We have already calculated these values, so storing them should not cause
significant performance degradation.
This is a prep chenge for sending a few more flags through internal_data,
namely if `sender_is_muted`.
* In `event_queue.py`, only the sender and recipient users who have muted
the sender will have the "read" flag set.
* We already skip enqueueing notifications for users who've muted the sender
after 58da384da3.
* The queue consume functions for email and push notifications already
check filter messages which have been read before sending notifications.
* So, the "read" logic in `event_queue.py` is unnecessary, and the
processing power saved from not enqueueing notifications for a single
user should be insignificant, so we remove these checks all toghether.
Earlier, the notification-blocking for messages from muted senders
was a side-effect of we never sending notifications for messages
with the "read" flag.
This commit decouples these two things, as a prep for having new
settings which will allow users to **always** receive email
notifications, including when/if they read the message during the
time the notifications is in the queue.
We still mark muted-sender messages as "read" when they are sent,
because that's desirable anyways.
The old name `push_notify_user_ids` was misleading, because
it does not contain user ids which should be notified for
the current message, but rather user ids who have the online
push notifications setting enabled.
When the Tornado server is restarted during an upgrade, if
server has old events with the `push_notify_user_ids` fields,
the server will throw error after this rename. Hence, we need
to explicitly handle such cases while processing the event.
This ensures it is present for all requests; while that was already
essentially true via process_client being called from every standard
decorator, this allows middleware and other code to rely on this
having been set.
This help mobile and terminal clients understand whether a server
restart changed API feature levels or not, which in turn determines
whether they will need to resynchronize their data.
Also add tests and documentation for this previously undocumented
event type.
Fixes: #18205.
This extends the /json/typing endpoint to also accept
stream_id and topic. With this change, the requests
sent to /json/typing should have these:
* `to`: a list set to
- recipients for a PM
- stream_id for a stream message
* `topic`, in case of stream message
along with `op`(start or stop).
On receiving a request with stream_id and topic, we send
typing events to clients with stream_typing_notifications set
to True for all users subscribed to that stream.
Since c3a8a15bae removed the last
instance of code using the dictionary code path, we actually need to
wait until one can no longer upgrade directly from 4.x to master in
order to avoid breakage should we remove this compatibility code,
since only today did we stop generating the old event format.
The bulk deletion codepath was using dicts instead of user ids in the
event, as opposed to the other codepath which was adjusted to pass just
user ids before. We make the bulk codepath consistent with the other
one. Due to the dict-type events happening in 3.*, we move the goal for
deleting the compat code in process_notification to 5.0.
django.utils.translation.ugettext is a deprecated alias of
django.utils.translation.gettext as of Django 3.0, and will be removed
in Django 4.0.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
A bug in the implementation of the all_public_streams API feature
resulted in guest users being able to receive message traffic to public
streams that should have been only accessible to members of the
organization.
This makes it much more clear that this feature does JSON encoding,
which previously was only indicated in the documentation.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
The Session middleware only adds `Vary: cookie` if it sees an access
to the from inside of it. Because we are effectively, from the Django
session middleware's point of view, returning the static content of
`request.saved_response` and never accessing the session, it does not
set `Vary: cookie` on longpoll requests.
Explicitly mark Tornado requests as varying by cookie.
A few internal fields used for tracking which types of notifications
have already been sent for a given message, like `hander_id` and the
`push_notified` bundle of fields were being incorrectly included in
message events delivered to clients clients.
One could argue these fields might be useful hints to clients, but
because notifications can be triggered later on via
`missedmessage_hook`, they have no useful purpose in the API.
This commit move these extended event field on a `internal_data`
object within the event object, and delete this field in `contents()`
for call points that would serve data to clients.
Tweaked by tabbott to provide a cleaner interface.
We're not bumping API_FEATURE_LEVEL because these fields have always
been documented as being present only due to a bug, so no clients
should be expecting or relying on them.
Fixes: #15947.
This logging is really only potentially interesting in a development
environment when the numbers are nonzero.
In production, it seems worth logging for consistency reasons.
Probably we'll eventually redo this block by change the log level, but
this is good enough to despam the development environment startup
output.
The `no_proxy` parameter does not work to remove proxying[1]; in this
case, since all requests with this adapter are to the internal Tornado
process, explicitly pass in an empty set of proxies to disable
proxying.
