2016-06-03 15:42:35 +02:00
|
|
|
# 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.
|
|
|
|
|
2016-07-20 17:34:00 +02:00
|
|
|
import logging
|
2021-10-12 14:50:34 +02:00
|
|
|
from typing import TYPE_CHECKING, Dict, List, Mapping, Optional, Tuple, Union, cast
|
|
|
|
|
|
|
|
from typing_extensions import TypedDict
|
2016-07-22 14:14:03 +02:00
|
|
|
|
2019-09-24 16:20:40 +02:00
|
|
|
from synapse.metrics.background_process_metrics import wrap_as_background_process
|
2019-12-04 16:09:36 +01:00
|
|
|
from synapse.storage._base import SQLBaseStore
|
2021-10-12 14:50:34 +02:00
|
|
|
from synapse.storage.database import (
|
|
|
|
DatabasePool,
|
|
|
|
LoggingDatabaseConnection,
|
|
|
|
LoggingTransaction,
|
|
|
|
make_tuple_comparison_clause,
|
|
|
|
)
|
|
|
|
from synapse.storage.databases.main.monthly_active_users import MonthlyActiveUsersStore
|
|
|
|
from synapse.types import JsonDict, UserID
|
2020-10-19 13:20:29 +02:00
|
|
|
from synapse.util.caches.lrucache import LruCache
|
2017-07-04 10:56:44 +02:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
if TYPE_CHECKING:
|
|
|
|
from synapse.server import HomeServer
|
|
|
|
|
2016-07-20 17:34:00 +02:00
|
|
|
logger = logging.getLogger(__name__)
|
2016-06-03 15:42:35 +02:00
|
|
|
|
|
|
|
# Number of msec of granularity to store the user IP 'last seen' time. Smaller
|
|
|
|
# times give more inserts into the database even for readonly API hits
|
|
|
|
# 120 seconds == 2 minutes
|
|
|
|
LAST_SEEN_GRANULARITY = 120 * 1000
|
|
|
|
|
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
class DeviceLastConnectionInfo(TypedDict):
|
|
|
|
"""Metadata for the last connection seen for a user and device combination"""
|
|
|
|
|
|
|
|
# These types must match the columns in the `devices` table
|
|
|
|
user_id: str
|
|
|
|
device_id: str
|
|
|
|
|
|
|
|
ip: Optional[str]
|
|
|
|
user_agent: Optional[str]
|
|
|
|
last_seen: Optional[int]
|
|
|
|
|
|
|
|
|
|
|
|
class LastConnectionInfo(TypedDict):
|
|
|
|
"""Metadata for the last connection seen for an access token and IP combination"""
|
|
|
|
|
|
|
|
# These types must match the columns in the `user_ips` table
|
|
|
|
access_token: str
|
|
|
|
ip: str
|
|
|
|
|
|
|
|
user_agent: str
|
|
|
|
last_seen: int
|
|
|
|
|
|
|
|
|
2019-12-04 16:09:36 +01:00
|
|
|
class ClientIpBackgroundUpdateStore(SQLBaseStore):
|
2021-12-13 18:05:00 +01:00
|
|
|
def __init__(
|
|
|
|
self,
|
|
|
|
database: DatabasePool,
|
|
|
|
db_conn: LoggingDatabaseConnection,
|
|
|
|
hs: "HomeServer",
|
|
|
|
):
|
2020-09-18 15:56:44 +02:00
|
|
|
super().__init__(database, db_conn, hs)
|
2016-06-03 15:42:35 +02:00
|
|
|
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates.register_background_index_update(
|
2016-07-22 14:14:03 +02:00
|
|
|
"user_ips_device_index",
|
|
|
|
index_name="user_ips_device_id",
|
|
|
|
table="user_ips",
|
|
|
|
columns=["user_id", "device_id", "last_seen"],
|
|
|
|
)
|
|
|
|
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates.register_background_index_update(
|
2018-03-28 13:03:13 +02:00
|
|
|
"user_ips_last_seen_index",
|
|
|
|
index_name="user_ips_last_seen",
|
|
|
|
table="user_ips",
|
|
|
|
columns=["user_id", "last_seen"],
|
|
|
|
)
|
|
|
|
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates.register_background_index_update(
|
2018-04-25 18:37:29 +02:00
|
|
|
"user_ips_last_seen_only_index",
|
|
|
|
index_name="user_ips_last_seen_only",
|
|
|
|
table="user_ips",
|
|
|
|
columns=["last_seen"],
|
|
|
|
)
|
|
|
|
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates.register_background_update_handler(
|
2019-04-03 11:07:29 +02:00
|
|
|
"user_ips_analyze", self._analyze_user_ip
|
2019-02-12 12:55:27 +01:00
|
|
|
)
|
|
|
|
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates.register_background_update_handler(
|
2019-04-03 11:07:29 +02:00
|
|
|
"user_ips_remove_dupes", self._remove_user_ip_dupes
|
2019-01-11 20:21:50 +01:00
|
|
|
)
|
|
|
|
|
|
|
|
# Register a unique index
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates.register_background_index_update(
|
2019-01-11 20:21:50 +01:00
|
|
|
"user_ips_device_unique_index",
|
|
|
|
index_name="user_ips_user_token_ip_unique_index",
|
|
|
|
table="user_ips",
|
|
|
|
columns=["user_id", "access_token", "ip"],
|
|
|
|
unique=True,
|
|
|
|
)
|
|
|
|
|
|
|
|
# Drop the old non-unique index
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates.register_background_update_handler(
|
2019-04-03 11:07:29 +02:00
|
|
|
"user_ips_drop_nonunique_index", self._remove_user_ip_nonunique
|
2019-01-11 20:21:50 +01:00
|
|
|
)
