class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
def __init__(self, db_conn, hs):
super(DeviceStore, self).__init__(db_conn, hs)
# Map of (user_id, device_id) -> bool. If there is an entry that implies
# the device exists.
self.device_id_exists_cache = Cache(name="device_id_exists",
keylen=2,
max_entries=10000)
self._clock.looping_call(self._prune_old_outbound_device_pokes,
60 * 60 * 1000)
self.register_background_index_update(
"device_lists_stream_idx",
index_name="device_lists_stream_user_id",
table="device_lists_stream",
columns=["user_id", "device_id"],
)
# create a unique index on device_lists_remote_cache
self.register_background_index_update(
"device_lists_remote_cache_unique_idx",
index_name="device_lists_remote_cache_unique_id",
table="device_lists_remote_cache",
columns=["user_id", "device_id"],
unique=True,
)
# And one on device_lists_remote_extremeties
self.register_background_index_update(
"device_lists_remote_extremeties_unique_idx",
index_name="device_lists_remote_extremeties_unique_idx",
table="device_lists_remote_extremeties",
columns=["user_id"],
unique=True,
)
# once they complete, we can remove the old non-unique indexes.
self.register_background_update_handler(
DROP_DEVICE_LIST_STREAMS_NON_UNIQUE_INDEXES,
self._drop_device_list_streams_non_unique_indexes,
)
@defer.inlineCallbacks
def store_device(self, user_id, device_id, initial_device_display_name):
"""Ensure the given device is known; add it to the store if not
Args:
user_id (str): id of user associated with the device
device_id (str): id of device
initial_device_display_name (str): initial displayname of the
device. Ignored if device exists.
Returns:
defer.Deferred: boolean whether the device was inserted or an
existing device existed with that ID.
"""
key = (user_id, device_id)
if self.device_id_exists_cache.get(key, None):
return False
try:
inserted = yield self._simple_insert(
"devices",
values={
"user_id": user_id,
"device_id": device_id,
"display_name": initial_device_display_name,
},
desc="store_device",
or_ignore=True,
)
self.device_id_exists_cache.prefill(key, True)
return inserted
except Exception as e:
logger.error(
"store_device with device_id=%s(%r) user_id=%s(%r)"
" display_name=%s(%r) failed: %s",
type(device_id).__name__,
device_id,
type(user_id).__name__,
user_id,
type(initial_device_display_name).__name__,
initial_device_display_name,
e,
)
raise StoreError(500, "Problem storing device.")
@defer.inlineCallbacks
def delete_device(self, user_id, device_id):
"""Delete a device.
Args:
user_id (str): The ID of the user which owns the device
device_id (str): The ID of the device to delete
Returns:
defer.Deferred
"""
yield self._simple_delete_one(
table="devices",
keyvalues={
"user_id": user_id,
"device_id": device_id
},
desc="delete_device",
)
self.device_id_exists_cache.invalidate((user_id, device_id))
@defer.inlineCallbacks
def delete_devices(self, user_id, device_ids):
"""Deletes several devices.
Args:
user_id (str): The ID of the user which owns the devices
device_ids (list): The IDs of the devices to delete
Returns:
defer.Deferred
"""
yield self._simple_delete_many(
table="devices",
column="device_id",
iterable=device_ids,
keyvalues={"user_id": user_id},
desc="delete_devices",
)
for device_id in device_ids:
self.device_id_exists_cache.invalidate((user_id, device_id))
def update_device(self, user_id, device_id, new_display_name=None):
"""Update a device.
Args:
user_id (str): The ID of the user which owns the device
device_id (str): The ID of the device to update
new_display_name (str|None): new displayname for device; None
to leave unchanged
Raises:
StoreError: if the device is not found
Returns:
defer.Deferred
"""
updates = {}
if new_display_name is not None:
updates["display_name"] = new_display_name
if not updates:
return defer.succeed(None)
return self._simple_update_one(
table="devices",
keyvalues={
"user_id": user_id,
"device_id": device_id
},
updatevalues=updates,
desc="update_device",
)
@defer.inlineCallbacks
def mark_remote_user_device_list_as_unsubscribed(self, user_id):
"""Mark that we no longer track device lists for remote user.
