1e5c49ad82 added support for shared channels -- but some users may
only currently exist in DMs or MPIMs, and not in channel membership.
Walk the list of MPIM subscriptions and messages, as well as DM users,
and add any such users to the set of mirror dummy users.
This leads to significant speedups. In a test, with 100 random unique
event classes, the old code processed a batch of 100 rows (on average
66-ish unique in the batch) in 0.45 seconds. Doing this in a single
query processes the same batch in 0.0076 seconds.
The previous query suffered from bad corner cases when the user had
received a large number of direct messages but sent very few,
comparatively. This mean that the first half of the UNION would
retrieve a very large number of UserMessage rows, requiring fetching a
large number of Message rows, merely to throw them away upon
determining that the recipient was the current user.
Instead of merging two queries of "last 1k received" + "last 1k sent",
we instead make better use of the UserMessage rows to find "last 1k
sent or received." This may change the list of recipients, as large
disparities in sent/received messages may result in pushing the
most-recently-sent users off of the list. These are likely uncommon
edge cases, however -- and the disparity is the whole reason for the
performance problem.
This also provides more correct answers. In the case where a user's
1001'th message sent was to person A today, but my most recent message
received was from them yesterday, the previous plan would show the
message I received yesterday message-id as the max, and not the more
recent message I sent today.
While we could theoretically raise the `RECENT_CONVERSATIONS_LIMIT` to
more frequently match the same recipient list as previously, this
increases the cost of the most common cases unreasonably. With a
1000-message limit, the common cases are slightly faster, and the tail
latencies are very much improved; raising `RECENT_CONVERSATIONS_LIMIT`
would increase the result similarity to the old algorithm, at the cost
of the p50 and p75.
| | Old | New |
| ------ | ------- | ------- |
| Mean | 0.05287 | 0.02520 |
| p50 | 0.00695 | 0.00556 |
| p75 | 0.05592 | 0.03351 |
| p90 | 0.14645 | 0.08026 |
| p95 | 0.20181 | 0.10906 |
| p99 | 0.30691 | 0.16014 |
| p99.9 | 0.57894 | 0.19521 |
| max | 22.0610 | 0.22184 |
On the whole, however, the much more bounded worst case are worth the
small changes to the resultset.
This is preparatory work towards adding a Topic model.
We plan to use the local variable name as 'topic' for
the Topic model objects.
Currently, we use *topic as the local variable name for
topic names.
We rename local variables of the form *topic to *topic_name
so that we don't need to think about type collisions in
individual code paths where we might want to talk about both
Topic objects and strings for the topic name.
Earlier, after a successful POST request on find accounts page
users were redirected to a URL with the emails (submitted via form)
as URL parameters. Those raw emails in the URL were used to
display on a template.
We no longer redirect to such a URL; instead, we directly render
a template with emails passed as a context variable.
Fixes part of #3128
When you click "Plan management", the desktop app opens
/self-hosted-billing/ in your browser immediately. So that works badly
if you're already logged into another account in the browser, since that
session will be used and it may be for a different user account than in
the desktop app, causing unintended behavior.
The solution is to replace the on click behavior for "Plan management"
in the desktop app case, to instead make a request to a new endpoint
/json/self-hosted-billing, which provides the billing access url in a
json response. The desktop app takes that URL and window.open()s it (in
the browser). And so a remote billing session for the intended user will
be obtained.
As explained in the comment, this is to prevent bugs where some strange
combination of codepaths could end up calling do_login without basic
validation of e.g. the subdomain. The usefulness of this will be
extended with the upcoming commit to add the ability to configure custom
code to wrap authenticate() calls in. This will help ensure that some
codepaths don't slip by the mechanism, ending up logging in a user
without the chance for the custom wrapper to run its code.
This test is ancient and patches so much that it's almost unreadable,
while being redundant considering we have comprehensive tests via the
SocialAuthBase subclasses. The one missing case was the one with the
backend we disabled. We replace that with a proper
test_social_auth_backend_disabled test in SocialAuthBase.
This is preparatory work towards adding a Topic model.
We plan to use the local variable name as 'topic' for
the Topic model objects.
Currently, we use *topic as the local variable name for
topic names.
We rename local variables of the form *topic to *topic_name
so that we don't need to think about type collisions in
individual code paths where we might want to talk about both
Topic objects and strings for the topic name.
This is preparatory work towards adding a Topic model.
We plan to use the local variable name as 'topic' for
the Topic model objects.
Currently, we use *topic as the local variable name for
topic names.
We rename local variables of the form *topic to *topic_name
so that we don't need to think about type collisions in
individual code paths where we might want to talk about both
Topic objects and strings for the topic name.
This is preparatory work towards adding a Topic model.