|
|
|
|
|
2019-09-23 16:59:43 +02:00
|
|
|
# Update the last seen info in devices.
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates.register_background_update_handler(
|
2019-09-23 16:59:43 +02:00
|
|
|
"devices_last_seen", self._devices_last_seen_update
|
|
|
|
)
|
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
async def _remove_user_ip_nonunique(
|
|
|
|
self, progress: JsonDict, batch_size: int
|
|
|
|
) -> int:
|
|
|
|
def f(conn: LoggingDatabaseConnection) -> None:
|
2019-01-11 20:21:50 +01:00
|
|
|
txn = conn.cursor()
|
2019-04-03 11:07:29 +02:00
|
|
|
txn.execute("DROP INDEX IF EXISTS user_ips_user_ip")
|
2019-01-11 20:21:50 +01:00
|
|
|
txn.close()
|
|
|
|
|
2020-08-07 18:17:17 +02:00
|
|
|
await self.db_pool.runWithConnection(f)
|
|
|
|
await self.db_pool.updates._end_background_update(
|
2020-08-05 22:38:57 +02:00
|
|
|
"user_ips_drop_nonunique_index"
|
|
|
|
)
|
2019-07-23 15:00:55 +02:00
|
|
|
return 1
|
2019-01-11 20:21:50 +01:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
async def _analyze_user_ip(self, progress: JsonDict, batch_size: int) -> int:
|
2019-02-12 12:55:27 +01:00
|
|
|
# Background update to analyze user_ips table before we run the
|
|
|
|
# deduplication background update. The table may not have been analyzed
|
|
|
|
# for ages due to the table locks.
|
|
|
|
#
|
|
|
|
# This will lock out the naive upserts to user_ips while it happens, but
|
|
|
|
# the analyze should be quick (28GB table takes ~10s)
|
2021-10-12 14:50:34 +02:00
|
|
|
def user_ips_analyze(txn: LoggingTransaction) -> None:
|
2019-02-12 12:55:27 +01:00
|
|
|
txn.execute("ANALYZE user_ips")
|
|
|
|
|
2020-08-07 18:17:17 +02:00
|
|
|
await self.db_pool.runInteraction("user_ips_analyze", user_ips_analyze)
|
2019-02-12 12:55:27 +01:00
|
|
|
|
2020-08-07 18:17:17 +02:00
|
|
|
await self.db_pool.updates._end_background_update("user_ips_analyze")
|
2019-02-12 12:55:27 +01:00
|
|
|
|
2019-07-23 15:00:55 +02:00
|
|
|
return 1
|
2019-02-12 12:55:27 +01:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
async def _remove_user_ip_dupes(self, progress: JsonDict, batch_size: int) -> int:
|
2019-01-22 12:55:53 +01:00
|
|
|
# This works function works by scanning the user_ips table in batches
|
|
|
|
# based on `last_seen`. For each row in a batch it searches the rest of
|
|
|
|
# the table to see if there are any duplicates, if there are then they
|
|
|
|
# are removed and replaced with a suitable row.
|
2019-01-11 20:21:50 +01:00
|
|
|
|
2019-01-22 12:55:53 +01:00
|
|
|
# Fetch the start of the batch
|
2021-10-12 14:50:34 +02:00
|
|
|
begin_last_seen: int = progress.get("last_seen", 0)
|
2019-01-11 20:21:50 +01:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
def get_last_seen(txn: LoggingTransaction) -> Optional[int]:
|
2019-01-11 20:21:50 +01:00
|
|
|
txn.execute(
|
|
|
|
"""
|
|
|
|
SELECT last_seen FROM user_ips
|
|
|
|
WHERE last_seen > ?
|
|
|
|
ORDER BY last_seen
|
|
|
|
LIMIT 1
|
|
|
|
OFFSET ?
|
|
|
|
""",
|
2019-04-03 11:07:29 +02:00
|
|
|
(begin_last_seen, batch_size),
|
2019-01-11 20:21:50 +01:00
|
|
|
)
|
2021-10-12 14:50:34 +02:00
|
|
|
row = cast(Optional[Tuple[int]], txn.fetchone())
|
2019-01-22 12:55:53 +01:00
|
|
|
if row:
|
|
|
|
return row[0]
|
|
|
|
else:
|
|
|
|
return None
|
|
|
|
|
|
|
|
# Get a last seen that has roughly `batch_size` since `begin_last_seen`
|
2020-08-07 18:17:17 +02:00
|
|
|
end_last_seen = await self.db_pool.runInteraction(
|
2019-01-11 20:21:50 +01:00
|
|
|
"user_ips_dups_get_last_seen", get_last_seen
|
|
|
|
)
|
|
|
|
|
2019-01-22 17:31:05 +01:00
|
|
|
# If it returns None, then we're processing the last batch
|
|
|
|
last = end_last_seen is None
|
2019-01-11 20:21:50 +01:00
|
|
|
|
2019-01-23 09:45:18 +01:00
|
|
|
logger.info(
|
|
|
|
"Scanning for duplicate 'user_ips' rows in range: %s <= last_seen < %s",
|
2019-04-03 11:07:29 +02:00
|
|
|
begin_last_seen,
|
|
|
|
end_last_seen,
|
2019-01-23 09:45:18 +01:00
|
|
|
)
|
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
def remove(txn: LoggingTransaction) -> None:
