Since essentially the first use of Tornado in Zulip, we've been
maintaining our Tornado+Django system, AsyncDjangoHandler, with
several hundred lines of Django code copied into it.
The goal for that code was simple: We wanted a way to use our Django
middleware (for code sharing reasons) inside a Tornado process (since
we wanted to use Tornado for our async events system).
As part of the Django 2.2.x upgrade, I looked at upgrading this
implementation to be based off modern Django, and it's definitely
possible to do that:
* Continue forking load_middleware to save response middleware.
* Continue manually running the Django response middleware.
* Continue working out a hack involving copying all of _get_response
to change a couple lines allowing us our Tornado code to not
actually return the Django HttpResponse so we can long-poll. The
previous hack of returning None stopped being viable with the Django 2.2
MiddlewareMixin.__call__ implementation.
But I decided to take this opportunity to look at trying to avoid
copying material Django code, and there is a way to do it:
* Replace RespondAsynchronously with a response.asynchronous attribute
on the HttpResponse; this allows Django to run its normal plumbing
happily in a way that should be stable over time, and then we
proceed to discard the response inside the Tornado `get()` method to
implement long-polling. (Better yet might be raising an
exception?). This lets us eliminate maintaining a patched copy of
_get_response.
* Removing the @asynchronous decorator, which didn't add anything now
that we only have one API endpoint backend (with two frontend call
points) that could call into this. Combined with the last bullet,
this lets us remove a significant hack from our
never_cache_responses function.
* Calling the normal Django `get_response` method from zulip_finish
after creating a duplicate request to process, rather than writing
totally custom code to do that. This lets us eliminate maintaining
a patched copy of Django's load_middleware.
* Adding detailed comments explaining how this is supposed to work,
what problems we encounter, and how we solve various problems, which
is critical to being able to modify this code in the future.
A key advantage of these changes is that the exact same code should
work on Django 1.11, Django 2.2, and Django 3.x, because we're no
longer copying large blocks of core Django code and thus should be
much less vulnerable to refactors.
There may be a modest performance downside, in that we now run both
request and response middleware twice when longpolling (once for the
request we discard). We may be able to avoid the expensive part of
it, Zulip's own request/response middleware, with a bit of additional
custom code to save work for requests where we're planning to discard
the response. Profiling will be important to understanding what's
worth doing here.
Hellosign now posts their callback as form/multipart, which Django only
permits to be read once. Attempts to access request.body after the
initial read throw "django.http.request.RawPostDataException: You
cannot access body after reading from request's data stream".
Fixes#13847.
This makes it possible to create a Zulip account from the mobile or
desktop apps and have the end result be that the user is logged in on
their mobile device.
We may need small changes in the desktop and/or mobile apps to support
this.
Closes#10859.
1) Created a new class `DatabaseType` and access its objects inside
`template_database_status()` instead of sending five arguments with
default values.
2) Made `check_files` and `setting_name` local variables instead of
function parameters since they had same value(None) for every call.
Fixes#13845.
webpack optimizes JSON modules using JSON.parse("{…}"), which is
faster than the normal JavaScript parser.
Update the backend to use emoji_codes.json too instead of the three
separate JSON files.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
This code was very useful when first implemented to help catch errors
where our backend templates didn't render, but has been superceded by
the success of our URL coverage testing (which ensures every URL
supported by Zulip's urls.py is accessed by our tests, with a few
exceptions) and other tests covering all of the emails Zulip sends.
It has a significant maintenance cost because it's a bit hacky and
involves generating fake context, so it makes sense to remove these.
Any future coverage issues with templates should be addressed with a
direct test that just accessing the relevant URL or sends the relevant
email.
In python 3.5-3.6, generic types had an __origin__ attribute which
indicated which generic they originated from; the code was reflecting on
that value to check types against the openapi spec. In python3.7, this
changed, and there's no longer an immediately simple way to get this
information in all cases. __origin__ appears to be the implementing
class now, returning `list` or `collections.abc.Iterator` rather than
`typing.List` and `typing.Iterator`. This adds a sloppy-but-effective
mechanism for inferring if a type maps to the List/Dict/Iterator/Mapping
types and gets the test suite passing again.
We now use realm_id for querying UserPresence
instead of building a big WHERE clause from the
list of user_ids.
This commit may be a bit hard to measure, since
we still get the list of user_ids for the PushToken
query in the same method.
It adds this index:
"zerver_userpresence_realm_id_timestamp_25f410da_idx" btree (realm_id, "timestamp")
We expect this index to provide a major performance improvement when
fetching presence data for the whole realm from the database on
servers like zulipchat.com hosting several realms.
We now validate streams with a separate
function from PM recipients.
It's confusing enough all the ways you can
encode a stream or encode the PM recipients,
but trying to do it all in one function was
hard to reason about and led to at least one
bug.
In particular, there was a bug where streams
with commas in them would get split. Now
we just don't ever split on commas inside
of `extract_stream_indicator`.
Fixes#13836
After removing internal_send_message() in a recent
commit, we now have only two callers for
extract_recipients, and they are both related
to our REQ mechanism that always passes strings
to converters. (If there are default values,
REQ does not call the converters.)
We therefore make two changes:
- use the more strict annotation of "str"
for the `s` parameter
- don't bother with the isinstance check
Note that while the test mocks the actual message
send, we now have a `get_stream` call in the queue
worker, so we have to set up a real stream for
testing (or we could have mocked that as well, but
it didn't seem necessary). The setup queries add
to the amount of queries reported by the test,
plus the `get_stream` call. I just made the
query count a digits regex, which is a little bit
lame, but I don't think it's worth risking test
flakes for this.