"""
yield self._simple_delete(
table="device_lists_remote_extremeties",
keyvalues={"user_id": user_id},
desc="mark_remote_user_device_list_as_unsubscribed",
)
self.get_device_list_last_stream_id_for_remote.invalidate((user_id, ))
def update_remote_device_list_cache_entry(self, user_id, device_id,
content, stream_id):
"""Updates a single device in the cache of a remote user's devicelist.
Note: assumes that we are the only thread that can be updating this user's
device list.
Args:
user_id (str): User to update device list for
device_id (str): ID of decivice being updated
content (dict): new data on this device
stream_id (int): the version of the device list
Returns:
Deferred[None]
"""
return self.runInteraction(
"update_remote_device_list_cache_entry",
self._update_remote_device_list_cache_entry_txn,
user_id,
device_id,
content,
stream_id,
)
def _update_remote_device_list_cache_entry_txn(self, txn, user_id,
device_id, content,
stream_id):
if content.get("deleted"):
self._simple_delete_txn(
txn,
table="device_lists_remote_cache",
keyvalues={
"user_id": user_id,
"device_id": device_id
},
)
txn.call_after(self.device_id_exists_cache.invalidate,
(user_id, device_id))
else:
self._simple_upsert_txn(
txn,
table="device_lists_remote_cache",
keyvalues={
"user_id": user_id,
"device_id": device_id
},
values={"content": json.dumps(content)},
# we don't need to lock, because we assume we are the only thread
# updating this user's devices.
lock=False,
)
txn.call_after(self._get_cached_user_device.invalidate,
(user_id, device_id))
txn.call_after(self._get_cached_devices_for_user.invalidate,
(user_id, ))
txn.call_after(
self.get_device_list_last_stream_id_for_remote.invalidate,
(user_id, ))
self._simple_upsert_txn(
txn,
table="device_lists_remote_extremeties",
keyvalues={"user_id": user_id},
values={"stream_id": stream_id},
# again, we can assume we are the only thread updating this user's
# extremity.
lock=False,
)
def update_remote_device_list_cache(self, user_id, devices, stream_id):
"""Replace the entire cache of the remote user's devices.
Note: assumes that we are the only thread that can be updating this user's
device list.
Args:
user_id (str): User to update device list for
devices (list[dict]): list of device objects supplied over federation
stream_id (int): the version of the device list
Returns:
Deferred[None]
"""
return self.runInteraction(
"update_remote_device_list_cache",
self._update_remote_device_list_cache_txn,
user_id,
devices,
stream_id,
)
def _update_remote_device_list_cache_txn(self, txn, user_id, devices,
stream_id):
self._simple_delete_txn(txn,
table="device_lists_remote_cache",
keyvalues={"user_id": user_id})
self._simple_insert_many_txn(
txn,
table="device_lists_remote_cache",
values=[{
"user_id": user_id,
"device_id": content["device_id"],
"content": json.dumps(content),
} for content in devices],
)
txn.call_after(self._get_cached_devices_for_user.invalidate,
(user_id, ))
txn.call_after(self._get_cached_user_device.invalidate_many,
(user_id, ))
txn.call_after(
self.get_device_list_last_stream_id_for_remote.invalidate,
(user_id, ))
self._simple_upsert_txn(
txn,
table="device_lists_remote_extremeties",
keyvalues={"user_id": user_id},
values={"stream_id": stream_id},
# we don't need to lock, because we can assume we are the only thread
# updating this user's extremity.
lock=False,
)
@defer.inlineCallbacks
def add_device_change_to_streams(self, user_id, device_ids, hosts):
"""Persist that a user's devices have been updated, and which hosts
(if any) should be poked.
