This commit makes get_recipient_info() faster by never creating
Django ORM objects. We use the ORM to create a values query
instead, and then we iterate over the rows to create various
collections of ids.
In order to avoid lots of code duplication, this commit unifies
how we query UserProfile for PMs and streams. Prior to this
commit we were getting "wide" UserProfile objects out of
our memcached cache. Now we just go to the database with our
list of userids. The new approach at worst adds one hop to the
database for PMs, which aren't really a performance bottleneck
(compared to streams). And the new approach actually saves a
hop when both partners aren't in cache (plus we don't pay the
penalty of hitting the cache itself).
The performance improvement here is easy to measure for messages
to streams with many users, even with all the other activity
that goes on inside do_send_messages(). I took test_performance()
in test_messages.py, set num_extra_users to 3000, and consistently
measured a ~20% speedup in do_send_messages().
This commit also eliminates fetching of emails. We probably
could have done that in a prior commit, but in this commit it
is very explicit that we don't need it. While removing email
from the query is a no-brainer, it actually had a negigible
impact on performance. Almost all the savings here comes from
not create UserProfile objects.
This function returns a summary of recipient data for a message
that's being sent. It's mostly just moving code into the
old function called get_recipient_user_profiles().
This commit is necessary to prevent bringing back emails from the
DB for all N recipients of a message just to see if the feedback
bot is being invoked.
We calculate `service_bot_tuples` earlier in the function, so that
we don't need "full" UserProfile objects later in the function.
This is part of consolidating code that basically just needs to
triage user_ids.
This starts to phase out the need for UserProfile objects in
do_send_messages(). UserProfile objects are expensive to create
for large streams with lots of users. The objects in the code
before this commit aren't even full UserProfile objects.
This change mostly sets up future performance improvements, but
we also get a minor speedup here when we run a test with 3000
stream subscribers.
There is no reason for either render_incoming_message() or
render_markdown() to require full UserProfile objects just to
triage alert words.
By only asking for user_ids, we save extra queries in two
callpaths and we make it easier to start using user_ids in
do_send_messages().
This function is essentially a copy of get_recipient_user_profiles,
which is about to go away. The new function enforces the contract of
typing indicators, which is that they don't apply to streams, which
allows us to use a relatively simple approach for getting user
profile objects.
We are diverging this code, because the send-message path needs
more optimizations.
This change introduces an extra hop to the database, but it is
generally faster due to nuances of the DB and the ORM. It
also sets us up to optimize get_recipient_user_profiles() by
avoiding creating ORM objects.
I measured the impact of this using a stream with 3000
subscribers, half of whom were idle, and it speeds things up
by 10%.
Avoid a join to UserProfile here speeds up the query from
86ms -> 28ms when you analyze it with about 2000 mobile users
in a 5000-user realm.
We also avoid some code duplication here, since we filter
UserPresence for the same group of users as we filter
PushDeviceToken.
This avoids an O(N-squared) hit during presence queries. The speedup
here is probably negligible compared to everything else going on, but
sets are more semantically correct, anyway.
Before this commit, postgres would choose a non-optimal query
plan to find all presence rows belonging to a realm. We now
do an extra query to get the list of relevant user_ids, which allows
the next query to take advantage of UserPresence's index on
user_profile_id.
Here is the query plan for the offending query (this particular query isn't
verbatim from the code, but it's representative of the problem):
explain analyze
select client_id
from zerver_userpresence
INNER JOIN zerver_userprofile ON
zerver_userprofile.id = zerver_userpresence.user_profile_id
WHERE
zerver_userprofile.is_active and
zerver_userprofile.realm_id = 3;
Hash Join (cost=149.66..506.82 rows=5007 width=4) (actual time=48.834..121.215 rows=5007 loops=1)
Hash Cond: (zerver_userprofile.id = zerver_userpresence.user_profile_id)
-> Seq Scan on zerver_userprofile (cost=0.00..260.11 rows=5369 width=4) (actual time=0.009..24.322 rows=5021 loops=1)
Filter: (is_active AND (realm_id = 3))
Rows Removed by Filter: 3
-> Hash (cost=87.07..87.07 rows=5007 width=8) (actual time=48.789..48.789 rows=5010 loops=1)
Buckets: 1024 Batches: 1 Memory Usage: 196kB
-> Seq Scan on zerver_userpresence (cost=0.00..87.07 rows=5007 width=8) (actual time=0.007..24.355 rows=5010 loops=1)
Total runtime: 145.063 ms
You can see above that we're filtering on realm_id instead of using an index.
