
"room_id IN (%s)" % (",".join(["?"] * len(room_ids)),)
)
args.extend(room_ids)
local_clauses = []
for key in keys:
local_clauses.append("key = ?")
args.append(key)
clauses.append(
"(%s)" % (" OR ".join(local_clauses),)
)
2015-10-09 08:48:31 -06:00
count_args = args
count_clauses = clauses
2015-10-13 06:47:50 -06:00
if isinstance(self.database_engine, PostgresEngine):
sql = (
2015-12-11 04:12:57 -07:00
"SELECT ts_rank_cd(vector, to_tsquery('english', ?)) AS rank,"
" room_id, event_id"
" FROM event_search"
" WHERE vector @@ to_tsquery('english', ?)"
2015-10-13 06:47:50 -06:00
)
2015-12-11 04:12:57 -07:00
args = [search_query, search_query] + args
count_sql = (
"SELECT room_id, count(*) as count FROM event_search"
" WHERE vector @@ to_tsquery('english', ?)"
)
count_args = [search_query] + count_args
2015-10-16 07:37:14 -06:00
elif isinstance(self.database_engine, Sqlite3Engine):
2015-10-13 06:47:50 -06:00
sql = (
2015-10-23 06:23:48 -06:00
"SELECT rank(matchinfo(event_search)) as rank, room_id, event_id"
" FROM event_search"
2015-10-13 06:47:50 -06:00
" WHERE value MATCH ?"
)
2015-12-11 04:12:57 -07:00
args = [search_query] + args
count_sql = (
"SELECT room_id, count(*) as count FROM event_search"
" WHERE value MATCH ?"
)
count_args = [search_term] + count_args
2015-10-16 07:37:14 -06:00
else:
# This should be unreachable.
raise Exception("Unrecognized database engine")
2015-10-09 08:48:31 -06:00
for clause in clauses:
sql += " AND " + clause
for clause in count_clauses:
count_sql += " AND " + clause
2015-10-16 04:24:02 -06:00
# We add an arbitrary limit here to ensure we don't try to pull the
# entire table from the database.
2015-10-13 08:50:56 -06:00
sql += " ORDER BY rank DESC LIMIT 500"
2015-10-09 08:48:31 -06:00
results = yield self._execute(
2015-12-02 04:38:51 -07:00
"search_msgs", self.cursor_to_dict, sql, *args
2015-10-09 08:48:31 -06:00
)
2015-10-22 09:54:56 -06:00
results = filter(lambda row: row["room_id"] in room_ids, results)
2015-10-09 08:48:31 -06:00
events = yield self._get_events([r["event_id"] for r in results])
event_map = {
ev.event_id: ev
for ev in events
}
highlights = None
if isinstance(self.database_engine, PostgresEngine):
2015-12-02 06:28:13 -07:00
highlights = yield self._find_highlights_in_postgres(search_query, events)
count_sql += " GROUP BY room_id"
count_results = yield self._execute(
"search_rooms_count", self.cursor_to_dict, count_sql, *count_args
)
count = sum(row["count"] for row in count_results if row["room_id"] in room_ids)
defer.returnValue({
"results": [
{
"event": event_map[r["event_id"]],
"rank": r["rank"],
}
for r in results
if r["event_id"] in event_map
],
"highlights": highlights,
"count": count,
})
2015-11-04 10:57:44 -07:00
@defer.inlineCallbacks
def search_rooms(self, room_ids, search_term, keys, limit, pagination_token=None):
2015-11-04 10:57:44 -07:00
"""Performs a full text search over events with given keys.