|
2019-01-11 20:21:50 +01:00
|
|
|
# This works by looking at all entries in the given time span, and
|
|
|
|
# then for each (user_id, access_token, ip) tuple in that range
|
|
|
|
# checking for any duplicates in the rest of the table (via a join).
|
|
|
|
# It then only returns entries which have duplicates, and the max
|
|
|
|
# last_seen across all duplicates, which can the be used to delete
|
|
|
|
# all other duplicates.
|
|
|
|
# It is efficient due to the existence of (user_id, access_token,
|
|
|
|
# ip) and (last_seen) indices.
|
2019-01-22 17:31:05 +01:00
|
|
|
|
|
|
|
# Define the search space, which requires handling the last batch in
|
|
|
|
# a different way
|
2021-10-12 14:50:34 +02:00
|
|
|
args: Tuple[int, ...]
|
2019-01-22 17:31:05 +01:00
|
|
|
if last:
|
|
|
|
clause = "? <= last_seen"
|
|
|
|
args = (begin_last_seen,)
|
|
|
|
else:
|
2021-10-12 14:50:34 +02:00
|
|
|
assert end_last_seen is not None
|
2019-01-22 17:31:05 +01:00
|
|
|
clause = "? <= last_seen AND last_seen < ?"
|
|
|
|
args = (begin_last_seen, end_last_seen)
|
|
|
|
|
2019-02-12 12:28:08 +01:00
|
|
|
# (Note: The DISTINCT in the inner query is important to ensure that
|
|
|
|
# the COUNT(*) is accurate, otherwise double counting may happen due
|
|
|
|
# to the join effectively being a cross product)
|
2019-01-11 20:21:50 +01:00
|
|
|
txn.execute(
|
|
|
|
"""
|
|
|
|
SELECT user_id, access_token, ip,
|
2019-02-12 12:28:08 +01:00
|
|
|
MAX(device_id), MAX(user_agent), MAX(last_seen),
|
|
|
|
COUNT(*)
|
2019-01-11 20:21:50 +01:00
|
|
|
FROM (
|
2019-02-12 12:28:08 +01:00
|
|
|
SELECT DISTINCT user_id, access_token, ip
|
2019-01-11 20:21:50 +01:00
|
|
|
FROM user_ips
|
2019-01-22 17:31:05 +01:00
|
|
|
WHERE {}
|
2019-01-11 20:21:50 +01:00
|
|
|
) c
|
|
|
|
INNER JOIN user_ips USING (user_id, access_token, ip)
|
|
|
|
GROUP BY user_id, access_token, ip
|
2019-01-22 17:31:05 +01:00
|
|
|
HAVING count(*) > 1
|
2019-04-03 11:07:29 +02:00
|
|
|
""".format(
|
|
|
|
clause
|
|
|
|
),
|
|
|
|
args,
|
2019-01-11 20:21:50 +01:00
|
|
|
)
|
2021-10-12 14:50:34 +02:00
|
|
|
res = cast(
|
|
|
|
List[Tuple[str, str, str, Optional[str], str, int, int]], txn.fetchall()
|
|
|
|
)
|
2019-01-11 20:21:50 +01:00
|
|
|
|
|
|
|
# We've got some duplicates
|
|
|
|
for i in res:
|
2019-02-12 12:28:08 +01:00
|
|
|
user_id, access_token, ip, device_id, user_agent, last_seen, count = i
|
|
|
|
|
|
|
|
# We want to delete the duplicates so we end up with only a
|
|
|
|
# single row.
|
|
|
|
#
|
|
|
|
# The naive way of doing this would be just to delete all rows
|
|
|
|
# and reinsert a constructed row. However, if there are a lot of
|
|
|
|
# duplicate rows this can cause the table to grow a lot, which
|
|
|
|
# can be problematic in two ways:
|
|
|
|
# 1. If user_ips is already large then this can cause the
|
|
|
|
# table to rapidly grow, potentially filling the disk.
|
|
|
|
# 2. Reinserting a lot of rows can confuse the table
|
|
|
|
# statistics for postgres, causing it to not use the
|
|
|
|
# correct indices for the query above, resulting in a full
|
|
|
|
# table scan. This is incredibly slow for large tables and
|
|
|
|
# can kill database performance. (This seems to mainly
|
|
|
|
# happen for the last query where the clause is simply `? <
|
|
|
|
# last_seen`)
|
|
|
|
#
|
|
|
|
# So instead we want to delete all but *one* of the duplicate
|
|
|
|
# rows. That is hard to do reliably, so we cheat and do a two
|
|
|
|
# step process:
|
|
|
|
# 1. Delete all rows with a last_seen strictly less than the
|
|
|
|
# max last_seen. This hopefully results in deleting all but
|
|
|
|
# one row the majority of the time, but there may be
|
|
|
|
# duplicate last_seen
|
|
|
|
# 2. If multiple rows remain, we fall back to the naive method
|
|
|
|
# and simply delete all rows and reinsert.
|
|
|
|
#
|
|
|
|
# Note that this relies on no new duplicate rows being inserted,
|
|
|
|
# but if that is happening then this entire process is futile
|
|
|
|
# anyway.
|
|
|
|
|
|
|
|
# Do step 1:
|
|
|
|
|
|
|
|
txn.execute(
|
|
|
|
"""
|
|
|
|
DELETE FROM user_ips
|
|
|
|
WHERE user_id = ? AND access_token = ? AND ip = ? AND last_seen < ?