This index is intended to optimize the performance of the very
frequently run query of "what is the presence status of all users in a
realm?".
Main changes:
- add realm_id to UserPresence
- add index for realm_id
- backfill realm_id for old rows
- change all writes to UserPresence to include
realm_id
The index is of this form:
"zerver_userpresence_realm_id_5c4ef5a9" btree (realm_id)
We will create an index on (realm_id, timestamp) in a
future commit, but I think it's a bit faster if you do
the backfill before the index.
There's also a minor tweak to the populate_db script.
This is just a refactoring to the more modern API
for sending internal messages.
To make this work we now plumb the email_gateway
flag through `internal_send_stream_message` instead
of `internal_send_message`.
We also change `send_zulip` to have its callers
pass in a full UserProfile object (which one of
them already had).
We prefer this to internal_send_message().
We are trying to deprecate `internal_send_message`,
which has extra moving parts related to
`extract_recipients` and `Addressee.legacy_build`.
There are two chunks of code that I touch here
that look pretty similar, but I'm not quite
sure they're worth de-duplicating, since they
use different topics and different message
content.
Instead of having `notify_new_user` delegate
all the heavy lifting to `send_signup_message`,
we just rename `send_signup_message` to be
`notify_new_user` and remove the one-line
wrapper.
We remove a lot of obsolete complexity:
- `internal` was no longer ever set to True
by real code, so we kill it off as well
as well as killing off the internal_blurb code
and the now-obsolete test
- the `sender` parameter was actually an
email, not a UserProfile, but I think
that got past mypy due to the caller
passing in something from settings.py
- we were only passing in NOTIFICATION_BOT
for the sender, so we just hard code
that now
- we eliminate the verbose
`admin_realm_signup_notifications_stream`
parameter and just hard code it to
"signups"
- we weren't using the optional realm
parameter
There's also a long ugly comment in
`get_recipient_info` related to this code
that I amended for now.
We should try to take action in a subsequent
commit.
This avoids an unnecessary join to UserProfile.
To verify this, you can do `print(queries)` in the
`test_get_custom_profile_fields_from_api` test. It's
kinda noisy, so I excerpted them below...
Before:
SELECT ...
FROM "zerver_customprofilefieldvalue"
INNER JOIN "zerver_userprofile" ON ("zerver_customprofilefieldvalue"."user_profile_id" = "zerver_userprofile"."id")
INNER JOIN "zerver_customprofilefield" ON ("zerver_customprofilefieldvalue"."field_id" = "zerver_customprofilefield"."id")
WHERE "zerver_userprofile"."realm_id" = 2
After:
SELECT ...
FROM "zerver_customprofilefieldvalue"
INNER JOIN "zerver_customprofilefield" ON ("zerver_customprofilefieldvalue"."field_id" = "zerver_customprofilefield"."id")
WHERE "zerver_customprofilefield"."realm_id" = 2'
I don't have any way to measure the two queries with
realistic data, but I would assume the second
query is significantly faster on most of our instances,
since CustomProfileField should be tiny.
I am trying to optimize a query in this endpoint.
I don't think I'll actually reduce the number of
queries, but I wanted to capture the query and
this was the easiest way to do it, so might as
well check in the code! :)
The line removed here is a noop, as both sides of the
immediately following conditional reassign the
same variable.
This harmless cruft was the result of the recent commit
1ae5964ab8, which added
support for single-user GETs.
This fixes a bug where our asynchronous requests were only copying the
Content-Type header (i.e. the one case where we're noticed) from the
Django HttpResponse. I'm not sure what the impact of this would be;
the rate-limiting headers rarely come up when breaking a long-polled
request. But it seems clearly an improvement to do this in a
consistent fashion.
Only the headers piece is a change; in Tornado
self.finish(x)
is equivalent to:
self.write(x)
self.finish()
Apparently, the arguments passed to template_database_status were
incorrect for the manual testing development database, in that we
didn't pass a status_dir when calling into that code from provision.
The result was that provisioning before running `test-backend` would
ignore changes to the list of check_files (etc.) made after rebasing,
and vice versa.
The cleanest fix is to compute status_dir from other values passed in;
I'm also going to open a follow-up issue for creating a better overall
interface here.
This adds a new API endpoint for querying basic data on a single other
user in the organization, reusing the existing infrastructure (and
view function!) for getting data on all users in an organization.
Fixes#12277.
This code is a bit flatter and just preps the data
for a single user. There is never any interaction
between the data for user A and user B, so we can
mostly avoid complicated nested data structures
and do most of the data-crunching on a per-user basis.
We also do an explicit sort of the data before
running it through groupby. The explicit sort
simplifies how we calculate `most_recent_info`
and also avoids needing to add `dt` to an intermediate
data structure.
Finally, when it comes to the individual client data,
the code has relied on the assumption that there is
only one row per client, which I believe to be true,
but now the code is more explicit about that.
The word "status" is vague, and this isn't
actually returning a list, so we now name it
get_presence_response.
I originally was gonna rename this to
get_presence_dict, but there's a function
called get_status_dict that returns a subset
of the response, so I think it's a bit more
clear that this is the bigger dict that
actually gets sent back.
We want to err on the side of server_timestamp being
old, since we may eventually use this to make responses
just include incremental changes, and we don't want a
time window (however small) when we miss presence rows.
The clients will be able to deal with duplicate data
to the extent that the time windows are overlapping.
Also, extracting the other local var here
(for `presences`) will set up a subsequent commit
where we re-format the data for clients with
slim_presence=True.
Tim Abbott
Chris Heald
Mateusz Mandera
Hashir Sarwar
Anders Kaseorg
Dinesh
Steve Howell
Vishnu KS
akashaviator