"""
with self._device_list_id_gen.get_next() as stream_id:
yield self.runInteraction(
"add_device_change_to_streams",
self._add_device_change_txn,
user_id,
device_ids,
hosts,
stream_id,
)
return stream_id
def _add_device_change_txn(self, txn, user_id, device_ids, hosts,
stream_id):
now = self._clock.time_msec()
txn.call_after(self._device_list_stream_cache.entity_has_changed,
user_id, stream_id)
for host in hosts:
txn.call_after(
self._device_list_federation_stream_cache.entity_has_changed,
host,
stream_id,
)
# Delete older entries in the table, as we really only care about
# when the latest change happened.
txn.executemany(
"""
DELETE FROM device_lists_stream
WHERE user_id = ? AND device_id = ? AND stream_id < ?
""",
[(user_id, device_id, stream_id) for device_id in device_ids],
)
self._simple_insert_many_txn(
txn,
table="device_lists_stream",
values=[{
"stream_id": stream_id,
"user_id": user_id,
"device_id": device_id
} for device_id in device_ids],
)
context = get_active_span_text_map()
self._simple_insert_many_txn(
txn,
table="device_lists_outbound_pokes",
values=[{
"destination":
destination,
"stream_id":
stream_id,
"user_id":
user_id,
"device_id":
device_id,
"sent":
False,
"ts":
now,
"opentracing_context":
json.dumps(context)
if whitelisted_homeserver(destination) else "{}",
} for destination in hosts for device_id in device_ids],
)
def _prune_old_outbound_device_pokes(self):
"""Delete old entries out of the device_lists_outbound_pokes to ensure
that we don't fill up due to dead servers. We keep one entry per
(destination, user_id) tuple to ensure that the prev_ids remain correct
if the server does come back.
"""
yesterday = self._clock.time_msec() - 24 * 60 * 60 * 1000
def _prune_txn(txn):
select_sql = """
SELECT destination, user_id, max(stream_id) as stream_id
FROM device_lists_outbound_pokes
GROUP BY destination, user_id
HAVING min(ts) < ? AND count(*) > 1
"""
txn.execute(select_sql, (yesterday, ))
rows = txn.fetchall()
if not rows:
return
delete_sql = """
DELETE FROM device_lists_outbound_pokes
WHERE ts < ? AND destination = ? AND user_id = ? AND stream_id < ?
"""
txn.executemany(delete_sql, ((yesterday, row[0], row[1], row[2])
for row in rows))
# Since we've deleted unsent deltas, we need to remove the entry
# of last successful sent so that the prev_ids are correctly set.
sql = """
DELETE FROM device_lists_outbound_last_success
WHERE destination = ? AND user_id = ?
"""
txn.executemany(sql, ((row[0], row[1]) for row in rows))
logger.info("Pruned %d device list outbound pokes", txn.rowcount)
return run_as_background_process(
"prune_old_outbound_device_pokes",
self.runInteraction,
"_prune_old_outbound_device_pokes",
_prune_txn,
)
@defer.inlineCallbacks
def _drop_device_list_streams_non_unique_indexes(self, progress,
batch_size):
def f(conn):
txn = conn.cursor()
txn.execute("DROP INDEX IF EXISTS device_lists_remote_cache_id")
txn.execute(
"DROP INDEX IF EXISTS device_lists_remote_extremeties_id")
txn.close()
yield self.runWithConnection(f)
yield self._end_background_update(
DROP_DEVICE_LIST_STREAMS_NON_UNIQUE_INDEXES)
return 1
class CacheTestCase(unittest.TestCase):
def setUp(self):
self.cache = Cache("test")
def test_empty(self):
failed = False
try:
self.cache.get("foo")
except KeyError:
failed = True
self.assertTrue(failed)
def test_hit(self):
self.cache.prefill("foo", 123)
self.assertEquals(self.cache.get("foo"), 123)
def test_invalidate(self):
self.cache.prefill("foo", 123)
self.cache.invalidate("foo")
failed = False
try:
self.cache.get("foo")
except KeyError:
failed = True
self.assertTrue(failed)
def test_eviction(self):
cache = Cache("test", max_entries=2)
cache.prefill(1, "one")
cache.prefill(2, "two")
cache.prefill(3, "three") # 1 will be evicted
failed = False
try:
cache.get(1)
except KeyError:
failed = True
self.assertTrue(failed)
cache.get(2)
cache.get(3)
def test_eviction_lru(self):
cache = Cache("test", max_entries=2, lru=True)
cache.prefill(1, "one")
cache.prefill(2, "two")
# Now access 1 again, thus causing 2 to be least-recently used
cache.get(1)
cache.prefill(3, "three")
failed = False
try:
cache.get(2)
except KeyError:
failed = True
self.assertTrue(failed)
cache.get(1)
cache.get(3)
class DeviceStore(DeviceWorkerStore, BackgroundUpdateStore):
def __init__(self, db_conn, hs):
super(DeviceStore, self).__init__(db_conn, hs)