|
|
|
|
""",
|
2019-04-03 11:07:29 +02:00
|
|
|
(user_id, access_token, ip, last_seen),
|
2019-02-12 12:28:08 +01:00
|
|
|
)
|
|
|
|
if txn.rowcount == count - 1:
|
|
|
|
# We deleted all but one of the duplicate rows, i.e. there
|
|
|
|
# is exactly one remaining and so there is nothing left to
|
|
|
|
# do.
|
|
|
|
continue
|
|
|
|
elif txn.rowcount >= count:
|
|
|
|
raise Exception(
|
2019-04-03 11:07:29 +02:00
|
|
|
"We deleted more duplicate rows from 'user_ips' than expected"
|
2019-02-12 12:28:08 +01:00
|
|
|
)
|
|
|
|
|
|
|
|
# The previous step didn't delete enough rows, so we fallback to
|
|
|
|
# step 2:
|
2019-01-11 20:21:50 +01:00
|
|
|
|
|
|
|
# Drop all the duplicates
|
|
|
|
txn.execute(
|
|
|
|
"""
|
|
|
|
DELETE FROM user_ips
|
|
|
|
WHERE user_id = ? AND access_token = ? AND ip = ?
|
|
|
|
""",
|
2019-04-03 11:07:29 +02:00
|
|
|
(user_id, access_token, ip),
|
2019-01-11 20:21:50 +01:00
|
|
|
)
|
|
|
|
|
|
|
|
# Add in one to be the last_seen
|
|
|
|
txn.execute(
|
|
|
|
"""
|
|
|
|
INSERT INTO user_ips
|
|
|
|
(user_id, access_token, ip, device_id, user_agent, last_seen)
|
|
|
|
VALUES (?, ?, ?, ?, ?, ?)
|
|
|
|
""",
|
2019-04-03 11:07:29 +02:00
|
|
|
(user_id, access_token, ip, device_id, user_agent, last_seen),
|
2019-01-11 20:21:50 +01:00
|
|
|
)
|
|
|
|
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates._background_update_progress_txn(
|
2019-01-22 12:55:53 +01:00
|
|
|
txn, "user_ips_remove_dupes", {"last_seen": end_last_seen}
|
2019-01-11 20:21:50 +01:00
|
|
|
)
|
|
|
|
|
2020-08-07 18:17:17 +02:00
|
|
|
await self.db_pool.runInteraction("user_ips_dups_remove", remove)
|
2019-01-22 17:20:33 +01:00
|
|
|
|
2019-01-11 20:21:50 +01:00
|
|
|
if last:
|
2020-08-07 18:17:17 +02:00
|
|
|
await self.db_pool.updates._end_background_update("user_ips_remove_dupes")
|
2019-01-11 20:21:50 +01:00
|
|
|
|
2019-07-23 15:00:55 +02:00
|
|
|
return batch_size
|
2019-01-11 20:21:50 +01:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
async def _devices_last_seen_update(
|
|
|
|
self, progress: JsonDict, batch_size: int
|
|
|
|
) -> int:
|
2021-02-16 23:32:34 +01:00
|
|
|
"""Background update to insert last seen info into devices table"""
|
2019-10-03 18:16:39 +02:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
last_user_id: str = progress.get("last_user_id", "")
|
|
|
|
last_device_id: str = progress.get("last_device_id", "")
|
2019-10-03 18:16:39 +02:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
def _devices_last_seen_update_txn(txn: LoggingTransaction) -> int:
|
2019-10-03 18:16:39 +02:00
|
|
|
# This consists of two queries:
|
|
|
|
#
|
|
|
|
# 1. The sub-query searches for the next N devices and joins
|
|
|
|
# against user_ips to find the max last_seen associated with
|
|
|
|
# that device.
|
|
|
|
# 2. The outer query then joins again against user_ips on
|
|
|
|
# user/device/last_seen. This *should* hopefully only
|
|
|
|
# return one row, but if it does return more than one then
|
|
|
|
# we'll just end up updating the same device row multiple
|
|
|
|
# times, which is fine.
|
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
where_args: List[Union[str, int]]
|
2020-04-08 00:06:39 +02:00
|
|
|
where_clause, where_args = make_tuple_comparison_clause(
|
|
|
|
[("user_id", last_user_id), ("device_id", last_device_id)],
|
|
|
|
)
|
2019-10-03 18:16:39 +02:00
|
|
|
|
|
|
|
sql = """
|
|
|
|
SELECT
|
|
|
|
last_seen, ip, user_agent, user_id, device_id
|
|
|
|
FROM (
|
|
|
|
SELECT
|
|
|
|
user_id, device_id, MAX(u.last_seen) AS last_seen
|
|
|
|
FROM devices
|
|
|
|
INNER JOIN user_ips AS u USING (user_id, device_id)
|
|
|
|
WHERE %(where_clause)s
|
|
|
|
GROUP BY user_id, device_id
|
|
|
|
ORDER BY user_id ASC, device_id ASC
|
|
|
|
LIMIT ?