We plan to use the local variable name as 'topic' for
the Topic model objects.
Currently, we use *topic as the local variable name for
topic names.
We rename local variables of the form *topic to *topic_name
so that we don't need to think about type collisions in
individual code paths where we might want to talk about both
Topic objects and strings for the topic name.
This is preparatory work towards adding a Topic model.
We plan to use the local variable name as 'topic' for
the Topic model objects.
Currently, we use *topic as the local variable name for
topic names.
We rename local variables of the form *topic to *topic_name
so that we don't need to think about type collisions in
individual code paths where we might want to talk about both
Topic objects and strings for the topic name.
Rename and restructure these comparison variables such that we don't
have a possibly impossible case for presence.last_connected_time being
None.
Fixes#25498.
We previously created the connection to the outgoing email server when
the EmailSendingWorker was first created. Since creating the
connection can fail (e.g. because of firewalls or typos in the
hostname), this can cause the `QueueProcessingWorker` creation to
raise an exception. In multi-threaded mode, exceptions in the worker
threads which are _not_ during the handling of a specific event
percolate out to `log_and_exit_if_exception` and trigger the
termination of the entire process -- stopping all worker threads from
making forward progress.
Contain the blast radius of misconfigured email servers by deferring
the opening of the connection until it is first needed. This will not
cause any overall performance change, since it only affects the
latency of the very first email after startup.
Creating the QueueProcessingWorker objects when the ThreadedWorker is
created can lead to a race which caused confusing error messages:
1. A thread tries to call `self.worker = get_worker()`
2. This call raises an exception, which is caught by
`log_and_exit_if_exception`
3. `log_and_exit_if_exception` sends our process a SIGUSR1, _but
otherwise swallows the error_.
4. The thread's `.run()` is called, which tries to access
`self.worker`, which was never set, and throws another exception.
5. The process handles the SIGUSR1, restarting.
Move the creation of the worker to when it is started, so the worker
object does not need to be stored, and possibly have a decoupled
failure.
Switches from Django's default error page to Zulip standard error
template. Also updates template for 405 error code to not use the 404
art.
Fixes#25626.
By default, `SELECT FOR UPDATE` will also lock any rows which are
`JOIN`ed into the selected rows; in the case of UserMessage rows, this
can mean arbitrary Message rows.
Since the messages themselves are not being changed, it is not
necessary to lock them -- and doing so may lead to deadlocks, in the
case that the UserMessage row is locked for update before the Message,
and some other request has already taken a read lock on the Message
and is blocked on the UserMessage write lock.
Change `select_for_update_query` to explicitly only lock UserMessage.
Updates title and main description to follow the general style
of the API endpoint documentation.
Updates `token` description to clarify suggested mobile client
behavior.
Adds a set of excluded endpoints for the test of generated curl
examples in the API documentation.
Currently, only the `api/test-notify` endpoint is excluded since
there would need to be a push notification bouncer set up to test
that generated curl example.
We return expected_end_timestamp as "None" for the plans to be
downgraded if number of users is not more than MAX_USERS_WITHOUT_PLAN
since they will be downgraded to self-managed plan and would
have push notifications enabled.
Requests to these endpoint are about a specified user, and therefore
also have a notion of the RemoteRealm for these requests. Until now
these endpoints weren't getting the realm_uuid value, because it wasn't
used - but now it is needed for updating .last_request_datetime on the
RemoteRealm.
For the RemoteRealm case, we can only set this in endpoints where the
remote server sends us the realm_uuid. So we're missing that for the
endpoints:
- remotes/push/unregister and remotes/push/unregister/all
- remotes/push/test_notification
This should be added in a follow-up commit.
os.path.getmtime needs to be mock.patched or otherwise the success of
the test depends on the filesystem state and breaks if version.py hasn't
been modified in a while.
`<time:1234567890123>` causes a "signed integer is greater than
maximum" exception from dateutil.parser; datetime also cannot handle
it ("year 41091 is out of range") but that is a ValueError which is
already caught.
Catch the OverflowError thrown by dateutil.
boto3 has two different modalities of making API calls -- through
resources, and through clients. Resources are a higher-level
abstraction, and thus more generally useful, but some APIs are only
accessible through clients. It is possible to get to a client object
from a resource, but not vice versa.
Use `get_bucket(...).meta.client` when we need direct access to the
client object for more complex API calls; this lets all of the
configuration for how to access S3 to sit within `get_bucket`. Client
objects are not bound to only one bucket, but we get to them based on
the bucket we will be interacting with, for clarity.
We removed the cached session object, as it serves no real purpose.
e883ab057f started caching the boto client, which we had identified
as slow call. e883ab057f went further, calling
`get_boto_client().generate_presigned_url()` once and caching that
result.