When you decompose the query into two queries, the total time is about 100ms, for a
savings of 33%. I imagine the savings would be even greater on an instance with lots
of realms. This was tested on dev with one really large realm and one tiny realm.
We were using `.order_by('user_profile_id', '-timestamp') in our
UserPresence query in get_status_dicts_for_query.
We don't need a full sort to produce the dictionary of statuses.
In fact the whole operation in Python is still O(N):
- divvy rows up to be per-user in an O(N) pass
- find max row for the 'aggregated' entry in an O(n) pass
per user
The one minor annoyance of this fix is that datetime_to_timestamp
is lossy, so if you naively call to_presence_dict before finding
the "max" row, you get test flakes if rows are created during the
same second. I decided to avoid calling to_presence_dict so there
are fewer moving parts, but there's still the ugly step of having
to remove the "dt" field from the final results.
The commit() call in fix() breaks migrations and tests (unless you
mock) due to outer transactions.
We now explicitly call commit() from the management command.
Usually a small minority of users are eligible to receive missed
message emails or mobile notifications.
We now filter users first before hitting UserPresence to find idle
users. We also simply check for the existence of recent activity
rather than borrowing the more complicated data structures that we
use for the buddy list.
This commit completely switches us over to using a
dedicated model called MutedTopic to track which topics
a user has muted.
This includes the necessary migrations to create the
table and populate it from legacy data in UserProfile.
A subsequent commit will actually remove the old field
in UserProfile.
Instead of peeking directly at the DB to verify our mutes are
set correctly, we now use the library function. This prepares
us to modify the DB internals while preserving the tests.
The double forward slash (//) after the protocol in URLs was being
mistakenly considered the beginning of an inline JS comment, causing
internationalization strings being cut unexpectedly.
Now the check for inline JS comments is only run in .js files.
Use this new variable to determine if the user already exists while
doing registration. While doing login through GitHub if we press
*Go back to login*, we pass email using email variable. As a result,
the login page starts showing the "User already exists error" if we
don't change the variable.
Admins need to know about private streams to delete them, even
if they are not subscribed. We send the minimal info possible
to the client to allow them to have a UI for that.
The refactor in b46af40bd3 didn't
correctly translate the code for managing request.user and
request._email, resulting in requests for the push notification
bouncer being rejected with this exception:
AttributeError: 'AnonymousUser' object has no attribute 'rate_limits'
This never made sense to be a flag on the UserMessage table, since
it's not per-user state. And in fact it doesn't need to be in a
database at all, since it's easily computed from content anyway.
Fixes#1099.
And it works!
A couple of things still to do:
* When a device token is no longer active, we'll get HTTP status 410.
We should then remove the token from the database so we don't keep
trying to push to it. This is fairly urgent.
* The library we're using has a nice asynchronous API, but this
version doesn't use it. This is OK now, but async will be
essential at scale.
This code empirically doesn't work. It's not entirely clear why, even
having done quite a bit of debugging; partly because the code is quite
convoluted, and because it shows the symptoms of people making changes
over time without really understanding how it was supposed to work.
Moreover, this code targets an old version of the APNs provider API.
Apple deprecated that in 2015, in favor of a shiny new one which uses
HTTP/2 to meet the same needs for concurrency and scale that the old
one had to do a bunch of ad-hoc protocol design for.