[1] https://github.com/psf/requests/issues/4600
Having both of these is confusing; TORNADO_SERVER is used only when
there is one TORNADO_PORT. Its primary use is actually to be _unset_,
and signal that in-process handling is to be done.
Rename to USING_TORNADO, to parallel the existing USING_RABBITMQ, and
switch the places that used it for its contents to using
TORNADO_PORTS.
In development and test, we keep the Tornado port at 9993 and 9983,
respectively; this allows tests to run while a dev instance is
running.
In production, moving to port 9800 consistently removes an odd edge
case, when just one worker is on an entirely different port than if
two workers are used.
tornado.web.Application does not share any inheritance with Django at
all; it has a similar router interface, but tornado.web.Application is
not an instance of Django anything.
Refold the long lines that follow it.
While urllib3 retries all connection errors, it only retries a subset
of read errors, since not all requests are safe to retry if they are
not idempotent, and the far side may have already processed them once.
By default, the only methods that are urllib3 retries read errors on
are GET, TRACE, DELETE, OPTIONS, HEAD, and PUT. However, all of the
requests into Tornado from Django are POST requests, which limits the
effectiveness of bb754e0902.
POST requests to `/api/v1/events/internal` are safe to retry; at worst,
they will result in another event queue, which is low cost and will be
GC'd in short order.
POST requests to `/notify_tornado` are _not_ safe to retry, but this
codepath is only used if USING_RABBITMQ is False, which only occurs
during testing.
Enable retries for read errors during all POSTs to Tornado, to better
handle Tornado restarts without 500's.
Without an explicit port number, the `stdout_logfile` values for each
port are identical. Supervisor apparently decides that it will
de-conflict this by appending an arbitrary number to the end:
```
/var/log/zulip/tornado.log
/var/log/zulip/tornado.log.1
/var/log/zulip/tornado.log.10
/var/log/zulip/tornado.log.2
/var/log/zulip/tornado.log.3
/var/log/zulip/tornado.log.7
/var/log/zulip/tornado.log.8
/var/log/zulip/tornado.log.9
```
This is quite confusing, since most other files in `/var/log/zulip/`
use `.1` to mean logrotate was used. Also note that these are not all
sequential -- 4, 5, and 6 are mysteriously missing, though they were
used in previous restarts. This can make it extremely hard to debug
logs from a particular Tornado shard.
Give the logfiles a consistent name, and set them up to logrotate.
These weren’t wrong since orjson.JSONDecodeError subclasses
json.JSONDecodeError which subclasses ValueError, but the more
specific ones express the intention more clearly.
(ujson raised ValueError directly, as did json in Python 2.)
Signed-off-by: Anders Kaseorg <anders@zulip.com>
The exception trace only goes from where the exception was thrown up
to where the `logging.exception` call is; any context as to where
_that_ was called from is lost, unless `stack_info` is passed as well.
Having the stack is particularly useful for Sentry exceptions, which
gain the full stack trace.
Add `stack_info=True` on all `logging.exception` calls with a
non-trivial stack; we omit `wsgi.py`. Adjusts tests to match.
By defaults, `requests` has no timeout on requests, which can lead to
waiting indefinitely. Add a half-second timeout on these; this is
applied _inside_ each retry, not overall -- that is, with retries any
of these functions may take a total of 1.5s.
Use the `no_proxy` proxy, which explicitly disables proxy usage for
particular hosts. This is a slightly cleaner solution than ignoring
all of the environment, as removing proxies is specifically what we
are attempting to accomplish.
The change in #2764 provided a better error message on one of the
three calls into Tornado, but left the other two with the old error
message. `raise_for_status` was used on two out of three.
Use a custom HTTPAdapter to apply this pattern to all requests from
Django to Tornado.
Apparently, `update_message` events unexpectedly contained what were
intended to be internal data structures about which users were
mentioned in a given message.
The bug has been present and accumulating new data structures for
years.
Fixing this should improve the performance of handling update_message
events as well as cleaning up this API's interface.
This was discovered by our automated API documentation schema checking
tooling detecting these unexpected elements in these event
definitions; that same logic should prevent future bugs like this from
being introduced in the future.
There is still some miscellaneous cleanup that
has to happen for things like analytics queries
and dead code in node tests, but this should
remove the main use of pointers in the backend.