|
|
|
|
) c
|
|
|
|
INNER JOIN user_ips AS u USING (user_id, device_id, last_seen)
|
|
|
|
""" % {
|
|
|
|
"where_clause": where_clause
|
|
|
|
}
|
|
|
|
txn.execute(sql, where_args + [batch_size])
|
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
rows = cast(List[Tuple[int, str, str, str, str]], txn.fetchall())
|
2019-10-03 18:16:39 +02:00
|
|
|
if not rows:
|
|
|
|
return 0
|
|
|
|
|
|
|
|
sql = """
|
|
|
|
UPDATE devices
|
|
|
|
SET last_seen = ?, ip = ?, user_agent = ?
|
|
|
|
WHERE user_id = ? AND device_id = ?
|
|
|
|
"""
|
|
|
|
txn.execute_batch(sql, rows)
|
|
|
|
|
|
|
|
_, _, _, user_id, device_id = rows[-1]
|
2020-08-05 22:38:57 +02:00
|
|
|
self.db_pool.updates._background_update_progress_txn(
|
2019-10-03 18:16:39 +02:00
|
|
|
txn,
|
|
|
|
"devices_last_seen",
|
|
|
|
{"last_user_id": user_id, "last_device_id": device_id},
|
|
|
|
)
|
|
|
|
|
|
|
|
return len(rows)
|
|
|
|
|
2020-08-07 18:17:17 +02:00
|
|
|
updated = await self.db_pool.runInteraction(
|
2019-10-03 18:16:39 +02:00
|
|
|
"_devices_last_seen_update", _devices_last_seen_update_txn
|
|
|
|
)
|
|
|
|
|
|
|
|
if not updated:
|
2020-08-07 18:17:17 +02:00
|
|
|
await self.db_pool.updates._end_background_update("devices_last_seen")
|
2019-10-03 18:16:39 +02:00
|
|
|
|
|
|
|
return updated
|
|
|
|
|
|
|
|
|
2020-10-07 17:27:56 +02:00
|
|
|
class ClientIpWorkerStore(ClientIpBackgroundUpdateStore):
|
2021-12-13 18:05:00 +01:00
|
|
|
def __init__(
|
|
|
|
self,
|
|
|
|
database: DatabasePool,
|
|
|
|
db_conn: LoggingDatabaseConnection,
|
|
|
|
hs: "HomeServer",
|
|
|
|
):
|
2020-10-07 17:27:56 +02:00
|
|
|
super().__init__(database, db_conn, hs)
|
|
|
|
|
2021-09-29 12:44:15 +02:00
|
|
|
self.user_ips_max_age = hs.config.server.user_ips_max_age
|
2020-10-07 17:27:56 +02:00
|
|
|
|
2021-09-13 19:07:12 +02:00
|
|
|
if hs.config.worker.run_background_tasks and self.user_ips_max_age:
|
2020-10-07 17:27:56 +02:00
|
|
|
self._clock.looping_call(self._prune_old_user_ips, 5 * 1000)
|
|
|
|
|
|
|
|
@wrap_as_background_process("prune_old_user_ips")
|
2021-10-12 14:50:34 +02:00
|
|
|
async def _prune_old_user_ips(self) -> None:
|
2021-02-16 23:32:34 +01:00
|
|
|
"""Removes entries in user IPs older than the configured period."""
|
2020-10-07 17:27:56 +02:00
|
|
|
|
|
|
|
if self.user_ips_max_age is None:
|
|
|
|
# Nothing to do
|
|
|
|
return
|
|
|
|
|
|
|
|
if not await self.db_pool.updates.has_completed_background_update(
|
|
|
|
"devices_last_seen"
|
|
|
|
):
|
|
|
|
# Only start pruning if we have finished populating the devices
|
|
|
|
# last seen info.
|
|
|
|
return
|
|
|
|
|
|
|
|
# We do a slightly funky SQL delete to ensure we don't try and delete
|
|
|
|
# too much at once (as the table may be very large from before we
|
|
|
|
# started pruning).
|
|
|
|
#
|
|
|
|
# This works by finding the max last_seen that is less than the given
|
|
|
|
# time, but has no more than N rows before it, deleting all rows with
|
|
|
|
# a lesser last_seen time. (We COALESCE so that the sub-SELECT always
|
|
|
|
# returns exactly one row).
|
|
|
|
sql = """
|
|
|
|
DELETE FROM user_ips
|
|
|
|
WHERE last_seen <= (
|
|
|
|
SELECT COALESCE(MAX(last_seen), -1)
|
|
|
|
FROM (
|
|
|
|
SELECT last_seen FROM user_ips
|
|
|
|
WHERE last_seen <= ?
|
|
|
|
ORDER BY last_seen ASC
|
|
|
|
LIMIT 5000
|
|
|
|
) AS u
|
|
|
|
)
|
|
|
|
"""
|
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
timestamp = self._clock.time_msec() - self.user_ips_max_age
|
2020-10-07 17:27:56 +02:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
def _prune_old_user_ips_txn(txn: LoggingTransaction) -> None:
|
2020-10-07 17:27:56 +02:00
|
|
|
txn.execute(sql, (timestamp,))
|
|
|
|
|
|
|
|
await self.db_pool.runInteraction(
|
|
|
|
"_prune_old_user_ips", _prune_old_user_ips_txn
|
|
|
|
)
|
|
|
|
|
2021-01-13 18:35:40 +01:00
|
|
|
async def get_last_client_ip_by_device(
|
|
|
|
self, user_id: str, device_id: Optional[str]
|
2021-10-12 14:50:34 +02:00
|
|
|
) -> Dict[Tuple[str, str], DeviceLastConnectionInfo]:
|
2021-01-13 18:35:40 +01:00
|
|
|
"""For each device_id listed, give the user_ip it was last seen on.