So, rip this code out. We'll build a pathway to the new API from
scratch; it's not that complicated.
We'd been getting errors from APNs that appeared to say that the
device tokens we were trying to send to were invalid. It turned out
that the device tokens didn't match the "topic" (i.e. app ID) we were
sending, which was because the topic was wrong, which was because we
were using the wrong SSL cert. But for a while we thought it might be
that we were somehow messing up the device tokens we put into the
database. This logging helped us work out that wasn't the issue, and
would have helped our debugging sooner.
This brings type-checking to the last place we fetch
data from Redis, with the exception of our APNs code
which is being replaced (with a Redis-free version,
thanks to improvements in Apple's APNs API) shortly.
This gives us type-checking, to help prevent bugs like the
last couple of commits fixed in our Tornado code and our
missed-message email handling. Fortunately no behavior
changes are needed here.
Because the Redis client returns exclusively bytes -- even for
hash keys -- even on Python 3, the test `'response' in status`
was always returning false, and the line that tries to decode
as JSON was never running, so we were passing `response`
through as a `bytes` object encoding some JSON.
I'm not sure what the impact of this bug was, and in particular
whether something downstream would have fudged it to make up for
this error.
Redis and the Redis client know nothing but bytes. When we take a
`bytes` object it returns and pass it down as `subject` here, it
causes an exception deep inside message processing if the realm has
any filters, when `bugdown.subject_links` attempts to search the
subject for the filters, which are of course `str` patterns.
For symmetry, make the conversion to bytes on the storing side
explicit too.
Previously, we didn't pass customized HTTP_HOST headers when making
network requests. As we move towards a world where everything is on a
subdomain, we'll want to start doing that.
The vast majority of our test code is written to interact with the
default "zulip" realm, which has a subdomain of "zulip". While
probably longer-term, we'll wish this was the root domain, for now, we
need to make our HTTP requests match what is expected by the test
code.
This commit almost certainly introduces some weird bugs where code was
expecting a different subdomain but the tests doesn't fail yet. It's
not clear how to find all of these, but I've done some grepping.
get_realm_by_email_domain was intended to be registration flow code
not used in other code, but it was leaked to a few places. This
removes one of the main remaining references to it outside the
registration code path.
The older fixture for this event assumed that the "assignee" key
had a value of '{}' if no one was assigned to a PR anymore.
However, that is no longer true because testing with requestbin
showed that in the latest JSON payload for this event, the key
"assignee" now has the value of 'null' (None when converted to
Python) when a user is unassigned from a PR. The current code
didn't handle this correctly. This commit makes sure it does!
Its unclear as to whether the old fixture was simply wrong or
whether GitHub changed its payloads in any way.
Previously, Zulip's server logs would not show which user or client
was involved in login or user registration actions, which made
debugging more annoying than it needed to be.
This is mostly pure code extraction.
It also removes some dead code in update_muted_topic, where
were updating muted_topics spuriously before calling
do_update_muted_topic.
Previously, realm.uri and realm.host didn't support using a subdomain
of the empty string (""), aka using the root domain.
Also, since we're already accessing self.subdomain, we don't need to
check REALMS_HAVE_SUBDOMAINS again.
Unlike creating a stream, there's really no reason one would want to
call the function to create a realm while uncertain whether that realm
already existed.
For filters like has:link, where the web app doesn't necessarily
want to guess whether incoming messages meet the criteria of the
filter, the server is asked to query rows that match the query.
Usually these queries are search queries, which have fields for
content_matches and subject_matches. Our logic was handling those
correctly.
Non-search queries were throwing an exception related to tuple
unpacking. Now we recognize when those fields are absent and
do the proper thing.
There are probably situations where the web app should stop hitting
this endpoint and just use its own filters. We are making the most
defensive fix first.
Fixes#6118
Most of the code in show_unreads is for diagnosising unread
counts issues, and we may not use that often.
