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).
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.
Co-authored-by: Patrick Cloke <clokep@users.noreply.github.com>
Co-authored-by: Sean Quah <8349537+squahtx@users.noreply.github.com>
To perform an emulated upsert into a table safely, we must either:
* lock the table,
* be the only writer upserting into the table
* or rely on another unique index being present.
When the 2nd or 3rd cases were applicable, we previously avoided locking
the table as an optimization. However, as seen in #14406, it is easy to
slip up when adding new schema deltas and corrupt the database.
The only time we lock when performing emulated upserts is while waiting
for background updates on postgres. On sqlite, we do no locking at all.
Let's remove the option to skip locking tables, so that we don't shoot
ourselves in the foot again.
Signed-off-by: Sean Quah <seanq@matrix.org>
When a local device list change is added to
`device_lists_changes_in_room`, the `converted_to_destinations` flag is
set to `FALSE` and the `_handle_new_device_update_async` background
process is started. This background process looks for unconverted rows
in `device_lists_changes_in_room`, copies them to
`device_lists_outbound_pokes` and updates the flag.
To update the `converted_to_destinations` flag, the database performs a
`DELETE` and `INSERT` internally, which fragments the table. To avoid
this, track unconverted rows using a `(stream ID, room ID)` position
instead of the flag.
From now on, the `converted_to_destinations` column indicates rows that
need converting to outbound pokes, but does not indicate whether the
conversion has already taken place.
Closes#14037.
Signed-off-by: Sean Quah <seanq@matrix.org>
* Add tests for StreamIdGenerator
* Drive-by: annotate all defs
* Revert "Revert "Remove slaved id tracker (#14376)" (#14463)"
This reverts commit d63814fd73, which in
turn reverted 36097e88c4. This restores
the latter.
* Fix StreamIdGenerator not handling unpersisted IDs
Spotted by @erikjohnston.
Closes#14456.
* Changelog
Co-authored-by: Nick Mills-Barrett <nick@fizzadar.com>
Co-authored-by: Erik Johnston <erik@matrix.org>
Remove type hints from comments which have been added
as Python type hints. This helps avoid drift between comments
and reality, as well as removing redundant information.
Also adds some missing type hints which were simple to fill in.
This matches the multi instance writer ID generator class which can
both handle advancing the current token over replication and by calling
the database.
c.f. #12993 (comment), point 3
This stores all device list updates that we receive while partial joins are ongoing, and processes them once we have the full state.
Note: We don't actually process the device lists in the same ways as if we weren't partially joined. Instead of updating the device list remote cache, we simply notify local users that a change in the remote user's devices has happened. I think this is safe as if the local user requests the keys for the remote user and we don't have them we'll simply fetch them as normal.
By always using delete_devices and sometimes passing a list
with a single device ID.
Previously these methods had gotten out of sync with each
other and it seems there's little benefit to the single-device
variant.
This is a first step in dealing with #7721.
The idea is basically that rather than calculating the full set of users a device list update needs to be sent to up front, we instead simply record the rooms the user was in at the time of the change. This will allow a few things:
1. we can defer calculating the set of remote servers that need to be poked about the change; and
2. during `/sync` and `/keys/changes` we can avoid also avoid calculating users who share rooms with other users, and instead just look at the rooms that have changed.
However, care needs to be taken to correctly handle server downgrades. As such this PR writes to both `device_lists_changes_in_room` and the `device_lists_outbound_pokes` table synchronously. In a future release we can then bump the database schema compat version to `69` and then we can assume that the new `device_lists_changes_in_room` exists and is handled.
There is a temporary option to disable writing to `device_lists_outbound_pokes` synchronously, allowing us to test the new code path does work (and by implication upgrading to a future release and downgrading to this one will work correctly).
Note: Ideally we'd do the calculation of room to servers on a worker (e.g. the background worker), but currently only master can write to the `device_list_outbound_pokes` table.
For users with large accounts it is inefficient to calculate the set of
users they share a room with (and takes a lot of space in the cache).
Instead we can look at users whose devices have changed since the last
sync and check if they share a room with the syncing user.
Debug for #8631.
I'm having a hard time tracking down what's going wrong in that issue.
In the reported example, I could see server A sending federation traffic
to server B and all was well. Yet B reports out-of-sync device updates
from A.
I couldn't see what was _in_ the events being sent from A to B. So I
have added some crude logging to track
- when we have updates to send to a remote HS
- the edus we actually accumulate to send
- when a federation transaction includes a device list update edu
- when such an EDU is received
This is a bit of a sledgehammer.
