As we have seen no further cases of this in production since #23215,
increase the severity to an error, and switch from returning a
list (which is not type-safe if the function declares a QuerySet
return) to returning the QuerySet without caching.
Failing to store the result in the cache, with an error, seems
superior to raising an exception; in both cases the next request will
redo the work, but we are guaranteed a worse user experience if we 500
the request.
Ref https://github.com/zulip/zulip/pull/23215#discussion_r994186493
Prior to 53231aa, the `ignore_unhashable_lru_cache` decorator had
a check for the development environment so that changes could be
seen on refresh.
Puts that check back in IgnoreUnhashableLruCacheWrapper class.
As noted in the previous commit, this causes bloat in memcached, for
no purpose. Log a warning when `cache_with_key` sees a QuerySet
returned from the function it is decorating.
Storing this key is superfluous, as it will be the same for all users,
and definitionally already known to fetch the cache for the realm. It
is also not currently used by the callsites that read rows from the
cache.
The pattern of using the same variable to apply filters
or alter the `QuerySet` in other ways might produce `QuerySet`s
with incompatible types. This behavior is not allowed by mypy.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
The `get_link_embed_data` / `link_embed_data_from_cache` pair as
introduced in c93f1d4eda uses the cache
as a temporary store inside of the `embed_links` worker; this means
that it must be durable storage, or the worker will stall and re-fetch
the same links to preview them.
Switch to plumbing through the fetched URL embed data as an parameter
to the Markdown evaluation which uses them, rather than using the
cache as an intermediary. This frees up the cache to be merely a
non-durable cache.
As a side-effect, this removes get_cache_with_key, and
link_embed_data_from_cache which was its only callsite.
This demonstrates a way to resolve the long-standing issue
of typing higher-order identity functions without using
`cast` and in a type-safe manner for decorators in `cache.py`.
Signed-off-by: Zixuan James Li <359101898@qq.com>
The previous logic was incorrect and was not flushing the stream from
cache after deletion.
```
stream = get_realm_stream("Verona", realm.id)
stream.delete()
get_realm_stream("Verona", realm.id)
```
In the above example, the last line of code would have returned
the stream from cache instead of throwing a Stream.DoesNotExist
error. This is fixed in the commit.
I have verified that this commit indeed fix the issue by verifying
that calling get_realm_stream again after deleting the stream
results in Stream.DoesNotExist error.
model__id syntax implies needing a JOIN on the model table to fetch the
id. That's usually redundant, because the first table in the query
simply has a 'model_id' column, so the id can be fetched directly.
Django is actually smart enough to not do those redundant joins, but we
should still avoid this misguided syntax.
The exceptions are ManytoMany fields and queries doing a backward
relationship lookup. If "streams" is a many-to-many relationship, then
streams_id is invalid - streams__id syntax is needed. If "y" is a
foreign fields from X to Y:
class X:
y = models.ForeignKey(Y)
then object x of class X has the field x.y_id, but y of class Y doesn't
have y.x_id. Thus Y queries need to be done like
Y.objects.filter(x__id__in=some_list)
This commit defines a new function `get_muting_users`
which will return a list of IDs of users who have muted
a given user.
Whenever someone mutes/unmutes a user, the cache will be
flushed, and subsequently when that user sends a message,
the cache will be populated with the list of people who
have muted them (maybe empty).
This data is a good candidate for caching because-
1. The function will later be called from the message send
codepath, and we try to minimize database queries there.
2. The entries will be pretty tiny.
3. The entries won't churn too much. An average user will
send messages much more frequently than get muted/unmuted,
and the first time penalty of hitting the db and populating
the cache should ideally get amortized by avoiding several
DB lookups on subsequent message sends.
The actual code to call this function will be written in
further commits.
This is no longer used in any important place,
get_user_profile_by_email is meant to be used only in manage.py shell
now and thus there's no point in this function being cached.
Emails are not unique, so we can only sensibly cache using keys formed
with both email and realm.
This requires adding a new cache key function for caching by delivery
email - user_profile_delivery_email_cache_key.
By registering a post_delete handler to clear appropriate caches in a
nicer way, we can get rid of the ugly flush-memcached call in the
delete_realm command.