This makes the inner cache on the client useless. Remove it.
Adds a support action for updating the minimum licenses on a
customer object once a default discount has also been set.
In the case that the current billing entity has a current active
plan or a scheduled upgrade to a new plan, then the minimum
licenses will not be updated.
This protects us from incorrectly handling situations where someone
tested and upgrade to 8.0 for a backup on a separate hostname, and
left the test system live while upgrading the main system, in a way
that results in duplicate RemoteRealm objects that are all marked as
locally deleted.
Further word is required to figure out how to avoid the original
duplication problem.
If we `.distinct("delivery_email")` then we must also
`.order_by("delivery_email")`; adc987dc43 added the `.order_by`
call, which broke the newsletter codepath, since it did not contain
the `delivery_email` in the ordering fields.
Add a flag to distinct on emails in `send_custom_email`.
The set of `enable_marketing_emails=True` are those that have opted
into getting marketing newsletter emails -- but we previously limited
further to only those users active in the last month.
Broaden that to "opted in, and either recently active or an owner or
an admin," with the goal of providing information to folks who may
have tried out Zulip in the past.
Co-authored-by: Tim Abbott <tabbott@zulip.com>
Earlier, 'topic' parameter length for
'/users/me/subscriptions/muted_topics' and '/user_topics' endpoints
were not validated before DB operations which resulted in exception:
'DataError: value too long for type character varying(60)'.
This commit adds validation for the topic name length to be
capped at 'max_topic_length' characters.
The doc is updated to suggest clients that the topic name should
have a maximum length of 'max_topic_length'.
Fixes#27796.
Old RemotePushDeviceTokens were created without this attribute. But when
processing a notification, if we have remote_realm, we can take the
opportunity to to set this for all the registrations for this user.
This moves the function which computes can_push and
expected_end_timestamp outside RemoteRealmBillingSession
because we might use this function for RemoteZulipServer
as well and also renames it.
For remote servers, we cannot advertise `List-Unsubscribe=One-Click`,
which is specified in RFC 8058[^1] to mean that the `List-Unsubscribe`
URL supports a POST request with no arguments to unsubscribe. Because
we show an interstitial and confirmation page, as this is not just a
mailing list which is disabled if you click the link, it does not
support the mail system performing the unsubscribe for the user.
Remove the inaccurate header for remote servers.
[^1]: https://datatracker.ietf.org/doc/html/rfc8058
612f2c73d6 started passing add_context to
`send_custom_server_email`, but did not make it make use of it.
Also add the `hostname` as a built-in value, since that is most likely
the most useful property.
This fixes the exception case on the initial
`/api/v1/remotes/server/analytics/status` case. Other exceptions from
`send_to_push_bouncer` are allowed to escape.
Co-authored-by: Alex Vandiver <alexmv@zulip.com>
Previously, passing a url longer than 200 characters for
jitsi_server_url caused a low-level failure at DB level. This
commit adds this restriction at API level.
Fixes part of #27355.
While the query parameter is properly excaped when inlined into the
template (and thus is not an XSS), it can still produce content which
misleads the user via carefully-crafted query parameter.
Validate that the parameter looks like an email address.
Thanks to jinjo2 for reporting this, via HackerOne.
We previously used get_accessible_user_ids to check whether the
sender can access all DM recipients, which was not efficient as
it queries the Message table. This commit updates the code to
make sure we use get_inaccessible_user_ids which is much more
efficient as it limits the queries to only DM recipients and
also queries the Message table only if needed.
This can still be optimized further as mentioned in #27835 but
this commit is a nice first step.
Saying `**options: str` is a lie, since it contains bools. We pluck
out the two bools that we need properly typed because we will be
pushing them into function calls, and type them explicitly as bools.
As premonitioned in c741c527d7, it is
indeed possible for `get_handler_by_id` to error out by cause the
handler has been unset elsewhere.
Protect the callsites of `get_handler_by_id` to be able to gracefully
handle when the handler has already done away.
This fixes a bug introduced in
6f93ab72c0 where deactivating a realm
would fail with an exception that sessions cannot be cleared inside
database transactions.
If the exception was because the channel closed, attempting to NAK the
events will just raise another error, and is pointless, as the server
already marked the pending events as NAK'd.
4af00f61a8 claimed that `on_finish` and
`on_connection_close` were mutually exclusive. In cases where a
`DELETE` is called on the queue while a longpoll is in progress, this
can cause _both_ to happen:
- The `DELETE` pushes a `cleanup_queue` event, which triggers
`finish_handler` to begin pushing out an empty event response to the
longpoll connection.
- In the midst of that, in an `await`, the longpoll connection drops,
and `on_connection_close` clears the handler.
- The `await` resumes, calls `finish`, and attempts to clear the
handler.