# Map of (user_id, device_id) -> bool. If there is an entry that implies
# the device exists.
self.device_id_exists_cache = Cache(
name="device_id_exists", keylen=2, max_entries=10000
)
self._clock.looping_call(self._prune_old_outbound_device_pokes, 60 * 60 * 1000)
self.register_background_index_update(
"device_lists_stream_idx",
index_name="device_lists_stream_user_id",
table="device_lists_stream",
columns=["user_id", "device_id"],
)
# create a unique index on device_lists_remote_cache
self.register_background_index_update(
"device_lists_remote_cache_unique_idx",
index_name="device_lists_remote_cache_unique_id",
table="device_lists_remote_cache",
columns=["user_id", "device_id"],
unique=True,
)
# And one on device_lists_remote_extremeties
self.register_background_index_update(
"device_lists_remote_extremeties_unique_idx",
index_name="device_lists_remote_extremeties_unique_idx",
table="device_lists_remote_extremeties",
columns=["user_id"],
unique=True,
)
# once they complete, we can remove the old non-unique indexes.
self.register_background_update_handler(
DROP_DEVICE_LIST_STREAMS_NON_UNIQUE_INDEXES,
self._drop_device_list_streams_non_unique_indexes,
)
@defer.inlineCallbacks
def store_device(self, user_id, device_id, initial_device_display_name):
"""Ensure the given device is known; add it to the store if not
Args:
user_id (str): id of user associated with the device
device_id (str): id of device
initial_device_display_name (str): initial displayname of the
device. Ignored if device exists.
Returns:
defer.Deferred: boolean whether the device was inserted or an
existing device existed with that ID.
"""
key = (user_id, device_id)
if self.device_id_exists_cache.get(key, None):
defer.returnValue(False)
try:
inserted = yield self._simple_insert(
"devices",
values={
"user_id": user_id,
"device_id": device_id,
"display_name": initial_device_display_name,
},
desc="store_device",
or_ignore=True,
)
self.device_id_exists_cache.prefill(key, True)
defer.returnValue(inserted)
except Exception as e:
logger.error(
"store_device with device_id=%s(%r) user_id=%s(%r)"
" display_name=%s(%r) failed: %s",
type(device_id).__name__,
device_id,
type(user_id).__name__,
user_id,
type(initial_device_display_name).__name__,
initial_device_display_name,
e,
)
raise StoreError(500, "Problem storing device.")
@defer.inlineCallbacks
def delete_device(self, user_id, device_id):
"""Delete a device.
Args:
user_id (str): The ID of the user which owns the device
device_id (str): The ID of the device to delete
Returns:
defer.Deferred
"""
yield self._simple_delete_one(
table="devices",
keyvalues={"user_id": user_id, "device_id": device_id},
desc="delete_device",
)
self.device_id_exists_cache.invalidate((user_id, device_id))
@defer.inlineCallbacks
def delete_devices(self, user_id, device_ids):
"""Deletes several devices.