|
|
|
|
|
|
|
|
The result might be slightly out of date as client IPs are inserted in batches.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
user_id: The user to fetch devices for.
|
|
|
|
device_id: If None fetches all devices for the user
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
A dictionary mapping a tuple of (user_id, device_id) to dicts, with
|
|
|
|
keys giving the column names from the devices table.
|
|
|
|
"""
|
|
|
|
|
|
|
|
keyvalues = {"user_id": user_id}
|
|
|
|
if device_id is not None:
|
|
|
|
keyvalues["device_id"] = device_id
|
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
res = cast(
|
|
|
|
List[DeviceLastConnectionInfo],
|
|
|
|
await self.db_pool.simple_select_list(
|
|
|
|
table="devices",
|
|
|
|
keyvalues=keyvalues,
|
|
|
|
retcols=("user_id", "ip", "user_agent", "device_id", "last_seen"),
|
|
|
|
),
|
2021-01-13 18:35:40 +01:00
|
|
|
)
|
|
|
|
|
|
|
|
return {(d["user_id"], d["device_id"]): d for d in res}
|
|
|
|
|
2021-10-25 14:01:04 +02:00
|
|
|
async def get_user_ip_and_agents(
|
|
|
|
self, user: UserID, since_ts: int = 0
|
|
|
|
) -> List[LastConnectionInfo]:
|
|
|
|
"""Fetch the IPs and user agents for a user since the given timestamp.
|
|
|
|
|
|
|
|
The result might be slightly out of date as client IPs are inserted in batches.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
user: The user for which to fetch IP addresses and user agents.
|
|
|
|
since_ts: The timestamp after which to fetch IP addresses and user agents,
|
|
|
|
in milliseconds.
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
A list of dictionaries, each containing:
|
|
|
|
* `access_token`: The access token used.
|
|
|
|
* `ip`: The IP address used.
|
|
|
|
* `user_agent`: The last user agent seen for this access token and IP
|
|
|
|
address combination.
|
|
|
|
* `last_seen`: The timestamp at which this access token and IP address
|
|
|
|
combination was last seen, in milliseconds.
|
|
|
|
|
|
|
|
Only the latest user agent for each access token and IP address combination
|
|
|
|
is available.
|
|
|
|
"""
|
|
|
|
user_id = user.to_string()
|
|
|
|
|
|
|
|
def get_recent(txn: LoggingTransaction) -> List[Tuple[str, str, str, int]]:
|
|
|
|
txn.execute(
|
|
|
|
"""
|
|
|
|
SELECT access_token, ip, user_agent, last_seen FROM user_ips
|
|
|
|
WHERE last_seen >= ? AND user_id = ?
|
|
|
|
ORDER BY last_seen
|
|
|
|
DESC
|
|
|
|
""",
|
|
|
|
(since_ts, user_id),
|
|
|
|
)
|
|
|
|
return cast(List[Tuple[str, str, str, int]], txn.fetchall())
|
|
|
|
|
|
|
|
rows = await self.db_pool.runInteraction(
|
|
|
|
desc="get_user_ip_and_agents", func=get_recent
|
|
|
|
)
|
|
|
|
|
|
|
|
return [
|
|
|
|
{
|
|
|
|
"access_token": access_token,
|
|
|
|
"ip": ip,
|
|
|
|
"user_agent": user_agent,
|
|
|
|
"last_seen": last_seen,
|
|
|
|
}
|
|
|
|
for access_token, ip, user_agent, last_seen in rows
|
|
|
|
]
|
|
|
|
|
2020-10-07 17:27:56 +02:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
class ClientIpStore(ClientIpWorkerStore, MonthlyActiveUsersStore):
|
2021-12-13 18:05:00 +01:00
|
|
|
def __init__(
|
|
|
|
self,
|
|
|
|
database: DatabasePool,
|
|
|
|
db_conn: LoggingDatabaseConnection,
|
|
|
|
hs: "HomeServer",
|
|
|
|
):
|
2019-10-03 18:16:39 +02:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
# (user_id, access_token, ip,) -> last_seen
|
|
|
|
self.client_ip_last_seen = LruCache[Tuple[str, str, str], int](
|
2021-05-24 15:02:01 +02:00
|
|
|
cache_name="client_ip_last_seen", max_size=50000
|
2019-10-03 18:16:39 +02:00
|
|
|
)
|
|
|
|
|
2020-09-18 15:56:44 +02:00
|
|
|
super().__init__(database, db_conn, hs)
|
2019-10-03 18:16:39 +02:00
|
|
|
|
|
|
|
# (user_id, access_token, ip,) -> (user_agent, device_id, last_seen)
|
2021-10-12 14:50:34 +02:00
|
|
|
self._batch_row_update: Dict[
|
|
|
|
Tuple[str, str, str], Tuple[str, Optional[str], int]
|
|
|
|
] = {}
|
2019-10-03 18:16:39 +02:00
|
|
|
|
|
|
|
self._client_ip_looper = self._clock.looping_call(
|
|
|
|
self._update_client_ips_batch, 5 * 1000
|
|
|
|
)
|
|
|
|
self.hs.get_reactor().addSystemEventTrigger(
|
|
|
|
"before", "shutdown", self._update_client_ips_batch