We're creating a dedicated fix_unreads management command with
less clutter.
This change is mostly based on a similar commit from hackerkid
in a feature branch. It borrows both code and ideas. Some of
it's my own stuff, as I was working on a newer branch.
We now call get_user_including_cross_realm_email() inside of
user_profiles_from_unvalidated_emails(), instead of using
get_user_profile_by_email.
This requires a few of our callers to pass down sender into us.
One consequence of this change is that we change the symptoms
for trying to send to emails outside of your realm. In some
cases, we simply raise an error that an email is invalid to us
instead of getting into the deeper validate_recipient_user_profiles
check.
We are trying to convert emails to user_profiles earlier in
the codepath. This may cause subtle changes in which errors
appear, but it's probably generally good to report on bad
addressees sooner than later.
This class simplifies the calling sequence to methods like
check_message and _internal_prep_message, and it's also more
type safe.
Checking for message types is encapsulated with calls to is_stream()
and is_private(). There are also shortcut constructors when you
know that the type of the address (stream vs. private), which is often.
In this we basically seed a single message for the user which will
be soft deactivated by sending a stream message / group PM to
ensure that is has at least one UserMessage row, since in real
world every human user will always have at least one User Message
row.
This causes `upgrade-zulip-from-git`, as well as a no-option run of
`tools/build-release-tarball`, to produce a Zulip install running
Python 3, rather than Python 2. In particular this means that the
virtualenv we create, in which all application code runs, is Python 3.
One shebang line, on `zulip-ec2-configure-interfaces`, explicitly
keeps Python 2, and at least one external ops script, `wal-e`, also
still runs on Python 2. See discussion on the respective previous
commits that made those explicit. There may also be some other
third-party scripts we use, outside of this source tree and running
outside our virtualenv, that still run on Python 2.
An expression like `force_bytes(chr(...))`, on Python 3 where the
`force_bytes` finds itself with something to do because `chr` returns
a text string, gives the UTF-8 encoding of the given value as a
Unicode codepoint.
Here, we don't want that -- rather we want the given value as a
single byte. We can do that with `struct.pack`.
This fixes an issue where the "Link with Webathena" flow was producing
invalid credential caches when run on Python 3, breaking the Zephyr
mirror for any user who went through it anew.
This management command creates the same indexes as migrations
82, 83, and 95, which are all indexes on the huge UserMessage
table. (*)
This command quickly no-ops with clear messaging when the
indexes already exist, so it's idempotent in that regard. (If
somebody somehow creates an index by the same name incorrectly,
they can always drop it in dbshell and re-run this command.)
If any of the migrations have not been run, which we detect simply
by the existence of the indexes, then we create them using a
`CREATE INDEX CONCURRENTLY` command. This functionality in
postgres allows you to create indexes against large tables
without disrupting queries against those tables. The tradeoff
here is that creating indexes concurrently takes significantly
longer than doing them non-concurrently.
Since most tables are small, we typically just use regular
Django migrations and run them during a brief interval while
the app is down.
For indexes on big tables, we will want to run this command
as part of the upgrade process, and we will want to run
it while the app is still up, otherwise it's pointless.
All the code in create_indexes() is literally copy/pasted
from the relevant migrations, and that scheme should work
going forward. (It uses a different implementation of
create_index_if_not_exist than the migrations use, but the
code is identical lexically in the function.)
If we ever do major restructuring of our large tables, such
as UserMessage, and we end up droppping some of these indexes,
then we will need to make this command migrations-aware. For
now it's safe to assume that indexes are generally additive in
nature, and the sooner we create them during the upgrade process,
the better.
(*) UserMessage is huge for large installations, of course.
Before this change, server searches for both
`is:mentioned` and `is:alerted` would return all messages
where the user is specifically mentioned (but not
at-all mentions).