There are three functional side effects:
• Correct an insignificant but mathematically offensive bias toward
repeated characters in generate_api_key introduced in commit
47b4283c4b4c70ecde4d3c8de871c90ee2506d87; its entropy is increased
from 190.52864 bits to 190.53428 bits.
• Use the base32 alphabet in confirmation.models.generate_key; its
entropy is reduced from 124.07820 bits to the documented 120 bits, but
now it uses 1 syscall instead of 24.
• Use the base32 alphabet in get_bigbluebutton_url; its entropy is
reduced from 51.69925 bits to 50 bits, but now it uses 1 syscall
instead of 10.
(The base32 alphabet is A-Z 2-7. We could probably replace all of
these with plain secrets.token_urlsafe, since I expect most callers
can handle the full urlsafe_b64 alphabet A-Z a-z 0-9 - _ without
problems.)
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Django 3.0 removed private Python 2 compatibility APIs
so used lru_cache() directly from functools.
We cast lru_cache to Any to avoid attr-defined error in mypy since we
are adding extra field, 'key_prefix', to this object later.
A few major themes here:
- We remove short_name from UserProfile
and add the appropriate migration.
- We remove short_name from various
cache-related lists of fields.
- We allow import tools to continue to
write short_name to their export files,
and then we simply ignore the field
at import time.
- We change functions like do_create_user,
create_user_profile, etc.
- We keep short_name in the /json/bots
API. (It actually gets turned into
an email.)
- We don't modify our LDAP code much
here.
This was hiding an actual type error in test_cache: a mismatch between
the object ID type, which is str, and the default id_fetcher, which
returns int.
Mypy’s insufficient support for default generic arguments basically
means we can’t use them without a lot of overloading, and there are
not enough callers here to justify that.
https://github.com/python/mypy/issues/3737
We avoid this being super messy where the code calls this by adding
some less generic wrappers for generic_bulk_cached_fetch.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
We need this field to avoid O(N) database operations
while fetching realm user data for clients with
`user_avatar_url_field_optional` flag enabled.
Part of #15287.
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>
Generated by pyupgrade --py36-plus --keep-percent-format, but with the
NamedTuple changes reverted (see commit
ba7906a3c6, #15132).
Signed-off-by: Anders Kaseorg <anders@zulip.com>
We pipe realm_id through functions where it is available,
this helps us avoid doing query for realm_id in loop when
multiple messages are being processed.
We remove the `owner` field from `page_params/realm_bots`
and bot-related events.
In the recent commit 155f6da8ba
we added `owner_id`, which we now use everywhere we need
bot owners for.
We also bump the `API_FEATURE_LEVEL` to 5 here. We
had already documented this in the prior commit to
add `owner_id`.
Note that we don't have to worry about mobile/ZT clients
here--we only deal with bot data in the webapp.
For the below payloads we want `owner_id` instead
of `owner`, which we should deprecate. (The
`owner` field is actually an email, which is
not a stable key.)
page_params.realm_bots
realm_bot/add
realm_bot/update
IMPORTANT NOTE: Some of the data served in
these payloads is cached with the key
`bot_dicts_in_realm_cache_key`.
For page_params, we get the new field
via `get_owned_bot_dicts`.
For realm_bot/add, we modified
`created_bot_event`.
For realm_bot/update, we modified
`do_change_bot_owner`.
On the JS side, we no longer
look up the bot's owner directly in
`server_events_dispatch` when we get
a realm_bot/update event. Instead, we
delegate that job to `bot_data.js`.
I modified the tests accordingly.
Previously, alert words were a JSON list of strings stored in a
TextField on user_profile. That hacky model reflected the fact that
they were an early prototype feature.
This commit migrates from that to a separate table, 'AlertWord'. The
new AlertWord has user_profile, word, id and realm(denormalization so
we can provide a nice index for fetching all the alert words in a
realm).
This transition requires moving the logic for flushing the Alert Words
caches to their own independent feature.
Note that this commit should not be cherry-picked without the
following commit, which fixes case-sensitivity issues with Alert Words.
Anders Kaseorg
Alex Vandiver
Lauryn Menard
Zixuan James Li
Zixuan James Li
Mateusz Mandera
Vishnu KS
Abhijeet Prasad Bodas
Dinesh
arpit551
Steve Howell
Hashir Sarwar
Anders Kaseorg
Aman Agrawal
Abhishek-Balaji