c.f. #15264
The two changes are:
1. Add indexes so that the select / deletes don't do sequential scans
2. Don't repeatedly call `SELECT count(*)` each iteration, as that's slow
* Scaffolding for background process to refresh profiles
* Add scaffolding for background process to refresh profiles for a given server
* Implement the code to select servers to refresh from
* Ensure we don't build up multiple looping calls
* Make `get_profile` able to respect backoffs
* Add logic for refreshing users
* When backing off, schedule a refresh when the backoff is over
* Wake up the background processes when we receive an interesting state event
* Add tests
* Newsfile
Signed-off-by: Olivier Wilkinson (reivilibre) <oliverw@matrix.org>
* Add comment about 1<<62
---------
Signed-off-by: Olivier Wilkinson (reivilibre) <oliverw@matrix.org>
* Remove special-case method for new memberships only, use more generic method
* Only collect profiles from state events in public rooms
* Add a table to track stale remote user profiles
* Add store methods to set and delete rows in this new table
* Mark remote profiles as stale when a member state event comes in to a private room
* Newsfile
Signed-off-by: Olivier Wilkinson (reivilibre) <oliverw@matrix.org>
* Simplify by removing Optionality of `event_id`
* Replace names and avatars with None if they're set to dodgy things
I think this makes more sense anyway.
* Move schema delta to 74 (I missed the boat?)
* Turns out these can be None after all
---------
Signed-off-by: Olivier Wilkinson (reivilibre) <oliverw@matrix.org>
* Fix a long-standing bug where non-ASCII characters in search terms,
including accented letters, would not match characters in a different
case.
* Fix a long-standing bug where search terms using combining accents
would not match display names using precomposed accents and vice
versa.
To fully take effect, the user directory must be rebuilt after this
change.
Fixes#14630.
Signed-off-by: Sean Quah <seanq@matrix.org>
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>
Fixes#13655
This change uses ICU (International Components for Unicode) to improve boundary detection in user search.
This change also adds a new dependency on libicu-dev and pkg-config for the Debian packages, which are available in all supported distros.
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>
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.
* Formally type the UserProfile in user searches
* export UserProfile in synapse.module_api
* Update docs
Co-authored-by: Sean Quah <8349537+squahtx@users.noreply.github.com>
The following scenarios would halt the user directory updater:
- user joins room
- user leaves room
- user present in room which switches from private to public, or vice versa.
for two classes of users:
- appservice senders
- users missing from the user table.
If this happened, the user directory would be stuck, unable to make forward progress.
Exclude both cases from the user directory, so that we ignore them.
Co-authored-by: Eric Eastwood <erice@element.io>
Co-authored-by: reivilibre <oliverw@matrix.org>
Co-authored-by: Sean Quah <8349537+squahtx@users.noreply.github.com>
Co-authored-by: Brendan Abolivier <babolivier@matrix.org>
There are two steps to rebuilding the user directory:
1. a scan over rooms, followed by
2. a scan over local users.
The former reads avatars and display names from the `room_memberships`
table and therefore contains potentially private avatars and
display names. The latter reads from the the `profiles` table which only
contains public data; moreover it will overwrite any private profiles
that the rooms scan may have written to the user directory. This means
that the rebuild could leak private user while the rebuild was in
progress, only to later cover up the leaks once the rebuild had completed.
This change skips over local users when writing user_directory rows
when scanning rooms. Doing so means that it'll take longer for a rebuild
to make local users searchable, which is unfortunate. I think a future
PR can improve this by swapping the order of the two steps above. (And
indeed there's more to do here, e.g. copying from `profiles` without
going via Python.)
Small tidy-ups while I'm here:
* Remove duplicated code from test_initial. This was meant to be pulled into `purge_and_rebuild_user_dir`.
* Move `is_public` before updating sharing tables. No functional change; it's still before the first read of `is_public`.
* Don't bother creating a set from dict keys. Slightly nicer and makes the code simpler.
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* Introduce `should_include_local_users_in_dir`
We exclude three kinds of local users from the user_directory tables. At
present we don't consistently exclude all three in the same places. This
commit introduces a new function to gather those exclusion conditions
together. Because we have to handle local and remote users in different
ways, I've made that function only consider the case of remote users.
It's the caller's responsibility to make the local versus remote
distinction clear and correct.
A test fixup is required. The test now hits a path which makes db
queries against the users table. The expected rows were missing, because
we were using a dummy user that hadn't actually been registered.
We also add new test cases to covert the exclusion logic.
----
By my reading this makes these changes:
* When an app service user registers or changes their profile, they will
_not_ be added to the user directory. (Previously only support and
deactivated users were excluded). This is consistent with the logic that
rebuilds the user directory. See also [the discussion
here](https://github.com/matrix-org/synapse/pull/10914#discussion_r716859548).