Now we follow the JS semantics:
is:mentioned -- all mentions, including wildcards
is:alerted -- has an alert word
Here is one relevant JS snippet:
} else if (operand === 'mentioned') {
return message.mentioned;
} else if (operand === 'alerted') {
return message.alerted;
And here you see that `mentioned` is OR'ed over both mention flags:
message.mentioned = convert_flag('mentioned') || convert_flag('wildcard_mentioned');
The `alerted` flag on the JS side is a simple mapping:
message.alerted = convert_flag('has_alert_word');
Fixes#5020
Given typeahed and the fact that this only worked if the person had a
full name that didn't contain whitespace, this side effect of the
original @shortname mentionfeature that we removed was experienced by
users as a bug.
Fixes#6142.
We apparently were using the default of num_before=1, not
num_before=0, which meant that if the very last randomly generated
message was one by cordelia mentioning lunch,
test_get_messages_with_search would fail because there were actually 3
matches.
This adds the authors to the Zulip repository on GitHub from
/authors/ along with re-styling the page to fit the same
aesthetic as /for/open-source/ and other product-pages.
This fixes the significant duplication of code between the
authenticate_log_and_execute_json code path and the `validate_api_key`
code path.
These's till a bit of duplication, in the form of `process_client` and
`request._email` interactions, but it is very minor at this point.
The old iOS app has been gone from the app store for 8 months, never
had a huge userbase, and its latest version didn't need this hack. So
this code is unlikely to do anything in the future; remove it to
declutter our authentication decorators codebase.
The check itself was correct, but the error message was in fact the
opposite of what this check is for. In other words, the only things
these users can do is post messages, and the error message when you
tried to do something else was to tell you that the user can't post
messages.
This technically changes the behavior in the case that
!settings.ZILENCER_ENABLED but is_remote_zulip_server(role).
Fortunately, that case is mostly irrelevant (in that remote zulip
servers is a Zilencer feature). The old behavior was also probably
slightly wrong, in that you'd get a zilencer-specific error message in
that case.
The new endpoints are:
/json/mark_stream_as_read: takes stream name
/json/mark_topic_as_read: takes stream name, topic name
The /json/flags endpoint no longer allows streams or topics
to be passed in as parameters.
This function optimizes marking streams and topics as read,
by using UserMessage.where_unread(), which uses a partial
index on the "read" flag.
This also simplifies the code path for ordinary message
flag updates.
In order to keep 100% line coverage, I simplified the
logging in update_message_flags, so now all requests
will show the "actually" format.
This is an interim step toward creating dedicated endpoints
for marking streams/topics as reads, so we do error checking
with asserts for flag/operation, so we don't introduce a
temporary translation string.
This is mostly a pure code extraction, except that we now
disregard the `messages` option for stream/topic updates,
since the web app always passes in an empty list (and this
commit is really just an incremental step toward creating
new endpoints.)
This is the first part of a larger migration to convert Zulip's
reactions storage to something based on the codepoint, not the emoji
name that the user typed in, so that we don't need to worry about
changes in the names we're using breaking the emoji storage.
We recently changed the populate_db data set to include more variable
message content, which happened to include the possibility of the word
"lunch" appearing in the test messages. This caused occasional
failures of the search tests that looked for messages containing
"lunch" starting at the beginning of time, not the beginning of the
test.
This commits adds new helper functions which are:
* get_users_for_soft_deactivation(): This function can be used to
fetch a list of human users which pass the criteria of minimum
inactivity period (in days) passed as a parameter to the function.
* do_soft_activate_users(): Given a list of users this function
reactivates them and help them catch up with the missing message
rows for them in the UserMessage table.
Steve Howell
Umair Khan
Brock Whittaker
Greg Price
Tim Abbott
Rishi Gupta
Yago González
Aditya Bansal
Preston Hansen
Vishnu Ks
rht
Eeshan Garg
Robert Hönig
David Taylor
James Rowan
Steve Howell
Vaida Plankyte
neiljp (Neil Pilgrim)
Jack Zhang