(We will also still need to drop the DB field.)
This is designed to have no user-facing change unless the client
declares bulk_message_deletion in its client_capabilities.
Clients that do so will receive a single bulk event for bulk deletions
of messages within a single conversation (topic or PM thread).
Backend implementation of #15285.
Fixes#2665.
Regenerated by tabbott with `lint --fix` after a rebase and change in
parameters.
Note from tabbott: In a few cases, this converts technical debt in the
form of unsorted imports into different technical debt in the form of
our largest files having very long, ugly import sequences at the
start. I expect this change will increase pressure for us to split
those files, which isn't a bad thing.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Automatically generated by the following script, based on the output
of lint with flake8-comma:
import re
import sys
last_filename = None
last_row = None
lines = []
for msg in sys.stdin:
m = re.match(
r"\x1b\[35mflake8 \|\x1b\[0m \x1b\[1;31m(.+):(\d+):(\d+): (\w+)", msg
)
if m:
filename, row_str, col_str, err = m.groups()
row, col = int(row_str), int(col_str)
if filename == last_filename:
assert last_row != row
else:
if last_filename is not None:
with open(last_filename, "w") as f:
f.writelines(lines)
with open(filename) as f:
lines = f.readlines()
last_filename = filename
last_row = row
line = lines[row - 1]
if err in ["C812", "C815"]:
lines[row - 1] = line[: col - 1] + "," + line[col - 1 :]
elif err in ["C819"]:
assert line[col - 2] == ","
lines[row - 1] = line[: col - 2] + line[col - 1 :].lstrip(" ")
if last_filename is not None:
with open(last_filename, "w") as f:
f.writelines(lines)
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
This allows straight-forward configuration of realm-based Tornado
sharding through simply editing /etc/zulip/zulip.conf to configure
shards and running scripts/refresh-sharding-and-restart.
Co-Author-By: Mateusz Mandera <mateusz.mandera@zulip.com>
Since production testing of `message_retention_days` is finished, we can
enable this feature in the organization settings page. We already had this
setting in frontend but it was bit rotten and not rendered in templates.
Here we replaced our past text-input based setting with a
dropdown-with-text-input setting approach which is more consistent with our
existing UI.
Along with frontend changes, we also incorporated a backend change to
handle making retention period forever. This change introduces a new
convertor `to_positive_or_allowed_int` which only allows positive integers
and an allowed value for settings like `message_retention_days` which can
be a positive integer or has the value `Realm.RETAIN_MESSAGE_FOREVER` when
we change the setting to retain message forever.
This change made `to_not_negative_int_or_none` redundant so removed it as
well.
Fixes: #14854
Generated by autopep8, with the setup.cfg configuration from #14532.
I’m not sure why pycodestyle didn’t already flag these.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
Generated by `pyupgrade --py3-plus --keep-percent-format` on all our
Python code except `zthumbor` and `zulip-ec2-configure-interfaces`,
followed by manual indentation fixes.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
We use retry_event in queue_processors.py to handle trying on failures,
without getting stuck in permanent retry loops if the event ends up
leading to failure on every attempt and we just keep sending NACK to
rabbitmq forever (or until the channel crashes). Tornado queues haven't
been using this, but they should.
The change in 180d8abed6, while correct
for the Django part of the codebase, had the nasty side effect of
exposing a failure mode in the process_notification logic if the users
list was empty.
This, in turn, could cause our process_notification code to fail with
an IndexError when trying to process the event, which would result in
that tornado process not automatically recovering, due to the outer
try/except handler for consume triggering a NACK and thus repeating
the event.
When more than one outgoing webhook is configured,
the message which is send to the webhook bot passes
through finalize_payload function multiple times,
which mutated the message dict in a way that many keys
were lost from the dict obj.
This commit fixes that problem by having
`finalize_payload` return a shallow copy of the
incoming dict, instead of mutating it. We still
mutate dicts inside of `post_process_dicts`, though,
for performance reasons.
This was slightly modified by @showell to fix the
`test_both_codepaths` test that was added concurrently
to this work. (I used a slightly verbose style in the
tests to emphasize the transformation from `wide_dict`
to `narrow_dict`.)
I also removed a deepcopy call inside
`get_client_payload`, since we now no longer mutate
in `finalize_payload`.
