Rather than use a bulk insert via Django, use the faster
`bulk_insert_all_ums` that we already have. This also adds a `ON
CONFLICT` clause, to make the insert resilient to race conditions.
There are currently two callsites, with different desired `ON
CONFLICT` behaviours:
- For `notify_reaction_update`, if the `UserMessage` had already been
created, we would have done nothing to change it.
- For `do_update_message_flags`, we would have ensured a specific bit
was (un)set.
Extend `create_historical_user_messages` and `bulk_insert_all_ums` to
support `ON CONFLICT (...) UPDATE SET flags = ...`.
Translators benefit from the extra information in the field names, and
need the reordering freedom that isn’t available with multiple
positional fields.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Black 23 enforces some slightly more specific rules about empty line
counts and redundant parenthesis removal, but the result is still
compatible with Black 22.
(This does not actually upgrade our Python environment to Black 23
yet.)
Signed-off-by: Anders Kaseorg <anders@zulip.com>
We do not create historical UserMessage rows, for messages that didn't
have one, while marking messages as read and simply ignore those messages.
We do so because there is no user of creating UserMessage rows and it just
wastes storage.
Note that we still allow to mark messages from unsubscribed streams as
read but only those which have UserMessage rows for them to handle the
case when the unread messages were not marked as read while unsubscribing
from the stream due to some race condition. In such cases, messages
will not be included in the unread count shown in "All messages" menu
(and stream is anyways not present in the left sidebar), but the message
border on the left is green if viewing the stream after unsusbcribing it.
So, to avoid the confusion for users, the messages will be marked as read
when user scrolls down.
Zulip's unread messages design has an invariant that all unread stream
messages must be in streams the user is subscribed to. For example, We
do not include the unread messages from unsubscribed streams in the
"unread_msgs" data structure in "/register" response and we mark all
unread messages as read when unsubscribing a user from a stream.
Previously, the mark as unread endpoint allowed violating that
invariant, allowing you to mark messages in any stream as unread.
Doing so caused the "message_details" data structures sent with
"update_message_flags" events to not contain messages from
unsubscribed streams, even though those messages were present in the
set of message IDs. These malformed events, in turn, caused exceptions
in the frontend's processing of such an event.
This change is paired with a separate UI change to not offer the "Mark
as unread" feature in such streams; with just this commit, that will
silently fail.
With some additions to the tests by tabbott.
Previously, an active production Zulip server would experience a class
of deadlocks caused by two or more concurrent bulk update operations
on the UserMessage table.
This is because UPDATE ... SET ... WHERE statements that execute in
parallel take row-level UPDATE locks as they get results; since the
query plans may result in getting rows in different orders between two
queries, this can result in deadlocks.
Some databases allow ORDER BY on their UPDATE ... WHERE statements;
PostgreSQL does not. In PostgreSQL, the answer is to do a sub-select
with an ORDER BY ... FOR UPDATE to ensure consistent ordering on row
locks.
We do this all code paths using bitand or bitor as part of bulk
editing message flags, which should ensure that these concurrent
operations obtain row level locks on the table in the same order.
Fixes#19054.
This reverts commit 40fcf5a633.
This commit triggers bug that we haven't fully tracked down, where web
app clients will continually send `update_message_flags` requests,
that then send out via the events system "0 messages were marked as
read" notices, eventually leading to a load spike.
The Tornado part can likely be fixed by checking if
updated_message_ids is empty, but we need to track down the frontend
bug as well.
We were blindly adding / removing flag from UserMessages without
check if they even need to be updated.
This caused server to repeatedly update flags for messages which
already had been updated, creating a confusion for other clients
like mobile.
Fixes#22164
Alex Vandiver
Anders Kaseorg
Sahil Batra
Tim Abbott
Mateusz Mandera
Christopher Chong
Aman Agrawal