|
|
|
|
)
|
|
|
|
|
2020-08-06 14:30:06 +02:00
|
|
|
async def insert_client_ip(
|
2021-10-12 14:50:34 +02:00
|
|
|
self,
|
|
|
|
user_id: str,
|
|
|
|
access_token: str,
|
|
|
|
ip: str,
|
|
|
|
user_agent: str,
|
|
|
|
device_id: Optional[str],
|
|
|
|
now: Optional[int] = None,
|
|
|
|
) -> None:
|
2017-06-27 16:53:45 +02:00
|
|
|
if not now:
|
|
|
|
now = int(self._clock.time_msec())
|
|
|
|
key = (user_id, access_token, ip)
|
2016-06-03 15:42:35 +02:00
|
|
|
|
|
|
|
try:
|
|
|
|
last_seen = self.client_ip_last_seen.get(key)
|
|
|
|
except KeyError:
|
|
|
|
last_seen = None
|
2020-08-06 14:30:06 +02:00
|
|
|
await self.populate_monthly_active_users(user_id)
|
2016-06-03 15:42:35 +02:00
|
|
|
# Rate-limited inserts
|
|
|
|
if last_seen is not None and (now - last_seen) < LAST_SEEN_GRANULARITY:
|
2017-06-27 14:37:04 +02:00
|
|
|
return
|
2016-06-03 15:42:35 +02:00
|
|
|
|
2020-10-19 13:20:29 +02:00
|
|
|
self.client_ip_last_seen.set(key, now)
|
2016-06-03 15:42:35 +02:00
|
|
|
|
2017-06-27 14:37:04 +02:00
|
|
|
self._batch_row_update[key] = (user_agent, device_id, now)
|
|
|
|
|
2019-09-24 16:20:40 +02:00
|
|
|
@wrap_as_background_process("update_client_ips")
|
2020-09-01 14:39:04 +02:00
|
|
|
async def _update_client_ips_batch(self) -> None:
|
2018-08-13 08:47:46 +02:00
|
|
|
|
|
|
|
# If the DB pool has already terminated, don't try updating
|
2020-08-05 22:38:57 +02:00
|
|
|
if not self.db_pool.is_running():
|
2018-08-13 08:47:46 +02:00
|
|
|
return
|
|
|
|
|
2019-09-24 16:20:40 +02:00
|
|
|
to_update = self._batch_row_update
|
|
|
|
self._batch_row_update = {}
|
2018-07-18 15:35:24 +02:00
|
|
|
|
2020-09-01 14:39:04 +02:00
|
|
|
await self.db_pool.runInteraction(
|
2019-09-24 16:20:40 +02:00
|
|
|
"_update_client_ips_batch", self._update_client_ips_batch_txn, to_update
|
|
|
|
)
|
2016-07-20 17:34:00 +02:00
|
|
|
|
2021-10-12 14:50:34 +02:00
|
|
|
def _update_client_ips_batch_txn(
|
|
|
|
self,
|
|
|
|
txn: LoggingTransaction,
|
|
|
|
to_update: Mapping[Tuple[str, str, str], Tuple[str, Optional[str], int]],
|
|
|
|
) -> None:
|
2020-08-05 22:38:57 +02:00
|
|
|
if "user_ips" in self.db_pool._unsafe_to_upsert_tables or (
|
2019-01-24 11:31:54 +01:00
|
|
|
not self.database_engine.can_native_upsert
|
|
|
|
):
|
|
|
|
self.database_engine.lock_table(txn, "user_ips")
|
2017-06-27 14:37:04 +02:00
|
|
|
|
2020-06-15 13:03:36 +02:00
|
|
|
for entry in to_update.items():
|
2017-06-27 14:37:04 +02:00
|
|
|
(user_id, access_token, ip), (user_agent, device_id, last_seen) = entry
|
|
|
|
|
2021-01-08 15:15:20 +01:00
|
|
|
self.db_pool.simple_upsert_txn(
|
|
|
|
txn,
|
|
|
|
table="user_ips",
|
|
|
|
keyvalues={"user_id": user_id, "access_token": access_token, "ip": ip},
|
|
|
|
values={
|
|
|
|
"user_agent": user_agent,
|
|
|
|
"device_id": device_id,
|
|
|
|
"last_seen": last_seen,
|
|
|
|
},
|
|
|
|
lock=False,
|
|
|
|
)
|
|
|
|
|
|
|
|
# Technically an access token might not be associated with
|
|
|
|
# a device so we need to check.
|
|
|
|
if device_id:
|
|
|
|
# this is always an update rather than an upsert: the row should
|
|
|
|
# already exist, and if it doesn't, that may be because it has been
|
|
|
|
# deleted, and we don't want to re-create it.
|
|
|
|
self.db_pool.simple_update_txn(
|
2018-09-20 12:14:34 +02:00
|
|
|
txn,
|
2021-01-08 15:15:20 +01:00
|
|
|
table="devices",
|
|
|
|
keyvalues={"user_id": user_id, "device_id": device_id},
|
|
|
|
updatevalues={
|
2019-01-11 20:21:50 +01:00
|
|
|
"user_agent": user_agent,
|
2018-09-20 12:14:34 +02:00
|
|
|
"last_seen": last_seen,
|
2021-01-08 15:15:20 +01:00
|
|
|
"ip": ip,
|
2018-09-20 12:14:34 +02:00
|
|
|
},
|
|
|
|
)
|
2019-09-23 15:16:10 +02:00
|
|
|
|
2020-08-07 18:17:17 +02:00
|
|
|
async def get_last_client_ip_by_device(
|
|
|
|
self, user_id: str, device_id: Optional[str]
|
2021-10-12 14:50:34 +02:00
|
|
|
) -> Dict[Tuple[str, str], DeviceLastConnectionInfo]:
|
2016-07-20 17:34:00 +02:00
|
|
|
"""For each device_id listed, give the user_ip it was last seen on
|
|
|
|
|
|
|
|
Args:
|
2020-08-07 18:17:17 +02:00
|
|
|
user_id: The user to fetch devices for.