The easiest solution is to make `clear_handler_by_id` tolerant to
multiple attempts to clear it. Since these processes run in parallel,
it means that parts may have a `handler_id` but `get_handler_by_id`
may error in attempting to look it up. We have not observed this in
testing, and I cannot currently prove it is impossible.
This ensures determinism in these tests doing mock_send.assert_called
with - avoids producing test flakes due to a different order of
retrieval of these objects from the database.
- The server sends the list of registrations it believes to have with
the bouncer.
- The bouncer includes in the response the registrations that it doesn't
actually have and therefore the server should delete.
This commit creates a RealmAuditlog entry with a new event_type
'RealmAuditLog.REALM_IMPORTED' after the realm is reactivated.
It contains user count data (using realm_user_count_by_role)
stored in extra_data.
This helps to have an accurate user count data for the billing
system if someone tries to signup just after doing an import.
This partially reverts 579bdc18f85ea8599c8cf1f53ddb02fd41d97993; it
assumed (based on its documentation) that `on_finish` was called for
all requests, even client-terminated ones. This is not accurate; it
is only called when the request calls `finish`, which only happens for
successful requests. This caused every client-closed connection to
leak a handler (ironically, exactly re-introducing the bug previously
fixed in 12a5a3a6e1).
This behaviour was obscured by the development environment's proxy;
see comment added in the previous commit.
Instead of replacing the `clear_handler_by_id` call into
`ClientDescriptor.disconnect_handler`, we instead place it on
`AsyncDjangoHandler.on_connection_close`. This is more correct for
a few reasons:
- `on_connection_close` will be called if the client goes away during
a request without a client descriptor. If the handler garbage
collection of handlers runs inside the ClientDescriptor, we leak
handlers.
- `disconnect_handler` also runs when successfully sending an event,
which already calls `on_finish`. We avoid double-calling
`clear_handler_by_id` by doing it in two clearly exclusive cases,
`on_finish` and `on_connection_close`.
- It combines the creation and garbage collection logic into one
file, decreasing action at a distance which causes memory leaks.
We call 'send_server_data_to_push_bouncer' just after registering
server for push notification.
This helps to have a current state of the user counts when first
logging in after the RemoteRealm flow.
Actions that change the number of user counts adds a deferred_work
queue processor job immediately update the billing service about your
change.
This helps to avoid having users see stale state for how many
users they have when trying to pay.
This is a rename of the previous
enqueue_register_realm_with_push_bouncer_if_needed but is clearer
about the fact that this will also upload audit logs if available.
Given that most of the use cases for realms-only code path would
really like to upload audit logs too, and the others would likely
produce a better user experience if they upoaded audit logs, we
should just have a single main code path here i.e.
'send_analytics_to_push_bouncer'.
We still only upload usage statistics according to documented
option, and only from the analytics cron job.
The error handling takes place in 'send_analytics_to_push_bouncer'
itself.
This is the only operating editing audit logs not already using a
transaction, and having it do so will simplify an upcoming interface
to be able to assume it is always inside a transaction.
Earlier, it was passing tests because the deffered_work queue
that calls send_realms_only_to_push_bouncer didn't update the
realms propery based on response received from bouncer.
This prep commit removes the invalid "dummy-uuid" used, as any
call to send_realms_only_to_push_bouncer will update realms
properties too.
We return an empty realms array as the realm is created midway in
do_create_realm, so the uuid is not already available. Also, our
intent here is not to verify the behaviour of the
send_realms_only_to_push_bouncer function because we'll have
separate tests for that. Here, we verify that deffered_work event
was sent and eventually it made call to send_to_push_bouncer
with appropriate data.
I accidentally free trials for both cloud and self hosted
enabled while testing, hence didn't catch it.
This mostly involves fixing `is_free_trial_offer_enabled` to
return the correct value and providing it the correct input.
When a self-hosted Zulip server does a data export and then import
process into a different hosting environment (i.e. not sharing the
RemoteZulipServer with the original, we'll have various things that
fail where we look up the RemoteRealm by UUID and find it but the
RemoteZulipServer it is associated with is the wrong one.
Right now, we ask user to contact support via an error page but
might develop UI to help user do the migration directly.
This commit adds code to not include original details of senders like
name, email and avatar url in the message objects sent through events
and in the response of endpoint used to fetch messages.
This is the last major commit for the project to add support for
limiting guest access to an entire organization.
Fixes#10970.
Alex Vandiver
ColeBurch
swayam0322
Prakhar Pratyush
Mateusz Mandera
Lauryn Menard
Tim Abbott
Sahil Batra
sanchi-t
Evgenii
Anders Kaseorg
Aman Agrawal
Karl Stolley
Sayam Samal
Vector73