Args:
room_id (list): The room_ids to search in
2015-11-04 10:57:44 -07:00
search_term (str): Search term to search for
keys (list): List of keys to search in, currently supports
"content.body", "content.name", "content.topic"
pagination_token (str): A pagination token previously returned
Returns:
list of dicts
2015-11-04 10:57:44 -07:00
"""
clauses = []
2015-12-02 04:38:51 -07:00
2015-12-02 06:28:13 -07:00
search_query = search_query = _parse_query(self.database_engine, search_term)
2015-12-11 04:12:57 -07:00
args = []
# Make sure we don't explode because the person is in too many rooms.
# We filter the results below regardless.
if len(room_ids) < 500:
clauses.append(
"room_id IN (%s)" % (",".join(["?"] * len(room_ids)),)
)
args.extend(room_ids)
2015-11-04 10:57:44 -07:00
local_clauses = []
for key in keys:
local_clauses.append("key = ?")
args.append(key)
clauses.append(
"(%s)" % (" OR ".join(local_clauses),)
)
# take copies of the current args and clauses lists, before adding
# pagination clauses to main query.
count_args = list(args)
count_clauses = list(clauses)
2015-11-04 10:57:44 -07:00
if pagination_token:
2015-11-05 09:10:54 -07:00
try:
origin_server_ts, stream = pagination_token.split(",")
origin_server_ts = int(origin_server_ts)
2015-11-05 09:10:54 -07:00
stream = int(stream)
except:
raise SynapseError(400, "Invalid pagination token")
2015-11-04 10:57:44 -07:00
clauses.append(
"(origin_server_ts < ?"
" OR (origin_server_ts = ? AND stream_ordering < ?))"
2015-11-04 10:57:44 -07:00
)
args.extend([origin_server_ts, origin_server_ts, stream])
2015-11-04 10:57:44 -07:00
if isinstance(self.database_engine, PostgresEngine):
sql = (
2015-12-11 04:12:57 -07:00
"SELECT ts_rank_cd(vector, to_tsquery('english', ?)) as rank,"
" origin_server_ts, stream_ordering, room_id, event_id"
2015-12-11 04:12:57 -07:00
" FROM event_search"
2015-11-04 10:57:44 -07:00
" NATURAL JOIN events"
2015-12-11 04:12:57 -07:00
" WHERE vector @@ to_tsquery('english', ?) AND "
2015-11-04 10:57:44 -07:00
)
args = [search_query, search_query] + args
count_sql = (
"SELECT room_id, count(*) as count FROM event_search"
" WHERE vector @@ to_tsquery('english', ?) AND "
)
count_args = [search_query] + count_args
2015-11-04 10:57:44 -07:00
elif isinstance(self.database_engine, Sqlite3Engine):
2015-11-12 08:19:56 -07:00
# We use CROSS JOIN here to ensure we use the right indexes.
# https://sqlite.org/optoverview.html#crossjoin
2015-11-12 08:33:47 -07:00
#
# We want to use the full text search index on event_search to
# extract all possible matches first, then lookup those matches
# in the events table to get the topological ordering. We need
2015-11-12 08:36:43 -07:00
# to use the indexes in this order because sqlite refuses to
2015-11-12 08:33:47 -07:00
# MATCH unless it uses the full text search index
2015-11-04 10:57:44 -07:00
sql = (
"SELECT rank(matchinfo) as rank, room_id, event_id,"
" origin_server_ts, stream_ordering"
2015-11-12 08:09:45 -07:00
" FROM (SELECT key, event_id, matchinfo(event_search) as matchinfo"
" FROM event_search"
" WHERE value MATCH ?"
" )"
" CROSS JOIN events USING (event_id)"
" WHERE "
2015-11-04 10:57:44 -07:00
)
args = [search_query] + args
count_sql = (
"SELECT room_id, count(*) as count FROM event_search"
" WHERE value MATCH ? AND "
)
count_args = [search_term] + count_args
2015-11-04 10:57:44 -07:00
else:
# This should be unreachable.
raise Exception("Unrecognized database engine")
sql += " AND ".join(clauses)
count_sql += " AND ".join(count_clauses)
2015-11-04 10:57:44 -07:00
# We add an arbitrary limit here to ensure we don't try to pull the
# entire table from the database.
sql += " ORDER BY origin_server_ts DESC, stream_ordering DESC LIMIT ?"