Args:
user_id (str): The ID of the user which owns the devices
device_ids (list): The IDs of the devices to delete
Returns:
defer.Deferred
"""
yield self._simple_delete_many(
table="devices",
column="device_id",
iterable=device_ids,
keyvalues={"user_id": user_id},
desc="delete_devices",
)
for device_id in device_ids:
self.device_id_exists_cache.invalidate((user_id, device_id))
def update_device(self, user_id, device_id, new_display_name=None):
"""Update a device.
Args:
user_id (str): The ID of the user which owns the device
device_id (str): The ID of the device to update
new_display_name (str|None): new displayname for device; None
to leave unchanged
Raises:
StoreError: if the device is not found
Returns:
defer.Deferred
"""
updates = {}
if new_display_name is not None:
updates["display_name"] = new_display_name
if not updates:
return defer.succeed(None)
return self._simple_update_one(
table="devices",
keyvalues={"user_id": user_id, "device_id": device_id},
updatevalues=updates,
desc="update_device",
)
@defer.inlineCallbacks
def mark_remote_user_device_list_as_unsubscribed(self, user_id):
"""Mark that we no longer track device lists for remote user.
"""
yield self._simple_delete(
table="device_lists_remote_extremeties",
keyvalues={"user_id": user_id},
desc="mark_remote_user_device_list_as_unsubscribed",
)
self.get_device_list_last_stream_id_for_remote.invalidate((user_id,))
def update_remote_device_list_cache_entry(
self, user_id, device_id, content, stream_id
):
"""Updates a single device in the cache of a remote user's devicelist.
Note: assumes that we are the only thread that can be updating this user's
device list.
Args:
user_id (str): User to update device list for
device_id (str): ID of decivice being updated
content (dict): new data on this device
stream_id (int): the version of the device list
Returns:
Deferred[None]
"""
return self.runInteraction(
"update_remote_device_list_cache_entry",
self._update_remote_device_list_cache_entry_txn,
user_id,
device_id,
content,
stream_id,
)
def _update_remote_device_list_cache_entry_txn(
self, txn, user_id, device_id, content, stream_id
):
if content.get("deleted"):
self._simple_delete_txn(
txn,
table="device_lists_remote_cache",
keyvalues={"user_id": user_id, "device_id": device_id},
)
txn.call_after(self.device_id_exists_cache.invalidate, (user_id, device_id))
else:
self._simple_upsert_txn(
txn,
table="device_lists_remote_cache",
keyvalues={"user_id": user_id, "device_id": device_id},
values={"content": json.dumps(content)},
# we don't need to lock, because we assume we are the only thread
# updating this user's devices.
lock=False,
)
txn.call_after(self._get_cached_user_device.invalidate, (user_id, device_id))
txn.call_after(self._get_cached_devices_for_user.invalidate, (user_id,))
txn.call_after(
self.get_device_list_last_stream_id_for_remote.invalidate, (user_id,)
)
self._simple_upsert_txn(
txn,
table="device_lists_remote_extremeties",
keyvalues={"user_id": user_id},
values={"stream_id": stream_id},
# again, we can assume we are the only thread updating this user's
# extremity.
lock=False,
)
def update_remote_device_list_cache(self, user_id, devices, stream_id):
"""Replace the entire cache of the remote user's devices.
Note: assumes that we are the only thread that can be updating this user's
device list.
Args:
user_id (str): User to update device list for
devices (list[dict]): list of device objects supplied over federation
stream_id (int): the version of the device list
Returns:
Deferred[None]
"""
return self.runInteraction(
"update_remote_device_list_cache",
self._update_remote_device_list_cache_txn,
user_id,
devices,
stream_id,
)
def _update_remote_device_list_cache_txn(self, txn, user_id, devices, stream_id):
self._simple_delete_txn(
txn, table="device_lists_remote_cache", keyvalues={"user_id": user_id}
)
self._simple_insert_many_txn(
txn,
table="device_lists_remote_cache",
values=[
{
"user_id": user_id,
"device_id": content["device_id"],
"content": json.dumps(content),
}
for content in devices
],
)
txn.call_after(self._get_cached_devices_for_user.invalidate, (user_id,))
txn.call_after(self._get_cached_user_device.invalidate_many, (user_id,))
txn.call_after(
self.get_device_list_last_stream_id_for_remote.invalidate, (user_id,)
)
self._simple_upsert_txn(
txn,
table="device_lists_remote_extremeties",
keyvalues={"user_id": user_id},
values={"stream_id": stream_id},
# we don't need to lock, because we can assume we are the only thread
# updating this user's extremity.
lock=False,
)
@defer.inlineCallbacks
def add_device_change_to_streams(self, user_id, device_ids, hosts):
"""Persist that a user's devices have been updated, and which hosts
(if any) should be poked.
