Fixes https://github.com/element-hq/synapse/issues/17274, hopefully.
Basically, old versions of Synapse could advance streams without
persisting anything in the DB (fixed in #17229). On restart those
updates would get lost, and so the position of the stream would revert
to an older position. If this happened across an upgrade to a later
Synapse version which included #17215, then sync could get blocked
indefinitely (until the stream advanced to the position in the token).
We fix this by bounding the stream positions we'll wait for to the
maximum position of the underlying stream ID generator.
Fixes up #17333, where we failed to actually send less data (the
`DISTINCT` didn't work due to `stream_id` being different).
We fix this by making it so that every device list outbound poke for a
given user ID has the same stream ID. We can't change the query to only
return e.g. max stream ID as the receivers look up the destinations to
send to by doing `SELECT WHERE stream_id = ?`
Reduce the replication traffic of device lists, by not sending every
destination that needs to be sent the device list update over
replication. Instead a "hosts to send to have been calculated"
notification over replication, and then federation senders read the
destinations from the DB.
For non federation senders this should heavily reduce the impact of a
user in many large rooms changing a device.
If the stream ID in the unconverted table is ahead of the device lists
ID gen, then it can break all /sync requests that had an ID from ahead
of the table.
The fix is to make sure we add the unconverted table to the list of
tables we check at start up.
Broke in https://github.com/element-hq/synapse/pull/17229
Re-introduces #17191, and includes #17197 and #17214
The basic idea is to stop calling `get_rooms_for_user` everywhere, and
instead use the table `device_lists_changes_in_room`.
Commits reviewable one-by-one.
It's almost always more efficient to query the rooms that have device
list changes, rather than looking at the list of all users whose devices
have changed and then look for shared rooms.
During the migration the automated script to update the copyright
headers accidentally got rid of some of the existing copyright lines.
Reinstate them.
If simple_{insert,upsert,update}_many_txn is called without any data
to modify then return instead of executing the query.
This matches the behavior of simple_{select,delete}_many_txn.
This avoids calling cursor_to_dict and then immediately
unpacking the values in the dict for other users. By not
creating the intermediate dictionary we can avoid allocating
the dictionary and strings for the keys, which should generally
be more performant.
Additionally this improves type hints by avoid Dict[str, Any]
dictionaries coming out of the database layer.
Old device entries for the same user were being removed in individual
SQL commands, making the batch take way longer than necessary.
This combines the commands into a single one with a IN/ANY clause.
Example of log entry before the change, regularly observed with
"log_min_duration_statement = 10000" in PostgreSQL's config:
LOG: duration: 42538.282 ms statement:
DELETE FROM device_lists_stream
WHERE user_id = '@someone' AND device_id = 'someid1'
AND stream_id < 123456789
;
DELETE FROM device_lists_stream
WHERE user_id = '@someone' AND device_id = 'someid2'
AND stream_id < 123456789
;
[repeated for each device ID of that user, potentially a lot...]
With the patch applied on my instance for the past couple of days, I
no longer notice overly long statements of that particular kind.
Signed-off-by: pacien <pacien.trangirard@pacien.net>
* Revert "Fix registering a device on an account with lots of devices (#15348)"
This reverts commit f0d8f66eaa.
* Revert "Delete stale non-e2e devices for users, take 3 (#15183)"
This reverts commit 78cdb72cd6.
This should help reduce the number of devices e.g. simple bots the repeatedly login rack up.
We only delete non-e2e devices as they should be safe to delete, whereas if we delete e2e devices for a user we may accidentally break their ability to receive e2e keys for a message.
AbstractStreamIdTracker (now) has only a single sub-class: AbstractStreamIdGenerator,
combine them to simplify some code and remove any direct references to
AbstractStreamIdTracker.
It's important that collections returned from `@cached` methods are not
modified, otherwise future retrievals from the cache will return the
modified collection.
This applies to the return values from `@cached` methods and the values
inside the dictionaries returned by `@cachedList` methods. It's not
necessary for the dictionaries returned by `@cachedList` methods
themselves to be read-only.
Signed-off-by: Sean Quah <seanq@matrix.org>
Co-authored-by: David Robertson <davidr@element.io>
The previous version of the code could mutate a cached value,
but only if the input requested all devices of a user *and* a specific
device.
To avoid this nonsensical situation we no longer fetch a specific
device ID if all of a user's devices are returned.
On startup, the `_device_list_id_gen` stream id generator is initialized
using the maximum stream id seen in a list of tables. When we started
populating the `device_list_remote_pending` table in #13913, we forgot
to add it to the aforementioned list of tables, so the stream id
generator can hand out old stream ids after a restart. The end result is
that Synapse can fail to handle device list update EDUs after a restart
when a partial state join is in progress.
Add the `device_list_remote_pending` table to the list of tables to
consider when initializing the `_device_list_id_gen` stream id generator.
Signed-off-by: Sean Quah <seanq@matrix.org>
This creates a new store method, `process_replication_position` that
is called after `process_replication_rows`. By moving stream ID advances
here this guarantees any relevant cache invalidations will have been
applied before the stream is advanced.
This avoids race conditions where Python switches between threads mid
way through processing the `process_replication_rows` method where stream
IDs may be advanced before caches are invalidated due to class resolution
ordering.
See this comment/issue for further discussion:
https://github.com/matrix-org/synapse/issues/14158#issuecomment-1344048703
This should help reduce the number of devices e.g. simple bots the repeatedly login rack up.
We only delete non-e2e devices as they should be safe to delete, whereas if we delete e2e devices for a user we may accidentally break their ability to receive e2e keys for a message.
StreamChangeCache.get_all_changed_entities can return None to signify
it does not have information at the given stream position. Two callers (related
to device lists and presence) were treating this response the same as an empty
list (i.e. there being no updates).
