The query plan for fetching recent messages from the arbitrary set of
streams formed by the intersection of 30 random users can be quite
bad, and can descend into a sequential scan on `zerver_recipient`.
Worse, this work of pulling recent messages out is redone if the
stream appears in the next batch of 30 users.
Instead, pull the recent messages for a stream on a one-by-one basis,
but cache them in an in-memory cache. Since digests are enqueued in
30-user batches but still one-realm-at-a-time, work will be saved both
in terms of faster query plans whose results can also be reused across
batches.
This requires that we pull the stream-id to stream-name mapping for
_all_ streams in the realm at once, but that is well-indexed and
unlikely to cause performance issues -- in fact, it may be faster
than pulling a random subset of the streams in the realm.
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>
This adds a helper based on testing patterns of using the "queries_captured"
context manager with "assert_length" to check the number of queries
executed for preventing performance regression.
It explains the rationale of checking the query count through an
"AssertionError" and prints the queries captured as assert_length does,
but with a format optimized for displaying the queries in a more
readable manner.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
This commit adds the OPTIONAL .realm attribute to Message
(and ArchivedMessage), with the server changes for making new Messages
have this set. Old Messages still have to be migrated to backfill this,
before it can be non-nullable.
Appropriate test changes to correctly set .realm for Messages the tests
manually create are included here as well.
Just using values 1 and 2 as stream ids is not good, because there's no
idea in which realm these streams are (or hypothetically if they exist).
This can create weird Messages with sender being a user of "zulip" realm
and the stream being in another realm - which would be a corrupted
state.
In English, compound adjectives should essentially always be
hyphenated. This makes them easier to parse, especially for users who
might not recognize that the words “web public” go together as a
phrase.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
We now complain if a test author sends a stream message
that does not result in the sender getting a
UserMessage row for the message.
This is basically 100% equivalent to complaining that
the author failed to subscribe the sender to the stream
as part of the test setup, as far as I can tell, so the
AssertionError instructs the author to subscribe the
sender to the stream.
We exempt bots from this check, although it is
plausible we should only exempt the system bots like
the notification bot.
I considered auto-subscribing the sender to the stream,
but that can be a little more expensive than the
current check, and we generally want test setup to be
explicit.
If there is some legitimate way than a subscribed human
sender can't get a UserMessage, then we probably want
an explicit test for that, or we may want to change the
backend to just write a UserMessage row in that
hypothetical situation.
For most tests, including almost all the ones fixed
here, the author just wants their test setup to
realistically reflect normal operation, and often devs
may not realize that Cordelia is not subscribed to
Denmark or not realize that Hamlet is not subscribed to
Scotland.
Some of us don't remember our Shakespeare from high
school, and our stream subscriptions don't even
necessarily reflect which countries the Bard placed his
characters in.
There may also be some legitimate use case where an
author wants to simulate sending a message to an
unsubscribed stream, but for those edge cases, they can
always set allow_unsubscribed_sender to True.
This makes us more efficient when handling
multiple users. We don't have to keep
sending the same two queries to the database.
Note that as part of this we eliminated
a failure mode for the obscure population
of users from whom both `user.is_guest` and
`user.can_access_public_streams()` returns
False. We know this would have only affected
Zephyr users (by looking at the code), and
we know we don't actually process Zephyr
users for email digests (or else we would
have raised exceptions in the old code).
We mostly need realm_id, but when we go to build
message lists, we need realm.uri.
We could probably be more aggresive about using
`only` here, but for now I am just trying to
reduce hops to the database.
Note that we are much more efficient about finding
active users here:
- we do one query per realm (instead of per-user)
- we pass the cutoff date to the database
- we get back just a list of distinct ids
The query counts increase here for somewhat
contrived reasons. The tests before this
commit reflected a successful trip to the
UserProfile cache, but that's not actually
realistic in practice.
We don't need to mock the dates here. We also
explicitly clear out all streams first, and then
we explicitly test with both the stream being
current and the stream being old.
We can use the _enqueue_emails_for_realm helper
to avoid all the Tuesday-related logic here.
We also don't bother to create UserActivity
records, since the bot gets excluded by virtue
of its being a bot. (Also, the date ranges
here were sketchy due to the time mocking.)
We can avoid all the date mocking now for all
but a couple tests that exercise the is-it-Tuesday
logic.
And this test now correctly tests that we exclude
recently active users.
And this allows us to remove the other test.
I now use sets for stream_ids in more of the digest
code.
As part of this I replaced exclude_subscription_modified_streams
with streams_recently_modified_for_user.
It's easier for the caller to just ask for ids
to delete from its callee than it is to pass
in a set/list to mutate.
The simpler boundary between the functions makes
the tests easier to write--you can see the
`filtered_streams` logic goes away in this diff.
I also make the tests a bit more thorough by using
combinations of Cordelia/Othello and Verona/Denmark
to try to find multiple possible flaws.
And I make the time intervals longer than 1s to
avoid false negatives from slow CI boxes.
If we have multiple users, this reduces the amount
of queries we need to do, because we get all
subscriptions for all users in a single query
to Subscription.
For the single-user case, we are introducing an
extra query hop, but the database is doing
roughly the same work, because we are just breaking
up this complex query into two hops:
messages =
select ... from message
where recipient__type_id in (
select stream_id from subscription
where ...
)
Now it's more like:
stream_ids =
select stream_id from subscription
where ...
messages =
select ... from message
where recipient__type_id in stream_ids
Note that we are not changing anything semantically
or algorithmically yet. The only overhead here
for the single-user case is boxing and unboxing
data into single-item dicts and lists.
The interfaces for callers in the view and the
queue processor remain the same for now.
Anders Kaseorg
Alex Vandiver
Steve Howell
Zixuan James Li
Mateusz Mandera
Adam Sah
Steve Howell
akshatdalton
shanukun
Vishnu KS