* When rebuilding the directory, exclude support and disabled users from
room sharing tables. Previously only appservice users were excluded.
* Exclude all three categories of local users when rebuilding the
directory. Previously `_populate_user_directory_process_users` didn't do
any exclusion.
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* Pull out GetUserDirectoryTables helper
* Don't rebuild the dir in tests that don't need it
In #10796 I changed registering a user to add directory entries under.
This means we don't have to force a directory regbuild in to tests of
the user directory search.
* Move test_initial to tests/storage
* Add type hints to both test_user_directory files
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* Improve typing in user_directory files
This makes the user_directory.py in storage pass most of mypy's
checks (including `no-untyped-defs`). Unfortunately that file is in the
tangled web of Store class inheritance so doesn't pass mypy at the moment.
The handlers directory has already been mypyed.
Co-authored-by: reivilibre <olivier@librepush.net>
It's a simplification, but one that'll help make the user directory logic easier
to follow with the other changes upcoming. It's not strictly required for those
changes, but this will help simplify the resulting logic that listens for
`m.room.member` events and generally make the logic easier to follow.
This means the config option `search_all_users` ends up controlling the
search query only, and not the data we store. The cost of doing so is an
extra row in the `user_directory` and `user_directory_search` tables for
each local user which
- belongs to no public rooms
- belongs to no private rooms of size ≥ 2
I think the cost of this will be marginal (since they'll already have entries
in `users` and `profiles` anyway).
As a small upside, a homeserver whose directory was built with this
change can toggle `search_all_users` without having to rebuild their
directory.
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
Now that we have `simple_upsert` that should be used in preference to
trying to insert and looking for an exception. The main benefit is that
we ERROR message don't get written to postgres logs.
We also have tidy up the return value on `simple_upsert`, rather than
having a tri-state of inserted/not-inserted/unknown.
Part of #9744
Removes all redundant `# -*- coding: utf-8 -*-` lines from files, as python 3 automatically reads source code as utf-8 now.
`Signed-off-by: Jonathan de Jong <jonathan@automatia.nl>`
This PR remove the cache for the `get_shared_rooms_for_users` storage method (the db method driving the experimental "what rooms do I share with this user?" feature: [MSC2666](https://github.com/matrix-org/matrix-doc/pull/2666)). Currently subsequent requests to the endpoint will return the same result, even if your shared rooms with that user have changed.
The cache was added in https://github.com/matrix-org/synapse/pull/7785, but we forgot to ensure it was invalidated appropriately.
Upon attempting to invalidate it, I found that the cache had to be entirely invalidated whenever a user (remote or local) joined or left a room. This didn't make for a very useful cache, especially for a function that may or may not be called very often. Thus, I've opted to remove it instead of invalidating it.
This PR adds a homeserver config option, `user_directory.prefer_local_users`, that when enabled will show local users higher in user directory search results than remote users. This option is off by default.
Note that turning this on doesn't necessarily mean that remote users will always be put below local users, but they should be assuming all other ranking factors (search query match, profile information present etc) are identical.
This is useful for, say, University networks that are openly federating, but want to prioritise local students and staff in the user directory over other random users.
- Update black version to the latest
- Run black auto formatting over the codebase
- Run autoformatting according to [`docs/code_style.md
`](80d6dc9783/docs/code_style.md)
- Update `code_style.md` docs around installing black to use the correct version
Autocommit means that we don't wrap the functions in transactions, and instead get executed directly. Introduced in #8456. This will help:
1. reduce the number of `could not serialize access due to concurrent delete` errors that we see (though there are a few functions that often cause serialization errors that we don't fix here);
2. improve the DB performance, as it no longer needs to deal with the overhead of `REPEATABLE READ` isolation levels; and
3. improve wall clock speed of these functions, as we no longer need to send `BEGIN` and `COMMIT` to the DB.
Some notes about the differences between autocommit mode and our default `REPEATABLE READ` transactions:
1. Currently `autocommit` only applies when using PostgreSQL, and is ignored when using SQLite (due to silliness with [Twisted DB classes](https://twistedmatrix.com/trac/ticket/9998)).
2. Autocommit functions may get retried on error, which means they can get applied *twice* (or more) to the DB (since they are not in a transaction the previous call would not get rolled back). This means that the functions need to be idempotent (or otherwise not care about being called multiple times). Read queries, simple deletes, and updates/upserts that replace rows (rather than generating new values from existing rows) are all idempotent.
3. Autocommit functions no longer get executed in [`REPEATABLE READ`](https://www.postgresql.org/docs/current/transaction-iso.html) isolation level, and so data can change queries, which is fine for single statement queries.