|
|
|
|
device_id: If None fetches all devices for the user
|
2016-07-20 17:34:00 +02:00
|
|
|
|
|
|
|
Returns:
|
2020-08-07 18:17:17 +02:00
|
|
|
A dictionary mapping a tuple of (user_id, device_id) to dicts, with
|
|
|
|
keys giving the column names from the devices table.
|
2016-07-20 17:34:00 +02:00
|
|
|
"""
|
2021-01-13 18:35:40 +01:00
|
|
|
ret = await super().get_last_client_ip_by_device(user_id, device_id)
|
2017-04-11 17:24:31 +02:00
|
|
|
|
2021-10-12 12:24:05 +02:00
|
|
|
# Update what is retrieved from the database with data which is pending
|
|
|
|
# insertion, as if it has already been stored in the database.
|
2017-06-27 15:46:12 +02:00
|
|
|
for key in self._batch_row_update:
|
2021-10-12 12:24:05 +02:00
|
|
|
uid, _access_token, ip = key
|
2017-06-27 15:46:12 +02:00
|
|
|
if uid == user_id:
|
|
|
|
user_agent, did, last_seen = self._batch_row_update[key]
|
2021-10-12 12:24:05 +02:00
|
|
|
|
|
|
|
if did is None:
|
|
|
|
# These updates don't make it to the `devices` table
|
|
|
|
continue
|
|
|
|
|
2017-06-27 15:46:12 +02:00
|
|
|
if not device_id or did == device_id:
|
2021-10-12 12:24:05 +02:00
|
|
|
ret[(user_id, did)] = {
|
2017-06-27 15:46:12 +02:00
|
|
|
"user_id": user_id,
|
|
|
|
"ip": ip,
|
|
|
|
"user_agent": user_agent,
|
|
|
|
"device_id": did,
|
|
|
|
"last_seen": last_seen,
|
|
|
|
}
|
2019-07-23 15:00:55 +02:00
|
|
|
return ret
|
2016-07-20 17:34:00 +02:00
|
|
|
|
2020-12-29 23:42:10 +01:00
|
|
|
async def get_user_ip_and_agents(
|
2021-09-22 15:09:43 +02:00
|
|
|
self, user: UserID, since_ts: int = 0
|
2021-10-12 14:50:34 +02:00
|
|
|
) -> List[LastConnectionInfo]:
|
2021-10-25 14:01:04 +02:00
|
|
|
"""Fetch the IPs and user agents for a user since the given timestamp.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
user: The user for which to fetch IP addresses and user agents.
|
|
|
|
since_ts: The timestamp after which to fetch IP addresses and user agents,
|
|
|
|
in milliseconds.
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
A list of dictionaries, each containing:
|
|
|
|
* `access_token`: The access token used.
|
|
|
|
* `ip`: The IP address used.
|
|
|
|
* `user_agent`: The last user agent seen for this access token and IP
|
|
|
|
address combination.
|
|
|
|
* `last_seen`: The timestamp at which this access token and IP address
|
|
|
|
combination was last seen, in milliseconds.
|
|
|
|
|
|
|
|
Only the latest user agent for each access token and IP address combination
|
|
|
|
is available.
|
2021-09-22 15:09:43 +02:00
|
|
|
"""
|
2021-10-25 14:01:04 +02:00
|
|
|
results: Dict[Tuple[str, str], LastConnectionInfo] = {
|
|
|
|
(connection["access_token"], connection["ip"]): connection
|
|
|
|
for connection in await super().get_user_ip_and_agents(user, since_ts)
|
|
|
|
}
|
2017-06-27 15:46:12 +02:00
|
|
|
|
2021-10-25 14:01:04 +02:00
|
|
|
# Overlay data that is pending insertion on top of the results from the
|
|
|
|
# database.
|
|
|
|
user_id = user.to_string()
|
2017-06-27 15:46:12 +02:00
|
|
|
for key in self._batch_row_update:
|
2021-10-25 14:01:04 +02:00
|
|
|
uid, access_token, ip = key
|
2017-06-27 15:46:12 +02:00
|
|
|
if uid == user_id:
|
|
|
|
user_agent, _, last_seen = self._batch_row_update[key]
|
2021-09-22 15:09:43 +02:00
|
|
|
if last_seen >= since_ts:
|
2021-10-25 14:01:04 +02:00
|
|
|
results[(access_token, ip)] = {
|
|
|
|
"access_token": access_token,
|
|
|
|
"ip": ip,
|
|
|
|
"user_agent": user_agent,
|
|
|
|
"last_seen": last_seen,
|
|
|
|
}
|
2017-06-27 15:46:12 +02:00
|
|
|
|
2021-10-25 14:01:04 +02:00
|
|
|
return list(results.values())
|