Finally, I added some comments here and there.
For testing, I mostly protect against the root
cause of the bug happening again, by adding a line
to make sure that `sender_realm_id` does not get
wiped out from the "wide" dictionary.
A better test would exercise the actual code that
exposed the bug here by sending a message to a bot
with two or more services attached to it. I will
do that in a future commit.
Fixes#14384
If we have an old event that's missing the field
`sender_delivery_email`, we now patch it at the top
of `process_message_event`, rather than for each call
to `get_client_payload`. This will make an upcoming
commit a bit easier to reason about. Basically, it's
simpler to shim the incoming event one time rather
than doing it up to four times. We know that
`get_client_payload` is non-destructive, because it
does a deepcopy.
Instead of trying to set the _requestor_for_logs attribute in all the
relevant places, we try to use request.user when possible (that will be
when it's a UserProfile or RemoteZulipServer as of now). In other
places, we set _requestor_for_logs to avoid manually editing the
request.user attribute, as it should mostly be left for Django to manage
it.
In places where we remove the "request._requestor_for_logs = ..." line,
it is clearly implied by the previous code (or the current surrounding
code) that request.user is of the correct type.
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.
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()
In e3ad9baf1d, we introduced yet another
bug where we incorrectly shared event dictionaries between multiple
queues.
Fortunately, the logging that reports on "event was not in the queue"
issues worked and detected this on chat.zulip.org, but this is a clear
indication that the comments we have around this system were not
sufficient to produce correct behavior.
We fix this by changing event_queue.push, the code that mutates the
event dictionaries, to do the shallow copies itself. The only
downside here is process_message_event, a relatively low-traffic code
path, does an extra per-queue dictionary copy. Given that presence,
heartbeat, and message reading events are likely more traffic and
dealing with HTTP is likely much more expensive than a dictionary
copy, this probably doesn't matter performance-wise.
(And if profiling later finds it is, there are potential workarounds
like passing a skip_copy argument we can do).
This flag affects page_params and the
payload you get back from POSTs to this
url:
users/me/presence
The flag does not yet affect the
presence events that get sent to a
client.
Zulip has had a small use of WebSockets (specifically, for the code
path of sending messages, via the webapp only) since ~2013. We
originally added this use of WebSockets in the hope that the latency
benefits of doing so would allow us to avoid implementing a markdown
local echo; they were not. Further, HTTP/2 may have eliminated the
latency difference we hoped to exploit by using WebSockets in any
case.
While we’d originally imagined using WebSockets for other endpoints,
there was never a good justification for moving more components to the
WebSockets system.
This WebSockets code path had a lot of downsides/complexity,
including:
* The messy hack involving constructing an emulated request object to
hook into doing Django requests.
* The `message_senders` queue processor system, which increases RAM
needs and must be provisioned independently from the rest of the
server).
* A duplicate check_send_receive_time Nagios test specific to
WebSockets.
* The requirement for users to have their firewalls/NATs allow
WebSocket connections, and a setting to disable them for networks
where WebSockets don’t work.
* Dependencies on the SockJS family of libraries, which has at times
been poorly maintained, and periodically throws random JavaScript
exceptions in our production environments without a deep enough
traceback to effectively investigate.
* A total of about 1600 lines of our code related to the feature.
* Increased load on the Tornado system, especially around a Zulip
server restart, and especially for large installations like
zulipchat.com, resulting in extra delay before messages can be sent
again.
As detailed in
https://github.com/zulip/zulip/pull/12862#issuecomment-536152397, it
appears that removing WebSockets moderately increases the time it
takes for the `send_message` API query to return from the server, but
does not significantly change the time between when a message is sent
and when it is received by clients. We don’t understand the reason
for that change (suggesting the possibility of a measurement error),
and even if it is a real change, we consider that potential small
latency regression to be acceptable.
If we later want WebSockets, we’ll likely want to just use Django
Channels.
Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
Abhijeet Prasad Bodas
Tim Abbott
orientor
Anders Kaseorg
m-e-l-u-h-a-n
Dinesh
Mateusz Mandera
Alex Vandiver
shanukun
Steve Howell
Aman Agrawal
Anders Kaseorg
Tim Abbott
Pragati Agrawal
Mateusz Mandera
Udit107710
Steve Howell