2015-11-04 10:57:44 -07:00
args.append(limit)
results = yield self._execute(
"search_rooms", self.cursor_to_dict, sql, *args
)
results = filter(lambda row: row["room_id"] in room_ids, results)
2015-11-04 10:57:44 -07:00
events = yield self._get_events([r["event_id"] for r in results])
event_map = {
ev.event_id: ev
for ev in events
}
highlights = None
if isinstance(self.database_engine, PostgresEngine):
2015-12-02 06:28:13 -07:00
highlights = yield self._find_highlights_in_postgres(search_query, events)
count_sql += " GROUP BY room_id"
count_results = yield self._execute(
"search_rooms_count", self.cursor_to_dict, count_sql, *count_args
)
count = sum(row["count"] for row in count_results if row["room_id"] in room_ids)
defer.returnValue({
"results": [
{
"event": event_map[r["event_id"]],
"rank": r["rank"],
"pagination_token": "%s,%s" % (
r["origin_server_ts"], r["stream_ordering"]
),
}
for r in results
if r["event_id"] in event_map
],
"highlights": highlights,
"count": count,
})
2015-12-02 06:28:13 -07:00
def _find_highlights_in_postgres(self, search_query, events):
"""Given a list of events and a search term, return a list of words
that match from the content of the event.
This is used to give a list of words that clients can match against to
highlight the matching parts.
Args:
2015-12-02 06:28:13 -07:00
search_query (str)
events (list): A list of events
Returns:
deferred : A set of strings.
"""
def f(txn):
highlight_words = set()
for event in events:
# As a hack we simply join values of all possible keys. This is
# fine since we're only using them to find possible highlights.
values = []
for key in ("body", "name", "topic"):
v = event.content.get(key, None)
if v:
values.append(v)
if not values:
continue
value = " ".join(values)
# We need to find some values for StartSel and StopSel that
# aren't in the value so that we can pick results out.
start_sel = "<"
stop_sel = ">"
while start_sel in value:
start_sel += "<"
while stop_sel in value:
stop_sel += ">"
2015-12-02 06:28:13 -07:00
query = "SELECT ts_headline(?, to_tsquery('english', ?), %s)" % (
_to_postgres_options({
"StartSel": start_sel,
"StopSel": stop_sel,
"MaxFragments": "50",
})
)
2015-12-02 06:28:13 -07:00
txn.execute(query, (value, search_query,))
headline, = txn.fetchall()[0]
# Now we need to pick the possible highlights out of the haedline
# result.
matcher_regex = "%s(.*?)%s" % (
re.escape(start_sel),
re.escape(stop_sel),
)
res = re.findall(matcher_regex, headline)
highlight_words.update([r.lower() for r in res])
return highlight_words
return self.runInteraction("_find_highlights", f)
def _to_postgres_options(options_dict):
return "'%s'" % (
",".join("%s=%s" % (k, v) for k, v in options_dict.items()),
)
2015-12-02 04:38:51 -07:00
def _parse_query(database_engine, search_term):
2015-12-02 04:38:51 -07:00
"""Takes a plain unicode string from the user and converts it into a form
that can be passed to database.
We use this so that we can add prefix matching, which isn't something
that is supported by default.
2015-12-02 04:38:51 -07:00
"""
# Pull out the individual words, discarding any non-word characters.
2015-12-02 04:38:51 -07:00
results = re.findall(r"([\w\-]+)", search_term, re.UNICODE)
if isinstance(database_engine, PostgresEngine):
return " & ".join(result + ":*" for result in results)
2015-12-02 06:50:43 -07:00
elif isinstance(database_engine, Sqlite3Engine):
return " & ".join(result + "*" for result in results)
2015-12-02 06:50:43 -07:00
else:
# This should be unreachable.
raise Exception("Unrecognized database engine")