"""
with self._device_list_id_gen.get_next() as stream_id:
yield self.runInteraction(
"add_device_change_to_streams",
self._add_device_change_txn,
user_id,
device_ids,
hosts,
stream_id,
)
defer.returnValue(stream_id)
def _add_device_change_txn(self, txn, user_id, device_ids, hosts, stream_id):
now = self._clock.time_msec()
txn.call_after(
self._device_list_stream_cache.entity_has_changed, user_id, stream_id
)
for host in hosts:
txn.call_after(
self._device_list_federation_stream_cache.entity_has_changed,
host,
stream_id,
)
# Delete older entries in the table, as we really only care about
# when the latest change happened.
txn.executemany(
"""
DELETE FROM device_lists_stream
WHERE user_id = ? AND device_id = ? AND stream_id < ?
""",
[(user_id, device_id, stream_id) for device_id in device_ids],
)
self._simple_insert_many_txn(
txn,
table="device_lists_stream",
values=[
{"stream_id": stream_id, "user_id": user_id, "device_id": device_id}
for device_id in device_ids
],
)
self._simple_insert_many_txn(
txn,
table="device_lists_outbound_pokes",
values=[
{
"destination": destination,
"stream_id": stream_id,
"user_id": user_id,
"device_id": device_id,
"sent": False,
"ts": now,
}
for destination in hosts
for device_id in device_ids
],
)
def _prune_old_outbound_device_pokes(self):
"""Delete old entries out of the device_lists_outbound_pokes to ensure
that we don't fill up due to dead servers. We keep one entry per
(destination, user_id) tuple to ensure that the prev_ids remain correct
if the server does come back.
"""
yesterday = self._clock.time_msec() - 24 * 60 * 60 * 1000
def _prune_txn(txn):
select_sql = """
SELECT destination, user_id, max(stream_id) as stream_id
FROM device_lists_outbound_pokes
GROUP BY destination, user_id
HAVING min(ts) < ? AND count(*) > 1
"""
txn.execute(select_sql, (yesterday,))
rows = txn.fetchall()
if not rows:
return
delete_sql = """
DELETE FROM device_lists_outbound_pokes
WHERE ts < ? AND destination = ? AND user_id = ? AND stream_id < ?
"""
txn.executemany(
delete_sql, ((yesterday, row[0], row[1], row[2]) for row in rows)
)
# Since we've deleted unsent deltas, we need to remove the entry
# of last successful sent so that the prev_ids are correctly set.
sql = """
DELETE FROM device_lists_outbound_last_success
WHERE destination = ? AND user_id = ?
"""
txn.executemany(sql, ((row[0], row[1]) for row in rows))
logger.info("Pruned %d device list outbound pokes", txn.rowcount)
return run_as_background_process(
"prune_old_outbound_device_pokes",
self.runInteraction,
"_prune_old_outbound_device_pokes",
_prune_txn,
)
@defer.inlineCallbacks
def _drop_device_list_streams_non_unique_indexes(self, progress, batch_size):
def f(conn):
txn = conn.cursor()
txn.execute("DROP INDEX IF EXISTS device_lists_remote_cache_id")
txn.execute("DROP INDEX IF EXISTS device_lists_remote_extremeties_id")
txn.close()
yield self.runWithConnection(f)
yield self._end_background_update(DROP_DEVICE_LIST_STREAMS_NON_UNIQUE_INDEXES)
defer.returnValue(1)